close

Вход

Забыли?

вход по аккаунту

?

Estimating tallgrass prairie soil moisture using active satellite microwave imagery and optical sensor inputs

код для вставкиСкачать
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI films
the text directly from the original or copy submitted. Thus, some thesis and
dissertation copies are in typewriter face, while others may be from any type of
computer printer.
The quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality illustrations
and photographs, print bleedthrough, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete manuscript
and there are missing pages, these will be noted.
Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and continuing
from left to right in equal sections with small overlaps.
Photographs included in the original manuscript have been reproduced
xerographically in this copy.
Higher quality 6* x 9” black and white
photographic prints are available for any photographs or illustrations appearing
in this copy for an additional charge. Contact UMI directly to order.
Bell & Howell Information and Learning
300 North Zeeb Road, Ann Arbor, Ml 48106-1346 USA
800-521-0600
0
UMI
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ESTIMATING TALLGRASS PRAIRIE SOIL MOISTURE USING
ACTIVE SATELLITE MICROWAVE IMAGERY AND OPTICAL SENSOR INPUTS
by
J. M. SHAWN HUTCHINSON
B.S.. Colorado State University, 1990
M.A., Kansas State University, 1997
A DISSERTATION
submitted in partial fulfillment of the
requirements for the degree
DOCTOR OF PHILOSOPHY
Department of Geography
College of Arts and Sciences
KANSAS STATE UNIVERSITY
Manhattan, Kansas
2000
Approved by:
ft Professor
C_J
fohn A. Harrington, Jr.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number 9984870
UMI*
UMI Microform9984870
Copyright 2000 by Bell & Howell Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
Bell & Howell Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ABSTRACT
Recent advances in active microwave remote sensing techniques provide the
potential for monitoring soil moisture conditions at the spatial and temporal scales
required for detailed local modeling efforts. The goal o f this research was to produce
accurate and spatially distributed estimates of soil moisture using a time series of ERS-2
images for the Konza Prairie, a tallgrass environment in northeast Kansas. The methods
used in this research involve field data collection o f soil moisture, digital image
interpretation of optical (NOAA AVHRR and LANDSAT TM) and radar (ERS-2)
imagery, and environmental modeling in a raster GIS environment.
To accomplish the research goals, the effect o f variable terrain on radar image
backscatter values was quantified and reduced.
Next, the scattering behavior o f the
overlying vegetation canopy was simulated using a water cloud model that estimated the
contribution of vegetation backscatter (o0ves) to the total backscatter coefficient (o°,0ui)Critical to this process were estimates of aboveground primary production made using the
normalized difference vegetation index from a combination of AVHRR and LANDSAT
TM images. With o°veg removed from o°ioui. the amount o f backscatter contributed by the
soil surface (o ^ i) was calculated and the linear relationship between o°„,i and volumetric
soil moisture was determined. This regression model was then inverted and solved for
volumetric soil moisture to quantify near surface soil moisture conditions across the study
area.
Local incidence angle had the strongest relationship on SAR image backscatter
values (r = -0.35) and when used in an empirical correction function reduced image
variance by at least 8%.
Backscatter modeling to separate the vegetation and soil
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
components of the radar signal performed worse than expected, resulting in a weak
correlation between composite a0*,,! and volumetric soil water content (r = 0.21) and
different values for burned and unburned watersheds (r = 0.09 and r = 0.32, respectively).
Soil backscatter values were estimated without accounting for canopy and litter layer
moisture conditions, causing a reduction in the effectiveness of the cloud model output.
The model performed very well, however, on a daily basis with single date correlations for
burned and unbumed watersheds being among the highest yet reported when using radar
satellite data.
While many studies have questioned the sensitivity of C-band radars,
operating at moderate incidence angles, to near surface soil moisture conditions, results
here show that the ERS-2 data are capable of monitoring soil moisture conditions over
even dense natural vegetation characteristic of tallgrass prairie.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF CONTENTS
LIST OF FIGURES..........................................................................................................iv
LIST OF TABLES......................................................................................................... viii
ACKNOWLEDGEMENTS............................................................................................ xi
1 INTRODUCTION_____________________________________________
1
2 LITERATURE REVIEW_____________________________________________ 6
2.1
The Microwave Spectrum: Characteristics and Properties...................... 10
2.1.1 Wavelength and Frequency.................................................................. 11
2.1.2 Polarization and Coherency................................................................. 14
2.1.3 Penetration Depth.................................................................................18
2.2
Origins and History of Radar........................................................................21
2.3
Side-Looking Radar Operation.....................................................................23
2.3.1 Spatial Resolution of SLR Systems..................................................... 25
2.3.2 Range Resolution................................................................................ 25
2.3.3 Azimuth Resolution............................................................................. 28
2.3.4 Image Geometry.................................................................................. 29
2.4
Radar Remote Sensing from Space.............................................................32
2.4.1 SEASATSAR......................................................................................34
2.4.2 Shuttle Imaging Radar (SIR)................................................................ 36
2.4.3 ALMAZ Program................................................................................ 42
2.4.3.1 COSMOS-1870.................................................................... 42
2.4.3.2 ALMAZ-1............................................................................. 44
2.4.4 European Remote Sensing Satellite (ERS) Series................................45
2.4.5 Japanese Environmental Remote Sensing Satellite (JERS-1).............49
2.4.6 Canadian Radar Satellite (RADARSAT)............................................50
2.5
The Radar Equation....................................................................................52
2.5.1 Radar Scattering Cross-Section............................................................ 53
2.5.2 Differential Scattering Coefficient...................................................... 54
2.6
Factors Influencing SAR Backscatter Response ........................................56
2.6.1 Geometric Factors............................................................................... 57
2.6.1.1 Topography and Comer Reflectors.....................................57
2.6.1.2 Surface Roughness...............................................................61
2.6.1.3 Surface and Volume Scattering...........................................64
2.6.2 Electrical Factors..................................................................................69
2.6.2.1 Soil Properties..................................................................... 70
2.6.2.2 Vegetation Effects.................................................................72
2.6.2.3 Temperature Effects..............................................................74
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.7 Radar-Based Soil Moisture Investigations .................................................76
2.7.1 Soil Moisture Estimates from Bare Soil Surfaces............................... 77
2.7.2 Soil Moisture Estimates from Vegetated Surfaces...............................85
2.8 Summary of the Literature Review ............................................................91
3
STUDY AREA_____________________________________________________94
3.1 Grasslands and Tallgrass Prairie................................................................94
3.2 Origins and Research History of Konza Prairie Biological Station...........98
3.3 Climate of Konza Prairie Biological Station............................................. 101
3.4 Geomorphoiogy and Soils of the Konza Prairie Biological Station......... 108
3.5 Flora of the Konza Prairie Biological Station............................................ 114
4
DATA AND METHODS____________________________________________121
4.1 Image Pre-Processing................................................................................ 125
4.1.1 Image Registration............................................................................ 125
4.1.2 Digital Number (DN) Conversion......................................................126
4.1.3 Topographic Correction..................................................................... 129
4.2 Cloud Model Design.................................................................................. 132
4.2.1 Model Description............................................................................. 132
4.2.2 Cloud Model Processing....................................................................136
4J
Landscape Specification............................................................................ 139
4.3.1 Study Area Delineation...................................................................... 139
4.3.2 Aboveground Biomass Estimation.................................................... 148
4.4 Soil Moisture Estimation........................................................................... 155
4.4.1 Soil Moisture Sampling..................................................................... 155
4.4.2 Correlation and Regression Analysis................................................ 160
4.5 Summary o f Data and Methods................................................................ 161
5
RESULTS AND DISCUSSION______________________________________ 164
5.1 Soil Moisture Sampling............................................................................. 164
5.2 Radar Image Rectification and Calibration............................................. 170
5J
Topographic Correction............................................................................ 173
5.4 Cloud Model Parameterization................................................................. 180
5.5 Vegetation Biomass Estimation................................................................ 191
5.6 Radar Baclucatter-Soil Moisture Relationships...................................... 198
5.6.1 Total Backscatter and Soil Moisture................................................. 198
5.6.2 Soil Backscatter and Soil Moisture...................................................205
ii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.7
Radar-Based Soil Moisture Estimates...................................................... 216
6
CONCLUSION__________________________________________________ 220
7
REFERENCES__________________________________________________ 228
8
APPENDIX A: SOIL MOISTURE DATA (MEASURED)_______________ 239
9
APPENDIX B: TRANSECT SAMPLE POINT LOCATIONS____________244
10 APPENDIX C: KONZA PRAIRIE BURN HISTORY__________________ 245
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF FIGURES
Figure 2.01. Impact o f water clouds on microwave transmission (from Ulaby et al.
1981).............................................................................................................. 7
Figure 2.02. Impact o f precipitation on microwave transmission (from Ulaby et al.
1981).............................................................................................................. 7
Figure 2.03. An electromagnetic wave (from Lillesand and Kiefer 1994)....................... 11
Figure 2.04. The electromagnetic spectrum (from Suits 1983)........................................ 13
Figure 2.0S. A linear-polarized electromagnetic wave (from Elachi 1987)..................... IS
Figure 2.06. Electrical fields combine to form circular-polarized wave (from
Elachi 1987).................................................................................................16
Figure 2.07. Penetration depth for a loamy soil at three different microwave
frequencies (from Ulaby et a l 1982)...........................................................20
Figure 2.08. Side-looking radar operation (from Lillesand and Kiefer 1994)................. 24
Figure 2.09. Relationship between pulse length and slant range on the range
resolution o f a radar image (from Lillesand and Kiefer 1994).................... 26
Figure 2.10. Relationship between radar image geometry and range resolution (Rr)
27
Figure 2.11. Dependence of azimuth resolution (R«), represented at two different
points (Ri and R2) in the azimuth direction, on antenna beamwidth
and ground range (from Lillesand and Kiefer 1994)................................... 29
Figure 2.12. Side-looking radar (SLR) image geometry (after Campbell 1987).............30
Figure 2.13. Three types of radar image distortion: layover, radar shadow, and
foreshortening (from Lillesand and Kiefer 1994)........................................32
Figure 2.14. Timeline o f operational imaging SAR satellites..........................................33
Figure 2.15. SEASAT satellite and operating characteristics (Source: Kramer
1996)............................................................................................................ 35
Figure 2.16. SIR-A sensor and operating characteristics (Source: Kramer 1996).......... 37
Figure 2.17. SIR-B sensor and operating characteristics (Source: Kramer 1996).......... 39
Figure 2.18. SIR-C/X-SAR sensor and operating characteristics (Source: Kramer
1996)............................................................................................................ 40
Figure 2.19. Cosmos-1870 operating characteristics (Source: Kramer 1996)................ 43
Figure 2.20. Almaz-1 satellite and operating characteristics (Source: Kramer
1996)............................................................................................................44
Figure 2.21. ERS-1/2 satellite and operating characteristics (Source: Kramer
1996)............................................................................................................47
Figure 2.22. JERS-1 satellite and operating characteristics (Source: Kramer
1996)............................................................................................................ 50
Figure 2.23. RADARSAT satellite and operating characteristics (Source: Kramer
1996)............................................................................................................ 51
Figure 2.24. Local incidence angle (8,) over flat ground (A) and the impact of
sloping terrain on local incidence angle values (B and C)...........................58
Figure 2.25. Comer-reflecting radar calibration targets: (a) dihedral, (b) trihedral
(from Ulaby et ed. 1982)..............................................................................60
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 2.26. Microwave scattering patterns from three surface roughness
conditions (after Ulaby et al. 1982).............................................................62
Figure 2.27. Vegetation structural categories with associated size, shape, and
orientation o f canopy components (after Dobson et a l 1995).................... 66
Figure 2.28. Sensor-target interactions affecting scattering coefficient o°
information content (after Engman and Chauhan 1995)..............................68
Figure 2.29. Dielectric constant (e) values at 1.4 GHz for five soil types and
increasing volumetric water content (from Ulaby et a l 1986).................... 71
Figure 2.30. Temperature-dielectric constant relationship for a silt loam soil at 3
GHz (from Ulaby et a l 1996)......................................................................75
Figure 2.31. Valid regions of three theoretical scattering models with respect to
surface roughness parameters (k = 2jc / X) (after Oh et a l 1992).................78
Figure 3.01. North American grassland variation along gradient of precipitation
and temperature (from Knapp and Seastedt 1998)...................................... 95
Figure 3.02. Estimated extent of tallgrass prairie prior to European settlement
(from Kuchler 1974a).................................................................................. 96
Figure 3.03. Location o f the Kansas Flint Hills (from Briggs et al. 1997)..................... 97
Figure 3.04. Location o f Konza Prairie Biological Station............................................99
Figure 3.05. Experimental design implement at Konza Prairie Biological Station
(from Knapp and Seastedt 1998)................................................................100
Figure 3.06. Mean monthly temperatures at KPBS from historical record (18911996).......................................................................................................... 103
Figure 3.07. Long-term temperature record at KPBS (1891-1996)............................... 104
Figure 3.08. Mean monthly precipitation at KPBS from historical record (18911996).......................................................................................................... 105
Figure 3.09. Mean monthly precipitation at KPBS from historical record (18911996).......................................................................................................... 106
Figure 3.10. Long-term precipitation record at KPBS (1891-1996). One hundred
year drought and flood conditions translate to approximately 460 mm
(e.g., 1966) and 1,400 mm (e.g., 1951) o f annual rainfall,
respectively............................................................................................... 107
Figure 3.11. Drainage networks of Konza Prairie Biological Station (from Oviatt
1998).......................................................................................................... 109
Figure 3.12. (a) Elevation, (b) aspect, and (c) slope of KPBS (from Hutchinson
1998).......................................................................................................... 110
Figure 3.13. Soils o f the Clime-Sogn association and their normal positions on the
Konza Prairie landscape (from Jantz et a l 1975)...................................... 111
Figure 3.14. Soils o f the Benfield-Florence association and their normal positions
on the Konza Prairie landscape (from Jantz et al. 1975)............................112
Figure 3.15. Generalized illustration of soils and their characteristic position in the
Konza Prairie landscape (from Ransom et a l 1998)................................. 113
v
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.16. Rank-order of the 10 most species rich families found at Konza
Prairie (from Freeman 1998): Asteraceae (AST), Poaceae (POA),
Fabaceae (FAB), Brassicaceae (BRA), Euphorbiaceae (EUP),
Cyperaceae (CYP), Lamiaceae (LAM), Scrophulariaceae (SCR),
Polygonaceae (POL), and Rosaceae (ROS)...............................................116
Figure 3.17. Life histories of vascular plants at Konza Prairie (from Freeman
1998). Other = combinations o f life history states....................................116
Figure 3.18. Life forms o f vascular plants at Konza Prairie (from Freeman 1998):
H = hemicryptophytes, T = therophytes, G = geophytes
(cryptophytes), P = phanaerophytes, O = combinations o f other life
forms, and C = chamaephytes...................................................................117
Figure 3.19. Habitats of the vascular plants o f Konza Prairie (from Freeman
1998): P = prairie, D = disturbed sites, F = forests, W = wetland, A =
aquatic; combinations of letters indicates species commonly found in
more than one habitat type.........................................................................118
Figure 4.01. ERS-2 radar image scene locations, A-ascending pass, B-descending
pass............................................................................................................ 123
Figure 4.02. Conceptual model of the conversion of raw image DN values to o°dB...... 129
Figure 4.03. Schematic of the derivation o f the empirical topographic correction
function based here on local incidence angle (after Bayer et al. 1991)..... 131
Figure 4.04. Conceptual model of the procedure used to calculate o°so,i....................... 139
Figure 4.0S. Effect o f fee on ANPP in ungrazed Konza Prairie watersheds
(modified from Knapp et al. 1998). Asterisks indicate significant
differences at P < 0.05............................................................................... 141
Figure 4.06. Effect o f topographic position on ANPP in ungrazed Konza Prairie
watersheds (modified from Knapp et al. 1998). Different letters
indicate significant differences at P < 0.05................................................143
Figure 4.07. The relationship between ANPP and topographic position in burned
(A) and unburned (B) watersheds (modified from Knapp et al. 1998)......143
Figure 4.08. Aboveground primary production during a wet (1993) and dry (1989)
year for adjacent burned and unbumed sites (from Knapp et al.
1998)..........................................................................................................145
Figure 4.09. 19% monthly and long-term mean rainfall totals measured at Konza
Prairie headquarters (data from Konza LTER dataset APT01)................. 146
Figure 4.10. Burned and unbumed watersheds of Konza Prairie as determined for
this study...................................................................................................148
Figure 4.11. Interaction o f electromagnetic energy with leaf tissue (modified from
Campbell 1987)......................................................................................... 150
Figure 4.12. Illustration o f the relationship between burned and unbumed
watersheds with the components of the radar cloud m odel..................... 155
Figure 4.13. Location o f transects and soil sampling points in watersheds ID
(burned) and 20B (unbumed).................................................................... 156
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.14. Looking from west to east along the soil moisture transect in
watershed ID, a burned watershed located within the Konza Prairie
study site.................................................................................................... 157
Figure 4.15. Looking from west to east along the soil moisture transect in
watershed 20B, an unbumed watershed located within the Konza
Prairie study site.........................................................................................157
Figure 4.16. Elevation and distance between soil sampling points, from west to
east, along the transects in watershed ID (burned) and 20B
(unbumed)..................................................................................................158
Figure 5.01. Recorded total daily precipitation and mean volumetric water content
for watersheds ID (burned) and 20B (unbumed). Rainfall data from
Konza LTER dataset AWEO1....................................................................166
Figure 5.02. Mean volumetric water content for watersheds ID (burned) and 20B
(unbumed) as compared to estimated soil water content calculated
using the Penman combination method with a starting value o f 30%
water by volume on June 28 (DOY 180)................................................... 169
Figure 5.03. Measured volumetric water content on August I (DOY 214) for
watersheds ID (burned) and 20B (unbumed)............................................ 170
Figure 5.04. Scatterplot showing the negative correlation between radar
backscatter and local incidence angle before topographic correction....... 175
Figure 5.05. Post-correction scatterplot o f radar backscatter and local incidence
angle........................................................................................................... 178
Figure 5.06. SAR image (July 23,1996) before and after topographic correction........178
Figure 5.07. Simulated vegetation backscatter (o°vcg) for cloud particles with radii
of0.75 cm, 1.0 cm, and 1.25 cm............................................................... 181
Figure 5.08. Simulated values o f the two-way loss factor (L2) for cloud particles
with radii of0.75 cm, 1.0 cm, and 1.25 c m ...............................................182
Figure 5.09. Cloud model simulation results showing the impact of vegetation on
the quantity of g0**! - o°vcg for a particle radius of 0.75 cm (assumes
o°l0i*i = -9.0 dB)..........................................................................................183
Figure 5.10. Cloud model simulation results showing the impact of vegetation on
the quantity of o0iaai - o°vcg for a particle radius o f 1.0 cm (assumes
o°.o*i = -9.0 dB)..........................................................................................184
Figure 5.11. Cloud model simulation results showing the impact of vegetation on
the quantity of o0toui - o°vcg for a particle radius o f 1.25 cm (assumes
° o«»i = -9.0dB)..........................................................................................185
Figure 5.12. Influence o f vegetation backscatter (o°vc«) and two-way loss factor
(L2) on estimated soil backscatter (o°*u) as a function of
aboveground primary production for a particle size of 1.0 cm
(assumes o0**! = -9.0 dB).......................................................................... 187
Figure 5.13. Simulated backscatter response for vegetation canopy simulated by a
water cloud with a particle radius of 1.0 cm ............................................. 188
Figure 5.14. Long-term record of ANPP in upland and lowland sites according to
fire frequency at KPBS (from fciapp etcU.1998).......................................189
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure S. 1S. Simulated values o f vegetation backscatter (o0^ and the two-way
loss factor (L2) for the litter layer based on a particle size of 0.7S cm...... 190
Figure 5. 16. Mean and estimated mean AVHRR-derived NDVI values for KPBS
(1996).........................................................................................................192
Figure 5.17. AVHRR NDVI image from composite period 13 showing its inherent
limitations with respect to spatial resolution..............................................194
Figure 5.18. TM false-color composite and NDVI image. Note the presence of the
cloud in the false color composite and differences in reflectance
between burned and unbumed watersheds................................................. 195
Figure 5.19. Relationship between ANPP and AVHRR-modified TM NDVI
values for watersheds 1A (burned) and 20A (unbumed)........................... 196
Figure 5.20. Mean total corrected backscatter versus precipitation............................... 199
Figure 5.21. Comparison between mean total backscatter (o°totai) and estimated
soil moisture content..................................................................................200
Figure 5.22. Precipitation and minimum temperatures during the 1996 study
period......................................................................................................... 203
Figure 5.23. Scatterplot of total backscatter (o°toiai) versus volumetric water
content........................................................................................................204
Figure 5.24. Scatterplot of total backscatter (o°l0ui) versus volumetric water
content by watershed burning treatment.................................................... 205
Figure 5.25. Relationship between soil backscatter (o°Wj|) and volumetric soil
moisture..................................................................................................... 206
Figure 5.26. Scatterplot of soil backscatter (o0wii) versus soil moisture by
watershed burning treatment......................................................................207
Figure 5.27. Regression relationship between soil backscatter (a°„ii) and
volumetric soil moisture on a daily basis for watershed ID (burned
watershed).................................................................................................. 210
Figure 5.28. Regression relationship between soil backscatter (o°»a) and
volumetric soil moisture on a daily basis for watershed 20B
(unburned watershed)................................................................................ 213
Figure 5.29. Scatterplot of soil backscatter (o°»ii) versus volumetric soil moisture
by watershed burning treatment, excluding outlying image dates.............217
Figure 5.30. Estimated near surface volumetric soil moisture conditions for KPBS
on August 27 (DOY 240) and September 5 (DOY 249) after
application o f a 3 x 3 low pass filter..........................................................218
Figure 5.31. Changes in estimated near surface volumetric soil moisture
conditions for KPBS from August 27 (DOY 240) to September 5
(DOY 249) after application of a 3 x 3 low pass filter..............................219
viii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF TABLES
Table 2.01. Energy-matter interactions by spectral region and their applications
(after Elachi 1987)........................................................................................ 9
Table 2.02. Units of length used in remote sensing (after Campbell 1987)..................... 12
Table 2.03. Frequencies used in remote sensing (after Campbell 1987)..........................12
Table 2.04. Active microwave frequency designations by wavelength and
frequency (after Lillesand and Kiefer 1994)................................................14
Table 2.0S. Definitions of radar surface roughness based on the Rayleigh criterion
for three wavelengths and local incidence angles (modified from
64
Lillesand and Kiefer 1994)..............................
Table 2.06. Linear regression statistics between soil moisture and o° from a
combined dataset using C-band (5.3 GHz) radar with both VV and
HH polarization (from Gogineni et al. 1991).............................................. 87
Table 2.07. Correlation coefficients for o° versus volumetric soil moisture; B =
burned, U = unbumed, # = significant at the a = 0.05 level, * =
significant at the a = 0.01 level (from Martin et al. 1989).......................... 88
Table 3.01. Approximate area and proportionate extent o f the Prairie Parkland
(Temperate) Province sections
...................................................... 97
Table 3.02. Description o f soil pedons identified by Wehmueller (1996) for KPBS
(modified from Ransom et al. 1998)..........................................................114
Table 4.01. ERS-2 images acquired for the KPBS study area during the summer
and foil of 1996. Acquisition times are local, look and incidence
angles are in degrees, A = ascending pass, D = descending pass...............122
Table 4.02. ERS-2 radiometric confidence intervals (probability percentage) for N
number of looks (after Laurel al. 1998)....................................................124
Table 4.03. USGS 7.5 minute topographic map sheets used for image registration.......125
Table 4.04. Required variables for cloud model processing.......................................... 137
Table 4.05. Biweekly phytomass statistics for sample dates during the study
period from Konza dataset PAB01 (* current and previous year's
dead).......................................................................................................... 149
Table 4.06. Spectral bands o f the AVHRR/3 sensor (from Kramer 1996).....................152
Table 4.07. Biweekly AVHRR NDVI composite periods (1996) used in this study..... 153
Table 4.08. Spectral bands o f the LANDSAT 5 TM sensor (from Campbell 1987)...... 154
Table 5.01. Mean gravimetric (g/g) and percent volumetric water content by radar
image date for watershed ID and 20B. Values are significantly
different (P=0.35)...................................................................................... 165
Table 5.02. Mean backscatter values and standard deviations by image date for the
entire study area (All) and burned (B) and unbumed (UB)
watersheds................................................................................................. 172
Table 5.03. Correlation coefficients (r) for topographic variables and
representative image pass dates. LIA = local incidence angle, SL =
slope, AS - aspect relative to northing, AR = aspect relative to radar
look direction, EL = elevation................................................................... 174
ix
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 5.04. Summary statistics for topographic variables in the study area...................176
Table S.05. Mean values and variances of o°taai before and after topographic
correction.................................................................................................. 179
Table 5.06. Biweekly AVHRR NDVI composite periods (1996) used in this study..... 192
Table 5.07. Biweekly AVHRR NDVI composite periods for 1996 used in this
study...........................................................................................................193
Table 5.08. Summary statistics for estimated post-peak biomass production on
radar image dates....................................................................................... 196
Table 5.09. Estimated biomass production and measured volumetric soil moisture
(6v) for watershed ID (burned watershed) on October 10........................ 211
Table 5.10. Comparison o f total backscatter (o°I0U|) and soil backscatter (a°so!i)
with volumetric water content data for watershed ID (burned
watershed) on September 5,1996 (DOY 249).......................................... 212
Table 5.11. Comparison of total and soil backscatter with soil moisture data for
watershed 20B (unbumed watershed) on July 23,1996 (DOY 205).........214
Table 5.12. Slope and y-intercept values for linear regression equations between
soil backscatter and volumetric soil moisture........................................... 215
x
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ACKNOWLEDGEMENTS
Funding for the research presented in this dissertation was provided through the
generous award o f a NASA Earth System Science Fellowship. Additional support for the
purchase of radar imagery was made possible by an ERS Research and Development
Grant sponsored by the European Space Agency. Also, access to the study site and the
availability of several descriptive datasets was supported by the Konza Prairie LTER
Program (NSF BSR-9011662).
Special thanks to all of the members o f my graduate committee for their guidance
and significant commitment of time while mentoring me through the doctoral process:
Dr. John Briggs, Dr. Douglas Goodin, Dr. Lisa Harrington, Dr. James Koetliker (and Dr.
Kyle Mankin), Dr. Loyd Stone (outside chair), and especially Dr. John A. Harrington, Jr.
Finally, the successful completion of this endeavor would not have been possible had it
not been for the encouragement of my wife, Dr. Stacy Lewis Hutchinson, who is my most
enthusiastic supporter, my toughest critic, and most importantly, my best friend.
xi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1
INTRODUCTION
Soil moisture is a critical variable that contributes to the physical processes,
biogeochemistry, and human systems that influence global change (Henderson-Sellers
1996). Soil moisture is a state variable that establishes the link between the Earth's
surface and atmosphere through its impact on surface energy and moisture fluxes (O'Neill
et al. 1996).
Increasingly, remotely sensed data are being used in land surface
climatology research and modeling efforts (Greenland 1994). In addition, antecedent soil
moisture conditions affect the hydrologic behavior o f an area through the partitioning of
precipitation into runoff and storage terms (O'Neill et al.
1993).
Because of this,
quantified soil moisture conditions represent critical input for models in meteorology,
climatology, hydrology, ecology, and agriculture.
However, the value of soil moisture as an environmental descriptor or as model
input is lessened by our inability to measure it in a consistent and spatially
comprehensive manner.
Traditionally, soil moisture is measured on a point basis.
Through interpolation, these point measurements can then be extrapolated across space to
produce maps o f the spatial distribution o f soil moisture conditions.
Over complex
terrain, however, interpolated results are subject to significant error. At the root of this
problem is the natural spatial and temporal variability o f soil moisture conditions, caused
by the "inhomogeneity of soil properties, topography, land cover, and precipitation”
(Wood et al. 1993, p. 167).
I
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Recent advances in active microwave remote sensing techniques have shown the
potential for monitoring soil moisture conditions at the spatial and temporal scales
required for large scale modeling efforts (Schmugge 1983, Engman and Chauhan 1995,
Ulaby et al. 1996).
Researchers, working at fine scales and/or over small areas with
ground-based or airborne radar systems, have been successful in documenting the linear
relationship between the radar backscatter coefficient (a0) and volumetric soil moisture
content (Martin et al. 1989, Pultz et al. 1990, Wang et al. 1992, Lin et al. 1994,
Bennallegue et al. 1995). More recently, research has shifted to using data gathered by
spacebome synthetic aperture radar (SAR) sensors to estimate soil moisture conditions
over large study areas (Cognard et al. 1995, Griffiths and Wooding 1996, Henebry and
Knapp 1996, Moran et al. 1997, Sano et al. 1998, Weimann et al. 1998, Biftu and Gan
1999, Tansey et al. 1999).
Two factors, (1) vegetation and surface structural characteristics and (2) the
relative dielectric constants o f the vegetation and soil media, have been shown to be the
primary controls over the amount o f backscattered radar energy. These factors can have
a significant influence on the resulting relationship between radar backscatter and surface
soil moisture. Simulation models have been and are currently being developed to better
understand the interaction between radar energy and the complex land cover of the
Earth's surface (Fung et al. 1992, Saatchi et al. 1994). Models such as these, based upon
a series of complex mathematical computations, provide a means to assess environmental
change and, in doing so, offer insight into whether that change was brought on through
natural variability or human alterations of the environment (Steyaert 1993).
2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Use o f multifrequency and multipolarimetric radar data are among the most
promising techniques available today to estimate soil moisture using microwave imagery
(e.g., Dubois et al. 1995a, 1995b; Rajat and Barros 2000). Unfortunately, sources for this
kind o f data are limited to certain ground-based or airborne sensors or the latest version
of space shuttle imaging radar (SIR-C) technology. The collection of a global dataset of
this type is either impractical or too expensive. Currently, the only radar data acquired on
a repetitive and global basis are from satellite sensors operating at a single frequency and
polarization. Therefore, developing methods for increasing the utility o f this valuable
resource should also be a priority (Kasischke and Bourgeau-Chavez 1997).
The goal of the research described herein is to develop and apply techniques that
will generate an accurate and spatially distributed estimate o f soil moisture over a densely
vegetated study site (Konza Prairie Biological Station) using radar satellite imagery. In
support o f this goal are five objectives.
The first is the collection o f soil moisture
samples from the study area concurrently with image acquisition.
The second is to
quantify and reduce the "topographic effect" of variable terrain that is readily apparent in
active microwave images (i.e., image restoration). Third is the development of a cloud
model simulation that is used to quantify the components of the total backscatter
coefficient (<J°toiai).
By modeling the vegetative contribution to the total amount o f
backscattered radar energy (o0^ , the amount o f microwave energy scattered only by the
soil surface (o°MI|) can be calculated.
The fourth objective, and important to the success of the simulation, is the
accurate estimation of aboveground net primary production (ANPP) that can be used as
3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
model input to determine vegetation-derived backscatter (o°veg).
Aboveground
production is approximated using the relationship between ANPP and normalized
difference vegetation index (NDVI) values calculated from imagery acquired by two
optical sensors during the study period. A time series o f fine resolution NDVI values was
produced by identifying the general change in study area NDVI using biweekly
Advanced Very High Resolution Radiometer (AVHRR) composite images, then
modifying the NDVI information from a single LANDSAT Thematic Mapper (TM)
scene to match the trend.
Finally, the fifth objective is to determine the linear
relationship between o°*,,i and volumetric soil moisture so it may be inverted to quantify
near surface soil moisture conditions across the study area and over time.
The accurate determination of input parameters has been an important focus o f
geographic research in the area of hydrologic modeling (Maidment 1993). Many o f the
contributions o f geography to science, in general, and hydrologic modeling, in particular,
lie in improvements regarding the spatial representation o f space-time phenomena such
as land cover (National Research Council 1997). Accurate quantification of soil moisture
over space and time, and at a relatively fine spatial resolution, would be a significant
advancement in the spatial representation of soil moisture conditions.
Such
advancements, made possible in part by the application o f tools such as remote sensing
and geographic information systems (GIS), will continue to strengthen the role o f
geography in the "new hydrology" (Hirschboeck 1999).
Jensen (1983) challenged geographers to assume a leading role in basic research
on biophysical remote sensing and noted that the increasingly accurate quantification of
4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
different biophysical variables will have a profound impact on the success of future
environmental modeling efforts. Bauer et al. (1999), however, makes the point that the
real challenge to geographers is not simply the quantification of biophysical variables,
performed in a strictly reductionist manner, but the use o f this information in
multidisciplinary synthesis that remains mindful of the spatial and temporal scale issues
such methodologies impose on our understanding of the world. This study attempts to
address the concerns of both Jensen and Bauer et al. by (1) testing the validity of using
satellite-based radar data to estimate surface soil moisture and (2) producing an
innovative representation o f soil moisture conditions that can be used in new integrative
methodologies.
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2
LITERATURE REVIEW
Radar remote sensing is increasingly being used as a tool to collect environmental
and natural resource information. However, research and practical experience with radar
remote sensing is very limited as compared to photographic or optical scanning systems.
Radar remote sensing is characterized by four key features that help distinguish it from
other remote sensing techniques. The first feature is that radar is an active microwave
sensor, meaning that it both transmits and receives energy within a define wavelength
region. Passive sensors, such as microwave or scanning radiometers, respond only to
energy that is naturally emitted and/or reflected from various surfaces. Because of their
active nature, imaging radars operate independently o f solar illumination and, thus, have
a day and night imaging capability.
Another distinctive feature of radar remote sensing is the ability of microwave
energy to penetrate clouds and rain.
Radar energy is capable o f penetrating the
atmosphere under most conditions, depending on the specific wavelength considered.
The impact of ice and water clouds is negligible, having a significant impact on image
quality only at wavelengths o f 2 cm or less (Figure 2.01). Intense rainfall poses slightly
more of a problem, but again only for radar wavelengths below 4 cm (Figure 2.02).
Current orbiting radar sensors are unimpeded by most haze, rain, smoke, and cloud
conditions.
6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
100
(cm)
30
10
Frequency (GHz)
1
Figure 2.01. Impact o f water clouds on microwave transmission (from Ulaby et al.
1981).
100
to
a
a
a
a
a
a
a
10
(cml
10
i
Fnqucncy (GHz)
Figure 2.02. Impact o f precipitation on microwave transmission (from Ulaby et al.
1981).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A further advantage offered by radar is the capability to penetrate vegetation and
soil (Ulaby et al. 1981). Penetration into the canopy depends on both the density of the
vegetation and its moisture content. Longer wavelength microwave energy can reach
more deeply into the vegetation canopy to obtain information about both the canopy and
ground surface. Shorter wavelength energy penetrates vegetation less, and only yields
information about the upper layer of the canopy. Microwave energy is also capable of
penetrating the ground itself, with penetration depth similarly related to wavelength and
moisture content.
Finally, and probably most importantly, radar remote sensing can be distinguished
from other forms in that microwave energy interacts with terrain features in a very
different manner than does energy from the visible or thermal spectrum. Microwave
responses, therefore, offer a significantly different view of the environment.
Understanding how various target surfaces influence the amount o f radar return, or
backscatter, is a key to radar-based environmental research.
Table 2.01 lists the
mechanisms by which energy from different regions of the electromagnetic spectrum
interacts with matter to generate specific biophysical information about the surface or
volume under study.
In the visible and near infrared spectrum, vibrational and electronic energy
transitions are the prime mode of energy exchange between the incident radiation waves
and the target material
In solid materials, the packing o f atoms in the crystalline
structure leads to a variety o f energy transfer phenomena with broad interaction bands
(Elachi 1987). These include molecular and ionic vibration, crystal field effects, charge
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
transfer, and electronic conduction. One o f the most important spectral features in this
wavelength region is the sharp variation in chlorophyll reflectivity near 0.7S pm,
frequently called the "red edge.” Thus, remote sensing in the visible and near infrared
provides clues as to the chemical composition of various surfaces.
Spectral Region
Gamma-, x-rays
Ultraviolet
Visible and nearinfrared
Interaction Mechanisms
Atomic processes
Electronic processes
Electronic and vibration
molecular processes
Mid-infrared
Vibrational, vibrationalrotational molecular
processes
Thermal emission,
vibrational and rotational
processes
Rotational processes,
thermal emission,
scattering, conduction
Scattering, conduction,
ionospheric effect
Thermal infrared
Microwave
Radio
Remote Sensing Applications
Mapping radioactive materials
Presence o f H and He in atmosphere
Surface chemical composition,
vegetation cover, biological
properties
Surface and atmospheric chemical
composition
Surface heat capacity, surface and
atmospheric temperature,
atmospheric and surface constituents
Atmospheric constituents, surface
temperature, surface physical
properties
Surface physical properties,
subsurface and ionospheric sounding
Table 2.01. Energy-matter interactions by spectral region and their applications (after
Elachi 1987).
At microwave wavelengths, interaction mechanisms do not correspond to energy
bands of specific structural components (Elachi 1987). This region is characterized by
collective interactions resulting from electronic conduction and nonresonant magnetic
and electric multipolar effects. As the incident wave interacts with a molecule, electrons
are displaced, causing the formation o f an oscillating dipole that generates an
9
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
electromagnetic field. This effect of the target medium is described by the dielectric
constant. Where there is an interface between two media (e.g., soii-vegetation interface),
the microwave energy front is reflected or scattered depending on the geometric shape of
the interface. The physical properties of the interface and the dielectric properties of the
two media are the major factors affecting the interaction o f wave and matter in the
microwave and radio parts of the spectrum. Remote sensing in these energy regions
provide information about the physical and electrical properties o f an object.
Because o f these different interaction mechanisms and the information that results
from them, the combined use o f microwave, visible, and infrared radiation is
complementary and allows the examination of geometric, bulk-dielectric, and molecular
resonance/chemical properties of a surface or volume (Ulaby et al. 1981).
2.1
The Microwave Spectrum: Characteristics and Properties
The term 'remote sensing' is most commonly used in conjunction with
electromagnetic techniques o f information acquisition (Fussell et al. 198S).
These
techniques make use of the entire electromagnetic spectrum, from low-frequency radio
waves through the microwave, submillimeter, far infrared, near infrared, visible,
ultraviolet,
x-ray,
and
gamma-ray regions of the
spectrum" (Elachi
1987).
Electromagnetic energy is the means by which biophysical information is transmitted
from an object o f interest to the sensor directly through space, or indirectly by reflection,
scattering, and re-radiation to the sensor.
This information can be encoded in the
frequency content, intensity, or polarization of the electromagnetic wave (Elachi 1987).
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.1.1
Wavelength and Frequency
Electromagnetic radiation consists of coupled electrical (E) and magnetic (M)
fields, oriented a right angles to each other, and varying in magnitude in a direction
perpendicular to the axis of propagation (Figure 2.03). Such energy is characterized by
three properties: wavelength, frequency, and amplitude (Campbell 1987). Wavelength is
the distance from one wave crest to the next. Frequency is measured as the number o f
crests passing through a fixed point in a given period of time. Amplitude is equivalent to
the height o f each peak (Campbell 1987).
E lactricfiald
-----
Velocity o f light
* ■ Frequency
(numtar of cydM par iwond
paving of a fiaad point)
Figure 2.03. An electromagnetic wave (from Lillesand and Kiefer 1994).
The speed o f electromagnetic energy (c) is constant at 299,893 km per second.
Frequency (v) and wavelength (k) are related:
11
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[Equation 2.01]
c = vX
Therefore, electromagnetic energy can be specified using either frequency or wavelength
(Campbell 1987).
Common units o f measurement for wavelength and frequency are
listed in Tables 2.02 and 2.03.
Unit
kilometer (km)
meter (m)
centimeter (cm)
millimeter (mm)
micrometer (tun)
nanometer (.run)
Angstrom unit (A)
Distance
1,000 m
1m
0.01 m
0.001 m
0.000001 m
nU»UUiH/wUU
n n n n n n n n i1 m
ul
|
= 10J m
= 10" m
= 10" m
= 10* m
=
— in-3
Ill m
ul
0.0000000001 m = 10*,J m
Table 2.02. Units o f length used in remote sensing (after Campbell 1987).
Unit
hertz (Hz)
kilohertz (kHz)
megahertz (MHz)
gigahertz (GHz)
Frequencyif
I
1,000 = 10J
1,000,000 = 10*
1,000,000,000 = 109
Table 2.03. Frequencies used in remote sensing (after Campbell 1987).
The electromagnetic spectrum is divided into a number o f spectral regions (Figure
2.04).
Exact definitions for the microwave and radar portions o f the electromagnetic
spectrum, but they are generally recognized as extending from 1 mm to 0.8 m in
12
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
wavelength (Suits 1983).
The term microwave energy is somewhat of a misnomer
because the microwave spectrum includes energy with wavelengths approximately
1,000,000 times longer than in the visible spectrum.
11
22
ii ti
i 'i
m
t I c*
WAVELENGTH
E
*2 •«
•<
*
S 3 '2 ft I ii•_i ni fi
I___
I I__ I__L
-L. J
GAM MA I
— :R a v j
K RA Y ■
■UV—
r
io'
I0 ‘
10 ’
IN F R A R E O
M ED IU M
<
o■? n‘ 1
n
1 I uML , wRI M , n
L
LRAo7m5iDI0A,AM0*
I—aiieft
WAVELENGTH (X)
AUOIO
AC
M ICROW AVE-1
1 GHz
T i— r~ r
io‘
ID
I
10"
10
“1— I— I— I— I— I— I
10*
1\0
to
FREQUENCY. Hz
Tm |
■LACK REPRESENTS ATMOSPHERIC ATTENUATION
Y-0
IMOIO atiotmiiow
MOIUI
Lawicgaq
o m iiiq ii
HEATING
__1__MOLECULAR
| VI (R A T IO N
ELECTRON
SHIRTS
INTERACTION MECHANISMS
OR PHENOMENA OETECTEO
[m m a m i
MOLECULAR
ROTATION ELECTRIC MAGNETIC
FIELD FLUCTUATIONS
Figure 2.04. The electromagnetic spectrum (from Suits 1983).
Imaging radars operate within a narrow range o f the microwave spectrum. Table
2.04 lists the primary subdivisions, or bands, o f the active microwave region as defined in
the United States. These bands follow what appears to be an arbitrary naming convention
that arose out o f the development o f military radars and the former need to keep these
frequency designations secret (Campbell 1987). The same convention is persists today
though the need for military security no longer exists. Common usage also identifies the
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
lower part o f a given frequency band with a subscript u and the upper part with a
subscript a (Ulaby et al. 1981).
Band
P
L
S
C
X
Ki
K
K.
Wavelength Range (cm)
100-30
30-15
15-7.5
7.5 - 3.75
3.75 - 2.4
2.4 -1.67
1.67-1.1
1.1 -0.75
Frequency Range (GHz)
0.3 - 1.0
1.0-2.0
2 .0 -4 .0
4.0 - 8.0
8.0 - 12.5
12.5 - 18.0
18.0-26.5
26.5-40.0
Table 2.04. Active microwave frequency designations by wavelength and frequency
(after Lillesand and Kiefer 1994).
2.1.2
Polarization and Coherency
Polarization defines the spatial orientation o f the electric field of an
electromagnetic wave that is emitted and/or received by a sensor (Campbell 1987,
Kramer 1996). The polarization o f electromagnetic radiation in general, and microwave
energy in particular, is an important feature in remote sensing. In addition to intensity
and frequency measures based on the various types o f energy-matter interaction
mechanisms, the comparison o f the polarization states o f emitted and reflected radiation
provides information regarding the properties o f a radiating or scattering object.
Polarization provides an increased ability to discriminate between different surface types
(Elachi 1987).
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For a given electromagnetic wave traveling in a positive z-direction, the direction
o f propagation, the electric field (E) vector must be located in the xy-plane that is
perpendicular to the z-axis (Ulaby et al. 1981). Over time, the leading edge o f that E
vector follows a characteristic sinusoidal curve in the x-plane and one o f three possible
curves in the ry-plane as the wave proceeds in the direction o f propagation. When the
curve is straight, the wave is linearly polarized (Figure 2.05) (Ulaby et al. 1981). Here,
the electric field (E) moves only in a fixed plane in the z-direction.
I
PROPAGATION
DIRECTION
H
Figure 2.05. A linear-polarized electromagnetic wave (from Elachi 1987).
Two electromagnetic waves o f the same frequency and traveling in the same
direction o f propagation, but with different polarization directions, superimpose
themselves into a resulting wave with an electric vector equaling the sum o f the two
constituent electric fields (Figure 2.06) (Elachi 1987). In this case, the curve formed in
15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the xy-plane can be circular or elliptical in shape, giving rise to a circular or elliptical
polarization state (Ulaby et al. 1981, Elachi 1987).
OIRCCTION OP PROPAGATION
Figure 2.06. Electrical fields combine to form circular-polarized wave (from Elachi
1987).
When considering more than a single electromagnetic wave at the same time,
coherency becomes an issue. Two waves are said to be coherent if there is a systematic
relationship between the instantaneous amplitudes o f their component fields (Elachi
1987). Completely polarized plane waves such as the single wave case described above,
are typically emitted by man-made energy sources such as lasers or radars.
Electromagnetic radiation emitted or reflected by natural objects, variable topography, or
the interface o f inhomogeneous media, however, can consist o f the superposition o f many
statistically independent energy waves o f different polarizations (Elachi 1987). If there is
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
no relationship between the component fields o f such energy, it is called an incoherent or
unpolarized wave (Elachi 1987).
In addition to emitting energy at different microwave frequencies, or wavelengths,
active radar sensors can also be designed to transmit and receive energy in various
polarization modes. Consider the field vectors of a uniform plane wave with respect to
the plane o f incidence, or the plane containing the normal vector to the emitting/reflecting
surface, and the propagation vector of the incoming electromagnetic wave.
If the
incident electric vector is perpendicular to the plane o f incidence, it is called a
horizontally polarized wave (Ulaby et al. 1981). In the case where the electric vector is
parallel to, or in, the plane o f incidence, it is said to be vertically polarized (Ulaby et al.
1981, Elachi 1987).
Although all radars do not share the same design characteristics, some microwave
sensors can selectively transmit and receive in either a horizontally (H) or vertically (V)
polarized mode.
This capability provides for four possible combinations o f signal
transmission and reception (Lillesand and Kiefer 1994): H send, H receive (HH); H send,
V receive (HV); V send, H receive (VH); and V send, V receive (W ). Imagery derived
from HH or VV signal transmission and reception is like-polarized. Cross-polarized
imagery results from a HV or VH mode o f operation (Lillesand and Kiefer 1994).
Various objects can modify the polarization o f reflected energy, and the resulting
like- or cross-polarized imagery can look very different (Lillesand and Kiefer 1994).
Image interpretation allows an analyst to identify areas in a region o f interest that act as
good or poor depolarizers (Campbell 1987). In like-polarized imagery, bright regions
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
represent areas of poor depolarizers and dark areas good depolarizers. The opposite is
true for cross-polarized imagery. Causes o f depolarization include, (1) quasi-specular
reflection from smoothly undulating surfaces, (2) multiple scattering caused by surface
roughness conditions, (3) multiple volume scattering, and (4) the anisotropic physical
properties o f the illuminated target area (Fung and Ulaby 1983).
2.1.3
Penetration Depth
Microwave energy is capable of penetrating homogeneous surface features,
including vegetation canopies and soil profiles, because o f its long wavelength nature.
However, successful interpretation o f penetration depths in microwave imagery is
difficult, and often impractical, due to the many other confounding factors that influence
the brightness values within a radar scene (Campbell 1987).
Penetration depth 5P is defined as l/e, the inverse o f the radar emissivity of a
surface. Emissivity is I - 1%, where Tr is the reflectivity o f a scattering surface (Ulaby et
al. 1982). Alternatively, 5P can be expressed as (Ulaby et al. 1982, van Oevelen and
Hoekman 1999):
Sp = 1 / 2a =*« (e'),/2 / 2 jc e"
[Equation 2.02]
where;
a = field attenuation coefficient
= wavelength
e' = the real part o f the dielectric constant o f the scattering surface, and
e" - the imaginary part o f the dielectric constant o f the scattering surface
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Regardless o f its mathematical expression, the importance o f penetration depth is
that, along with resolution, it provides a measure o f the dimensions o f the effective
scattering layer, or that volume of the illuminated surface that is most responsible for
intercepting and scattering the emitted radar signal.
Another term used to describe the sensing depth o f radar energy is skin depth S„
which is defined as (Ulaby et al. 1982, van Oevelen and Hoekman 1999):
8S= 1 / a =I / [(2 ji / Xo) |Im (e)iy2|
[Equation 2.03]
where;
e = e' - ie", or the relative complex dielectric constant
By definition, skin depth is half o f the penetration depth. Therefore, the two terms are not
synonymous, though their usage in the literature is often confused. Campbell (1987)
describes penetration depth in a general sense, stating that Sp is the depth at which the
microwave signal is reduced to 1/e (=37%) o f its emitted strength at the boundary
between two inhomogeneous surfaces.
However, experimental field values for
penetration depth are highly variable, and are controlled by a number o f factors.
Penetration depth tends to increase when using low frequency (long wavelength) radar
and when environmental conditions are dry. In addition, the angle at which the radar
energy strikes a target surface influences penetration, with steeper "incidence" angles
permitting greater penetration. This relationship results in generally deeper penetration
depths in the near range portion o f a radar image than at the far-range edge (Campbell
1987).
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Indirect estimates o f 5P have been made through empirical studies of radar
brightness values at a variety o f wavelengths and incidence angles. Over bare soils, radar
penetration varies widely according to frequency and soil moisture conditions (Figure
2.07). Even for high frequency (short wavelength) microwave energy and wet soils,
radar penetration in the centimeter range is still much deeper than that which would be
obtained using visible or infrared radiation.
1.3GHZ
4.0 GHz
ur*
Oil
&2
VoIu m M c Moisture Content m, <| cur*)
Figure 2.07. Penetration depth for a loamy soil at three different microwave frequencies
(from Ulaby et al. 1982).
Penetration depth into a vegetation canopy is also affected by factors such as
canopy geometry, canopy density, and the vegetation moisture content (Ulaby et al.
1982). Ulaby et al. (1982) report that the Sp into mature, green crop canopies can be up
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
to several meters at the I GHz frequency and is limited to one meter or less at microwave
frequencies above 10 GHz. As vegetation senesces and the moisture content o f plant
canopies declines significantly, penetration o f the radar wave can be much deeper (Ulaby
etal. 1982).
2.2
Origins and History o f Radar
RADAR, or "radio detection and ranging," was originally developed as a tool using
radio waves to detect the presence o f objects and to determine their distance (range) and
location. The development o f today's imaging radars is based on the pioneering work o f
scientists striving to better understand the nature o f microwave energy.
Heinrich Hertz (1857-1894), whose name represents the measurement unit for
frequencies, extended the research o f James Clerk Maxwell (1831-1879) in describing the
characteristics o f electromagnetic radiation through his research on the propagation o f
radiation in the microwave and radio spectrum (Campbell 1987). Hertz's groundbreaking
work showing how radio waves reflect from metallic surfaces laid the foundation for
research leading to the development o f modem radios and radar.
The father o f modem radio, Guglielmo M. Marconi (1874-1937), continued the
work o f Hertz by creating working antennas for transmitting and receiving radio signals
(Campbell 1987). His demonstration o f the practicality o f trans-Atlantic communications
proved the feasibility o f wireless communications. For this work, he shared the Nobel
prize in physics in 1909 (Campbell 1987). Credit for the development o f modem radars
is often given to A.H. Taylor and L.C. Young, scientists working at the U.S. Naval
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Research Laboratory (NRL).
In what amounts to an accidental discovery, their
experiments with high frequency radio transmissions in 1922 along the Anacostia River
near Washington, D.C. showed the potential for using radio signals to detect the presence
o f ships at long distances (Campbell 1987). This unexpected finding held promise for not
only improving maritime navigation safety but also had important military implications.
The radar developed by Taylor and Young was a continuous-wave system that
depended on the separate placement of a transmitter and receiver (Skolnik 1980). The
first operational pulse radar design was used in 192S by Taylor, G. Breit, and M. Tuve to
measure the height o f the ionosphere, but work on pulse radars for detecting objects on
the Earth's surface did not begin until the early 1930s (Ulaby et al. 1981). Successful
pulse radars were developed nearly simultaneously in the United States, Great Britain,
and Germany. Robert Watson-Watt, a British scientist, is frequently credited with this
design o f the first modem radar, though others recognize Taylor and Young o f the NRL
for this accomplishment. Research in this area progressed rapidly during the period of
1933-1935, and the design evolved into a single device that coupled the transmission and
reception functions.
Design improvements proceeded quickly both during and after World War II
because o f the military significance o f radar systems. Radar is frequently credited as
evening the odds between the undermanned and under-equipped British Royal Air Force
(RAF) and the German Luftwaffe during the Battle o f Britain. Primary enhancements of
radar sensing systems during this time included a narrowing o f the wavelength interval
(improved spectral resolution), better timing o f the pulsed radar energy, and higher power
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
transmission capabilities (Campbell 1987). Side-looking radars were also developed for
military reconnaissance aircraft. Side-looking airborne radars (SLAR) allowed aircraft to
gather valuable intelligence information during varying weather conditions and from safe
distances without the fear o f enemy fire. Cumulative experiences with radar drawn from
both the World War II and the Cold War years also illustrated the potential of using radar
as a remote sensing tool for civilian scientific use.
2.3
Side-Looking Radar Operation
The general scheme o f the operation o f side-looking radar (SLR) systems is shown
in Figure 2.08 (Lillesand and Kiefer 1994). Microwave energy is transmitted from the
antenna in a series o f short bursts over a time measuring in milliseconds (IO'6 seconds).
Figure 2.08a (Lillesand and Kiefer 1994) shows the progress o f such a burst, or pulse,
over time. Solid lines represent the emitted radar energy as it moves away from the
antenna. At time 6, the wavefront is approaching the house, and at time 7 the return echo
(dashed line) begins the trip back to the receiver. At time 12, the return signal reaches
the antenna and is recorded on the response graph (Figure 2.08b).
At time 9, the
transmitted radar wave is reflected off o f the tree and this echo is recorded at time 17.
Because the tree is less reflective than the house, a weaker response is recorded at time
17.
The distance, or range, between the transmitter and reflecting object can be
determined by measuring the return tune o f the radar echoes. Since radar energy is
propagated at near the velocity o f light, the slant range (direct distance between the
23
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
transmitter and object) to an illuminated object can be found using (Lillesand and Kiefer
1994):
SR. = c t / 2
[Equation 2.04]
where:
SR = slant range
c = velocity o f electromagnetic energy (3 x 10® m s'1)
t = time between pulse transmission and echo reception
(u)
Radar puli* m m from aircraft
.. Raturn signal from ire*
Return signal from house
lb)
I
Raturn from houm
High energy
output puls*
i
Return from trs t
8
I
£
0
2
4
6
8
Time
10
12
14
16
IB
m
Figure 2.08. Side-looking radar operation (from Lillesand and Kiefer 1994).
24
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
The product o f c and t is divided by two because time is measured for both how
long is takes the radar pulse to travel to and from the target.
This principle o f
determining range electronically by measuring transmission echo time is central to the
operation o f imaging radars (Lillesand and Kiefer 1994).
2.3.1
Spatial Resolution o f SLR Systems
The ground resolution o f a SLR system is dependent on two system parameters:
pulse length (PL) and antenna beamwidth (P) (Lillesand and Kiefer 1994). Pulse length
is the amount o f time that the radar antenna is actively emitting a burst o f energy and
determines the spatial resolution o f the radar image in the range direction, or the direction
in which the radar energy is being propagated. Antenna beamwidth, or the width o f the
emitted radar pulse, controls resolution in the azimuth, or flightline, direction.
2.3.2
Range Resolution
In order for a SLR system to separately image two features that are located close
to each other, the signal reflected from each illuminated object must be received
separately by the antenna. If the two return signals overlap, the objects will appear to be
a single feature. Figure 2.09 shows the relationship between pulse length and the direct
distance between the sensor and target, or slant range.
Because the slant range distance is less than PL/2, the radar pulse is able to strike
the furthest object, B, and have the return signal reach back to the closest object, A, while
the pulse at A continues to be reflected. As a result, the return signals overlap and the
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
two separate objects appear as a single large feature on the resulting image. However, if
the slant range distance is greater than PL/2, the return signals are received separately and
the two features will be imaged independently.
Front of return
from A
Roar of outgoing
Front of return wave from
(overlapi return from A)
Figure 2.09. Relationship between pulse length and slant range on the range resolution o f
a radar image (from Lillesand and Kiefer 1994).
The slant range resolution, therefore, is independent o f the actual sensor-target
distance (slant range) and is equal to PL/2. Although slant range resolution does not vary
with distance, the corresponding ground distance is affected by sensor-target range.
Ground resolution (in the range direction) varies inversely with the cosine o f the
depression angle (0<i), or the angle formed by a line perpendicular to the radar sensor at a
26
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
right angle to the flight path and a line connecting the radar to the object being imaged
(Figure 2.10).
Slant Range
Ground Range
Figure 2.10. Relationship between radar image geometry and range resolution (R,).
The complement o f the depression angle is called the look angle (0i) and is the
off-nadir angle at which the radar sensor is illuminating the ground. The look angle is
also commonly referred to as the illumination angle, pointing angle, off-nadir angle, or
viewing angle (Kramer 1996).
Because o f the inverse relationship between ground
resolution and the cosine o f the depression angle, the ground resolution in the range
direction (R,) can be found using (Lillesand and Kiefer 1994):
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[Equation 2.05]
Rr = C T / 2 COS 0d
where:
= pulse length
t
2.3.3 Azimuth Resolution
Along with range resolution, the overall spatial resolution o f SLR systems is
determined by azimuth resolution.
Figure 2.11 shows azimuth resolution (R*) is
determined by the angular beamwidth ((3) o f the radar antenna and the ground range
(GR). With increasing ground range, azimuth resolution becomes coarser as the emitted
radar beam "fens out" from the antenna. At ground range I (GRi) the objects at points A
and B would reflect energy independently o f one another because they are not located in
the same "wedge" o f transmitted energy. Thus, A and B would be imaged separately. At
GR2, objects A and B would be illuminated simultaneously and, therefore, will not be
resolved separately. Azimuth resolution is calculated as (Lillesand and Kiefer 1994):
[Equation 2.06]
R. = G R * p
The beamwidth (P) capability of an antenna is directly proportional to the
wavelength o f the radar pulse (X) and inversely proportional to the length of the antenna
(AL) (Lillesand and Kiefer 1994):
[Equation 2.07]
p=X/AL
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 2.11. Dependence o f azimuth resolution (R«), represented at two different points
(Ri and R2) in the azimuth direction, on antenna beamwidth and ground range (from
Lillesand and Kiefer 1994).
2.3.4 Image Geometry
The geometry o f a radar image is fundamentally different from that o f a
photograph or scanner-based imagery. The key difference is that a radar image is based
on a distance rather than an angular measuring system (Lillesand and Kiefer 1994).
Because o f this and the active nature o f radar, active microwave imagery is subject to
three important types o f distortion: radar shadows, layover, and foreshortening.
Figure 2.12 shows the image geometry for a SLR system, with the upper and lower
edges o f the transmitted signal defining the edges o f the radar image in the across-track
direction. In the mid-range section o f the image, steep terrain intercepts the radar beam
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and prevents some areas from being illuminated. Such areas are referred to as radar
shadows. The presence of radar shadows are a function o f topographic relief and the
direction o f the flight path or orbital track in relation to the orientation o f the topography
(Campbell 1987).
Depression angle also plays a factor in the development of radar
shadows. Given similar conditions o f relief, radar shadowing will be more severe in farrange portions o f an image, where the depression angle is smallest, or for radar sensors
using shallower depression angles.
Radar Shadow
Figure 2.12. Side-looking radar (SLR) image geometry (after Campbell 1987).
As described earlier, radar sensors measure distance based on the time delay
between the transmitted signal and the return echo. Because o f this, these distances are in
slant range format, or the straight-line distance between the sensor and the target
(Campbell 1987, Lillesand and Kiefer 1994). The inherent slant range nature of radar
data creates geometric distortions in the resulting imagery.
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
One kind o f distortion is called the radar layover effect. Often, when vertical
objects such as tall buildings or mountains are illuminated by a radar pulse, the radar
wave front strikes the top o f the object before reaching the base. The reflected signal
from the top o f the feature, therefore, will be captured by the receiver before that o f the
base. This action causes the taller feature furthest from the sensor to appear to be leaning
inward toward the nadir point, "laying over" the closer, but shorter, feature. This layover
effect is most prominent at the near-range and is analogous to relief displacement in
aerial photography (Campbell 1987, Lillesand and Kiefer 1994).
A second type o f geometric distortion, radar foreshortening, occurs in the mid - to
far-range parts o f the image in terrain o f moderate to high relief (Campbell 1987). With
foreshortening, the transmitted radar pulse reaches the base o f an object before the top, so
that the slant range depiction o f distance and is correct with regard to the ordering o f
features from nearest to farthest. However, the relative distance between specific points
in the image may be artificially compressed or expanded. Near-range slopes will appear
steeper than they actually are while far-range slopes are shown to be longer and shallower
(Campbell 1987). This foreshortening effect becomes more pronounced as the near­
range slope approaches perpendicularity to the depression angle and illuminating radar
beam (Lillesand and Kiefer 1994). Figure 2.13 shows the dependence o f these distortions
on important factors such as across-track distance and the angular relationship between
the depression angle and illuminated feature.
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
' Mvdorarianin#*
■MMtffeMian
T m ln Uopa ttm ttr tfwn m a t Kn
«M to Inapt «Mh layawr
ftMultip in u p :
|
(«raund rin(0 (am p) I
Wookroturn
Figure 2.13. Three types o f radar image distortion: layover, radar shadow, and
foreshortening (from Lillesand and Kiefer 1994).
2.4
Radar Remote Sensing from Space
The operating characteristics and geometric considerations explained with respect
to side-looking radars apply also to radar images acquired from space. Because only
synthetic aperture radars can be operated from orbital altitudes, side-looking radars
operating from space are often referred to only by the acronym SAR. It should be noted
that slight differences in look angle and incidence angle exits with regard to spacebome
SAR because o f the earth's curvature (Lillesand and Kiefer 1994).
The history of spacebome radar remote sensing dates back to June 28, 1978 with
the launch o f the experimental radar satellite SEASAT-1, only six years after the launch
32
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o f the first optical earth observation satellite LANDSAT-l on July 23, 1972. Though
SEASAT was operational for only 106 days, its innovative design was the model from
which future radar satellites were designed. In addition, SEASAT SAR data is often
credited for igniting interest in the research potential o f terrestrial radar applications.
Despite the success o f SEASAT, major gaps in continuous global radar coverage have
existed through most o f recent history (Figure 2.14).
IM IURI 1971
1910
1913
1914
19M
1911
1990
199]
1994
1994
1999
MM
SO]
SATSLUTI
COVWAOI
|
9A3AT
SR-A
SR-B
COSMOS-1170
ALMAZ-1
WS-I
J9RS-1
3m-C(3L*-D
sntcoLft-a
n»i
RADAMAT-I
HUROOA
AM3R-I*
RADARSAT-2
7/J7/79-1079/79
11/12/91 | 11/1*91
10/1/94 | 10/12/94
7/21/9? ■ ■ ■
7/20/99
VMl § ■ ■ 10/17/92
7/1791 -STANDBY4/94
«MI | 4/20/94
9/30/94 | 10/11/94
4/2195-CURRINT
11/495. CU1UUNT
«2494 . C U m i N T l ^
FLAMNID12/2009 1881
FLAMNID2002 ^
Of WAT!ONAL
BgggSfliT A H D ir MODI
fOMCASTtD OrtRATIOM
BSS8S8 9LAMNID MMBOM
Figure 2.14. Timeline o f operational imaging SAR satellites.
Until the launch o f the ERS-l satellite in 1991, radar coverage o f the Earth was
incomplete, spatially and/or temporally, with information collected only in short windows
o f opportunity: 70 days for SEASAT-1, 3 days for SIR-A, I week for SIR-B and two
years for Cosmos-1870. In addition, data collected by Soviet, then Russian, platforms
were difficult to obtain and have had little world-wide circulation.
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Since 1991, the Earth has been imaged in the microwave spectrum in a repetitive,
continuous, and global manner. The period from 1991-1998 may well represent the highwater mark for space-based radar remote sensing. During this time, more satellites with
the widest array o f radar operating characteristics were generating data simultaneously.
This wealth of radar imagery included the ERS series o f satellites with low incidence
angle C-band (VV) data, JERS-l with high resolution (18 m) L-band (HH) data at a
relatively steep incidence angle, and RADARSAT with C-band (HH) imagery available
in a range o f off-nadir viewing angles, incidence angle, and image resolutions.
2.4.1
SEASATSAR
The SEASAT platform, launched on June 27, 1978, was the first orbiting satellite
to contain an imaging radar system (Figure 2.15). SEASAT was the first o f a proposed
series o f earth observation satellites designed primarily for oceanographic research
(Lillesand and Kiefer 1994). The satellite was launched into a near-polar orbit at an
altitude o f 800 km but, after 106 operational days, the satellite suffered an abrupt power
system failure on October 8, 1978 (Kramer 1996, Winokur 1996).
O f these 106
operational days, only during 70 were radar imagery acquired.
The SAR onboard SEASAT, just one sensor in a four instrument suite, operated at
a wavelength o f 23 cm (L-band) with HH polarization and a fixed look angle o f 20°.
Pixel resolution was approximately 25 m in range and azimuth. Image swath width was
100 km. Imagery obtained from the SEASAT SAR clearly displayed a wide range o f
oceanic and atmospheric phenomena, including the directional spectra o f ocean waves,
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
bathymetric features, polar ice motion, storms, and rainfall. Though the SEASAT SAR
was included for the m ain purpose of ocean wave imaging, the orbital characteristics o f
the satellite led to the acquisition o f many terrestrial radar images o f the northern
hemisphere (Winokur 1996).
These images demonstrated a sensitivity to surface
roughness, slope, and land-water boundaries which hinted at applications in geology,
forestry, hydrology, and land cover mapping (Winokur 1996, Kramer 1996).
/
A
Satellite:
SEASAT
6/27/78 - 10/9/78
Operational Date:
Sensor:
L-band SAR
Wavelength:
23 cm
Polarization
HH
Look Angle:
20°(fixed)
Spatial Resolution: 25m x 25m
800 km
Altitude:
100 km
Swath Width:
Notes:
1. First spacebome SAR mission.
2. 106 operational days, 70 days o f data generation.
Figure 2.15. SEASAT satellite and operating characteristics (Source: Kramer 1996).
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Despite the scientific successes leading from the mission, the short lifetime of
SEASAT prevented the acquisition o f a seasonal data set. In addition, design limitations
became evident. The SEASAT SAR was a single-parameter instrument using a fixed
wavelength, polarization, and incidence angle. While the near-nadir incidence angle was
ideal for imaging oceanic features, it produced severe geometric layover distortions in
regions o f high terrain relief.
2.4.2 Shuttle Imaging Radar (SIR)
The next space-based SAR missions following SEASAT were from two, o f a total
of four, shuttle imaging radar flights.
The first, SIR-A (shuttle imaging radar with
payload A), was launched on November 12, 1981 from the space shuttle Columbia
(Kramer 1996). The SIR-A SAR was based on the same SAR technology used for
SEASAT, but was focused on geological research applications (Winokur 1996). The
SIR-A SAR was an L-band (23.S cm) system with HH polarization and was flown at an
altitude o f260 km (Figure 2.16) (Kramer 1996).
The primary difference between SIR-A and SEASAT was incidence angle, with
the SIR-A sensor incorporating a larger look angle o f approximately 50°. The steeper
look angle generated imagery with less layover distortion in areas o f high relief an
important consideration given the geological research focus (Winokur 1996). SIR-A had
a swath width o f SO km and a spatial resolution o f 40 m in range and azimuth (Kramer
1996).
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
x
Shuttle Mission:
STS-2 (Columbia)
11/12/81-11/15/81
Operational Date:
Sensor:
SIR-A (L-band SAR)
Wavelength:
23.5 cm
Polarization:
HH
Look Angle:
47°-53°
Spatial Resolution: 40 m x 40 m
Altitude:
260 km
Swath Width:
50 km
Notes:
1. First shuttle imaging radar mission.
Figure 2.16. SIR-A sensor and operating characteristics (Source: Kramer 1996).
During its three day mission, SIR-A completed five passes over the United States
and imaged portions o f all continents, with the exception o f Antarctica (Campbell 1987).
Data for over 10 million km2 o f the earth's surface were acquired, and many tropical, arid,
and mountainous regions were imaged for the first time in the microwave spectrum
(Lillesand and Kiefer 1994).
One o f the major accomplishments o f SIR-A was the
demonstration that L-band radar could penetrate dry desert soils to a depth o f several
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
meters when buried, and previously uncharted, dry river beds were identified beneath the
Sahara Desert (Winokur 1996).
The second SIR flight, SIR-B, was flown onboard the Challenger spacecraft from
October 5-12, 1984 (Figure 2.17). Like the SIR-A sensor, the SIR-B SAR was based on
the SEASAT radar design, however, the SIR-B SAR featured a moveable antenna that
allowed for adjustments to the look angle over a 15° to 60° range (Lillesand and Kiefer
1994, Kramer 1996, Winokur 1996). As with previous radar missions, the SIR-B SAR
operated at L-band (23.5 cm) and HH polarization. Altitude was 225 km and the swath
width ranged from 30 to 60 km. Spatial resolution was 25 m in the azimuth direction and
14 m or 46 m in the range direction at a look angle o f 60° and 15°, respectively (Kramer
1996).
SIR-B provided the first multi-incidence angle radar data set and was the first
space-based sensor capable o f generating stereo radar image pairs (Lillesand and Kiefer
1994, Winokur 1996). Data generated from this mission demonstrated the capability for
land cover mapping (especially for forested areas), as well as the sensitivity o f L-band
radar to soil moisture and geologic, structural, and lithographic features (Winokur 1996).
The success o f the SIR-A and SIR-B missions, complemented by advances in
aircraft-based SAR research, led to the design o f a second generation SAR instrument for
the Space Radar Laboratory (SRL) mission (Winokur 1996).
This newly designed
sensor, the SIR-C/X-SAR, was originally scheduled for launch in 1989, but was
cancelled because of the space shuttle Challenger explosion (Lillesand and Kiefer 1994).
The launch was rescheduled and two missions using the SIR-C/X-SAR sensor, SRL-1
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and SRL-2, took place from April 9-20, 1994 and September 30-October 11, 1994
(Kramer 1996).
Shuttle Mission:
Operational Date:
Sensor:
Wavelength:
Polarization:
Look Angle:
Spatial Resolution:
Altitude:
Swath Width:
STS-13 (Challenger)
10/5/84- 10/12/84
SIR-B (L-band SAR)
23.5 cm
HH
15° - 60°
25 m x 14 m or 46 m
225 km
30 km - 60 km
Figure 2.17. SIR-B sensor and operating characteristics (Source: Kramer 1996).
The
innovative
SIR-C/X-SAR
sensor
incorporated
a
multifrequency,
multipolarization, and variable incidence angle design that allowed the acquisition of the
optical equivalent o f multispectral radar imagery from space for the first time (Figure
2.18).
This "multiparameter" capability began a new direction in SAR-based remote
sensing investigation. The SIR-C/X-SAR radar operated at three frequencies, L-band
(23.S cm), C-band (5.8 cm), and X-band (3.1 cm) and four polarization modes (HH, HV,
39
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
W , VH). The radar look angle could be adjusted in 1° increments within the range of
15° - 60° with swath widths o f 15-90 km while the shuttle spacecraft orbited at an altitude
o f 225 km (Kramer 1996).
Shuttle Mission:
Operational Date:
Sensor:
Wavelength:
Polarization
1. STS-59 (Endeavour), SLR-1
2. STS-68 (Endeavour), SLR-2
1. SLR-1 (4/9/94-4/20/94)
2. SLR-2 (9/30/94-10/11/94)
SIR-C/X-SAR (multifrequency)
L-band SAR: 23.5 cm
C-band SAR: 5.8 cm
X-bandSAR: 3.1cm
L-band SAR: H H , H V , W , V H
C-band SAR: HH, HV, W , VH
X-bandSAR: W
15° -60° (variable)
30mx30m
225 km
15-90 km
Look Angle:
Spatial Resolution:
Altitude:
Swath Width:
Notes:
I. First space-based multifrequency and
muftipolarimetric radar
Figure 2.18. SIR-C/X-SAR sensor and operating characteristics (Source: Kramer 1996).
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Multiparameter data allowed for improved identification o f vegetation, biomass
estimates, crop monitoring, and flooding conditions over areas with a wide range o f land
cover conditions (Winokur 1996).
The availability o f multipolarimetric data, in
particular, permitted major advancements in measuring the moisture content o f soils and
plant canopies from space (e.g., Dubois et al. 1995a, 1995b). Also, using calibrated
cross-polarization L-band data, the aerodynamic roughness o f the ground was measured,
allowing researchers to assess the ability o f winds to initiate dust and sand storms
(Winokur 1996).
Though many o f these scientific breakthroughs were the result o f
extensive airborne SAR research, the SIR-C missions confirmed that these capabilities
could be extended to orbiting platforms (Winokur 1996).
The biggest disadvantage regarding the four SIR missions was their ephemeral
nature. Though much o f the earth's surface was imaged, and various capabilities and
applications were discovered or advanced significantly, the combined time in space for
the four SIR missions was just over one month. Approximately 66% of this time took
place during the months o f October and November. Only one mission, SLR-1, was
conducted outside o f this time frame (April). These limitations precluded the acquisition
o f a continuous, repetitive, global, and seasonal radar data set. However, the flexibility
o f the shuttle platform allowed for relatively low cost experimentation in optimizing the
design configuration o f future SAR sensors onboard orbiting satellites (Campbell 1987).
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.4.3 ALMAZ Program
The former Soviet Union developed the Almaz (Russian for diamond) earth
resource satellite program because of their requirement for an all-weather remote sensing
platform. This capability was required to image the northern regions of Russia which are
frequently shrouded by clouds (Kramer 1996). The Almaz family of satellites was based
on the successful Salyut space station program o f the 1960's and I970's (Kramer 1996,
Winokur 1996). These spacecraft were very large and could be visited by crews in the
event o f operational difficulties (Winokur 1996).
Objectives of the program were to
support natural resource exploration, collect remotely sensed data for environmental
studies, and provide a means to monitor natural disasters and other catastrophic situations
(Kramer 1996).
2.4.3.1 COSMOS-1870
Cosmos-1870 was the initial Almaz mission and is considered the USSR's first
attempt at developing a space-based radar remote sensing capability. The satellite was
launched on July 25, 1987 and ended operations on July 30, 1989 (Kramer 1996).
Orbiting at an altitude o f 275 km, Cosmos-1870 housed two S-band (9.6 cm) SAR
sensors mounted on each side of the platform (Figure 2.19). This configuration allowed
for imaging on the east and west sides of the orbital track.
The Salyut-based satellite had the capability to roll on its axis and extend the
pointing range o f the radars in the cross-track direction upwards to a 250 km field o f view
(Kramer 1996). By extending this view, ground controllers could obtain imagery that
42
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
would normally have been outside of the either sensor's illumination range. The swath
width o f Cosmos-1870 was 20 km with a spatial resolution of approximately 25 m in the
azimuth and range directions (Kramer 1996). The mode of polarization used in the dual
sensor setup was believed to be HH.
Cosmos-1870
Satellite:
Operational Date:
7/25/87 - 7/30-89
Dual S-band SAR
Sensor:
Wavelength:
9.6 cm
Polarization:
HH
Unknown
Look Angle:
Spatial Resolution: 25 m x 25 m
Altitude:
275 km
Swath Width:
20 km
Notes:
1. First Soviet radar remote sensing platform.
Figure 2.19. Cosmos-1870 operating characteristics (Source: Kramer 1996).
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.4.3.2 ALMAZ-1
The second, and final, satellite o f the Almaz program was Almaz-1. Launched on
March 31, 1991, the satellite had an operational phase lasting from May 1991 through
October 17, 1992 when it ran out o f fuel and crashed into the Pacific Ocean (Kramer
1996). Like its prototype Cosmos-1870, Almaz-1 contained two S-band (9.6 cm) HH
SAR sensors mounted on each side o f the satellite (Figure 2.20). Imagery could be
acquired simultaneously on the east and west sides of the orbital track with a swath width
o f 40 km. Again, the satellite could be rolled about on its axis to extend its field of view,
this time up to 350 km.
Satellite:
Operational Date:
Sensor:
Wavelength:
Polarization:
Look Angle:
Spatial Resolution:
Altitude:
Swath Width:
Almaz-1
3/31/91 -10/17/92
S-band
9.6 cm
HH
30°-60° (variable)
I0mxl5m
270 km (initial), 380 km (later)
40 km
Figure 2.20. Almaz-1 satellite and operating characteristics (Source: Kramer 1996).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Incidence angles ranged between 30° and 60°, and spatial resolution varied
between 10 m and IS m depending on the range and azimuth of the specific area imaged
(Kramer 1996). The Almaz-1 satellite was initially launched into an orbit altitude of 270
km that was later increased to 380 km in an attempt to prolong the life o f the mission
(Lillesand and Kiefer 1994, Kramer 1996).
Almaz-1 was designed for a wide variety of applications including oceanography,
geology and geophysics, and agriculture and forestry (Kramer 1996). The data produced
by Almaz-1 was of high quality and complementary to that o f SPOT and LANDSAT
thematic mapper (TM) data, and is considered to compete directly with radar imagery
acquired from contemporary radar satellites (Kramer 1996).
Unlike the data from
Cosmos-1870, which was handled with a great deal of secrecy, the former Soviet space
agency made arrangements through the Almaz Corporation to market Almaz-1 data in
western nations (Kramer 1996, Winokur 1996).
A follow-on mission to Almaz-1, Almaz-IB, was being developed prior to the
dissolution o f the USSR.
This planned satellite included a three-frequency suite that
included S-, X-, and P-hand radars with a multipolarimetric capability (Kramer 1996).
Original plans called for a mid-1998 launch, but the program apparently was cancelled.
2.4.4 European Remote Sensing Satellite (ERS) Series
The European Remote Sensing Satellite (ERS) program was initiated in 1981
under the auspices o f the thirteen nation European Space Agency (ESA). This effort led
to the launch o f two radar remote sensing satellites, as well as the development of a
45
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ground support infrastructure to receive and process SAR imagery (Kramer 1996). The
first satellite in the series, ERS-l, was launched on July 17, 1991 into an altitude o f 785
km and was placed in standby mode in June 1996 (Kramer 1996).
The ERS-l satellite contains a suite o f three primary sensors, the centerpiece of
which is a C-band (S.66 cm) SAR referred to as the Active Microwave Instrument (AMI)
(Figure 2.21).
The remaining complement o f instruments includes an Along-Track
Scanning Radiometer and Microwave Sounder (ATSR) and a K-band radar altimeter
(RA-1) (Kramer 1996). Two separate radars comprise the AMI, with one sensor capable
of operating in IMAGE and WAVE modes and one for WIND mode operations (Kramer
1996, Winokur 1996). In IMAGE mode, the AMI produces radar imagery with a 100 km
swath width and a spatial resolution o f approximately 30 m in the azimuth and range
directions. The single-frequency radar incorporates a linear vertical (W ) polarization
scheme and a fixed look angle of 23° at beam center (Lillesand and Kiefer 1994, Kramer
1996, Winokur 1996).
In WAVE mode, the AMI is used to measure changes in the radar reflectivity of
sea surfaces caused by wave action and provides information about the lengths and
directions o f ocean wave systems (Kramer 1996). Operations in WIND mode are used to
calculate the speed and direction o f surface winds.
46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Satellite:
Operational Date:
Sensor:
Wavelength:
Polarization
Look Angle:
Spatial Resolution:
Altitude:
ERS-l
ERS-2
ERS-l: 7/17/91-6/96
ERS-2: 4/21/95-Current
AMI (C-band SAR)
5.66 cm
W
23°
30 m x 30 m
ERS-l: 785 km
ERS-2: 780 km
100 km
Swath Width:
Notes:
1. ERS-l currently operating in STANDBY mode.
Figure 2.21. ERS-1/2 satellite and operating characteristics (Source: Kramer 1996).
During its five years o f operation, ERS>1 AMI proved to be an extremely reliable
and stable instrument. The sensor was designed to provide data for a wide variety of
applications including ocean observation, polar ice monitoring, terrestrial ecology,
geology, and studying ocean wave phenomenon (Kramer 1996). Data collected from the
ERS-l SAR were used to monitor flood water levels during the severe 1993 flooding in
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the midwestem United States, has been proven to be a powerful tool in measuring the
Greenland ice sheet, and can detect and monitor changes in ocean oil spills (Winokur
1996). In addition, the ERS-l SAR has shown the capability o f monitoring agricultural
crops using multitemporal datasets, and has been effective in studying boreal forest issues
related to the terrestrial carbon cycle (Winokur 1996).
The second satellite in the ERS series, ERS-2, was launched on April 21, 199S
and continues to generate data today. The design and operational characteristics of ERS2 are identical to those of ERS-l with the exception of a slightly lower altitude (780 km)
and the inclusion of an additional instrument, the Global Ozone Monitoring Experiment
(GOME) (Kramer 1996). The ERS satellites were placed in a tandem orbit configuration,
beginning in mid-August 1995, so that the ERS-2 track is 24 hours earlier and identical to
that taken by ERS-l (Kramer 1996).
The benefit o f the tandem orbits is that they provide for an interferometric
capability.
ERS SAR interferometry can be used to generate detailed (10 m) digital
elevation model (DEM) data for a given area.
In addition, it is possible to create
"coherence images" that highlight areas of change.
For example, bright areas on a
coherence image indicate a strong correlation in radar reflectivity between two dates,
while dark regions suggest an alteration to the illuminated surface (Kramer 1996).
The continuous operation o f the ERS satellites since 1991 represents the first
long-term presence of radar remote sensing platforms generating data that are readily
available to the scientific community.
Both ERS-l and ERS-2 have collected an
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
extended time series of data over several seasons, permitted the measurement of long­
term and seasonal variations of land features, vegetation cover, and sea ice.
2.4.5 Japanese Environmental Remote Sensing Satellite (JERS-1)
In 1984, the National Space Development Agency of Japan began development of
the JERS-1 satellite that was successfully launched on February 11, 1992 (Lillesand and
Kiefer 1994). Though JERS-1 contains two 8-band optical sensors (OPS), the principal
payload instrument is an L-band SAR with HH polarization, a fixed 35° incidence angle,
and an 18 m spatial resolution (Figure 2.22) (Lillesand and Kiefer 1994). The repeat
cycle o f the JERS-1 orbit is 44 days.
The longer wavelength L-band radar was selected in order to provide greater
penetration of vegetative cover and sand layers, while the relatively large incidence angle
was incorporated to reduce the layover distortions in mountainous and high-retief
regions. Unlike the SEASAT and ERS-l and 2 satellites, the JERS-1 spacecraft includes
two tape recorders that may be used to store SAR data for subsequent playback to
selected ground stations. As a result, global SAR data can be acquired independent of a
global network o f ground stations. Unfortunately, a reduction in transmitter power has
limited the use o f JERS-l data.
49
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Satellite:
Operational Date:
Sensor:
Wavelength:
Polarization
Look Angle:
Spatial Resolution:
Altitude:
Swath Width:
JERS-1
2/11/92- 1998
L-band SAR
1.275 GHz
HH
35.21°
18 m x 18 m
568 km
75 km
Figure 2.22. JERS-1 satellite and operating characteristics (Source: Kramer 1996).
2.4.6
Canadian Radar Satellite (RADARSA T)
The Canadian Space Agency (CSA) developed RADARSAT, which was
launched by NASA from Vandenberg AFB, California on November 4, 1995.
This
spacecraft carries a single-frequency (C-band), single-polarization (W ) SAR which
offers a variety of beam selections that can image swaths ranging from 35 km to 500 km
with resolutions from 10 meters to 100 meters, respectively (Figure 2.23). Incidence
angles range from less than 20 degrees to more than 50 degrees.
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Satellite:
Operational Date:
Sensor:
Wavelength:
Polarization
Look Angle:
Spatial Resolution:
Altitude:
Swath Width:
RADARSAT
11/4/95 - Current
C-band SAR
5.6 cm
HH
10° - 60° (variable)
Variable (7 options)
Highest Resolution: 9 m x 9 m
Standard: 25 m x 28 m
Lowest Resolution: 100m x 100 m
798 km
Variable (6 options)
Highest Resolution: 50 km
Standard: 100 km
Lowest Resolution: 500 km
Motes:
1. Multi-mode SAR sensor that can produce 8 image
products.
2 . Three transmit pulses and use o f different radar
beams permit a wide range o f swath widths,
incidence angles, and image resolutions.
Figure 2.23. RADARSAT satellite and operating characteristics (Source: Kramer 1996).
This satellite is the first in a proposed series o f RADARSAT spacecraft. The
RADARSAT program is operated as a quasi-commercial program, with the Canadian
Government paying for the development and operation of the spacecraft and a private
51
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
company, RADARSAT International (RSI), will market data with its revenue helping to
defray the cost o f operations and production.
2.5
The Radar Equation
In radar remote sensing, the ability to detect and differentiate objects on the ground
is caused by variations in the return signal reflected by various surfaces back towards the
radar receiver.
The basic relationship that exists between the radar platform, the
illuminated target, and the received microwave signal is given by the radar equation
(Ulaby et al. 1982):
Pr = (Pt G2 A.2 o) / (4«)3 R4
[Equation 2.08]
where:
Pr = signal power returned to the radar antenna from the ground,
Pt = power transmitted toward the ground surface,
G = antenna gain,
X = wavelength o f the radar energy, and
R = the range from the transmitter to the target
Antenna gain is a measure of the radar system's ability to focus the emitted
microwave energy from the transmitter to the ground. Range is incorporated into the
radar equation to account for spreading loss, l/4itR2, or the reduction in power density
realized at the ground surface caused by the spreading o f the emitted microwave energy
about a sphere o f radius R surrounding the radar antenna (Ulaby et al. 1982). Note that
spreading loss occurs twice, both as energy is emitted from the antenna, as well as when
the radar signal is scattered back to the receiver.
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.5.1 Radar Scattering Cross-Section
It is clear that the operating characteristics of the radar sensor itself play an
important role in the amount o f radar energy available to be scattered by the ground
surface. All of the factors identified to this point are determined by the design of the
radar system and, therefore, are known quantities. The remaining variable a, or the radar
scattering cross-section, is not controlled by the radar design. Rather, o is dependent on
characteristics o f the ground surface and the type o f scattering that occurs when a target
is illuminated by incoming microwave energy. From the remote sensing perspective, a is
the key variable o f the radar equation as it is the component containing biophysical
information from the ground surface.
The radar scattering cross-section can be expressed in equation form as (Ulaby et
a l 1982):
o = An (1-&) Gu
[Equation 2.09]
where:
A« = the effective receiving area of the scatterer,
= the gain towards the receiving antenna, and
& = the fraction o f intercepted power not absorbed by the scatterer
Given the definition of
I- & represents the fraction o f intercepted power
reradiated back towards the sensor. The variables in [Equation 2.09] are combined into
the single factor o because they are both difficult to measure and their individual
contributions to the final value o f c are generally unimportant or not of interest.
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Essentially, the scattering cross-section represents the area o f an isotropic scatterer that is
sufficient to generate the measured return radar signal (Campbell 1987).
2.5.2 Differential Scattering Coefficient
The definition o f the scattering cross-section o is not based on the characteristics
o f specific types of targets, rather it is the scattering cross-section of a single target
(Ulaby et al. 1982). Because most remote sensing investigations study areas rather than
point targets, a radar equation developed for area scattering is needed.
The power
received from a single target is given by [Equation 2.08]. This equation can be modified
to reflect the average power received (Pr, lvg) by summing the scattering behavior of
multiple individual scatterers [Equation 2.10] (Ulaby et al. 1982). The variables to the
left of the summation symbol are the fixed system characteristics relating to wavelength
and spreading loss.
Pr..vg = [X2 /
(4k)3] l N . i - 1
[Equation 2.10]
(Pti Gi2 a ) / R,4
In this spatial expression of the radar equation, the scattering cross-section a can
be replaced by the differential scattering coefficient o°, also referred to as the scattering
or backscattering coefficient, which is defined as the average value of the scattering
cross-section per unit area (Ulaby et al. 1982):
[Equation 2.11]
o° = o, / AAj
54
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Alternatively, the scattering coefficient can be considered the amount o f power projected
onto an intercepting surface that produces an echo at the radar antenna equivalent to that
received from the target (Ridley et al. 1996). In radar images, pixel brightness values are
often expressed in terms o f the backscattering coefficient o°. Because the backscattering
coefficient can exhibit a large dynamic range, o° values are most often reported in
logarithmic decibel (dB) units.
Assuming the observed ground contains many point scatterers where no single
scattering element dominates the backscattered signal, and the radar equation variables P(,
G, and R are essentially constant over the illuminated area, a reasonable spatial average
of the scattering cross-section o can be calculated (Ulaby et al. 1982). In this case, the
scattering cross-section o can is replaced in [Equation 2.11] by o°AAj and substituted into
[Equation 2.10] which gives:
Pr, ivg = {K2 / (4it)3] £
n>
.
(P«i Gi2 o°AAi) / R*4
[Equation 2.12]
Again, assuming ground conditions permit the use o f o°,[Equation 2.12] can be
converted to integral form (Ulaby et al. 1982):
Pr,»vg = [^> / (4it) ] /■(«■iiiumiMicd (Pt G2 o dA) / R
[Equation 2.13]
Ulaby et al. (1982) note that [Equation 2.13] is the "area-extensive form o f the radar
equation" and is the most common form used when working with imaging radars.
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.6
Factors Influencing SAR Backscatter Response
The backscatter coefficient a 0 is a complex value that represents the cumulative
impact of several factors that control the ability of different surfaces to reflect microwave
energy and how much biophysical information is embedded in the reflected signal In the
discussion of the radar equation, it was shown that o° is related to specific operating
characteristics of a radar system such as transmitted power and antenna gain, but other
system variables such as surface roughness and incidence angle also are important. In
addition, various landscape properties are also significant factors governing o° which, in
turn, determines the tone or intensity o f a radar image.
The impact o f radar system parameters and landscape characteristics on the
scattering coefficient o° are very much interrelated. Therefore, a detailed description of
each individual factor is difficult to accomplish, a task made even more difficult when
considering the complexity of real-worki landscapes. A more convenient approach is to
consider system parameters and landscape properties in terms of their geometric and
electrical control over microwave scattering.
Geometric controls are largely related to image geometry, discussed previously, but
also includes structural attributes o f the Earth's surface and any overlying vegetation.
The other category is electrical controls which reside in the electrical properties o f the
target itself and are determined by the relative dielectric constants of the illuminated soil
and vegetation at a given microwave wavelength. While geometric factors shape the
three-dimensional distribution o f the scattered field, electrical properties help to
determine the «nwgnitiiA» of the radar backscatter response (Dobson et al. 1995).
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Temporal variations in these categorical controls can also be introduced through
land management and disturbances, the phenological development o f vegetation, and
climatic fluctuations at both daily and seasonal timescales (Dobson et al. 1995). In
addition, specific target variables such as vegetation type and canopy water status serve
to alter the scattering response in a nonrandom manner (Pultz et al. 1990).
"The
challenge in extracting usefill information from SAR data resides in how to deconvolve
the geometrical and electrical effects on radar backscatter" (Dobson et al. 1995, p. 200).
2.6.1 Geometric Factors
2.6.1.1 Topography and Corner Reflectors
One o f the most obvious features o f radar imagery is its "sidelighted" appearance
(Lillesand and Kiefer 1994).
Variations in local relief produce different radiometric
responses throughout a radar image due to the side-looking operation o f the active sensor
(Bayer et al. 1991).
Although this "topographic effect" can assist in the visual
interpretation o f SAR images, it is difficult to separate the influence o f changing slope
and aspect characteristics from variations caused by vegetation and soil conditions
(Franklin et al. 1995). This problem is magnified when considering that up to 40% or
more of the total variance in gray tones within a radar scene can be attributed to varying
terrain (Bayer etal. 1991, Verhoest et al. 1998).
Variations in terrain slope force changes to the angle, or local incidence angle (00,
at which the emitted microwave energy strikes the target surface. The local incidence
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
angle is defined as the angle o f the line formed between the sensor and target and a
perpendicular to the illuminated surface (Figure 2.24) (Campbell 1987, Kramer 1996).
As a result, slopes facing towards the sensor tend to generate high scattering coefficients
while those facing away from the sensor produce little scatter in a manner described
earlier with regard to image geometry and radar shadow effects.
Figure 2.24. Local incidence angle (0,) over flat ground (A) and the impact of sloping
terrain on local incidence angle values (B and C).
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Related to topographic effects is the frequent presence o f particularly strong
scattering features, or hard targets, often found in radar imagery (Ulaby et al. 1982). The
size o f a hard target in a given radar image is disproportionately large and does not reflect
the true size o f the surface feature (Campbell 1987). Such large return signals are caused
by two mechanisms: the comer-reSector effect and half-wavelength resonance (Ulaby et
al. 1982).
The comer reflector effect is exploited by using ground targets to externally
calibrate a radar sensor. These ground targets, or calibration targets, provide a radar
reflection o f a known scattering coefficient and can be compared to the recorded o° in an
image. Calibration targets have a constant dielectric characteristic and surface roughness,
and their response to changing incidence angles is a slowly varying function of incidence
angle (Ulaby et al. 1982). As a result, calibration targets are relatively insensitive to their
orientation to the radar sensor.
Calibration targets in the form o f dihedral or trihedral corner reflectors are often
installed for use in the field to support microwave remote sensing projects. The basic
form o f a comer reflector is the presence o f a series of flat metallic surfaces, each
arranged at 90° angles to the other (Figure 2.2S). The orientation of the metal surfaces
encourages specular reflection o f the incoming radar wave so that it is eventually
reflected back to the sensor at original magnitude and polarization.
The presence of comer-reflecting hard targets in a radar image is not limited to
calibration targets. The comer reflector effect is often responsible for the characteristic
radar signature o f urban areas where the angular orientation o f metal, concrete, and
59
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
masonry surfaces mimic the behavior of calibration targets (Campbell 1987). Comer
reflectors also appear in images o f rural areas, the cause o f which can frequently be
traced back to the presence o f metallic farm buildings.
Figure 2.25. Comer-reflecting radar calibration targets: (a) dihedral, (b) trihedral (from
Ulaby etal. 1982).
Hard targets in radar imagery can also be caused by half-wavelength resonance
effects.
Though less common than comer reflection, resonance effects can be very
strong. Resonant elements are characterized by metallic objects, or other objects with a
high dielectric constant, of lengths that are integer multiples of XU (Ulaby et al. 1982).
The magnitude o f the reradiated microwave energy caused by half-wavelength resonance
is similar to that which would be expected from a radar antenna o f the same size (Ulaby
et al. 1982).
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.6.1.2 Surface Roughness
Incidence angle determines the factor that assumes the primary geometric control
on the scattering coefficient:
topography or surface roughness.
For local incidence
angles ranging from 0° (normal to the surface) to 30°, the backscatter coefficient is
dominated by topographic slope and aspect. At incidence angles from 30° to 70°, surface
roughness becomes the dominant control
For incidence angles greater than 70°,
distortion in the form o f radar shadows will predominate (LQlesand and Kiefer 1994).
At intermediate incidence angles, where the impact o f surface roughness on the
scattering coefficient o° increases in significance, defining the "smoothness" of a surface
becomes critical. Figure 2.26 shows how the relative smoothness of the ground surface
affects the shape o f the scattered microwave signal.
Given a smooth surface with
relatively homogeneous electrical properties, the angle o f the reflection (0S) is equal to
the angle of incidence (Ulaby et al. 1982). In addition, most o f the reflected energy will
be sent in a single direction (i.e., isotropic reflection).
As the illuminated surface
becomes "rougher", incidence angle plays less o f a role in determining how and where
the radar wave is reflected. As the surface becomes very rough, the shape of the scattered
field approaches that caused by a Lambertian surface and the resulting reflection is
diffuse (i.e., anisotropic) (Ulaby etal. 1982).
For intermediately rough surfaces, the radiation pattern consists o f two parts: a
reflected (coherent) component and a scattered (diffuse or incoherent) component (Ulaby
et al. 1982). Again, the reflected component is specular, though the amount o f energy
reflected in the direction o f 0, is less than the case o f a smooth surface. As the ground
61
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
surface becomes increasingly rough, the radiation pattern becomes entirely incoherent,
with the reflected component gradually decreasing in importance.
The degree of
coherency o f the radiation pattern impacts the polarization of the scattered or reflected
microwave signal, as previously discussed. Therefore, the depolarization o f microwave
energy is affected by surface roughness conditions.
/
/
— Cohcm nt Com ponent
/
I
Diflbic Sciltcnat
Figure 2.26. Microwave scattering patterns from three surface roughness conditions
(after Ulaby et al. 1982).
An illuminated surface that appears smooth to a radar sensor may be very rough
to an optical sensor.
The degree of smoothness exhibited by the ground surface is
relative and based on statistical relationships measured in units o f wavelength (Ulaby et
al. 1982). The most common measure used to describe surface roughness is the standard
deviation o f the surface height variation or root-mean-square (rms) height o f the surface
62
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
variation, which measures the statistical variation of the random component o f ground
surface height (Ulaby et al. 1982).
Surface height variation is used in the process of classifying the smoothness of
various surfaces. One such classification is the Rayleigh criterion which identifies rough
surfaces as those where the rms height (S h ) o f surface variations is greater than 1/8 of the
wavelength (X ) of the remote sensor divided by the cosine of the local incidence angle (0 j)
[Equation 2.14].
(smooth surface)
< X / 8 COS
[Equation 2.14]
0 j < S h (rough surface)
The Rayleigh criterion shows that as the wavelength o f the incoming radar signal
decreases, surfaces will appear to become rougher given the same angle of incidence.
Similarly, for a constant wavelength value, surfaces become smoother in appearance as
the incidence angle increases, assuming the impact o f slope and aspect is negligible.
Table 2.05 shows how variations in wavelength and incidence angle influence the
classification of a surface.
While acknowledging the usefulness o f the Rayleigh criterion as a first-order
estimate of surface smoothness, Ulaby et aL (1982) present a more stringent definition
called the Fraunhofer criterion [Equation 2-15]. The basis for this modification resides in
the need for an improved surface roughness classifier when modeling the scattering
behavior o f natural surfaces, where the radar wavelengths tend to be very similar in
magnitude to the rms height of the ground surface (Ulaby et al. 1982).
63
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Roughness
Category
0 = 20°
Smooth
Intermediate
Rough
0 = 45°
Smooth
Intermediate
Rough
0 = 70°
Smooth
Intermediate
Rough
K« Band (X.=.86cm)
RMS Height (cm)
X Band (X.=3.2cm)
L Band (X=23.5cm)
<0.04
0.04-0.21
>0.21
<0.14
0.14-0.77
>0.77
<1.00
1.00-5.68
>5.68
<0.05
0.05 - 0.28
>0.28
<0.18
0.18 -1.03
> 1.03
<1.33
1.33 - 7.55
>7.55
<0.10
0.10-0.57
>0.57
<0.37
0.37-2.13
>2.13
<2.75
2.75 -15.6
> 15.6
Table 2.0S. Definitions of radar surface roughness based on the Rayleigh criterion for
three wavelengths and local incidence angles (modified from Lillesand and Kiefer 1994).
S h (sm ooth surface) < X. / 32 C O S 0 < S h (rough surface)
[ E q u a t i o n 2 . 15 ]
2.6.1.3 Surface and Volume Scattering
At this point, it is useful to make a clear delineation between surface scattering
and volume scattering. As microwave energy emitted from a radar sensor strikes the
boundary surface between two inhomogeneous media (e.g., air and bare soil), part o f that
emitted energy is reflected away from the boundary surface and the remainder is
transmitted into the lower medium (Ulaby et al. 1982). This is what is referred to as
"scattering."
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In a scenario where the lower medium is homogeneous, or nearly so, scattering
only occurs at the boundary surface itself and is considered a surface scattering event
(Ulaby et al. 1982). Conversely, if the lower medium is a mixture of materials with
varying dielectric properties (e.g., vegetation canopy), some o f the energy transmitted
into this inhomogeneous lower medium is reflected back through the upper medium and
to the radar receiver (Ulaby et al. 1982). This type of scattering is known as volume
scattering.
Much work has been done studying the surface scattering o f radar energy from
bare and vegetated surfaces. Empirical models are frequently used to develop inversion
algorithms to account for the scattering effects of bare soils with surface roughness
parameters o f sufficient magnitude to influence the backscatter coefficient (Oh et al.
1992, Benallague et al. 199S, Altese et al. 1996). However, the impact o f surface
roughness might only be significant when considering bare soils or cultivated fields,
where row structure and the lack o f a continuous vegetation canopy maximizes its
importance (Pultz et al. 1990).
Altese et al. (1996) suggest that the complexity of surface roughness conditions
over natural landscapes should minimize any decline in sensitivity of o° for soil moisture.
In fact, surface roughness conditions might only play an important role in the
interpretation of o° when the emitted radar energy is not appreciably absorbed or
scattered by a canopy layer (r.e., volume scattering) (Engheta and Elachi 1982).
In attempting to estimate soil moisture over vegetated terrain using active
microwave remote sensing, radar energy will eventually have to interact in some fashion
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
with a plant canopy to obtain information about the soil surface.
The structure of
vegetation is characterized by the size, shape, and orientation o f its main stems, branches,
and foliage (Dobson et al. 1995).
Simple structural models of plants have been
successfully used in conjunction with radiative transfer models to document the
distinctive scattering patterns of broad categories of plant growth forms (Figure 2.27).
tf)
Trunk
None
Branches
Non-woody stems
Foliage
Blade-like,
erect
Broadleaf
None
Non-woody stems
Broadleaves
Shrub
Small
Small
Blade-like or
broadleaves
Excurrent
Conical
Size and orientation
varies with height,
long and thin
Needles
Decurrent
Cylindrical,
forked
Forked, horizontal,
short and thick
Broadleaves
Columnar
Cylindrical
None
Blade-like,
clumped
Structural Category
n /
Blade-like
i t !
■e
£
£
l
5
£
Figure 2.27. Vegetation structural categories with associated size, shape, and orientation
o f canopy components (after Dobson et a l 1995).
The effect of vegetation structure on o° is greatly dependent on the incidence
angle o f the sensor, its operational frequency, and polarization of the radar energy
66
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
(Engman and Chauhan 1995). In general, vegetation structural components that are large
relative to the wavelength o f the incoming microwave signal produce radiation patterns
that are specular in nature (Dobson et al. 1995). As the objects within the vegetation
canopy become smaller with respect to wavelength, the radiation pattern becomes more
diffuse. The relationships between o° and radar wavelength, vegetation structure, and
polarization are complex.
The general rules described above indicate that canopy
conditions that result in more difiiise scattering patterns also serve as good depolarizers
o f microwave energy, while those permitting more specular reflection to occur will have
less impact on the coherency of the signal
With the presence o f a canopy, radar backscatter will consist of energy returned to
the sensor from four types of target interactions: direct scattering (volume scattering),
reflection and direct reflection (surface-volume), and surface scattering (Figure 2.28)
(Chauhan et al. 1994, Engman and Chauhan 1995). These interactions act to both reduce
the amount of measured backscattered energy and add information from the vegetation
(e.g., moisture content).
At incidence angles near nadir, o° is dominated by the surface scattering
component, hence the embedded soil moisture information is derived from the soil
surface directly. However, at larger incidence angles the canopy volume and surfacevolume component increase in importance (Fung and Eom 1985). Given the incidence
angles of current operational SAR satellites, the backscatter contribution o f vegetation
canopies is likely o f enough significance to require a correction factor in order to extract
accurate soil moisture estimates (Cognard et al. 1995).
67
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Diect Scattering
Reflected
Drect Reflected
Sieface Scattering
Figure 2.28. Sensor-target interactions affecting scattering coefficient o° information
content (after Engman and Chauhan 1995).
Again, the system variable-canopy structure relationship plays a role in governing
the relative importance of each type of target interaction on the resulting backscatter
coefficient. Structural components o f a vegetation canopy that are oriented in a similar
manner tend to be more transparent to the incoming radar signal. For example, in grass
canopies with an erect, blade-like foliage characteristics illuminated with a W-polarized
C-band SAR, surface-volume scattering was found to be the dominant target interaction
at low angles o f incidence (Saatchi and van Zyl 1995). However, as the incidence angle
increases, the contribution o f the volume scattering term becomes more important.
68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.6.2 Electrical Factors
The relative dielectric constant (e) o f a target determines its transmission,
absorption, and scattering behavior. An increase in the dielectric constant of a target
surface results in an increase in its radar reflectivity (Fung and Ulaby 1983). Changes in
the dielectric constant of natural materials, such as plant biomass and soils, is usually
associated with variations in moisture content (Fung and Ulaby 1983).
Microwave
radiation is sensitive to the moisture content o f soils because of the large difference
between the dielectric properties of water and dry soil. This difference is caused by
alignment of the water molecule dipole in response to an applied electromagnetic field
(Engman and Chauhan 1995).
The dielectric constant e of a substance is comprised of both a real s' and
imaginary e" part. Active microwave sensors are sensitive to the real component o f the
dielectric constant and passive microwave radiometers to the imaginary part (Zerdev and
Kulemin 1993). For oven dried soils,
e'
values range between 2 and 4, depending on the
soil bulk density pb, while e" is generally less than 0.05 (Ulaby et al. 1996). For dry soils
such as these, the dielectric constant is independent of both temperature and microwave
frequency (Ulaby et al. 1996).
The dielectric constant o f water, however, is e* =80 and e" = 4 at room
temperature and a frequency of I GHz (Ulaby et al. 1996). O f significance for active
microwave estimates of soil moisture, the contrasting dielectric constants o f soils varying
in moisture content can result in increases in the measured radar scattering coefficient of
as much as 10 dB (Schmugge 1983, Zerdev and Kulemin 1993). In microwave imagery,
69
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
surfaces with high dielectric properties appear brighter (higher dB) than those with lower
values.
2.6.2.1 Soil Properties
The exact scattering response of different soils varies among textural classes.
Specifically, the particle size distribution o f a soil determines the increase or decrease in
its dielectric constant as moisture conditions vary. The process by which this occurs is
similar to those forces which act to shape the soil-water characteristic curve of a given
soil (see Hillel 1982).
As water is added to a dry soil, the forces of attraction between water molecules
and soil particles decrease until such a point where any more water that is added becomes
"free water" occupying the pore space o f the soil volume. Water that is held tightly by a
soil particle contributes little in terms o f increasing its relative dielectric constant. When
sufficient water is added so that the molecules are free to rotate in response to the applied
microwave energy, significant increases in the soil dielectric constants are observed
(Schmugge 1983).
As a result, the volume fraction o f water in the soil mixture is
responsible for the increase in its dielectric constant (Ulaby et al. 1996).
At a given water content, clay soils will have a lower dielectric constant than
sandy soils because more water is adsorbed by the larger cumulative surface area of the
clay particles.
Differences in the surface area o f soil particles dictate that coarser-
textured soils (e.g., sandy soils) will show faster and larger increases (i.e., greater slope)
in dielectric constant than finer soils such as silts and clays.
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
However, experiments designed to examine the dielectric range o f different soils
varying in moisture content show that moisture content is the dominant control and that
there is less variation between soil types than one might expect (Figure 2.29). Although
Schmugge (1983) reported that the radar scattering response is relatively independent of
soil texture, Figure 2.29 indicates that for a given value of e', the range in volumetric
moisture content can vary from between l%-3% between similarly textured soils. That
range increases to as much as 7% between sandy loam and silty clay soils at volumetric
soil moisture levels above 0.3S cm3/cm3. Also, note how sandy soils exhibit a more rapid
increase in e' than do soils with a larger clay component.
Figure 2.29. Dielectric constant (e) values at 1.4 GHz for five soil types and increasing
volumetric water content (from Ulaby et al. 1986).
71
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.6.2.2 Vegetation Effects
In addition to soil and water characteristics, the dielectric properties of vegetation
have an important influence on the radar backscatter response.
Ulaby and Jedlicka
(1984) showed the dielectric properties o f vegetation to be proportional to moisture
content. Thus, dielectric properties in vegetation are subject to variations induced by the
progression o f the seasons as well as atmospheric forcings that control moisture fluxes at
daily and seasonal time scales (Dobson et al. 1995).
Imaging radars are increasingly being used to estimate vegetation biomass and
leaf area index (LAI) for a variety of landcover types including grasslands (Ulaby et al.
1984, Martin et al. 1989, Smith et al. 1994), forests (Hussin et al. 1991, Dobson et al.
1992a, Wang et al. 1994), and croplands (Durden et al. 1995, Xu etal. 1996). Because •
many radar-based soil moisture studies use the same frequency radar, soil moisture
extraction algorithms must take into account vegetation effects. Given the same radar
system parameters, greater biomass levels serve to increase the importance o f volume and
surface-volume scattering contributions to the scattering coefficient o°. Because of this,
information from the vegetation, owing to the dielectric properties o f the canopy, can
comprise a significant portion o f the biophysical information inherent in the backscatter
coefficient.
Even relatively small amounts o f grassland vegetation can influence the
success of sofl moisture estimation. Martin et al. (1989) reported statistically significant
differences in the sensitivity o f radar backscatter to soil moisture in prairie sites subjected
to various prescribed burning treatments.
72
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The general dependence o f the dielectric constant e, and therefore o°, on canopy
water status is a nonrandom source o f noise for inversion models designed to estimate
surface soil moisture conditions (Dobson et al. 1995). Dobson et al. (1995) note that the
magnitude o f this noise is (1) both geographically and seasonally controlled, and (2) most
likely to occur in semi-arid environments during dry seasons with high transpirative
demand.
For loblolly pine (Pinus taeda) forests, diurnal variations in e during late summer,
as measured by airborne SAR, cause less than 1 dB variation in o° (Dobson et al. 1995).
This phenomenon, however, is relatively unstudied in grassland environments where
grass species are likely to exhibit a much greater diurnal e fluctuation than arboreal
vegetation. Evidence o f this can be found in the patterns o f pre-dawn and midday leaf
xylem pressure potential (y) for dominant grasses in the tallgrass prairie ecosystem
(Knapp et al. 1993). The magnitude o f this diurnal fluctuation quite possibly could
exceed the relative radiometric calibration accuracy o f spacebome SAR images. In this
event, it is important to use time series data acquired at approximately the same time of
day to avoid detecting changes in o° caused by diurnal variations in canopy water status
rather than surface soil moisture conditions.
In addition to its effect on soil moisture, precipitation also has the potential to
influence o° in a nonrandom fashion through interactions with plant canopies. Dobson et
al. (1992b) showed that intercepted precipitation in a forest environment increases o° by
1-2 dB at L- and C-bands. They noted that the magnitude of change in o° at these radar
wavelengths is dependent on forest structure.
73
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In grassland environments, plant canopies and the accumulation o f a dense litter
layer insulates soil surfaces from the rapid evaporation and runoff o f water that occurs on
bare soils.
Griffiths and Wooding (1996) observed that rainfall-induced increases in
radar scattering in a time series o f ERS-l images had a longer duration for grass and
wheat fields than for bare soils. Model simulations by Saatchi et al. (1994) showed that
litter layers can significantly increase the scattering coefficient o° as the layer becomes
both thicker and wetter. Similarly, Hutchinson and Rundquist (1999) indicated that the
lingering effect o f local precipitation events and changing dew point temperatures
affected landcover classification attempts by elevating ERS-2 image brightness values in
a manner inconsistent with what would be expected for soil moisture increases, given the
prevailing soil conditions and surface types.
2.6.2.3 Temperature Effects
The dielectric constant of water is generally unaffected by environmental
temperatures greater than 0°C (Ulaby et al. 1996). Therefore, the dielectric properties of
the soil mixture are also generally insensitive to changes in temperature. However, if
freezing conditions are present, a significant change in the dielectric constant results for
soils with different volumetric water contents (mv) (Figure 2.30). At temperatures below
0°C, t> for both moist (mv - 0.26) and dry (mv - 0.0S) soils are similar to those that
would be expected for very dry soils (=3).
As temperatures increase and pass the
freezing threshold, there is nearly a threefold increase in s' for moist soils. The increase
74
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
in the imaginary part
e"
is even larger.
In contrast, temperature exerts very little
influence on the dielectric response of relatively dry soils.
200
m
10.0
10 -
1--
OlO
Figure 2.30. Temperature-dielectric constant relationship for a silt loam soil at 3 GHz
(from Ulaby et al. 1996).
In the context o f soil moisture investigations, the temperature dependency o f the
dielectric properties o f soils and moisture content is significant. Freezing temperatures
essentially mask the presence of water in the soil mixture by causing a reduction in both
the real e' and imaginary e" parts of the dielectric constant. A high dielectric value results
in an increase in the radar scattering coefficient that, in turn, can translate into higher soil
moisture values. The opposite is also true: dry soils have lower dielectric constants that
tend to reduce the scattering coefficient. As temperatures drop below 0°C, the dielectric
properties o f moist soils are rapidly altered such that they appear very similar to those of
dry soils.
75
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
2.7
RaHar-Rasgd Soil Moisture Investigations
The foundation o f radar-based soil moisture investigations is the linear relationship
between o° and volumetric soil water content in the upper 5 cm o f the soil profile
(Dobson and Ulaby 1986, Pultz et al. 1990, Lin et al. 1994, Engman and Chauhan 1995).
However, exploiting this relationship through inversion is not a simple process because
the backscatter coefficient is also dependent on many other variables, many of which
already have been discussed.
Much o f what is known about the impact o f radar system parameters and
landscape variables on the scattering o f active microwave energy can be traced back to
the 1970's and the pioneering work o f F.T. Ulaby o f the Remote Sensing Laboratory at
the University o f Kansas. The results o f Ulaby*s research and that o f others with ties to
Ulaby such as R.K. Moore, A.K. Fung, and M.C. Dobson provides an essential
foundation from which to begin microwave remote sensing studies. The findings from
this early research are presented in the seminal three volume book series, Microwave
Remote Sensing:
Active and Passive (Ulaby et al. 1981, 1982, 1986).
Similar
contributions for passive microwave systems have been made by T.J. Schmugge, working
out o f the NASA Goddard Space Flight Center.
From the perspective o f active microwave remote sensing, a majority o f early
research efforts were based on simulations or used data acquired from ground- or
airborne SAR systems. Also, many simulation and experiments were not necessarily
designed with the objective o f deriving surface soil moisture estimates. The advent of
satellite-based SAR systems and the increasing need for repetitive and accurate spatial
76
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
estimates o f soil moisture conditions has created a rapidly expanding research effort to
exploit new data sources and improve existing modeling techniques. In addition, recent
research has called into question some o f the conventional wisdom regarding the optimal
radar operating characteristics for soil moisture research and the relative impact o f
different landscape variables on the sensitivity o f the backscatter coefficient to surface
soil moisture content.
Efforts to derive soil moisture estimates using radar remote sensing techniques
can be separated into two general categories: image simulation and empirical research
using radar data acquired from a variety o f sensors. Image or model-based simulations
are generally geared around one o f the major theoretical scattering models and are
designed to test the relationships between the backscatter coefficient o° and critical
system parameters and landscape variables. Simulations are generally based upon the
scattering behavior o f hypothetical idealized surfaces, often bare soils, in order to clarify
the relationship between o° and variables such as the dielectric constant (i.e., wetness) o f
soils, surface roughness, frequency, incidence angle. Similarly, much empirical research
focuses on bare surfaces to avoid the influence o f vegetation o f the resulting scattering
coefficient.
2.7.1 Soil Moisture Estimatesfrom Bare Soil Surfaces
Surface scattering processes are often examined using one o f three theoretical
scattering models: the small perturbation model (SPM)> physical optics model (POM),
and geometrical optics model (GOM). The geometrical and physical optics models are
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
each considered variations o f the Kirchhoff scattering model (Altese et al. 1996). Each
model has been shown effective for modeling radar scattering under different surface
roughness conditions (Figure 2.31) (Engman and Chauhan 1995). Given surfaces with
increasing roughness characteristics, the preferred scattering model begins with the SPM
for the smoothest surfaces to the POM for intermediate surface, and then the GOM for
very rough surfaces {e.g., low to high frequency ranges). The mathematical expressions
for each model as well as specific validity criteria can be found in Ulaby et al. (1982).
SPM
POM
GOM
34
3
5
10
15
20
Correlation length x 1c
Figure 2.31. Valid regions o f three theoretical scattering models with respect to surface
roughness parameters (k = lit I X) (after Oh et al. 1992).
The results from various experiments with these models have led to the general
conclusion that as surface roughness increases, radar scattering increases at the expense
o f a dim inished sensitivity o f o° to soil moisture (Engman and Chauhan 1995). However,
78
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
results that contradict this generalization exist. Narayanan et al. (1994) generated L- and
X-band image simulations for 100 m x 100 m bare soil surfaces at lm resolution using
the POM and GOM scattering models, respectively. They reported that simulated L-band
images showed more variations in tone for smoother surfaces (rms height = 0.8 cm,
coefficient of variation 26%) than for rougher surfaces (rms height = 4.0 cm, coefficient
o f variation 7%) under similar soil moisture conditions. As a result, the backscatter
coefficient was found to be more sensitive to soil moisture conditions given rough soil
surfaces because o f the increased image variability inherent in the simulated smooth
surface image.
Evidence has been offered as to why results using the SPM, POM, and GOM vary
widely between studies.
Oh et al. (1992) used the University of Michigan's truck-
mounted POLARSCAT scatterometer to provide actual L-, C-, and X-band data horn
bare soil surfaces with which to compare theoretical predictions.
Results from this
analysis indicated that surface roughness conditions typical for many natural surfaces
actually foil outside o f the valid usage criteria for the theoretical models.
Further, simulations conducted by Zerdev and Kulemin (1993) show that the
detection o f different soil moisture sensitivities in L, C, X, and K bands is a reflection o f
the limitation o f current scattering models to properly account for surface roughness
effects and the inherent variability o f natural landscapes. They note that the SPM should
be the preferred theoretical scattering model, despite its smooth surface criteria, because
SPM-derived scattering predictions agree best with experimental data even for surfaces
with roughness characteristics exceeding its validity range.
79
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
An alternative to the SPM, POM, and GOM is the Integral Equation Model (IEM)
(Fung and Chen 1992). The IEM has been used successful^ for bare soils with a wide
range of surface roughness conditions. Chen et al. (199S) used the IEM to generate
simulated backscatter coefficients from surfaces with a variety o f roughness
characteristics, incidence angles, and radar frequencies in order to develop an empirical
bare surface soil moisture model. As part o f this process, they performed a sensitivity
analysis based on a Monte Carlo simulation to study the effect o f surface parameters such
as rms height, correlation length, and dielectric constant on simulated o° values. The
objective o f the sensitivity analysis was to determine which radar parameters maximize
the sensitivity o f o° to soil moisture while minimizing the impact o f surface roughness in
soil moisture inversion studies.
The sensitivity analysis o f C-band radar at both HH and W polarizations showed
that error introduced by changing incidence angles and roughness characteristics
significantly affected the estimation o f the backscatter coefficient, while only altering soil
moisture conditions resulted in no significant differences. Chen et al. (1995) also found
that by using a ratio o f polarized waves (o°hh/w), derived backscatter coefficients were
relatively unaffected by changing incidence angle and roughness parameters.
When
using the ratio method, the backscatter coefficient o°mwv at larger incidence angles
showed more sensitivity to soil moisture, while at the same time being influenced more
by the impact o f surface roughness.
In their model formulation, Chen et al. 1995 used the IEM to produce 2000 sets o f
data based on multiple frequencies, incidence angles, rms heights, correlation lengths.
80
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
They then performed a multiple linear regression analysis on the IEM datasets. The final
inversion model incorporated the copolarized ratio <x°hji*v and performed very well when
used with data from a study by Oh et al. (1992).
The IEM was also used by Altese et al. (1996) to estimate soil moisture through
inversion and to test the sensitivity o f the backscatter coefficient to surface roughness
conditions typical o f bare agricultural fields and ERS-l radar parameters (S.3 GHz,
9=23°, VV). They found that the backscatter coefficient was very sensitive to surfaces
with low rms heights (< I cm) and that variations in o° decreased with increasing
roughness.
Altese et al. (1996) also found only a mild response from the backscatter
coefficient to variations in the dielectric constant e. Given a change in the dielectric
constant from S to 25 (i.e., dry to wet soil), o° changes by only 5 dB, holding roughness
conditions equal The variation o f the backscatter coefficient with dielectric constant,
though, is similar for a range o f frequencies and incidence angles.
As a result, the
sensitivity o f o° to dielectric constant may be relatively weak, but appears to be
independent o f radar operating characteristics.
The impact o f a range o f radar parameters outside o f the ERS-l configuration,
including polarization, incidence angle, and frequency was also examined (Altese et al.
1996).
If the dielectric constant was changed from 2.5 to 45, the responses o f the
backscatter coefficient for W
and HH polarized waves were similar and increased
rapidly to a value o f 10. After 10, the increase is only slight, though W polarization
showed slightly more sensitivity to increasing dielectric constant.
Regardless, the
81
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
possibility o f distinguishing between soil moisture classes for soils with a dielectric
constant above 20 would be very difficult.
The influence o f rms height was minimized at incidence angles o f approximately
20°.
Given similar values for the dielectric constant, incidence angle, and VV
polarization, rms height variations show little impact on the backscatter coefficient for
frequencies between 4.5 and 7.5 GHz. The effect o f correlation length is consistently
lower than that of rms height, but is minimized for frequencies lower than 6 GHz. The
significance o f the results reported by Altese et al. (1996) is that, at least for bare soil
surfaces, the operating parameters o f spacebome radar satellites, ERS-l and -2, in
particular, have been shown sufficient to detect at least coarse changes in soil moisture
conditions over a wide range o f surface roughness conditions.
Soil moisture estimates were also obtained by inverting the IEM equation using
observed backscatter coefficients derived from an ERS-l image. Inversion results in an
estimate o f the dielectric constant, which is then converted into volumetric water content
using previously developed empirical relationships that were validated for a wide range
o f soils varying in moisture status. The calculated moisture content was then compared
to field measurements to determine model agreement. The inversion results were poor,
with large errors in the estimated dielectric constant values. In many cases, the IEM
could not even derive a dielectric constant (i.e., model did not converge) from the
observed backscatter coefficient. Despite their conclusions from the sensitivity analysis,
Altese et al. (1996) attributed the poor soil moisture estimates to the high sensitivity of
82
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the ERS-l radar to surface roughness conditions and the weak relationship between the
backscatter coefficient and dielectric constant.
In a similar study conducted by van Oevelen and Hoekman (1999), a version of
an IEM, called INVIEM or "inverted IEM", was applied to AirSAR data from Spain and
ERS-l imagery from Niger. Though statistical evidence of the agreement between the
inverted IEM and measured soil moisture conditions was sparse, INVIEM provided a
range o f soil moisture values that compared well with the ground data measured in Spain
and Niger.
Visual analysis also showed that the INVIEM model did a good job o f
illustrating the spatial variation o f moisture conditions, especially with regard to local
irrigation activities (van Oevelen and Hoekman 1999).
Still focusing on bare soil surfaces, Benallegue et al. (199S) used the helicoptermounted ERASME radar to measure moisture content and surface roughness over a 10
month time period as part of the Orgeval '89 experiment. The ERASME radar operates at
two frequencies (X- and C-bands) and two polarization modes (HH and W ). For their
soil moisture measurements, they selected an incidence angle o f 20° to mimic as closely
as possible the 23° incidence angle o f the ERS-l sensor.
Comparisons between the backscatter coefficient and soil moisture measurements
made at the depth o f 0-5 cm were very poor. No correlation between moisture content
and any o f the four frequency/polarization combinations (C-HH, C -W , X-HH, and XW ) was found over agricultural fields subjected to different tillage practices. This poor
relationship was attributed to surface roughness complications, indicating that ERS-l and
-2 radar parameters might be ill-suited to estimate soil moisture. Though soil moisture
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
estimation at the fold scale was unsuccessful, Benallegue et al. (1995) noted that at the
scale o f approximately 1 km2, backscatter coefficients derived from a radar configuration
similar to that o f ERS-l were able to correctly detect a general decrease in soil moisture
conditions over time.
Though many o f the findings from studies attempting to estimate soil moisture
from bare surfaces are discouraging, some research in this area has been successful
Griffiths et al. (1996) evaluated a series o f ERS-l imagery to monitor soil moisture
conditions. Part o f their study area consisted o f three bare agricultural fields and, for
these fields, a strong positive correlation was found between <r° and measured volumetric
soil moisture. Correlation coefficients for the three fields were 0.99, 0.98, and 0.76. For
the field with the lowest correlation, internal soil variability and surface roughness
changes during the study period were credited with causing the weaker relationship.
Muhipolarimetric radar data is increasingly being used to estimate soil moisture
and surface roughness parameters.
Studies by Oh et al. (1992) and Dubois et al.
(I995ab) have shown that co-polarized
( a 'W w )
and cross-polarized ratios
( o V /w )
o f the
backscatter coefficient can be used to develop robust empirical models to calculate
dielectric constant and surface roughness values over bare soils and surfaces with a very
minor vegetative component (e.g., vegetation heights > 15 cm at L-band).
These
empirical models are then inverted in a manner similar to that described for the study by
Altese et al. (1996) to estimate volumetric soil moisture (nw) and rms height (Sh).
The major advantage o f the muhipolarimetric approach is that it can be applied to
surfaces with a wider range o f surface conditions, avoiding the sensitivity o f single-
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
channel radars to varying surface roughness conditions. Unfortunately, current orbiting
radar satellites operate only at single frequencies and single polarizations so
multipolarimetric methods depend on limited data acquired from past SIR-C missions or
data generated from specially designed vehicle and aircraft radar systems such as the
University of Michigan's truck-mounted POLARSCAT scatterometer.
In general, the co-polarized ratio (o°m^v) tends to be more sensitive to changing
soil moisture than to surface roughness. Conversely, the cross-polarized ratio
(o°hV/w)
is
strongly related to rms height while being only weakly dependent on soil moisture.
Interestingly, the cross-polarized ratio
(o°hV/w)
has also shown the promise o f being a
good index o f vegetation biomass. Though the modeling effort o f Dubois et al. (1995ab)
was focused on bare soil surfaces, the valid range o f the model included some sparsely
vegetated surfaces. The correlation between an L-band
o°hV/vv
ratio and a normalized
difference vegetation index (NDVI) derived from SPOT imagery showed surprisingly
good agreement between NDVI values in the range from 0.2 to 0.6. Dubois et al.
(1995ab) were able to exploit this relationship to "mask" radar pixels that were too
heavily vegetated to model
2.7.2 Soil Moisture Estimates from Vegetated Surfaces
Techniques designed to estimate soil moisture from bare soil surfaces are
obviously limited in their geographic application.
The impact of vegetation on soil
moisture estimates is highly dependent on several factors, including thickness and density
o f the vegetated layer (and any accumulated litter), radar frequency, and incidence angle.
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Combined, these factors control how much microwave energy is able to pass through the
vegetation canopy and make contact with the soil surface. For example, as canopies
become thicker and more dense, steeper incidence angles and lower frequency (longer
wavelength) radar is necessary. If vegetation is short and relatively sparse, the range o f
applicable radar frequencies and incidence angles increase.
With the presence o f a vegetated layer above the soil-air interface, volume
scattering becomes a key contributor to the backscatter coefficient. In addition, it appears
that vegetation contributes positively in active microwave remote sensing, making o°
more responsive to soil moisture while reducing the degrading effect o f surface
roughness (Fung and Eom 1985).
The beneficial impact o f vegetation and volume
scattering effects is generally more applicable to grassland environments than forested
areas. The difference between the size, distribution, and orientation o f the structural
components o f grasslands and forests, as compared to the wavelength o f most operational
imaging radars, makes grasslands much more "transparent" to radar energy. However,
radar-based soil moisture modeling is not limited to bare soils or grasslands and can be
performed over both coniferous and deciduous forests (Karam et al. 1992).
Blanchard et al. (1982) found that o° is more sensitive to soil moisture when more
o f the backscattered signal results from multiple scattering caused by vegetation. Also,
the presence o f a diffuse vegetative layer, where the radar signal is not appreciably
absorbed, has been shown to cause increases in o° even when the incidence angle is large
and the ground surface is smooth (Engheta and Elachi 1982). This effect is especially
strong if the smooth surface has a high dielectric constant (e.g., standing water).
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Gogineni et al. 1991, as part of the First ISLSCP (International Satellite Land
Surface Climatology Project) Field Experiment (FIFE) at Konza Prairie, reported
increasing correlations between soil moisture and o° with increasing incidence angles
when using the helicopter-mounted radar HELOSCAT (Table 2.06). They also noted that
microwave energy illuminating a vegetated surface may penetrate the canopy less at
higher incidence angles, but may be counterbalanced by a decrease in the degrading
impact o f terrain slope modulation.
Incidence Angle
0°
15°
30°
45°
R
0.48
0.61
0.70
0.73
Standard Error
2.35
2.40
1.87
1.67
Table 2.06. Linear regression statistics between soil moisture and o° from a combined
dataset using C-band (S.3 GHz) radar with both W and HH polarization (from Gogineni
etal. 1991).
Microwave remote sensing o f soil moisture over grassland and agricultural
regions has been well-studied. Most o f these studies, like that o f Gogineni et al. (1991),
have used data acquired from ground-based or airborne scatterometers and have reported
reasonable correlations between o° and soil moisture for a variety of radar wavelengths
and a range o f incidence angles.
More recently, research has shifted to using data
gathered by spacebome SAR sensors to estimate soil moisture conditions over large study
areas.
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Although exact correlation values differ between these studies, the evident trend
was for superior correlations at incidence angles o f 30°-45°, with HH polarization
yielding results similar to those obtained using W polarized radar. Also, the influence
o f attenuating factors such as vegetation biomass, the presence o f a litter layer, and
terrain slope modulation were documented for the first time in field studies.
The capability o f C-band radar to sense soil moisture over grassland areas was
also reported by Martin et al. (1989). Working at the Konza Prairie and using a truckmounted 4.75 GHz (6.3 cm) radar, backscatter coefficients were found to be linearly
related and highly correlated with measured soil moisture. The high correlations found
between the o° and soil moisture at relatively large incidence angles (e.g., 30-45°)
indicates that the effectiveness o f spacebome radars may not be limited to near-nadir look
angles (Table 2.07). However, the 6.3 cm wavelength used in this study is longer than
the C-band sensors in current orbiting satellites such as ERS-2 and RADARSAT.
Polarization
(Look Angle)
HH(15)
HH(30)
HH(45)
HV(15)
HV(30)
HV(45)
W (1 5 )
W (3 0 )
W (4 5 )
1985
1984
B
0.93*
0.94*
0.94*
0.96*
0.96*
0.93*
0.90*
0.92*
0.94*
U
0.92*
0.94*
0.92*
0.90*
0.93*
0.81*
0.871
0.831
0.85*
B
0.88*
0.82*
0.71*
0.68*
0.70*
0.69*
0.84*
0.79*
0.75*
U
0.86*
0.76*
0.45
0.76*
0.76*
0.601
0.93*
0.79*
0.23
1986
B
0.97*
0.97*
0.80*
0.71*
0.59
0.74*
0.97*
0.96*
0.94*
U
0.81*
0.75*
0.74*
0.69
0.86*
0.67
0.93*
0.89*
0.73*
Table 2.07. Correlation coefficients for o° versus volumetric soil moisture; B = burned,
U = unburned, # = significant at the a = 0.05 level, * = significant at the a - 0.01 level
(from Martin etal. 1989).
88
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Martin et al. (1989) reported that relatively small differences in vegetation
biomass and/or the existence o f a litter layer caused by prescribed burning treatments had
a clear impact on the o°-soil moisture relationship. The slopes from regression equations
between the backscatter coefficient and some expressions o f soil moisture (e.g.,
gravimetric) were statistically different between burned and unbumed areas. As a result,
accurate soil moisture predictions on an operational basis may require different equations
for different areas to account for variations in vegetation biomass and thatch
accumulation.
Pultz et al. (1990) reported similar results for agricultural fields in Saskatchewan
using an airborne C-band (S.3 GHz) HH polarized radar.
Correlation coefficients
between the o° and soil moisture for structurally similar wheat fields ranged from 0.80 to
0.9S. A poorer relationship (r = 0.64 to 0.79) was found for canola fields which, relative
to the radar wavelength, offer a more substantial canopy to attenuate the radar signal.
Pultz et al. (1990) also discovered a stronger relationship between o° and soil moisture
for the sub-surface profile (2-4 cm depth) than for surface soils (0-2 cm depth) and
offered an explanation as to why this happened. If soils are wet, moisture content at both
depth ranges is similar, but the surface layer contributes most to the received signal
However, as the soil dries, the sub-surface profile remains wetter for a longer time.
Radar energy then penetrates the dry surface layer and responds to the moisture content
o f the subsurface layer only.
Less encouraging results were reported by Taconet et al. (1994) for the wheat
fields in the Orgeval watershed o f France. Using the airborne ERASME scatterometer,
89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
they found that moisture estimates based on a linear relationship between o° and soil
moisture using C-band data at steep (20°) and intermediate (40°) incidence angles are
insufficient in the presence of a vegetation layer. Only a 6 dB difference was measured
between wet and dry soils, causing the introduction o f high amounts o f error into soil
moisture estimates. Because of this weak relationship between backscatter and moisture
content, only three classes o f soil water content in the range o f 0.10 - 0.40 cm3/cm3 could
be accurately estimated.
Cognard et al. (1995) found similarly disappointing results using ERS-l imagery
over the Naizin experimental watershed in northwestern France. Calculated correlation
coefficients between o° and measured soil moisture ranged from 0.17 for vegetable fields
to 0.44 for wheat.
Surprisingly, the o°*soil moisture relationship was only slightly
stronger for grass fields (r = 0.23) than for vegetable fields. These findings support the
notions that attenuation from vegetation canopy exists, and that the amount of
degradation that this attenuation causes varies among the different cover types. Both
Taconet et al. (1994) and Cognard et al. (1995) note that, given the operating
characteristics o f current radar satellites, the scattering contribution o f vegetation
canopies are likely o f enough significance to require a correction factor in order to extract
accurate soil moisture estimates.
Such a correction was applied by Taconet et al. (1994,1996). They used a simple
simulation based on the cloud model proposed by Attema and Ulaby (1978) to reproduce
backscatter coefficients at X- and C-bands and incidence angles from 20-40°. The cloud
model portrays a plant canopy as a layer o f water droplets, or clouds, o f varying density
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and size depending on the characteristics o f the actual plant canopy which it is
simulating. Inverting the cloud model permits estimation o f soil moisture from measured
backscatter coefficients. By adding another vegetation correction to the cloud model
equation, Taconet et al. (1996) reported estimated soil moisture accuracies o f ± 0.0S
cm3/cm3 (r = 0.86). This is a significant improvement over the uncorrected form (r =
0.75).
Griffiths et al. (1996) examined multitemporal ERS-l imagery in an agricultural
region o f the UK. For three fields with bare soil surfaces, they reported a strong positive
correlation between o° and volumetric soil moisture with r > 0.98 for two of three fields.
Meanwhile, correlation values for three grass fields were not significant.
2.8
Summary of the Literature Review
Radar remote sensing is increasingly being used as a tool to collect environmental
and natural resource information. The measurement o f biophysical variables is possible
by quantifying the backscatter coefficient o°, a complex value that represents the
cumulative impact o f several factors that control the ability o f different surfaces to reflect
microwave energy.
The impact o f radar system parameters and landscape characteristics on o° are
interrelated and can be considered in terms o f their geometric and electrical control over
microwave scattering. Geometric controls are largely related to image geometry, and
include the influence o f structural attributes o f the Earth's surface, as well as that o f any
overlying vegetation.
Electrical controls are determined by the relative dielectric
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
constants o f the illuminated soil and vegetation at a given microwave wavelength. While
geometric factors shape the three-dimensional distribution o f the scattered field, electrical
properties help to determine the magnitude o f the radar backscatter response (Dobson et
al. 1995).
The foundation o f radar-based soil moisture investigations is the linear relationship
between o° and volumetric soil water content in the upper 5 cm o f the soil profile
(Dobson and Ulaby 1986, Pultz et al. 1990, Lin et al. 1994, Engman and Chauhan 1995).
However, exploiting this relationship through inversion is not a simple process because
o f the impact o f other geometric and electrical controls over the backscatter coefficient.
Many o f the studies that have reported reasonable correlations between o° and soil
moisture have used data acquired from ground-based or airborne scatterometers. Only
recently has research shifted to using data gathered by spacebome SAR sensors to
estimate soil moisture conditions over large study areas.
One promising approach to estimating soil moisture using active microwave energy
is to use muhipolarimetric radar data, which can be applied over surfaces with a wider
range o f surface conditions. Unfortunately, current orbiting radar satellites operate only
at single frequencies and single polarizations, meaning that these muhipolarimetric
techniques foil to exploit one o f the most valuable sources o f SAR data, that acquired by
orbiting satellite sensors. The Earth has been imaged in the microwave spectrum in a
repetitive, continuous, and global manner since 1991.
The research results discussed above emphasize how the backscatter coefficient
o° is affected by the relationships between different radar operating characteristics and
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
landscape features. One o f the frustrations held by those involved in active microwave
remote sensing is the prevalence o f contradictory results.
Depending on the study
referenced, different parameters are more or less critical in shaping the final value o f o°
and, therefore, any soil moisture estimate. It is important to note here, however, that
there have been few comprehensive field-based radar studies that account for both
canopy influences and terrain slope modulation in deriving backscatter-based soil
moisture estimates.
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3
STUDY AREA
3.1
Grasslands and Tallgrass Prairie
Grasslands are the largest of the major vegetation formations on Earth and are
dominant in the mid-continent region of North America (Smith 1996). From the eastern
forest-grassiand ecotone in the east, more humid grasslands transition to the semi-arid
steppes o f the west. These Great Plains grasslands, which occupy over 1.5 million km2 of
land area, can be grouped into three major types: tallgrass prairie, mixed-grass prairie,
and shortgrass steppe (Knapp and Seastedt 1998). However, much diversity exists within
each group on both an east-west and north-south gradient (Brown and Gersmehl 1985)
(Figure 3.01). Of these general grassland types, tallgrass prairie is the most mesic and
productive grassland (Samson and Knopf 1994).
Two distinct features characterize grassland ecosystems in general and tallgrass
prairie in particular (Knapp and Seastedt 1998). The first is the spatial and temporal
variability o f the climate (Borchert 1950). The second is that grasslands have properties
that make them extremely susceptible to agricultural exploitation through the
management o f domesticated plants and herbivores (Harrington and Hannan 1991). The
exceptional fertility o f tallgrass prairie soils, a result o f the much greater belowground
storage o f organic matter than in forest cover types, has made these ecosystems attractive
for row-crop agriculture (Seastedt and Knapp 1993). Within the contiguous United
States, historical records indicate that nearly 68 million ha o f tallgrass prairie once
94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
extended from Kansas to Ohio and Texas to Canada (Figure 3.02). Today, more than
90% o f this once expansive tallgrass prairie region has been cultivated, with most states
experiencing a loss in excess o f 98 % (Samson and Knopf, 1994).
Regional Grassland Gradients
\C v « « rc M
--------------r ✓ *
V
Cj>c* * T
i— --A
I
\\ \
Figure 3.01. North American grassland variation along gradient o f precipitation and
temperature (from Knapp and Seastedt 1998).
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.02. Estimated extent o f tallgrass prairie prior to European settlement (from
Kuchler 1974a).
Despite this substantial loss, more than 2 million ha o f native tallgrass prairie
remains.
However, only in the Flint Hills region o f Kansas do significant tracts of
unplowed tallgrass prairie remain (Figure 3.03). The Flint Hills are a band o f streamdissected hills covering 44,000 km2 in the east-central part o f Kansas, extending south
from near the Kansas-Nebraska border into Oklahoma. Thu region is considered to
represent the largest contiguous area o f unplowed tallgrass prairie remaining in the
United States, encompassing an area o f approximately 1.6 million ha (Knapp and
Seastedt 1998). From an ecoregion perspective, the Flint Hills are the second smallest
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
section in the Prairie Parkland (Temperate) province within the Prairie Division o f the
Humid Temperate domain (Table 3.01) (Bailey 1983, Bailey et al. 1994).
Figure 3.03. Location o f the Kansas Flint Hills (from Briggs et al. 1997).
Number
251A
251B
251C
25 ID
251E
251F
21G
Section
Red River Valley
North-Central Glaciated Plains
Central Dissected Till Plains
Central TUI Plains
Osage Plains
Flint Hills
Central Loess Plains
Area (km^)
47,400
133,100
175,900
93,500
46,900
44,000
24,300
Extent o f U.S. (%)
0.5
1.4
1.9
1.0
0.5
0.5
0.3
Table 3.01. Approximate area and proportionate extent o f the Prairie Parkland
(Temperate) Province sections.
97
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Grazing by domestic livestock is the dominant land use in the Flint Hills because
the relatively steep slopes and rocky soils have prevented widespread establishment of
row-crop agriculture. In Kansas, tallgrass prairie pastures support a livestock industry
second only to Texas in animal-unit-months (Briggs et al. 1997). Although pastureland
is the single largest land use in the region, the area supports a surprising mix o f ranching
and fanning operations (Kollmorgen and Simonett 196S).
3.2
Origins and Research History o f Konza Prairie Biological Station
In this research, the relationship between radar backscatter and soil moisture will
be examined at the Konza Prairie Biological Station (KPBS), a 3487 ha site located
approximately 11 km south of Manhattan, Kansas, within the northern reaches o f the
Kansas Flint Hills (Figure 3.04). In 2000, the official name o f the site was changed from
the Konza Prairie Research Natural Area (KPRNA) to the KPBS. The original 371 ha of
what was to become the Konza Prairie was purchased by The Nature Conservancy in
1971 and then deeded to Kansas State University (Knapp and Seastedt 1998). Through
land acquisitions over the next eight years, Konza expanded to its current size and now
represents the largest parcel o f tallgrass prairie in North America dedicated to ecological
research (Konza LTERIV Proposal). Currently, the KPBS is managed by the Division of
Biology at Kansas State University.
An experimental plan established in 1971 assigned KPBS watersheds to different
treatments o f prescribed burning, ranging from annual bums to long-term (e.g., 20 years)
unbumed (Figure 3.05). In October 1987, bison were reintroduced to Konza to examine
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the effects of gracing on the prairie ecosystem and, as o f 1992, 1100 ha were being
actively grazed. Cattle also graze in selected watersheds.
RUty County
KANSAS
/
Koau Piaint
Biolofieit Stalina
----------- 1
A
0««7 County
N
10
0
10
20
30
40
30 Kilometer*
Figure 3.04. Location o f Konza Prairie Biological Station.
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
R attdrch Traatmant*:
N • Oraaaa I f aatlva graaara
K - Uagraaaa aarth branch ol King* Craak
C - Craaad I f eattla
S • Ungraaad Shana Craak walarahada
HQ » Haadguartara araa (ungraaad)
AL ■ Lavlaad agricultural land
U M J I l t - I of jaara balwaan burning
3038 ■ Unbarnad far 3 raara, bur nod for 3 foort
UiCJ) ■ Raplleailaaa (l-lall burn, a-aprlag burn)
r
^
In A T / s TR
I /
\ J — .
WHITE
PA S.
1
I TEXAS
J k )C p a;
K 20A
N 20A
K 04A
s
> J <02A
CO4 A
04B ^K 01B
N 02A
N 04B
N 2Q B
N 04C
C01C
N 04A
N 04D
° ° 2D*—
C 04D
020D
0206/T04
Figure 3.05. Experimental design implement at Konza Prairie Biological Station (from
Knapp and Seastedt 1998).
In 1980, Konza Prairie was selected as one o f the six original Long-Term
Ecological Research (LTER) sites funded by the National Science Foundation (NSF).
Core topics o f study for the entire LTER network include, but are not (united to: (I)
pattern and control o f primary production; (2) spatial and temporal distribution of
populations selected to represent trophic structure; (3) pattern and control o f organic
matter accumulation in surface layers and sediments; (4) patterns o f inorganic input and
100
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
movements through soils, groundwater, and surface water; and (S) patterns and frequency
o f disturbance to the ecosystem (Callahan 1984). The management regime implemented
at Konza Prairie supports the main research thrust, which has expanded from a focus on
the role o f fire in the tallgrass prairie ecosystem, to a more comprehensive research
program examining the spatial and temporal interaction of fire, grazing and climate
across the prairie landscape in a nonequilibrium context (Knapp and Seastedt 1998).
Konza Prairie was included as the site o f the intensive field campaigns related to
NASA's First International Satellite Land Surface Climatology Project (ISLSCP) Field
Experiment (FIFE) in 1987 and 1989. The primary goals o f the FIFE program were to:
(1) better understand the role o f biology in controlling interactions and the atmospherebiosphere interface; and (2) to investigate the use o f remote sensing techniques to
measure climatologically significant land surface parameters (Sellers et al. 1992). Given
the intensity and duration o f the LTER effort and the comprehensive measurements and
analysis completed during the FIFE years, Konza Prairie is arguably one o f the most
intensively studied grasslands on Earth.
3.3
Climate o f Konza Prairie Biological Station
The climate o f the Great Plain region is described as continental owing to the
distance between the area and a water body that could act as a buffer in the exchange o f
energy between the Earth and atmosphere. Any discussion o f the climate o f a region
often begins by describing the temperature and rainfall regime.
In the Great Plains,
however, climate almost defies such description because o f its variability at every time
101
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
scale. At shorter time scales (e.g., daily), the extreme and highly variable nature o f the
Great Plains climate lies not only in its remoteness from a large, controlling water body,
but also in the influence o f various air masses as they move through the mid-continent
region (Rosenberg 1986). Strong seasonal differences characteristic o f the mid-continent
climate are created by an annual cycle o f varying influence by two major atmospheric
circulation components:
the tropical Hadley cell and the mid-latitude Westerlies
(Harrington and Harman 1991).
The climate o f Konza Prairie typifies continentality with its warm, moister
summers and dry, colder winters. Temperature conditions for this part of Kansas are
generally mild, with mean monthly minimum and maximum temperatures ranging from 1.9°C in January to 26.5°C in July (Figure 3.06). The mean annual temperature for the
106 years spanning 1891-1996 is 12.9°C. During this period, 56 o f 106 years had mean
annual temperatures below the long-term average (Figure 3.07).
Growing season
temperatures are consistently high, while winter temperatures are lower and much more
variable (Figure 3.06).
102
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
25-
•3.0
•15
V >5 ■
•10
■1 . 5
•
1.0
■0. 5
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
MmmB
Q T apM
~ " iriMirriitini
Figure 3.06. Mean monthly temperatures at KPBS from historical record (1891-1996).
103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
15.5
1934
1S.0 ■
14.5 14.0 •
13.5 ■
13.0 •
H
12.5 ■
12.0
■
11.5 1951
s s s x l & s s i
Y var
Figure 3.07. Long-term temperature record at KPBS (1891-1996).
Prairies, in general, and Konza, in particular, rely heavily on spring and summer
precipitation. At KPBS, approximately 75% o f the entire annual average precipitation
M s during the growing season with most rainfall (53% o f total) occurring between the
months o f May and August (Figure 3.08). A large percentage o f the annual precipitation
M s in the form o f spring and summer thunderstorms (Hayden 1998). Interestingly,
much o f the warm season precipitation M s at night. For central and eastern Kansas,
nocturnal precipitation accounts for approximately 55-60% o f all warm season rainM
events (Balling 1985).
104
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
JAN FEB MAR APR MAY JIJN JUL AUG SEP OCT NOV DEC
Mwk
g P w e i |i H r t « •
Figure 3.08. Mean monthly precipitatk)n at KPBS from historical record (1891-1996).
The mean annual precipitation o f 835 mm is sufficient to support forest or
savanna vegetation, but climatic variability (drought), fire, and grazing play important
roles in maintaining this grassland (Borchert 1950, Axelrod 1985, Anderson 1990). The
absolute monthly variability in rainfall is indicated by the standard deviation which is
lowest during the winter months, the time with the least precipitation, and highest during
the growing season, when most precipitation falls on the prairie (Figure 3.08). However,
the coefficient o f variation (CV) o f monthly precipitation totals shows that the overall
relative variability for all months is very high (> 54%) and, when considered by season,
105
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
is highest during the winter and lowest during the spring and early summer months
(Figure 3.09). During the period o f 1891 to 1996 (Figure 3.10), a majority o f years (55 o f
106) had total precipitation amounts o f less than the 106 year average o f 835 mm, so the
distribution is skewed. The one hundred year drought and flood conditions are, in terms
o f annual rainfall amounts, 460 mm and 1,400 mm, respectively (Gray et al. 1998).
too
140
-90
•S O
100-
I
60
■■50
-40
■30
40-
•20
-10
JAN FES MAR APR MAY JUN JUL AUO SEP OCT NOV DEC
Mm *
Figure 3.09. Mean monthly precipitation at KPBS from historical record (1891*1996).
106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1500 -
1931
1300
j
1100 ■
700-
500
1966
Ym t
Figure 3.10. Long-term precipitation record at KPBS (1891-1996). One hundred year
drought and flood conditions translate to approximately 460 mm (e.g., 1966) and 1,400
mm («.#., 1951) o f annual rainfall, respectively.
Prairies receive less than 200 mm o f winter precipitation, in sharp contrast to the
significant winter rains associated with the forested southeastern United States (Borchert
1950). Average snowfalls on grasslands are generally less than 800 mm (approximately
80 mm liquid water content). Konza Prairie receives an average o f 521 mm o f snow each
year (approximately 52 mm o f liquid water) (Hayden 1998).
Annual evaporation averages 1,360 mm and the annual moisture deficit
(evaporation minus precipitation) averages 525 mm, but varies from -280 mm (a surplus)
to a deficit o f 1,140 mm (Gray et al. 1998).
Precipitation generally exceeds actual
107
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
evapotranspiration for all seasons except summer (Bark 1987). Irrigation experiments
conducted at Konza have shown that, on average, an additional 350 mm y r1of additional
rainfall is required to meet the evapotranspiration demand of Konza vegetation (Lewis
1996).
Evapotranspiration rates are often enhanced by the high wind velocities
characteristic of the Great Plains. The windiest months of the year are March and April
with daily wind speeds increasing through the daylight hours before declining at night
(Flora 1948).
3.4
Geo morphology and Soils o f the Konza Prairie Biological Station
Konza Prairie was formed through millions o f years o f weathering and erosion by
tributary streams o f the Kansas River. The dendritic drainage pattern o f the five major
drainage basins within Konza (Kings Creek, Shane Creek, Pressee Branch, Swede Creek,
and Deep Creek) are a dominant landform feature (Figure 3.11) (Oviatt 1998).
Limestone benches and shale slopes produce the terraces prominent in the topography o f
the upland areas. These forms are created by the layering o f erosion-resistant limestone
with less-resistant shale (Oviatt 1998). The limestone layers tend to be permeable to
water, but the shale is not, forcing water to move laterally where the layers meet. In
places where that interface intersects the soil surface, seeps and springs occur (Oviatt
1998).
The level uplands o f shallow rocky soils give way to bottomlands with deeper,
more permeable soils. Elevations on Konza range from 312 to 445 m above sea level and
soil type and depth vary substantially with topographic position (Figure 3.12a). The
108
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Konza landscape also exhibits strong east (90°) and west (270°) aspect components
(Figure 3.12b), with moderate slopes averaging nearly 6% (Figure 3.12c).
Figure 3.11. Drainage networks o f Konza Prairie Biological Station (from Oviatt 1998).
109
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Elevation (m)
g
310-340
341 -361
362-383
384-404
409-426
427 - 448
tapect (Degrees)
Slope (Percent)
M O -4
I
5 -8
9 -1 2
1 3 -1 6
1 7 -2 0
■
Figure 3.12. (a) Elevation, (b) aspect, and (c) slope o f KPBS (from Hutchinson 1998).
Drainage densities (stream length/drainage basin area) at Konza, a measure o f the
relationship between a watershed's climate and geology, range between 5.8 km km*2 and
7.2 km km*2 (Oviatt 1998). These relatively low values, though typical for the geology in
the eastern and central United States, indicate that precipitation within the Konza
landscape is consumed primarily by a high rate o f evapotranspiration rather than
contributing to overland flow or soQ infiltration (Oviatt 1998).
The soils o f Konza Prairie are ideally suited for native grasses and rangeland.
Two associations are identified: Clime-Sogn in the south and east, and Benfield-Florence
in the north and west (Figure 3.13 and 3.14) (Jantz et al. 1975). Lowland soils are deep,
flat to sloping soils formed from thick colluvial and alluvial deposits. Representative soil
series include Tully silty clay loam, Reading sflt loam, and to a smaller extent, Irwin silty
110
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
clay loam. Surface layer soils are generally 25 - 30 cm deep with subsoil depths in
excess o f 1.2 m. These soils are moderately well to well drained with generally slow
permeability rates and are rarely flooded.
Figure 3.13. Soils o f the Clime-Sogn association and their normal positions on the Konza
Prairie landscape (from Jantz et al. 1975).
Hillside, or mid-slope, soils consist o f moderately deep (0.5 m), sloping, and
calcareous soils with surface soil depths also in the 25 -30 cm range. These soils belong
almost exclusively to the Clime-Sogn complex, which take on water very slowly and are
subject to water erosion where the surface is unprotected by vegetation. Water holding
capacity is low and very low in Clime and Sogn soils, respectively.
I ll
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 3.14. Soils o f the Benfield-Florence association and their normal positions on the
Konza Prairie landscape (from Jantz et al. 1975).
Upland soils are similar, and are represented primarily by Benfield-Florence
complex soils with small inclusions of Dwight-Irwin complex soils. These soils consist
o f moderately deep, gently sloping to moderately steep soils weathered from shale and
limestone. Surface soil layer depths range from 2 to 10 cm, though subsoil depths can
extend as deep as 0.8 m.
Chert and limestone fragments are common in Benfield-
Florence soils, and can comprise up to 35% o f the total volume o f each soil horizon
(Jantz etal., 1975).
Ransom et al. (1998) describe the work o f Wehmueller (1996) who mapped soil
distributions at a scale o f 1:2,000 for watershed N04D within Konza Prairie. Four groups
o f soils were identified: 1) soils on and near the top o f interfluves and benches, 2) soils
112
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
on steep side slopes, 3) soils on foot slopes, and 4) soils on terraces and floodplains. A
total o f 11 different soil pedons, both named and unnamed, were identified and described
(Figure 3.IS and Table 3.02). Because soil development processes, parent materials, and
the relationship between soils and landscape position are similary across KPBS, the
distributions identified for watershed N04D are representative o f those across the Konza
site.
X
Figure 3.15. Generalized illustration o f soils and their characteristic position in the
Konza Prairie landscape (from Ransom et al. 1998).
113
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Landscape
Position
Interfluves
and Benches
Side Slopes
Foot Slopes
Terraces and
Floodplains
Pedon
Taxonomy/Description
Konza
Dwight
Florence
Fine, montmorillonitic, mesic Udertic Paleustoll
Fine, montmorillonitic, mesic Typic Natrustoll
Clayey-skeletal, montmorillonitic, mesic Udertic
Argiustoll
Thin loess over limestone residuum
Loamy-skeletal, mixed, mesic Pachic Calciustoll
Very fine, montmorillonitic, mesic Udertic Argiustoll
Colluvium over shale residuum
Fine, mixed, mesic Pachic Argiustoll
Fine, mesic Typic Calciaquoll (in and below springs)
Fine, mixed, mesic Cumulic Hapludoll (above springs)
Fine-loamy, mixed, mesic Cmulic Hapludoll
Labette
Tuttle
Benfield
Clime
Tully
Unnamed
Unnamed
Ivan
Table 3.02. Description o f soil pedons identified by Wehmueller (1996) for KPBS
(modified from Ransom et al. 1998).
Konza soils are characterized by generally low permeability rates and waterholding capacities. However, the shrink-swell characteristics o f local soils, combined
with near continuous vegetation coverage, greatly impact overland flow velocities and
infiltration rates (Gray et al. 1998). Often, no runoff is observed after infrequent, low to
moderate intensity rains.
However, consecutive days o f such events can lead to
significant runoff. Duell (1990) applied SO mm o f artificial rainfall at a rate o f 60 mm/h
to a Konza Prairie site and noted no runoff. The same artificial rainfall event applied to
the same site the following day led to 10 mm to 20 mm o f runoff.
3.5
Flora o f the Konza Prairie Biological Station
Controls over regional vegetation types and transition zones (ecotones) are
complex, though moisture is often considered the single most important determining
114
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
factor.
Harrington and Hannan (1991) summarized the development o f our
understanding o f climate-vegetation relationships.
Essentially, work in this area has
evolved from a statistical comparison o f vegetation boundaries and surface climate data,
to using derived climatic measures, to recognizing the influence o f synoptic-scale
atmospheric motion in shaping regional climates. The advantage o f using pressure and
windflow patterns in vegetation studies is that they are more closely related to the
ultimate cause o f climatic differences, the uneven distribution o f energy across the Earth,
rather than incorporating proxy variables that are simply measures o f that difference
(Corcoran 1982).
The expression o f flora at Konza results from regional climatic influences as well
as local-scale factors such as soil types, burning regime, and grazing. Over five hundred
species o f vascular plants have been reported on Konza Prairie since 1975 (Freeman
1998). The ten most species-rich families account for nearly 60% o f all species identified
at Konza Prairie and are comparable to those found in the entire Flint Hills or Bluestem
Prairie region (Kuchler 1974b) (Figure 3.16). This similarity indicates that the flora of
Konza Prairie is both taxonomically and ecologically representative o f the region as a
whole (Freeman 1998).
Perennial plants comprise 65% o f all the species on Konza
Prairie, with annuals comprising most o f the remaining species (Figure 3.17).
115
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Family
Figure 3.16. Rank-order o f the 10 most species rich families found at Konza Prairie
(from Freeman 1998): Asteraceae (AST), Poaceae (POA), Fabaceae (FAB),
Brassicaceae (BRA), Euphorbiaceae (EUP), Cyperaceae (CYP), Lamiaceae (LAM),
Scrophulariaceae (SCR), Polygonaceae (POL), and Rosaceae (ROS).
70
60
SO
40
30
20
10
0
A— M
l
Life history
Figure 3.17. Life histories o f vascular plants at Konza Prairie (from Freeman 1998).
Other - combinations o f life history states.
116
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Description o f the Konza Prairie flora on a life form basis permits a greater
functional understanding o f the local vegetation.
Raunkiaer’s life form classification
system (1934) categorizes plants according to the position of the perennating bud relative
to the soil surface. Five classes are identified: phanaerophytes (trees and shrubs with
buds at least 25 cm above ground); chamaephytes (shrubs and perennial herbs with buds
less than 25 cm above ground); hemicryptophytes (perennials and biennials with buds at
ground level or within the surface layer o f soil); geophytes or cryptophytes (perennials
with rhizomes, tubers, or bulbs located well belowground); and therophytes (annuals).
Hemicryptophytes are the most abundant life-form on Konza Prairie, followed in order
by therophytes, geophytes, phanaerophytes, and chamaephytes (Figure 3.18).
40
30
I
I
30
£
10
0
—*H
I I
T
II
o
M
p
i c=a
o
c
Life-form
Figure 3.18. Life forms o f vascular plants at Konza Prairie (from Freeman 1998): H =
hemicryptophytes, T = therophytes, G = geophytes (cryptophytes), P = phanaerophytes,
0 = combinations o f other life forms, and C = chamaephytes.
117
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Description o f species based on the two major habitat types o f Konza Prairie
permits some generalization o f the flora, though these habitats are not mutually exclusive
and species frequently occur in multiple habitats. The majority o f species occur in prairie
habitats (Figure 3.19).
Less common habitat types, such as forest and wetlands,
contribute a disproportionately large percentage to the total species composition
(Freeman 1998). Species found on disturbed sites include disturbance-dependent native
plants, but more than 70% o f these species are introduced (Freeman 1998).
P
D
F
W
PDFP
A
DW
Habitats
Figure 3.19. Habitats o f the vascular plants o f Konza Prairie (from Freeman 1998): P =
prairie, D - disturbed sites, F = forests, W - wetland, A - aquatic; combinations o f letters
indicates species commonly found in more than one habitat type.
3.5.1
Tallgrass Prairie
Tallgrass prairie comprises more than 90% o f the area of Konza Prairie and the
largest category (> 40%) o f the total species count (Freeman 1998). Nearly 2S0 native
118
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
species can be found in prairie habitats on Konza, o f which approximately 70% are
perennials and 30% are annuals (Freeman 1998). The most common species are native
perennial warm-season
(C 4)
grasses including big bluestem (Andropogon gerardii
Vitman), little bluestem (A. scoparius Michx.), indiangrass (Sorghastrum nutans [L.]
Nash), and switchgrass (Panicum virgatum L.) (Freeman and Hulbert 198S, Freeman
1998). In more xeric sites, typical mixed-grass prairie and shortgrass steppe species such
as blue grama (Bouteloua gracilis [H.B.K.] Lag. ex Griffiths), sideoats grama (B.
curtipendula [Michx.] Torr.), and buffalo grass (Buchloe dactyloides [Nutt.] Engeim.)
(Freeman and Hulbert 1985, Freeman 1998) are found.
In addition to grasses, forbs are commonly found throughout the Konza
landscape. Over 400 species of forbs have been identified, most o f which are Cj (cool
season) species (Kazmaier 1993). Common species on mesic sites are white aster (Aster
ericoides L. subsp. ericoides), daisy fleabane (Erigeron strigosus Muhl. ex WOld.), wild
alfalfa (Psoralea tenuiflora Pursh), blue sage (Salvia azurea Lam.), prairie goldenrod
0Solidago missouriensis Nutt. var. fasciculata Holz.), and rigid goldenrod (S. rigida L.
var. humilis Porter) (Freeman and Hulbert 1985, Freeman 1998). Species found on more
xeric locations include western ragweed (Ambrosia psilostachya DC.), white sage
(Artemisia ludoviciana Nutt. var. ludoviciana), aromatic aster (Aster oblongifolius Nutt,
var. oblongifolius), purple coneflower (Echinacea angustifolia L. punctata Hook.),
nosebum (Tragia betonicifolia Nutt.), and western ironweed (Vemonia baldwinii Torr.
Subsp. interior (Small) Faust) (Freeman and Hulbert 1985, Freeman 1998).
119
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In addition to grasses and forbs, woody shrubs can be locally important,
especially on unburned watersheds.
Shrubs species include lead plant (Amorpha
canescens [Nutt.] Pursh), rough-leaved dogwood (Comus drummondii C.A. Mey.),
fragrant sumac (Rhus aromatica Ait. subsp. serotina [Greene] RehdJ, smooth sumac
(Rhus glabra L.), and prairie wild rose (Rosa arkansana Porter) (Freeman and Hulbert
1985, Freeman 1998). Shrub thickets are common in upland draws and on slopes along
rock outcrops and are usually dominated by rough-leaved dogwood, fragrant sumac,
smooth sumac, and buckbrush (Symphoricarpos orbiculatus) (Freeman 1998).
3.5.2
Deciduous Forest
The forests o f Konza Prairie cover only 7% o f the area o f the site and most
commonly occur in bands, ranging in width from 10 m to 300 m, along intermittent and
permanent tributaries o f the Kansas River (Freeman 1998). These "galleryn forests are
best developed along King's and Shane Creek, in the northern part o f the research area.
Konza Prairie forests have lower species richness than eastern deciduous equivalents, but
support many woody and herbaceous species near the western limit o f their ranges
(Freeman 1998).
Approximately 100 species are restricted to these forest habitats,
making up 19% o f the total species found on Konza (Freeman 1998). Dominant tree
species are bur oak (Quercus macrocarpa Michx.), chinquapin oak (Q. muehlenbergii
Engelm.), hackberry (Celtis occidentalis L.), and American elm (Ulmus americana L.)
(Freeman and Hulbert 1985, Freeman 1998).
120
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4
DATA AND METHODS
The goal o f this research is to apply techniques that will generate an accurate and
spatially distributed estimate o f soil moisture using radar satellite imagery. In support o f
this goal are five objectives. The first is to quantify and reduce the "topographic effect”
o f variable terrain that is readily apparent in active microwave images (i.e., image
restoration). Second, a simulation model is developed to estimate the contribution o f
backscattered radar energy from the vegetated surface (o°vcg) to the total radar return
(o°totti)-
To accomplish this, a multi-sensor technique using normalized difference
vegetation index (NDVI) from the Advanced Very High Resolution Radiometer
(AVHRR) and LANDSAT Thematic Mapper (TM) sensors is applied to generate a tune
series of vegetation biomass production data for model input.
By modeling the
vegetative contribution to the total amount o f backscattered radar energy, the amount o f
microwave energy scattered only by the soil surface
(< j°soii)
can then be calculated.
Finally, the linear relationship between soil backscatter and volumetric soil moisture is
identified, then inverted, to quantify near surface soil moisture conditions across the
study area and over time.
A time-series o f eight precision image products (PRI) from the second European
Remote Sensing Satellite (ERS-2) were acquired for the summer and fall o f 1996 (Table
4.01). The PRI images o f the study area were processed by the European Space Research
Institute (ESRIN) in a multipie look (N - 3), or speckle-reduced, format, and are
corrected for SAR antenna pattern and range-spreading loss (Tomassini 1999). However,
121
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
no corrections for terrain-induced variations in the backscatter intensity were made at the
processing center. In addition, sources o f image distortion such as foreshortening and
layover effects were not removed.
—^ __Date
Info
07-22 08-01 08-26 09-05 09-30
Orbit
6572
6708
7073
7209
7574
7710
8075
8211
Frame
0783
2817
0783
2817
0783
2817
0783
2817
Path
205
341
205
341
205
341
205
341
A
D
A
D
A
D
A
D
2342
1212
2342
1212
2342
1212
2242
1112
75.9
283.9
75.9
283.9
75.9
283.9
75.9
283.9
Pass
Direction
Acquisition
Time
Look Direction
10-10 11-04 11-14
Table 4.01. ERS-2 images acquired for the KPBS study area during the summer and fall
o f 1996. Acquisition times are local, look and incidence angles are in degrees, A =
ascending pass, D - descending pass.
For ascending pass images, image dates reflect the difference between local and
UTC time. For example, ESRIN reports the first image was acquired on July 23 at 4:42
a.m. (UTC). The time of overpass o f Konza Prairie, however, was July 22 at 10:42 p.m.
LST (local standard time). For the purposes o f this study, the day o f acquisition for the
ascending pass images will be one day later (e.g., July 22 will be considered July 23).
Local acquisition times for the November images have been adjusted to account for the
transition from daylight savings to standard tune.
122
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Operating characteristics o f the ERS-2 sensor are shown in Figure 2.21. ERS-2
images have a pixel and line spacing o f 12.5 m, a nominal spatial resolution o f 30 m in
both range and azimuth directions, and a pixel depth, or radiometric resolution, o f 16 bits.
The dimensions o f the ERS-2 PRI scenes are 100 km in the range direction and at least
102.5 km in the azimuth direction (Figure 4.01).
-M
-M
M
Figure 4.01. ERS-2 radar image scene locations, A-ascending pass, B-descending pass.
The radiometric confidence intervals for the ERS-2 sensor are shown in Table
4.02. Radiometric confidence intervals are a measure o f the radiometric resolution o f an
image based on the probability that the image intensity o f an homogeneous target is
within a specified value range (Laur et al. 1998). For the three-look imagery used in this
study, there is a 90% probability that the measured pixel intensity in the radar image is
123
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
within an error range o f ± 4.S dB. However, the radiometric accuracy o f the ERS-2 SAR
sensor and the stability errors associated with it have been found to be less than a fraction
o f 1 dB (Laur et a i 1998).
Confidence Interval
Bounds (±dB)
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
6.0
1
8
16
24
32
40
47
53
59
64
68
72
75
Num jer o f Looks
4
2
3
15
17
12
34
30
24
49
35
43
46
55
62
66
73
56
74
64
81
87
71
81
77
86
91
81
89
93
85
92
95
94
88
97
95
90
98
5
19
38
54
68
78
86
90
94
96
97
98
98
Table 4.02. ERS-2 radiometric confidence intervals (probability percentage) for N
number o f looks (after Laur et al. 1998).
O f the eight images acquired, the ascending and descending pass images were
recorded in the same image frame from identical orbital paths. Each of these sets o f
images is very
sim ilar in
terms o f image geometry because they are acquired from
essentially the same point in space relative to the earth and illuminate the same ground
area at the same time o f day.
Such images are also known as repeat-pass images.
Because repeat-pass images are different only in what day they were recorded, they offer
some advantages in terms o f image pre-processing by greatly simplifying the process o f
image registration.
124
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.1
Image Pre-Processing
4.1.1
Image Registration
Each subset o f repeat-pass images was first registered to the others using a first-
order transformation to better align the images, then combined into four-image stacks.
Each image stack was then subjected to an image-to-map registration process using a
third-order polynomial transformation and no fewer than 10 ground control points
(GCPs) to convert GCP file coordinates to UTM coordinates (ERDAS 1994). Standard
USGS 7.5 minute topographic maps (Table 4.03), in the form of digital raster graphics
(DRGs), were used to determine the UTM coordinates for each GCP.
Kansas 1:24,000' opographic Maps
Olsburg SW
Flush
Fort Riley NE
Riley
St. George
Junction City
Swede Creek
Keats
Tuttle Creek Dam
Manhattan
Wamego SW
Milford
Westmoreland
Milford Dam
White City NE
Odgen
White City NW
Olsburg
OlsburgNW
Wreford
Table 4.03. USGS 7.5 minute topographic map sheets used for image registration.
125
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
After the images were registered and geocoded in UTM coordinates, the two
image stacks were combined into a single eight-image stack. The Konza study area was
then subset from the radar scenes using a GIS-based vector boundary file.
4.1.2
Digital Number (DN) Conversion
In ERS-2 PRI images, pixel intensities are representative o f the radar brightness
((3°) o f the illuminated scene.
The relationship between radar brightness and the
backscatter coefficient o° o f an image is (Laur et al. 1998):
o° = (3° * sin 0,-
[Equation 4.01]
where;
o° = radar backscatter coefficient,
(3° = radar brightness, and
0i - local incidence angle.
The digital number (DN) value o f PRI images is related to {3° and o° by (Laur et
al. 1998):
DN2 = Constant * P° = Constant * (o° / sin 00 = Constant (a) * o°
[Equation 4.02]
Constant (a) is a function that is dependent on localincidence angle (Laur et al.
1998):
Constant (a) - K * (sin 0 ^ / sin 00
[Equation 4.03]
126
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
where;
K. = calibration constant (944,061)
Oucf- reference incidence angle (23°)
The calibration constant, K, is specific to the sensor and processing center and can be
found in the SAR leader file, one of four header files that accompany each PRI image
(see Appendix D o f Laur et al. 1998).
Using [Equations 4.02 and 4.03], the relationship between the DN value and the
backscatter coefficient <x° can be derived:
o° = DN2 * (1 / K) * (sin 0i / sin 0i,ref)
[Equation 4.04]
From [Equation 4.04] it is evident thatthe conversion o f image DN values to their
corresponding backscatter coefficient valuesrequires knowledge o f pixel-specific local
incidence angle. To estimate this parameter, slope (SL) and aspect (AS) information was
derived from a standard 30 m x 30 m USGS digital elevation model (DEM) o f the study
area. This topographic information can then be combined with the known value for the
sensor depression angle (0* - 67°) to develop a raster layer o f local incidence angles (0i)
on a per-pixel basis (Goyal et al. 1998):
0c = (90° - (0d ± SL))
[Equation 4.05]
127
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Slope values were added to 0* if the aspect o f the pixel, relative to the look
direction o f the radar (AR), was less than or equal to 90°. If AR > 90°, then the slope
value was subtracted. Aspect relative to radar look direction ranges from 0° to 180° and
was determined using the aspect values obtained from the DEM coupled with the known
direction in which the radar sensor is "seeing" the illuminated terrain. These values were
approximately 76° and 284° for the ascending and descending pass images, respectively.
For example, if AS = 284, then AR for a descending pass image would be 180° and the
slope value for that pixel would be subtracted from the sensor depression angle in
[Equation 4.05].
Because the DEM-based slope and aspect data were the coarsest
resolution data used in this study, the eight SAR images were resampled, using the
nearest neighbor method, to the same spatial resolution o f 30 m x 30 m.
Image digital number (DN) values, now in backscatter coefficient form, were then
converted into decibel (dB) units (Laur et al. 1998):
o°dB = 10 * log o°
[Equation 4.06]
Overall, the process o f converting raw SAR image DN values to the radar
backscatter coefficient
( o °<ib )
is summarized in Figure 4.02. On an operational basis, this
conceptual model was applied using the spatial modeling module o f ERDAS Imagine
8.3.1.
128
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
SL
RartvLqw of Local Ioddeoct Anglt
AS
'©
- * 0
-» (w )
|an
s&a
DN Convtmcn to Badncattar
\w i
DN
t v
• ©
K
0
I
t««l
Backscattw to Dedbal Unita
Stacked ractv laytr 0
Single raster layer 0
Coaitaot value ( © Caiculatioo
Figure 4.02. Conceptual model o f the conversion o f raw image DN values to o°dB.
4.1.3
Topographic Correction
Undulating terrain produces significant radiometric variations throughout a SAR
scene due the active nature of the sensor and its direct effect on sensor-target geometry.
Bayer et al. (1991) estimate that between 14% and 40% o f the variance in image DN
values can be attributed to terrain slope modulation. The effect o f relief on SAR imagery
can be significant and may be confused with the variance caused by vegetation and soil
conditions.
129
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Five topographic variables, local incidence angle (LIA), aspect relative to
northing (AS), aspect relative to radar look direction (AR), slope (SL), and elevation (EL)
were examined using correlation analysis to see which had the strongest relationship with
total backscatter. The topographic variable most closely related to total backscatter is
then used in a polynomial equation designed to minimize the impact o f that variable on
the magnitude o f o0t0tii values (Bayer et al. 1991). The polynomial equation forms the
basis for the derivation o f an empirical correction function.
In Figure 4.03, f(i) represents the polynomial model function adjusted to the
image and topographic data, f(r) is the DN value at the correction reference value, and r is
the correction reference value (Bayer et al. 1991). The original set o f DN value data
pairs is represented by DN„, DNC is the corrected DN value where the influence o f the
topographic variable has been minimized, and K(i) is the empirical correction function
(Bayer etal. 1991):
DNC= DN„ * K(i)
[Equation 4.07]
K(i) = f(r) / f(Q
[Equation 4.08]
130
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
oo
A -°-o- - 2 .0 . - -<». -
r = 23*
Local Incidence Angle (9-,)
Figure 4.03. Schematic o f the derivation o f the empirical topographic correction function
based here on local incidence angle (after Bayer et al. 1991).
The correction o f image DN values is performed with reference to conditions on
level ground. For SAR imagery acquired from the ERS-2 sensor and with respect to
incidence angle, the reference correction value r is 23°. The correction reference value
f(r) is determined by finding the value o f f(i) at an incidence angle of 23°. This value
remains fixed when calculating the correction function K(0 for all other incidence angles.
For the flat ground case, K(i) - 1 and no adjustments to the uncorrected DN values (DNU)
will be made. However, K(i) # I for the remaining incidence angle classes. As a result, a
constant correction factor will be calculated for each discrete incidence angle value
(Bayer et al. 1991). The correction is applied by multiplying the uncorrected DN values
belonging to each incidence angle class by the derived correction function [Equation
4.08].
131
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.2
4.2.1
Cloud Modei Design
Model Description
A
general first-order model describes the components o f the total amount of
backscattered radar energy (o°to<ai) recorded on a S A R image (Ulaby et al.
1 9 8 2 ):
O °toal — 0°veg + (O°sofl / L 2)
[Equation 4.09]
where:
o°totti = total amount o f radar backscatter,
o°veg = backscatter contributed by vegetation,
o°soii - backscatter contributed by soil surface, and
L2 = two-way loss factor o f the vegetation canopy.
The total amount o f radar backscatter (o°iaai) is a known quantity represented by
the calibrated and converted DN values in the SAR PRI images. Because estimation of
soil moisture is the objective, a means o f quantifying the backscatter contribution o f the
overlying vegetation canopy (o°veg) is necessary. Once this value is calculated, the model
shown in [Equation 4.09] can be rearranged and solved for soil-contributed backscatter
[Equation 4.10]
In radar remote sensing, it is the three-dimensional distribution o f water
molecules at the earth's surface that exerts primary control over the behavior of
132
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
microwave energy (Dobson et al. 1995). For most illuminated surfaces, the distribution
and abundance o f water is controlled by vegetation and the uppermost layers o f the soil
profile. A simple cloud model, first introduced by Attema and Ulaby (1978), estimates
CT°ve* by simulating the vegetation canopy as a dense "cloud" o f water particles [Equation
4.11] (Ulaby e/a/. 1982):
<y°veg = ((<Tv * cos 0) / 2
Ke)
* (1 - 1 / L2)
[Equation 4.11]
where;
0V= volume backscattering coefiBcient
k« = volume extinction coefiBcient, and
0 = sensor incidence angle (23°).
The two-way loss factor (L2) is the squared value o f the one-way (one direction) loss
factor (L ):
L = exp (Ke * h * sec 0)
[Equation 4.12]
where,
h = vegetation canopy height.
Equations 4.13-4.21 describe how the remaining components of the cloud model,
ov and
Kc,
are calculated. First, the derivation o f the volume backscattering coefiBcient ov
is shown. The vegetation canopy volume is assumed to consist o f randomly distributed
uniform scatterers. Ignoring the effect o f multiple scattering, the volume backscattering
coefiBcient (ov) is determined by (Ulaby et al. 1982):
133
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
[Equation 4.13]
<jv - N «b
where;
N = number o f scattering particles per unit volume, and
Ob = backscattering cross-section o f particle.
The backscattering cross-section o f the water, or cloud, particle (Ob) is calculated
as shown in [Equation 4.14] and [Equation 4.1S] (Ulaby et al. 1981):
Ob= nr2 * §b
[Equation 4.14]
where;
4b - backscattering efficiency (normalized backscatter cross-section),
r = radius o f cloud particle,
[Equation 4.15]
4b= 4 * ( 2 n r / X ) 4 *((8c-l)/(Ee + 2))2
where;
X = sensor wavelength, and
6c = complex dielectric constant of water.
The remaining variable to be discussed is the volume extinction coefficient k«,
which is a component o f both the cloud model and the one-way loss factor (L):
Ke = N Qe
[Equation 4.16]
where;
Qe = extinction cross-section o f a water particle.
134
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The extinction cross-section (Qe) is the sum of the absorption (Qa) and scattering
cross-sections (Q,) o f the water particle (Ulaby et al. 1981):
Qe = Q, + Qj
[Equation 4.17]
Q. = (A.2 / 7t) * (2jtr / A.)3 * Im {-K}
[Equation 4.18]
where;
Im{-K} = imaginary portion o f the dielectric constant of water
Q, = (2X2 / 3tc) * (2n r / X)6 ♦ |K|2
[Equation 4.19]
where;
K = complex dielectric constant o f water
The term K is defined in terms o f the complex index o f refraction (n) o f the
particle to relative to the background medium (Ulaby et al. 1981):
K = (n2 - 1) / (n2 + 2) = (£c - 1) / (£c + 2)
[Equation 4.20]
The complex index o f refraction can be expressed as:
[Equation 4.21]
n = np/nb = (eq>/e Cb)l/2
where;
Dp = complex index o f refraction o f the particle sphere,
135
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
IU, = complex index o f refraction o f the background medium,
Ecp = complex dielectric constant o f the particle sphere, and
Scb= complex dielectric constant o f the background medium.
If the background medium is air, then nb = 1 and £eb = I (Ulaby et al. 1981, Weast 1984).
Therefore, n2 = £q, and K can be expressed in terms o f the complex dielectric constant o f
the particle (£«).
The solution for Qe is based on the Rayleigh approximations for the scattering (&)
and absorption (Ij.) efficiency factors which, in turn, are represented by the ratio o f Qa
and Qs to the physical cross-section (A) o f the scattering particle (A = n r2). The use o f
the Rayleigh approximations is preferred over the more computationally demanding Mie
scattering coefficients and yields acceptable accuracies if the particle size is much smaller
than the wavelength o f the incident electromagnetic wave (Ulaby et al. 1981).
4.2.2
Cloud Model Processing
Input variables required by the cloud model equations for use in the first-order
backscatter model fell into three categories (Table 4.04).
The first is fixed system
parameters related to the operating characteristics o f the SAR sensor.
These fixed
variables, wavelength and incidence angle, were set based upon the radar sensor used in
the analysis, in this case ERS-2, and may be the most critical for successful soil moisture
estimation. To a large degree, microwave wavelength and incidence angle determine
what influence vegetation has on attenuating the incoming radar energy.
136
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Category
System
Variables
Incidence angle
Wavelength
Water Cloud Radius o f water cloud particle
Real part o f dielectric constant
Imaginary part o f dielectric constant
No. o f scattering particles per unit volume
Vegetation
Canopy height
Symbol
e
X
r
K
Im(-K)
N
h
Table 4.04. Required variables for cloud model processing.
The second category o f input variables is focused on the properties o f the water
cloud used to simulate the vegetation canopy. The size o f the particles simulating the
green plant canopy was determined by comparing the output o f a spreadsheet model,
programmed with the [Equations 4.12-4.19], and selecting the radius measurement that
generated the best range o f vegetation backscatter (o°vci) values for the levels o f
aboveground biomass production expected within the study area. Dielectric constant
values representing live biomass and litter were taken from the literature (Saatchi et al
1994).
The final category is comprised o f the vegetation parameters canopy height (h)
and number o f scattering particles per unit o f canopy volume (N). Canopy height was set
at 46 cm for the entire study area based on the measured canopy values reported in
Saatchi et al. (1994). The number o f scattering particles per unit volume was derived
from estimates o f aboveground net primary production (ANPP) for the study area. The
basis for the calculation o f N, as well as the description o f the methods used to obtain the
biomass estimate, are outlined in the following section.
137
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A three-step procedure was used to estimate sod-contributed backscatter (o°soii).
All calculations were done in linear backscatter units and reported in decibels. First,
particle size was set based on the output o f the spreadsheet model and a curve o f expected
a0vc* values, given a range o f aboveground biomass production from 0-1000 g/m \ was
generated. Second, a third-order polynomial was then used to describe the line o f best fit
through the possible range of o°vcg values. Again, the amount o f aboveground production
is the key variable because it is used to determine the number o f scattering particles per
unit volume (N ) within the water cloud simulating the plant canopy. In turn, N is integral
to several o f the sub-calculations o f the cloud model, including the volume backscattering
coefficient (ov) and volume extinction coefficient (k«). Through k«, N also influences the
one-way (L) and two-way (L2) loss factors.
Finally, the last step in the calculation o f o°»ii used the previously calculated
polynomial describing o0,*, as a function o f biomass with estimates of aboveground net
primary production (ANPP) to determine o0* , across the study area for all image dates
(Figure 4.04). From these values, o0wa was determined using [Equation 4.10]. As was
the case with converting raw SAR DN values to radar backscatter coefficients, this
procedure was applied to the radar images using the graphic modeling module o f ERDAS
Imagine 8.3.1.
138
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Vagotatioa-caotributtd Backacattv
Valiant Extinction Coaffldaot
J* Order Polynom it
tec
Tto-wayLoaa Factor
Sdl-cantiibufttd Backacattv
Figure 4.04. Conceptual model o f the procedure used to calculate o0Wii.
4.3
Landscape Specification
4.3.1 Study Area Delineation
From the previous section, only one variable remains to be specified in order to
successfully derive soil-contributed backscatter (o°„u): the number o f scattering particles
per unit volume (N). Given the temporal dimension o f this study, however, using only a
single biomass estimate, such as the peak ANPP values which are routinely sampled at
139
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Konza, is not appropriate given the cycles of phenological development captured in the
time-series of radar imagery. Further, single aboveground biomass measurements fail to
account for spatial variations in ANPP that have been shown to exist within the study
area. Therefore, before an estimate o f soil moisture can be made, it is necessary to subset
Konza Prairie into areas o f
sim ilar
biomass characteristics in order to calculate more
realistic N values.
This delineation is advantageous because it (1) simplifies the construction of the
needed raster data sets, and (2) reduces the running tune required by the o°Wji model. By
generalizing the study area into regions o f similar production and canopy structure, the
size o f the required data files is reduced and the overall time required for model
computation is kept to a minimum.
Knapp et al. (1998) note that ANPP in tallgrass prairies can vary significantly
between years but, within any given year, ANPP is relatively consistent across areas of
similar topographic position, soil type, and foe frequency. The results from long-term
studies at Konza show that topographic position and the occurrence o f fire have
important effects on biomass production both individually and in an interactive manner as
realized in the availability o f resources such as water and nutrients (Knapp et al. 1998).
In general, findings show that ANPP is significantly greater (P < 0.05) at sites that are
burned annually as compared to areas with a low frequency o f fire (Figure 4.05). Here,
differences in fire frequency promote increased productivity by the dominant C* grasses
characteristic o f taUgrass prairie in annually burned sites, and increased forb production
in unburned areas.
140
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
500
Annual Fire
Frequency
Low Fim
Frequency
0
Total
Grass
Forbs
Figure 4.05. Effect o f fire on ANPP in ungrazed Konza Prairie watersheds (modified
from Knapp et al. 1998). Asterisks indicate significant differences at P < 0.05.
Also, as discussed in the previous chapter, grasslands, in general, and tallgrass
prairies, in particular, can accumulate large amounts o f detritus and standing dead plant
material (cumulatively termed here as "thatch" or "litter”). As reported in Knapp et al.
(1998), litter layers in tallgrass prairie can build-up over time to amounts approaching
1,000 g m'2 or a depth o f 30 cm in highly productive areas. It is the removal of this dead
material by fire that is the primary mechanism behind the increased aboveground
production in annually burned watersheds. This accumulation o f dead plant matter has
been shown to have many ecological effects at a variety o f environmental scales. From
the perspective o f radar remote sensing, the presence o f a dense litter layer can have a
significant impact on the magnitude o f the backscatter coefficient o°. This is especially
true if the litter layer is thick and/or holds large amounts o f water. Because burning
141
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
treatments at Konza are conducted at the watershed level, an appropriate means to
identify areas o f similar biomass production would be to differentiate between
watersheds based solely on fire frequency.
Up to this point, only the impact o f fire on ANPP has been considered important in
determining the biomass production. Other factors, such as topographic position, also
play a role in determining ANPP values. Figure 4.06 shows that a significant increase in
ANPP is found only in annually burned lowland watersheds. In addition to exhibiting a
distinct difference according to topographic position, annual ANPP is also much more
variable in the upland areas o f annually burned watersheds than in the associated lowland
sites, reflecting a gradient in resource availability within the same watershed (Figure
4.07).
No significant relationships between annual ANPP in watersheds with a low
frequency o f fire and topographic position were identified (Knapp et al. 1998). Mean
values o f ANPP calculated from samples collected over a four-year period at sites o f
intermediate elevation showed that mid-slope biomass production is more similar to that
o f lowlands than uplands (Knapp et al. 1998).
142
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
600
Annual Fire Frequency
Low Fire Frequency
Lowland Upland
^owland Upland
Figure 4.06. Effect of topographic position on ANPP in ungrazed Konza Prairie
watersheds (modified from Knapp et al. 1998). Different letters indicate significant
differences at P < 0.05.
1000
1000
B
400
&
3
soo
600
100
400
200
430
455
445
415
f
420
425
4)0
435
Elevation (m)
Figure 4.07. The relationship between ANPP and topographic position in burned (A) and
unburned (B) watersheds (modified from Knapp et al. 1998).
143
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Interannual differences in precipitation can cause differences from the general
trends o f biomass production outlined above, which are most applicable for years of
"normal" rainfall. Following a drought (e.g., 1989), ANPP can be significantly higher in
unburned prairie sites (Figure 4.08).
Also, during wet years (e.g., 1993), fire can
significantly increase aboveground production on the upland portions of annually burned
watersheds. However, despite rainfall anomalies during May and June of 1996, monthly
precipitation totals during the study period fall into the normal range (Figure 4.09), and
the general trends in ANPP should apply.
144
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1993
N
200
■
Grass
400
P-o
200
200' Forbs
100
100
130
200
230
300
Day of Year
• Annual fire frequency
o---o Low fire frequency
Figure 4.08. Aboveground primary production during a wet (1993) and dry (1989) year
for adjacent burned and unbumed sites (from Knapp et al. 1998).
145
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
200
180
160
140
120
9 100
80
60
40
20
0
JAM' FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Mm *
CZJMoolNy Total ♦ Lo^T— llw o
Figure 4.09. 1996 monthly and long-term mean rainfall totals measured at Konza Prairie
headquarters (data from Konza LTER dataset APT01).
Due to limitations associated with the spatial resolution o f the radar imagery used
in this study, making the distinction between biomass production in uplands and lowlands
is not practical.
Also, the number o f available "ground-truth" soil moisture
measurements prevents development o f a statistically sound quantitative method o f
relating soil-contributed backscatter (0°»ii) to soil moisture if the study area is delineated
by both burning regime and topographic position. This limitation, combined with the
consistent and significant difference between ANPP in burned and unbumed sites in
years with normal precipitation and the sensitivity o f microwave energy to the presence
146
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o f a litter layer, dictates that fire frequency must be the basis for differentiation o f study
area subsets.
The question then becomes one o f determining the appropriate fire frequency to
use in distinguishing between a "burned" and "unbumed" watershed given the prescribed
burning program implemented at Konza Prairie. Because of the typically high levels o f
ANPP associated with tallgrass prairies and because an equilibrium between the
accumulation and decomposition o f litter is reached in approximately three years (Knapp
et al. 1998), burned watersheds are identified as those burned after reaching peak
biomass levels from the previous year. These watersheds include those subjected to fell
burns in 199S as well as those burned on an annual basis in the spring o f 1996. All other
watersheds, those not burned in 1995 or 1996, are referred to as unburned (Figure 4.10).
The fire history o f Konza for the years 1995 and 1996 can be found in Appendix C. In
February o f 1996, wildfires accidentally burned 16 watersheds, as well as affecting
portions o f 3 other watersheds. For accounting purposes here, those watersheds only
partially affected by the February 1996 wildfires are classified as unburned.
147
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Untamed Watentaed
Buncd Watenhed
Figure 4.10. Burned and unburned watersheds o f Konza Prairie as determined for this
study.
4.3.2 Aboveground Biomass Estimation
As part o f the ongoing LTER activities at Konza, total aboveground biomass is
sampled in late August or early September, the time when peak biomass production is
generally achieved (Konza dataset PAB01). More importantly for this study, biomass in
watersheds 1A (burned) and 20A (unbumed) are sampled every two weeks from May IS
to September 15. In each watershed and for every sampling date, biomass is clipped at
148
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ground level in twenty 0.1-m2 quadrants and sorted into categories o f live grass, live
fbrbs, current year’s dead, and previous year’s dead. Sample data from watersheds 1A
and 20A are shown in Table 4.05.
DOY Watershed
206
219
233
247
261
1A
20A
1A
20A
1A
20A
1A
20A
1A
20A
&rmz
Mean Live Mean Dead*
77.90
394.90
596.80*
319.70
107.80
380.50
704.30*
385.40
132.20
326.40
336.10
752.90*
130.49
360.80
295.00
563.47*
139.03
307.40
279.20
547.19*
Table 4.05. Biweekly phytomass statistics for sample dates during the study period from
Konza dataset PAB01 (* current and previous year’s dead).
Because soil moisture estimates will be made using a time series o f radar imagery,
single estimates o f aboveground net primary production (ANPP), or estimates made on
days for which there is no radar imagery, are o f limited value as input into the cloud
model However, by using vegetation indices based upon other remotely sensed data it is
possible to characterize and evaluate vegetation conditions throughout the study period.
When combined with point measurements taken on the ground, an index such as the
normalized difference vegetation index (NDVI) can be very useful in inferring patterns
and processes across a given landscape.
149
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Essentially, NDVI is a ratio o f spectral bands that measures the "greenness" o f the
Earth's surface [Equation 4.22]. The NDVI equation produces values in the range o f -1.0
to 1.0, where increasing positive values represent increasing greenness and negative
values indicate nonvegetated surface features (e.g., water, barren lands, ice and snow, or
cloud cover). The specific bands comprising the near infrared and red wavelengths vary
by sensor. Green vegetation has low reflectance in the red electromagnetic spectrum due
to absorption by chlorophyll and other plant pigments in the chloroplasts o f leaf tissue.
However, due to scattering by the spongy mesophyll tissue within the leaf itsel£ healthy
vegetation is highly reflective in the near-infrared spectrum (Figure 4.11).
NDVI = (IR - Red) / (IR + Red)
[Equation 4.22]
IRmQQRRBB
GREEN
Chlaroflatft
Figure 4.11. Interaction o f electromagnetic energy with leaf tissue (modified from
Campbell 1987).
150
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Researchers have related NDVI on an empirical basis to many biophysical
measurements, including leaf area index, percent cover, and aboveground biomass
(Goward et al. 1985, Box et al. 1989, and Briggs and Nellis 1989). In this study, NDVI
information from two satellite sensors is used to indicate biomass production during the
study period. At Konza, NDVI estimates o f biomass production are aided by the simple
structure o f the plant canopy, which allows most aboveground biomass to contribute
directly to measured reflectance values (Briggs et al. 1998). In addition, the NDVI
values produced by tallgrass prairie vegetation respond very well to seasonal patterns
resulting from phonological development and climatic inputs, as well as the spatial
patterns caused by topography and fire (Briggs et al. 1998).
The first type o f remotely sensed data used in the biomass estimation procedure
was acquired by the Advanced Very High Resolution Radiometer (AVHRR), one of
several sensors onboard the improved TIROS series o f meteorological satellites operated
by the National Oceanographic and Atmospheric Administration (NOAA) (Campbell
1987). The AVHRR sensor collects reflectance data in five spectral bands at a spatial
resolution o f 1.1 km (Table 4.06). Beginning in 1990, the U.S. Geological Survey’s
EROS Data Center began acquiring daily AVHRR images to produce weekly and
biweekly maximum normalized difference vegetation index (NDVI) composites o f the
United States (Eidenshink 1992). Compositing periods are necessary to ensure cloudy
conditions do not obscure AVHRR observations o f the land surface. The result is a near
cloud-free image that depicts maximum vegetative greenness.
151
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The AVHRR-based NDVI is the difference of band 2 minus band 1 reflectance
values normalized over the sum o f bands I and 2. The biweekly composite products
distributed by the EROS Data Center are in the Lambert Azimuthual Equal Area
projection and contain NDVI values scaled to the range of 0 to 200, where computed -1.0
equals rescaled 0 ,0 equals 100, and 1.0 equals 200.
Band Number
I
2
3a
3b
4
5
Spectral Dimension (pm)
0.58 - 0.68
0.73 - 1.00
1.58 - 1.64
3.55 - 3.93
10.30-11.30
11.50-12.50
Region Name
Visible
Near infrared
Thermal infrared
Thermal infrared
Thermal infrared
Thermal infrared
Table 4.06. Spectral bands o f the AVHRR/3 sensor (from Kramer 1996).
Eleven total biweekly NDVI composites spanning the study period were used to
calculate mean NDVI values for the study area (Table 4.07). Mean NDVI values from
each composite period were assigned a mid-point date so that a line could be fit between
the data points permitting an estimate o f NDVI on days when biomass sampling
occurred.
The advantage o f AVHRR-based NDVI data is that it is a relatively low cost
source for temporal vegetation information. However, the coarse spatial resolution o f the
AVHRR sensor precludes estimating biomass at a level o f detail sufficient as input into
the radar cloud model. Therefore, a single image comprised o f finer resolution data from
152
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a second satellite sensor, the LANDSAT Thematic Mapper (TM), was used to extend
spatially the greenness information provided by the time-series o f AVHRR imagery.
Period Date o f Coverage (1996)
To
From
07-04
13
06-21
07-18
14
07-05
08-01
15
07-19
08-15
16
08-02
08-29
17
08-16
08-30
09-12
18
09-26
19
09-13
20
09-27
10-10
10-24
10-11
21
11-07
22
10-25
11-21
23
11-08
DOY
173-186
187-200
201-214
215-228
229-242
243-256
257-270
271-284
285-298
299-312
313-326
Mid-Point
DOY
179
193
207
221
235
249
263
277
291
305
319
Table 4.07. Biweekly AVHRR NDVI composite periods (1996) used in this study.
One TM image from July 22 (DOY 204) was acquired, rectified, and projected.
The LANDSAT TM sensor records reflectance data from seven spectral bands (Table
4.08) and TM-based NDVI is calculated as the difference of band 4 and band 3
reflectance values divided by then summed to tal
It is assumed that the TM-based NDVI values on DOY 204 are the same as on
DOY 206, the first day for which biweekly biomass sampling data are available. Using
estimated mean AVHRR NDVI on DOY 206 as the base, TM NDVI values were
adjusted linearly, based on the percent change in mean AVHRR NDVI values, so that
TM-based vegetation index information was generated for the remaining clipping dates.
Next, the mean TM NDVI for watersheds 1A and 20A, a burned and unbumed watershed
153
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
where biweekly biomass sampling was conducted, was determined and used to develop a
linear regression model relating the AVHRR-modified TM NDVI values to aboveground
net primary production (ANPP).
Band Number Spectral Dimension (pm)
I
0.45 - 0.52
2
0.52 - 0.60
3
0.63 - 0.69
4
0.76 - 0.90
5
1.55-1.75
10.4 - 12.5
6
7
2.08 - 2.35
Region Name
Blue-green
Green
Red
Near infrared
Mid infrared
Far-infrared
Mid-infrared
Table 4.08. Spectral bands o f the LANDSAT 5 TM sensor (from Campbell 1987).
Figure 4.12 illustrates how the delineation of the study area by burning treatment,
aboveground biomass estimates, and various components o f the radar cloud model
interact to generate soil backscatter (o0»ii) values.
Burned watersheds lack the
accumulated litter layer found in unburned watersheds and differ primarily in terms of
current year's aboveground production. The remotely-sensed ANPP estimates provide
the means to identify the spatial variation in aboveground biomass that, after cloud model
processing, impacts the magnitude o f the resulting o°»a value.
This relationship is
repeated within unbumed watersheds, with the added influence o f a litter layer. That
influence can be quantified and accounted for over the unburned watersheds within the
study area by describing the litter layer in a similar manner as the green plant canopy
154
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(e.g., height, biomass, dielectric constant) and processing this information using the
appropriate cloud model equations.
\
XS
1
V3
Bunted Waftnhedwtfh H i£ Abova^oundPhyo
z
C^92:92A
t f*:»
-• *t I :> *1
w?
i*
■t t - r s . .
i t j* .a » : • • » ! i»:.t s
ij
I
f, . .J t
I
.t
» t» •. t i- • •{ t»
. . .t
.i : . -i-i
• t- to
{►
.i !>■•»i B-* t »•■
I
£
Burned W atartw d wMi Low Aboveground Pkyto
3
l t» - K »•'* * t*-»:*l
t
•.**-;
t-t* ■: n » .* t s . t t t*.ii-»
;!•••» «♦,*•«j»
.| ...t l
•■•t-t
t- ti •
£
• »t t
I U.H-l i ’ •< t l>.*t'l't* *| l I* >1 t-t* • t
•
: t* • -i t*.>-t t*i»i»t t o . - i i * . t - : > - > t t - -t t
........... . .
- »t t t»:»t t
Unhened Wmnhed wih Ling L «y
Figure 4.12. Illustration o f the relationship between burned and unburned watersheds
with the components of the radar cloud model
4.4
Soil Moisture Estimation
4.4.1 Soil Moisture Sampling
Soil samples were collected at 11 points along two transects in both an annually
burned (ID ) and an unbumed watershed (20B) at Konza Prairie (Figures 4.13-4.15).
These transects also comprise the permanent time-domain reflectometry (TDR) network
155
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
used in the 1D/20B transect studies (Konza data set code PTN01) and span an uplandlowland-upland topographic gradient (Figure 4.16).
To minimize the introduction o f
error caused by small-scale soil variations, a total o f 12 soil samples were collected at
each sampling point along both transects. Samples were extracted from the top S cm o f
the soil profile with a total o f 66 individual samples being removed per transect per
collection period.
After extraction, the soil plugs were placed in sealed plastic bags and
stored in a chilled portable cooler.
20B
•
I
SmM * S m
I W ntnfc *4
100
Figure 4.13. Location o f transects and soil sampling points in watersheds ID (burned)
and 20B (unbumed).
1S6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.14. Looking from west to east along the soil moisture transect in watershed ID,
a burned watershed located within the Konza Prairie study site.
Figure 4.1S. Looking from west to east along the soil moisture transect in watershed
20B, an unbumed watershed located within the Konza Prairie study site.
157
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
440
435-
I 430'
425-
420-
415
0
50
100
200
ISO
Dt i PH
250
300
w
Figure 4.16. Elevation and distance between soil sampling points, from west to east,
along the transects in watershed ID (burned) and 20B (unburned).
Soil sampling was conducted during morning hours concurrent with satellite
overpass.
For ascending pass images, sampling took place the following morning,
approximately eight hours after image acquisition.
In the case o f descending pass
images, sampling occurred about two hours prior to the actual overpass time. After all
soil samples were collected for each overpass date, they were transported to laboratory
facilities where a series o f weight measurements were taken. The first measurement was
of the plastic bag and soil plugs. After weighing, the soil samples were removed from the
plastic bags, placed into paper containers, and oven-dried at a temperature o f 110°C for
158
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
three days. The empty plastic bags were then re-weighed. After oven-drying, the soil
samples were weighed once more in their paper containers, emptied, and the paper
containers re-weighed.
Mass wetness (w), or gravimetric water content on a percentage basis was then
calculated (Hillel 1982):
w = Mw/ M, = [(M, - Ms) / Ms] * 100
[Equation 4.22]
where;
Mw = mass o f water (g)
Mt = total mass (g) = wet sample - bag weight
Ms = mass o f sod solids (g) - dry sample • bag weight
Gravimetric soil moisture measurements (w) were then converted into volumetric
water content (0V). The use of 8Vrather than w is preferred for two reasons. First, and
most importantly, the linear relationship between 0Vand o° is the foundation o f radarbased soil moisture estimates. Second, volumetric moisture estimates are "more directly
adaptable to the computation of fluxes and water quantities" added to or removed from
the sod (Hillel 1982, p. 11).
Volume wetness is frequently calculated as a percentage value and, unlike mass
wetness, is based on the total volume o f the soil including the volume o f soil pores, which
takes into account the presence o f air within the soil media (Hillel 1982):
159
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0V= Vw/ V, = Vw/ (V, + Vf)
[Equation 4.23]
where;
Vw= volume of water (g/cm3)
Vt = total volume (g/citr)
V, = volume o f soil solids (g/cm3)
V f = volume o f soil pores (g/cm3)
Gravimetric water content measurements are easily transformed into volumetric
form by using estimates o f the bulk density o f soil (Hillel 1982):
0V= w pi, / pw
[Equation 4.24]
where;
Pb = dry bulk density
pw= density of water (= I g/cm3)
Approximate bulk density values for the transect sampling points were obtained from
data in the map unit interpretation database (MUIR) joined with a digital soils layer o f the
study area (Konza dataset KZGIS006). Soil type differences across the Riley/Geary
county line were reconciled with the aid o f an NRCS soil scientist (Hoffinan 1999). A
complete listing o f gravimetric measurements and volumetric soil moisture conversions
are listed in Appendix A.
4.4.2
Correlation and Regression Analysis
The association, or consistency in covariation, between both total backscatter
(o°tMi) and soil backscatter (o°»u) and estimated volumetric water content (0V) was
160
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
examined using the Pearson's product moment correlation coefficient (r). Correlations
between backscatter values and (0V) were compared according to watershed type and date
o f image acquisition in order to draw inferences about the performance of the cloud
model over burned and unbumed surfaces as the phenology o f the vegetation changes
over time.
Bivariate regression was used to establish the form o f the relationship between the
independent variable 6Vand the dependent variable o°soii:
a°soQ = (m* 0V) + b
[Equation 4.25]
The strength o f this regression relationship was evaluated using the coefficient o f
determination (r2), which measures the variation about the least-squares regression line
relative to the overall variation in the independent variable or the ratio of the explained
variation to the total variation in the dataset (Devore and Peck 1986, McGrew and
Monroe 1993).
Based on the backscatter-soil moisture relationships having the highest amount o f
explained variation, a spatially explicit estimation o f volumetric near surface soil
moisture is calculated by inverting [Equation 4.26] and solving for 6y.
4.5
Summary o f Data and Methods
The data and methods used in this research are geared towards producing an
accurate and spatially distributed estimate o f soil moisture using active microwave
161
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
imagery acquired by an orbiting SAR sensor. Prior to estimating soil moisture, four
major tasks must be accomplished. First, the effect o f topography on the radar imagery
will be quantified and reduced to minimizing the noise in the radar imagery caused by
variations in local terrain.
Topographic correction follows established techniques by
examining the correlation between image backscatter values and topographic variables,
then using the most closely related topographic variable to derive a polynomial correction
equation.
Second, a simulation model is developed to estimate the contribution o f
vegetation backscatter
(o °v e* )
to the total backscatter coefficient
( o 0^ ) .
The model is
derived from the radar cloud model introduced by Attema and Ulaby (1978), but is
applied here in a unique manner by using simulated o°veg to estimate the amount of
backscatter contributed directly by the soil surface (o°»u) to the total backscatter
coefficient.
To support the
modeling effort, a method was needed to estimate
aboveground net primary production for the entire study area over the 114 day study
period.
To accomplish this, a multi-sensor technique using normalized difference
vegetation index (NDVI) from the Advanced Very High Resolution Radiometer
(AVHRR) and LANDSAT Thematic Mapper (TM) sensors was used to generate a time
series o f vegetation biomass production data for input into the radar cloud model
Finally, the linear relationship between o°»a, as the dependent variable, and
volumetric soil moisture, as the independent variable, was identified using linear
regression. That regression relationship can be inverted and solved for volumetric soil
162
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
moistures to quantify near surface soil moisture conditions across the study area and over
time based on the magnitude o f o°»n values.
163
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5
RESULTS AND DISCUSSION
5.1
Soil Moisture Sampling
The soil samples that were collected concurrently with satellite overpass along
transects in watershed ID and 20B were oven-dried and weighed to determine
gravimetric water content. Mean mass wetness was determined for all 11 sampling sites
located along each transect. Given the spatial resolution o f the raster-based remotely
sensed data, and because of the resampling operations performed during image pre­
processing, some o f the sampling sites could not be independently resolved.
In
watershed ID, sample sites #6 and #8 (numbered from west to east) replicated the total
backscatter
(o°toui)
values for sites #5 and #7. Three sample points fell within the same
radar pixel as the preceding site in watershed 20B (#3, #6, and #11). Though soil
moisture values for each sample site in both transects are reported here, only the first o f
these points were used in later analyses while the "duplicate" samples were discarded.
The gravimetric soil moisture (g water/g dry soil) estimates were converted to
moisture on a percent volumetric basis [(cm3 water/cm3 dry soil)* 100] based on their
location and the corresponding bulk density values contained in the map unit
interpretation database (MUIR) o f the digital soils layer o f the study area. Bulk density
values in the MUIR database are representative values that are typically found in
polygons o f a given soil type and slope category, and often do not portray the desired
level o f local variation. For example, in watershed ID, the entire transect fell within a
164
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
single bulk density category (pb =1.35 g/cm3), despite the clear transition between upland
and lowland sites. Local variation was captured slightly better in watershed 20B where
soil polygons with two bulk density values were identified. Upland and lowland sample
sites in watershed 20B had bulk density values of pb = 1.35 g/cm3 and pb - 1.40 g/cm3,
respectively.
Mean volumetric water content by date for each watershed were found to be
significantly different (Table 5.01). Values ranged from a high of 47.1% on August 27 to
a low o f 18.9 % on September 5. The transect mean soil moisture levels were generally
higher in the unbumed versus the annually-burned watershed with two exceptions: July
23 and August 27.
Date
DOY
07-23-96
08-01-96
08-27-96
09-05-96
10-01-96
10-10-96
11-05-96
205
214
240
249
275
284
310
Waters led ID
w
(k/b )
9 v (% )
44.4
35.4
47.1
18.9
38.6
34.9
42.5
.33
.26
.35
.14
.29
.26
.32
Watershed 20B
w(K/g) 0 v ( % )
.30
41.7
.28
39.1
44.1
.32
.18
24.6
40.7
.30
40.4
.29
.34
46.1
Table 5.01. Mean gravimetric (g/g) and percent volumetric water content by radar image
date for watershed ID and 20B. Values are significantly different (P=0.35).
Measured soil moisture levels follow a trend related to total daily rainfall recorded
at the Konza headquarters building (Figure 5.01). Two o f the three highest volumetric
measurements for each watershed occurred on July 23 (DOY 205) and August 27 (DOY
165
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
240), the same two dates when soils were wetter on the annually burned watershed. Each
o f these days was preceded by rainM totals exceeding 37 mm for the previous 8 days.
The driest conditions were measured on September S (DOY 249) when only 0.6 mm o f
rain was recorded during the previous 12 days while the daily maximum temperature
averaged 30°C.
45-
•60
g40•50
1ij ”30-
L
r
■• 30 -I
-•20
3 t o-•
8
10
O
N
N
N
N
N
N
N
A
m
D«ytfY*«r
■ P n a p u i o e * W « m Im4 1 0 ( W t t n M l B
Figure 5.01. Recorded total daily precipitation and mean volumetric water content for
watersheds ID (burned) and 20B (unbumed). RainM data from Konza LTER dataset
AWE01.
An alternative method to evaluate the soil moisture conditions within the study
site is to compare measured volumetric values to estimated soil water content. Soil water
166
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
calculations were made by subtracting estimated actual evapotranspiration (AET) from an
initial soil water content of 300 mm o f water per 1 m of soil (i.e., 30% water by volume)
beginning on June 28 (DOY 180).
Actual evapotranspiration is the product o f potential evapotranspiration (PET), a
crop water-usage coefficient (Kc), and the soil moisture coefficient (Km) [Equation 5.01].
The Penman combination method was used to calculate PET while incorporating actual
water usage coefficients for tallgrass flora estimated for the previous year (Lamm et al.
1987, Lewis 1996). The input values required for the Penman combination calculations
included minimum and maximum temperature, rainfall, solar radiation, and wind data
(Konza dataset AWE01). Local wind speed data were not available, so a mean daily
value o f 130 miles day'1 (approximately 5 mph) was used. In addition, it was assumed
that no losses were caused by runoff or percolation and that water was not a limiting
factor (K*m= I).
AET = PET * Kc * K™
[Equation 5.01]
The curve generated by this estimation technique should be very similar to the
actual trend in soil moisture levels throughout the study period as it fluctuates in response
to changing weather conditions and vegetation phenological development. Comparing
mean volumetric water content to estimated soil water storage reinforces the previous
finding that measured soil moisture values are responding to local precipitation (Figure
5.02). Discrepancies between the absolute values o f measured and estimated soil water
167
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
content are enhanced by the assumption that water was not a factor limiting AET. This
assumption forces the minimum values o f estimated soil water content to drop well below
those that would be expected if Ksm were adjusted to reflect the relationship between
plant water usage and soil water content as soil moisture levels approach the approximate
permanent wilting point values for Konza soils. The purpose o f using estimated soil
water content here, however, is not to calculate specific soil moisture values but to
illustrate the general trend in soil moisture conditions during the study period. This trend
in estimated soil moisture conditions is more closely related to actual soil moisture
conditions than precipitation totals alone and provides a better comparison with measured
soil moisture values.
168
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
45
45
•40
40 •
I
•35
r
o
-3 0
30-
I -
25
20IS
-10
180 190 200 210 220 230 240 250 260 270 280 290 300 310 320 330
D a y rfY m
A V a ta W lD
■ W ataM SO B
CrtwMSadWitoCoMMr
Figure S.02. Mean volumetric water content for watersheds ID (burned) and 20B
(unbumed) as compared to estimated soil water content calculated using the Penman
combination method with a starting value o f 30% water by volume on June 28 (DOY
180).
Because the sampling transects in watersheds ID and 20B encompass an uplandlowland-upland topographic gradient, soil moisture values were expected to follow a lowhigh-low pattern, with higher values in the relatively low elevation locations. Figure S.03
shows measured volumetric water content for one day from each watershed. Although
only a single day is illustrated in this graphic, the remaining curves for the other six dates
trace a similar pattern, with the major difference being in the magnitude o f the moisture
values. The expected pattern is especially evident in the unburned watershed (20B)
169
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
where the accumulated litter layer characteristic o f these areas helps retain water and
retard evaporative loss from the soil surface. Variations in the expected trend can be
explained, in part, by the elevation differences of specific sampling locations (refer to
Figure 4.16).
50 ■
£
45 ■
40 -
20
■
t
2
3
4
6
5
7
8
9
10
11
T rw utct SaapkBf P w t
Figure 5.03. Measured volumetric water content on August 1 (DOY 214) for watersheds
ID (burned) and 20B (unburned).
5.2
Radar image Rectification and Calforatipp
The ascending and descending pass satellite image stacks (four dates each),
created using the first-order transformation to improve the alignment o f the repeat pass
170
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
images, were georectified separately.
Rectification, already a tedious and time-
consuming procedure, is further complicated when dealing with SAR imagery. Because
many o f the traditional ground control point locations (e.g., road intersections) are
difficult to locate on radar imagery, most o f the original SAR scene was needed to obtain
as many control point candidates as possible. During this study, the best control points
tended to be comer-reflecting and other metal objects such as buildings and flood control
spillway gates. Major road intersections along interstate highways and smooth reflective
surfaces such as airport runways were also acceptable.
The root mean square (RMS) error from the rectification procedure was 0.56 and
0.71 for the ascending and descending pass image stacks, respectively. Surprisingly,
GCP locations on the descending pass images were much more difficult to identify as
compared to the ascending pass images. This difficulty resulted in a considerably higher
RMS error, yet well within the acceptable range (< 1). Descending pass images are the
product o f slightly different image geometry.
Radar look direction is approximately
294°, compared to 76° for the ascending pass data. In addition, and probably more
importantly, there are minor changes in the incidence angle between the two pass types.
In ascending images, the Konza Prairie is essentially centered within the radar
scene, so the study area incidence angle is the same as the scene center incidence angle o f
23°. In the descending mode, however, Konza is in the extreme eastern portion o f the
scene. In fact, a small portion o f the southeast comer o f the study area is not contained in
the three descending pass images. As a result, the incidence angle in these images is
closer to the near range value o f the scene (21°).
However, the scene center local
171
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
incidence angle of 23° was used for both ascending and descending pass images in
calculations throughout the study.
After image calibration and conversion from raw digital numbers to decibels,
mean total backscatter (o0toui) values for the entire study area were found to range from a
high o f -7.533 dB on October 10 to a low o f -11.580 on September 5 (Table 5.02).
Single-factor analysis of variance (ANOVA) showed that a significant difference in o°toai
existed between the study area as a whole, burned watersheds, and unbumed watersheds
(F-value - 2.453, P = 0.110). In addition, a matched pairs t-test indicated that each
treatment was also different from the other two (P < 0.003).
Date
07-23-96
08-01-96
08-27-96
09-05-96
10-01-96
10-10-96
11-05-96
11-14-96
DOY
205
214
240
249
275
284
310
319
All
-8.30
-10.07
-10.49
-11.58
-10.26
-7.53
-9.84
-9.50
Mean
B
-7.52
-8.54
-9.36
-10.42
-9.22
-6.05
-8.41
-7.85
UB
-8.12
-9.70
-9.67
-10.69
-9.42
-7.38
-9.38
-9.22
Standard Deviation
All
B
UB
3.13
3.59
3.57
4.04
4.66
4.30
3.15
3.82
3.85
3.66
4.26
4.51
3.87
3.28
3.95
4.19
4.48
4.21
3.49
3.99
4.01
4.10
4.33
4.58
Table 5.02. Mean backscatter values and standard deviations by image date for the entire
study area (All) and burned (B) and unbumed (UB) watersheds.
172
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.3
Topographic Correction
Calibrated o°t0t*i values represent backscatter amounts from both the vegetation
canopy and soil surface, modified by the influence o f topography. A raw radar image
exhibits significant radiometric distortion caused, in part, by the variable topography of
the study area. To eliminate some of this unwanted "noise" from the desired signal,
topographic modulation o f the backscattered signal must be reduced. Five topographic
variables — local incidence angle (LIA), aspect relative to northing (AS), aspect relative
to radar look direction (AR), slope (SL), and elevation (EL) — were examined using
correlation analysis to see which had the strongest relationship with o0toai values.
Because o f the visual differences noticed first during the rectification procedure,
ascending and descending pass images were examined separately to determine whether
independent topographic correction equations were required to account for the different
viewing geometry. Two image dates for each pass were examined. In each case, days
with the lowest mean total backscatter values and drier soil moisture conditions were
selected because the influence o f vegetation on backscatter values would be at its
minimum on these dates. Dry soil moisture conditions lead to a lower plant canopy
moisture content that reduces the influence o f vegetation biomass on the radar return,
while permitting topography to exert greater relative control over the total backscattered
signal.
Table 5.03 shows the correlation coefficients for a single ascending and a single
descending pass image. Local incidence angle (LIA) was most strongly related to the
image a°t0««i values (r = -0.35 and *0.32 for ascending and descending pass images,
173
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
respectively). The negative correlation was expected and indicates that locations with
small local incidence angles, or terrain that is sloping towards the radar sensor, tend to
have a higher returns than those areas oriented in the opposite direction (Figure 5.04).
The correlation between o°ioai and the topographic variables examined are comparable to
those reported in studies conducted over much more rugged terrain (e.g., Bauer et al.
1991). The similarity between correction results shown here to those reported elsewhere
in the literature suggest that polynomial correction techniques, in concert with relatively
coarse resolution digital elevation data, reaches a maximum in its effectiveness over even
moderate relief. Further improvements in the topographic correction o f radar imagery
based upon measured topographic variables such as local incidence angle will depend
greatly on the availability o f fine resolution digital terrain data.
Topographic
Variable
LIA
SL
AS
AR
EL
Asc ending Pass
C18-27-915
All
B
UB
-0.35 -0.34 -0.35
-0.12 -0.13 -0.13
-0.24 -0.23 -0.25
0.24 0.23 0.25
-0.09 •0.04 -0.14
Descsending Pass
CI9-05-9)5
Ail
UB
B
-0.32 -0.37 -0.25
-0.03 -0.03 -0.04
-0.22 -0.26 -0.18
0.25 0.30 0.19
-0.21 -0.21 -0.21
Table 5.03. Correlation coefficients (r) for topographic variables and representative
image pass dates. LIA = local incidence angle, SL - slope, AS - aspect relative to
northing, AR= aspect relative to radar look direction, EL - elevation.
174
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
63365122
Local Incidence Angle (Degreei)
Figure 5.04. Scatterplot showing the negative correlation between radar backscatter and
local incidence angle before topographic correction.
In a comparison o f the five topographic variables, correlation coefficients
between the two image passes were very similar, with the exception o f elevation. In this
case, the descending pass image had a much stronger relationship with the elevation
variable than the ascending pass image.
The differences in correlation between the
ascending and descending pass images cannot be explained by topography as the
characteristics o f both the burned and unburned watersheds are similar to those o f the
study area as a whole (Table 5.04). Interestingly, the changes in correlation are only
noticeable for those topographic variables sensitive to changes in sensor incidence angle
(i.e., local incidence angle, slope, and elevation).
175
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Statistic
LIA
SL
AS
AR
EL
Watersheds
All
B
UB
All
B
UB
All
B
UB
All
B
UB
All
B
UB
Mean
23.6
23.3
23.8
5.6
5.8
5.3
186.7
179.8
190.6
85.2
91.0
81.4
393.2
393.8
393.2
Median
25
24
25
5
5
5
204
184
212
80
88
76
398
397
400
SD
6.5
6.8
6.1
3.3
3.4
3.2
106.9
105.6
107.2
55.1
57.4
53.1
28.7
28.2
29.5
Min
3
4
3
0
0
0
0
0
0
0
0
0
318
328
318
Max
42
42
42
20
19
20
361
361
361
180
180
180
445
445
444
Table 5.04. Summary statistics for topographic variables in the study area.
In the descending pass images, the incidence angle o f approximately 21° is closer
to a nadir view o f the study area than the 23° incidence angle assumed for images
acquired by the ERS sensors. The difference in incidence angle may be sufficient to
slightly reduce the influence o f vegetation that can attenuate or "diffuse" the radar
energy, magnifying the effect o f topography on recorded backscatter values. As a result,
the limited influence o f slope on backscatter values seen with ascending pass images is
even further reduced in the descending mode.
A slight reduction in the vegetation
attenuation means that radar energy is better able to penetrate the plant canopy. Because
o f this, elevation, and the differences in soil moisture associated with elevation changes,
exerts more influence on the magnitude o f the backscatter response in the descending
mode given the location o f the study area within these scenes.
176
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Because local incidence angle was the topographic variable most closely related
to o°totai. it was used to generate the polynomial correction equation. The correction
reference value f(r) o f *10.072 was found by inserting the reference local incidence angle
value o f 23° into the polynomial model function f(Q [Equation 5.02].
Next, the
correction function K(i) is calculated by dividing the model function f(Q by f(r), the
correction reference value.
f(i) = -0.0070(LIA)2 + 0.1433(LIA)-9.6650
[Equation 5.02]
The post-correction scatterplot o f backscatter and local incidence angle shows a
large reduction in the strength o f their association (Figure 5.05). There still exists a weak
negative relationship between LIA and o°dB magnitude. A visual comparison o f one SAR
image before and after topographic correction shows significant changes in the
backscatter response (Figure 5.06). The change is most pronounced in lowland areas
where backscatter values have increased.
177
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
i I
A
i
A
“■
-30 -
0
5
10
15
30
35
30
35
40
45
Local Incidence Angle (Degree*)
Figure S.OS. Post-correction scatterplot o f radar backscatter and local incidence angle.
Figure 5.06. SAR image (July 23,1996) before and after topographic correction.
178
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The exact performance o f the correction procedure was evaluated by measuring
the variance o f backscatter value before and after application of the polynomial model
(Table 5.05). The observed reduction o f image variance in the range o f 8.4% to 16.1% is
very good for this type o f correction model and compares favorably with others found in
the literature (e.g., Hinse et al. 1988, Bayer et al. 1991).
After topographic correction,
all dates showed a decline in image variance and reductions in mean o°ioai values.
Examination o f the scatterplot indicates that the polynomial function appeared to be more
successful in reducing o0(0«ai at the lower local incidence angles than increasing those at
higher angles. The increased slant range distance through the vegetation at higher local
incidence could be responsible for attenuating the radar signal to such a degree that a
correction, based on topographic variables alone, cannot address.
Date
07-23-96
08-01-96
08-27-96
09-05-96
10-01-96
10-10-96
11-05-96
11-14-96
Variance
After
Before
Correction
Correction
Difference (%)
Mean Variance Mean Variance
-14.17
•8.64
11.43
-8.30
9.81
-8.68
17.78
-9.68
16.28
-10.07
-16.10
-10.09
9.95
-10.49
11.86
15.20
13.36
-I2.lt
-11.15
-11.58
10.78
-15.32
-10.26
12.73
-9.86
-11.07
19.70
-7.18
17.52
-7.53
12.19
-13.30
-9.47
-9.84
14.06
-11.47
16.82
-9.50
19.00
-9.10
Table 5.05. Mean values and variances o f o°ioai before and after topographic correction.
179
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.4
Cloud Model Parameterization
Six input variables are required to process the cloud model estimate o f vegetation-
derived backscatter (o°veg). O f these six inputs, three are either fixed sensor parameters
or static estimates o f the canopy and litter layer height. Three variables remain to be
specified: the dielectric constant o f the green plant canopy and litter layer, size o f the
particle comprising the water cloud, and an estimate o f biomass that can be transformed
into a value representing the density o f particles within the water cloud (N).
The dielectric constant o f a green plant canopy at C-band wavelengths o f 15 +
i5.5 was taken from Saatchi et al. (1994) who calculated the value using the Ulaby-El
Rayes procedure and plant moisture and biomass measurements taken during FIFE. The
dielectric constant at C-band for the particles o f the thatch layer in unbumed watersheds *
(74.0 + i26.2) was also determined by Saatchi et al. (1994), this time using the Debye
formula for pure water.
The radius o f the water particle was selected by experimenting with droplet sizes
o f varying dimensions.
The modeled contribution to o°(oui by the vegetation canopy
(o°vct) when using particles with a radius o f 0.75 cm, 1.0 cm, and 1.25 cm in shown in
Figure 5.07. Particle size impacts simulated o°vcc values in several ways. First, larger
particle diameters results in both more backscatter from the vegetation canopy, at a given
biomass amount, and a greater rate o f change in the magnitude o f the backscatter
response as the scattering particle density, or biomass, increases.
180
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
-
10'
f
-15-
-25-
-30
0
100
200
300
400
500
600
700
800
900 1000
AlVPd/m1)
0 .75 cm a i m ------ 1 0 c m a i m -------1.25 cm a i m
Figure 5.07. Simulated vegetation backscatter (o°veg) For cloud particles with radii o f
0.75 cm, 1.0 cm, and 1.25 cm
Particle size is closely related to the amount o f radar energy lost through
extinction within the scattering volume or from scattering away from the sensor. The
amount o f energy lost is accounted for by the two-way loss factor (L2), which represents
the degradation o f radar energy as it enters and exits the plant canopy. The two-way loss
factor is used in both the o°vcg and soil-contributed backscatter (c°so ii) equations.
Vegetation backscatter and loss factor values also reflect the influence o f
increasing biomass through its impact on N, the number o f scattering particles per unit
volume. Simulated o°veg increases rapidly over a range of biomass values before reaching
181
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a plateau where increasing production has little effect. The rate o f change in simulated
o°veg curve, and the production level point where it reaches this plateau, is the result o f
the combined influence o f particle size and ANPP.
Meanwhile, the loss factor
experiences exponential growth as biomass production increases (Figure S.08).
180
160
1
140
I
120
li 100
I
J
6*
80
60
! »
0
100
200
300
400
500
600
700
800
900
1000
A N IF fcte1)
0.?3camifai
lC c m m ta -----
Figure S.08. Simulated values of the two-way loss factor (L2) for cloud particles with
radii o f 0.75 cm, 1.0 cm, and 1.2S cm.
O f additional interest is the relative contribution of o°veg and soQ backscatter
(a°soii) on the total backscatter coefficient. In other words, since the objective o f this
modeling process is to quantify soil backscatter
(o°soii),
it is important to understand the
182
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
amount o f "noise" in the desired backscatter signal introduced by the vegetation canopy.
Figures S.09-S.11 again show the simulated a°VCg curves generated for particle sizes o f
0.7S cm, 1.0 cm, and 1.2S cm. Assuming a o°t0ui value o f -9.0 dB, the remaining curve
represents the value o f a0toiai - o°veg. This curve, not to be confused with o°soii because the
effect o f the two-way loss factor is ignored, represents pure soil backscatter or the
amount o f backscatter that is not affected by scattering or extinction within the canopy
volume. Where these two curves intersect is the point where o°mtai is composed of an
even mix of vegetation and raw soil backscatter. When a0VCg exceeds o°t0ai - o0veg,
180
-5
■ 160
0
•30
0
100
200
300
400
500
600
700
800
900
1000
ANW fcfrtf)
Totol-VnrtrtiMi BrHriWw----- Two-War Lom Factor
Figure 5.09. Cloud model simulation results showing the impact o f vegetation on the
quantity o f o°toui - a0veg for a particle radius o f 0.75 cm (assumes a0toc1i = -9.0 dB).
183
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
180
•10
•140
I
•120
-15
'■
•
100
•60
•40
IVio-Way Lon Factor (UnM rlfcte)
•160
•20
-30
0
100
200
300
400
500
600
700
800
900
1000
ANFP(g/m2)
T«etalmiBacfaciMti —
Total-VifrtaiioaBaekicattai------ Two-Way low Tactor
Figure 5.10. Cloud model simulation results showing the impact of vegetation on the
quantity o f a°IoUi - o0veg for a particle radius o f 1.0 cm (assumes o°taai = *9.0 dB).
184
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
180
-
10-
•140
I
-120
-15•100
•80
1
I
]
r
•40
-25-
I
•20
•30
0
100
200
300
400
500
600
700
800
900
1000
ANFP(gfta2)
— A p U ta B a c k x iM ii
ToUl- ^ p U ta B x fa n ttH -----Two>W«yL o m Factor
Figure 5.11. Cloud model simulation results showing the impact o f vegetation on the
quantity of a°totii - o°VCg for a particle radius o f 1.25 cm (assumes a°toiai = -9.0 dB).
At particle sizes o f 0.75 cm and 1.0 cm (Figures 5.9 and 5.10), the relative
contribution of o°veg to o°,oui never exceeds 60%, even at production levels as high as
1000 g/m2. However, when the radius o f the water particle is increased to 1.25 cm
(Figure 5.11), o0* , is responsible for more than 90% o f the total backscatter coefficient at
a relatively modest production level o f 440 g/m2, and over 99% at 790 g/m2 (Figure
5.12).
O f the three particle sizes examined, a radius o f 1.0 cm was selected because
simulated o0^ values were best suited to the biomass range typical of tallgrass prairie,
185
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and resulted in the most reasonable estimated contribution o f a0** to the total backscatter
coefficient. The 1.0 cm o°vcg curve displayed characteristics that were good compromises
between the smallest and largest particle sizes examined. The potential magnitude of
o°veg, a slower rate o f change over a greater range o f biomass values, and the expected
dominance of o0^ over o°totai - o°veg at a higher production levels made the 1.0 cm
particle size the optimum choice.
Figure S.12 shows the simulated o°veg and o°Mii curves for a water cloud with a
particle radius of 1.0 cm. Again, o°wa is the product o f the quantity (o0t0iai - o°veg) and the
two-way loss factor (L2). Note that over bare ground both (o0,w»i - o°veg) and o°sou are the
same and identical to the assumed o°toui value o f -9.0 dB.
However, as biomass
production increases, (o 0totai - o°v«g) decreases as o °Veg begins to comprise a larger
percentage o f the total backscatter coefficient.
Meanwhile, estimated o°»a is at its
minimum value over bare soil and increases rapidly owing to the exponential increase in
the two-way loss factor.
186
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
180
■ 160
140
I
■120
- •
-
10-
100
r
i
■80 i
•60
■40
-25-
i
!
-20
•30
0
100
200
300 400
300 600
700
800
900 1000
ANFP(|te2)
V n rtrtiw BichniMw
SoJ Bffcif t t T
Two-Way Lon Factor
Figure 5.12. Influence o f vegetation backscatter (a0vcg) and two-way loss factor (L2) on
estimated soil backscatter ( ct0 soh) as a function o f aboveground primary production for a
particle size o f 1.0 cm (assumes o°totai = -9.0 dB).
To simplify the arithmetic o f subtracting a°veg from o0t0tai within the ERDAS
Imagine spatial modeling interface, the backscatter function created by the 1.0 cm radius
water particles were transformed back into their original linear form. The o°veg-ANPP
function could then be described by a third-order polynomial which, in turn, was used to
generate the aQVCg estimates for each image date (Figure 5.13).
187
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.08
0.07
f 0.06
.1
<± 0.05
k
| 0.04
« 0.03
k
y - -0.1636*!fl(»b<x))+OJ]370*tW»b<Jt))) -0.0024^1n(tb<x))) -H].2320
2 0.02
0.01
0.00
0
100 200 300 400 500 600 700 800 900 1000 1100 1200
AflPP (grin2)
Figure 5.13. Simulated backscatter response for vegetation canopy simulated by a water
cloud with a particle radius o f 1.0 cm.
For the same reasons that a radius o f 1.0 cm was used to simulate the green plant
canopy, a water particle radius o f 0.75 cm was selected for the litter layer in the unburned
watersheds. The litter layer considered here does not include the standing dead biomass
generated annually, but includes only the thick mat o f dead vegetation found near the soil
surface. Long-term research conducted at Konza has shown that the amount o f dead
biomass contained in the litter layer is approximately 382 g/m2 ± 19 (Knapp et al. 1998).
Since 1990, however, biomass production in both upland and lowland areas has generally
exceeded the long-term mean values (Figure 5.14).
For this reason, the dead plant
material in the litter layer in this study was set at 450 g/m2. This total results in the litter
188
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
layer contributing a constant -18.7 dB to the total backscattered signal, with a two-way
loss value o f 1.9 linear units (m m '1) (Figure 5.15).
Figure 5.14. Long-term record o f ANPP in upland and lowland sites according to fire
frequency at KPBS (from Knapp et al. 1998).
189
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
-10
■
•4
-18.7 dB
-
■3
20-
1.9
i
I
5
I
-35-40
0
100
200
300
400
500
600
700
800
900 1000
A M ? (fe ta l)
- Uttar Bwfautlw------Two-Wqr
Factor
Figure S. 15. Simulated values o f vegetation backscatter (a°veg) and the two-way loss
factor (L2) for the litter layer based on a particle size o f 0.75 cm.
The cloud model for unbumed watersheds differs from that of the burned
watersheds because it consists o f two layers rather than one comprised of the plant
canopy alone. In the two-layer model, the contributions o f the plant canopy and litter
layer are assumed to be additive and [Equation 4.10] must be modified to account for the
backscatter contributions and loss factors o f the two types o f scattering media [Equation
5.03].
190
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
O°soii = [(Li2 + L22) * G°(otai] - (o°veg * Lt2) - (o0hner * L22)
[Equation 5.03]
where:
Li = two-way loss factor o f the vegetation canopy
L2 = two-way loss factor o f the litter layer
<T0iiner- backscatter contribution o f the litter layer
5.5
Vegetation Biomass Estimation
The 1996 AVHRR NDVI product produced by the EROS Data Center arrived in a
geographically rectified format but were converted to UTM from the original Lambert
Azimuthal Equal Area projection. Eleven total biweekly NDVI composites spanning the
study period were used to generate mean NDVI values for the study area (Table 5.06).
Mean study area NDVI values were plotted against day of acquisition, which was
determined by identifying the mid-point within the 14-day composite period. From these
data, a second-order polynomial was used to generate a line o f best fit between those
dates (Figure 5.16). This allowed mean study area NDVI to be estimated for any day
within the composite period, and specifically on those days when biomass sampling was
conducted. With the July 24 (midpoint DOY 206) as the base image, the percent change
in mean AVHRR NDVI was calculated (Table 5.07)
191
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Period
13
14
15
16
17
18
19
20
21
22
23
Date of Coverage (1996)
From
To
06-21
07-04
07-05
07-18
07-19
08-01
08-02
08-15
08-16
08-29
08-30
09-12
09-13
09-26
09-27
10-10
10-11
10-24
10-25
11-07
11-08
11-21
DOY
Max
NDVI
0.790
0.785
0.790
0.755
0.760
0.750
0.745
0.710
0.675
0.645
0.630
Mean
NDVI
0.760
0.753
0.774
0.734
0.732
0.726
0.714
0.696
0.657
0.633
0.619
173-186
187-200
201-214
215-228
229-242
243-256
257-270
271-284
285-298
299-312
313-326
Min
NDVI
0.690
0.710
0.755
0.700
0.705
0.705
0.680
0.670
0.640
0.625
0.605
Table 5.06. Biweekly AVHRR NDVI composite periods (1996) used in this study.
0.80
y - 0*192 ♦ O.OOlli. Q.OOOOMs‘
R2 - 0 5 7
0.75 ■
i
1
3
0.70 ■
0.65 -
<
•
■
Mm
tta M
■••NDVI ft mb AVHRR
a s m NDVI (AVHRR) t e dippi
0.55
ISO
200
250
300
350
DarofYmt
Figure 5.16. Mean and estimated mean AVHRR-derived NDVI values for KPBS (1996).
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
DOY
206
219
233
247
261
AVHRR NDVI
0.75
0.75
0.74
0.72
0.71
% Change
N/A
-1.06
-2.39
-4.11
-6.10
Table 5.07. Biweekly AVHRR NDVI composite periods for 1996 used in this study.
The entire Konza study area is comprised o f fewer than 90 AVHRR pixels,
illustrating the coarse spatial resolution o f the AVHRR sensor (Figure 5.17). However,
using the temporal NDVI information provided by the series o f composite AVHRR
images a guide, a single LANDSAT TM image from July 22 (DOY 204) was used to
increase the spatial resolution o f the NDVI data. The TM image was rectified, the UTM
projection applied, and the NDVI calculation performed with ERDAS Imagine. The root
mean square (RMS) error from the rectification procedure was 0.56 and the mean NDVI
value for the study area was 0.51 ± 0.08 (Figure 5.18). It is assumed here that the TMbased NDVI values on DOY 204 are the same as on DOY 206, the first day o f post-peak
biomass clipping, and no adjustments were made to these values based on the percent
change in AVHRR NDVI values.
193
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 5.17. AVHRR NDVI image from composite period 13 showing its inherent
limitations with respect to spatial resolution.
194
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
NDVI 07-22-96)
■
0.00-0.19
■
0.20-0.29
■ I
■
0.40-0.49
0 .5 0 -0 59
■
0.60-0.72
0.30-0.39
Figure S. 18. TM false-color composite and NDVI image. Note the presence o f the cloud
in the false color composite and differences in reflectance between burned and unburned
watersheds.
On the remaining clipping days, each pixel value o f the TM NDVI product was
adjusted by the same percentage change calculated from the AVHRR NDVI values for
the same period. Next, the mean TM NDVI for watersheds IA and 20A, a burned and
unburned watershed where biweekly biomass sampling was performed, was calculated.
Comparing these values to clipped estimates of green aboveground biomass allowed for
the development o f a linear regression model relating the AVHRR-modified TM NDVI
values to aboveground net primary production (ANPP) (Figure S. 19). Summary statistics
195
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
for biomass estimates on days for which radar images were acquired are shown in Table
5.08.
420
4 00 •
380 ^
1
360-
•
■
W tttrA rtlA
Whtatri»420A
r-2X3.Ms.M3.36
Ra -0.70
,
340 320 300 •
y - 2 l6 2 * 7 z - <33.24
Ra - 0 3 l
280 260
0.51
0.52
0.54
0.53
0.55
0.56
IM NDVI
Figure 5.19. Relationship between ANPP and AVHRR-modified TM NDVI values for
watersheds 1A (burned) and 20A (unburned).
Date
07-23-96
08-01-96
08-27-96
09-05-96
10-01-96
10-10-96
11-05-96
DOY
205
214
240
249
275
284
310
s
113.71
112.94
109-91
108.57
103.95
102.07
95.85
Mean
356.77
348.32
317.32
303.45
255.40
235.86
171.11
Max
755.42
744.73
702.38
683.72
619.13
592.85
505.80
Min
169.11
162.50
136.31
124.77
84.82
68.57
14.73
Table 5.08. Summary statistics for estimated post-peak biomass production on radar
image dates.
196
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Estimates o f ANPP using NDVI data are most frequently performed during the
period of green-up to peak biomass production. At Konza, the relationship between
ANPP and NDVI during this stage of phenological development is generally strong (R2 >
0.80), but the slopes and y-intercepts o f the regressions can vary significantly between
years and watersheds. Post-peak NDVI statistics are less frequently reported since the
objective o f much NDVI-related research is as an estimator o f photosynthetically active
aboveground vegetative biomass. In this study, the post-peak biomass relationship with
NDVI is a critical component o f the success or failure o f the radar-based soil moisture
estimation. The single most important input variable o f the cloud model is biomass,
which serves as the basis from which the density of particles in the water cloud is
determined.
Separate regression models typically are required for burned and unburned
watersheds because o f the impact of the accumulated litter layer on reflectance values in
areas not subjected to fire. The importance o f separate regressions models based on fire
regime may be even more important as the vegetation begins to brown down. The
steadily increasing amount o f dead and standing dead plant material during senescence
likely influences reflectance to a greater extent than when the canopy is developing.
As could be expected, the explanatory power o f the post-peak linear regression
model is much greater for the annually burned watershed (R2 - 0.70) than the unbumed
watershed (R2 = 0.51).
Though neither relationship is especially strong, the linear
functions identified for watersheds 1A and 20A were used to estimate biomass
throughout the study area.
Optimally, individual LANDSAT TM images (or other
197
R e p ro du ced with permission o f the copyright owner. Further reproduction prohibited without permission.
hyperspatial resolution imagery) acquired near the clipping dates would be used to
develop the regression equations. But, given funding limitations for the purchase o f
additional satellite imagery, the kind o f multi-sensor approach used to estimate biomass
described here is an effective, and less expensive, alternative.
5.6
Radar Backscatter-Soil Moisture Relationships
S. 6.I
Total Backscatter and Soil Moisture
As was done with the volumetric soil moisture measurements, total backscatter
values (corrected for topography) were visually compared with precipitation totals to
determine whether o°t0ai is influenced by expected soil moisture conditions (Figure 5.20).
Mean o0toai for the entire study area, for burned watersheds, and unbumed watersheds
appeared to be closely related to rainfall events.
198
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
60
2M
au
t
•5 0
31)
-10
214
310
m
240 I
•4 0
-11
249
•3 0
•20
■Pnc^iUtiga* AflW ilmhiii A M W i t a U
■
W O ah it
Figure 5.20. Mean total corrected backscatter versus precipitation.
A better comparison can be performed using the estimated soil water content
curve because factors other than precipitation, such as temperature and solar radiation,
are included in the estimate (Figure 5.21). In this way, evapotranspirative demands can
be considered when evaluating the relationship between backscatter, precipitation, and
soil moisture. During the warmer summer and early foil months, mean (Ami follows very
closely the trend established by estimated soil water content. For example, on July 23
(DOY 205) mean o°to«i is very high, owing to the significant rainfall the previous day.
From this date until September 5 (DOY 249), both the soil water curve and mean tA m
199
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
values steadily decline, with the exception of August 27 (DOY 240) when o°ioai increases
because of three significant rainfall events that preceded image acquisition.
■ 30
f
-10
..to
■
•5
-13
180 190 300 310 330 330 340 3S0 300 370 380 390 300 310 330 330
D*r«rYmr
•
A lW tfM M i
* a— AWil iihifc
a
T M V M M
KibHMSoAWiMc
Figure S.21. Comparison between mean total backscatter (o°totti) and estimated soil
moisture content.
Beginning on October 1 (DOY 27S), the backscatter begins to increase as soil
moisture levels rise in response to reduced evaporative demand and additional rainfall.
Overall, total mean
during the cooler fall months is much higher than would be
expected given the estimated soil water content. Again, estimated soil water values are
affected by the assumption that Km - 1. However, the increase in a°tmi may be caused
by a drop in canopy moisture content as vegetation begins to senesce. Lower canopy
200
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
moisture reduces the attenuating effects o f volume scattering within the green plant
canopy, leading to higher backscatter as the incoming radar energy penetrates to and
reflects from the soil surface in a more direct manner.
Only two o f the eight total images used in this study, October 10 (DOY 284) and
November 14 (DOY 319), do not follow the trend suggested by the soil water storage
curve. The October 10 and November 14 images were acquired in the descending mode,
meaning that the image was recorded at approximately 1212 hours LST. Therefore, even
if the dew point temperature was reached that morning, it is unlikely that free water on
the vegetation and soil surface would still be present at the time o f image acquisition and,
therefore, would not be expected to play a role in elevating backscatter values. The best
explanation for the higher than expected backscatter values relates to the combined effect
o f large amounts o f rainfall and canopy senescence. Total precipitation in the three days
prior to October 10 measured in excess o f 30 mm.
Frequent and abundant rainfall,
combined with a dying or dead plant canopy results in larger mean a°IOtti, especially in the
burned watersheds where the soil surface is relatively exposed. Contributing to this could
be lower amounts o f organic matter in the near surface soil profile, leading to lower
infiltration rates and, perhaps, an increase in the surface ponding o f rainfall. Unbumed
watersheds have greater mean oVm as well, but the presence o f the litter layer attenuates
incoming microwave energy to a greater extent.
On October 10, in the unbumed
watersheds, the litter layer has had time to dry. If rain had occurred the day before, or
during, image acquisition, mean o°to«i for the burned and unbumed watersheds would be
much more similar, as it was on July 23 (DOY 205). Litter alone does not make for a
201
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
highly reflective surface, but when that layer of dead biomass is saturated with free water,
it can dominate the radar return.
On November 14 (DOY 319), the estimated soil water content curve indicates
drier conditions than those on November S (DOY 310). Instead, there is almost a linear
increase in all three mean o°u>t*i values. Here, rainfall is not an issue because only 4 mm
of rainfall was recorded in the preceding two-week period. Instead, cold temperatures
could be responsible for the increase (Figure S.22).
The minimum temperature was
below freezing on the morning the November 14 image was collected, as had been the
daily low temperature for the previous 6 days. If some or all of the water present in the
near surface soils or litter layer changed phase before acquisition, backscatter would
increase owing to the corresponding decrease in its dielectric constant. Higher values for
the real and imaginary portions of the dielectric constant translate into an increase in the
magnitude o f the backscattered energy. The increase in mean o0^ between November S
and November 14, despite the slight decrease in the estimated soil water storage value,
suggest at least partial freezing had taken place. Though the image from November 14
was used obtained for this study, the use o f any backscatter values from this data in an
inversion model is unwise given the potential influence o f freezing temperatures.
202
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
35 ■
33
>30
43 •
I "
”30
■
•10
i n 190 300 310 330 2 S 340 330 300 310 3 » 390 300 310 330 330
DqrtfYw
— Pm » *rton ■ Twwpwww
Figure S.22. Precipitation and minimum temperatures during the 1996 study period.
The role o f senescing vegetation on the ability of radar energy to better penetrate
the canopy has been discussed, but no measurements o f vegetation moisture status were
taken or used in the modeling phase. In this study, it is assumed that canopy moisture
remains constant throughout the study period.
This assumption has the potential of
significantly reducing the effectiveness o f the cloud model in estimating o°ve| and,
subsequently, the soil contribution to the total backscatter coefficient. Accordingly, a
strong linear relationship between o°M1| and volumetric soil moisture, the key to
successful soil moisture estimation, may be difficult to obtain.
The regression between o°umt and volumetric soil moisture shows the appropriate
trend, but the correlation between the two variables is weak
(Figure 5.23). Total
203
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
backscatter values still contain the influence o f the vegetation and litter contributions
made over the different watershed types. Separating a0totti values by burning treatment
results in distinct regression lines with different slope and y-intercept values (Figure
5.24). Not surprisingly, the correlation coefficient for the annually burned watershed (r =
0.25) is better compared to that calculated for composite o°toai (r = 0.21). Backscatter
from burned watersheds is more responsive to soil moisture content than in unburned
areas because no attenuating litter layer is present. When calculated independently, the
correlation between o°toai and soil moisture in unburned watersheds is lower due to the
attenuating effect o f the accumulated litter layer near the soil surface (r = 0.15).
•3.0
■4.0-
-o -*o-
10.0
-
-12.0• ♦
H -14.0 ■
•11.0
0
10
30
30
V ilM itrir Wattr CmM
40
SO
(H>
Figure 5.23. Scatterplot o f total backscatter (o°ux*i) versus volumetric water content.
204
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
-10
■4.0 ■
c
60'
1 -8.0 10.0
y -0 .0 5 4 3 k -11.369
r-0 .1 5
-
14.0 •
y - 0.0925*- 13.371
r-0 .2 5
18.0
0
10
20
30
40
50
60
VohM Mrir W aw r Cmmmk (•*)
A WdMWlD ■ WitaWia
Figure S.24. Scatterplot o f total backscatter (o°to(li) versus volumetric water content by
watershed burning treatment.
5 .6.2
Soil Backscatter and Soil Moisture
The relationship between total backscatter
(<r°ioai)
and soil moisture, whether
considered by watershed type or as a whole, is weak. Because a large portion o f the total
backscatter coefficient is comprised o f energy reflected by the vegetation canopy, an
estimate o f soil-derived backscatter (o°wu) should be better correlated with soil moisture
conditions. If the cloud model adequately represented the behavior o f the plant canopy,
o°nii would represent an unbiased quantity that is controlled directly by near surface soil
moisture conditions.
205
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The scatterplot o f soil backscatter (o0Mii) versus volumetric soil moisture (Figure
5.25) shows that the correlation coefficient (r = 0.21) is actually less than that for o°ioiii
and soil moisture. The weaker correlation indicates that the cloud model is not accurately
simulating the true behavior o f the vegetation canopy in burned and unburned
watersheds.
The poor performance o f the cloud model is confirmed by the large
difference in correlation coefficients between burned and unburned watersheds (Figure
5.26). If the cloud model was successfully simulating the influence of both the plant
canopy and litter layer, the strength o f the relationship between o°50j| and near surface soil
moisture should be very similar for the two watersheds.
4.0
0.0
•3.0
I
4 .0
•60
-
r-0.21
8.0
•« •
|,ao
/ •••
•no
16.0
18.0
0
10
30
30
40
SO
<0
Vlfc— llil WMN C M M t ( H )
Figure 5.25. Relationship between soil backscatter (o°»ii) and volumetric soil moisture.
206
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.0
2.0-
1
■
k* W
i
V I
-6 0
-8 0
-
&
-12.014 0
18.0
0
10
20
30
V i h n r t r W aitr Ch m
& WDm M IO
40
k
SO
60
(S )
■ W«tusMaQB
Figure 5.26. Scatterplot o f soil backscatter (o0wii) versus soil moisture by watershed
burning treatment.
The failure o f the cloud model can be traced back to three possible factors. The
first area o f concern that needs to be addressed is registration error. Though the rms error
associated with radar image georectification was well within the reasonable range (< 1), it
is not known how representative the point measurements o f soil moisture were o f the
actual conditions within the large 30 m x 30 m area o f the resampled radar image pixels.
Compounding possible registration and "mixed" pixel errors is the omnipresent radar
speckle that is reduced, but not eliminated, in ERS-2 PRI image products.
Another reason for poor model performance may be caused by errors in the
biomass estimates from the AVHRR-modified TM NDVI data that led to inaccurate
207
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
calculations for the density o f scattering particles in the water cloud simulating the
canopy.
In areas where biomass production is overestimated, o°»ii will also be
overestimated because o f the larger loss factor associated with higher biomass levels.
Because available water is one o f the main factors limiting production in the tallgrass
prairie, increased aboveground biomass production is often associated with a relative
abundance o f soil moisture. Therefore, errors in biomass estimation that lead to the
overestimation o f o°Wii are most likely to occur in areas with lower soil moisture levels.
Similarly, underestimates of biomass lead to underestimates of soil backscatter. Given
this situation, however, low o°soii are calculated for areas where conditions are actually
more moist. A series o f under- and overestimates o f soil backscatter act in combination
to significantly flatten what should be a 1:1 relationship between o°w,i and volumetric
water content. This kind of error cannot be corrected in this study, because it originates
in the technique used to generate the NDVI-biomass regression model.
However, known errors in biomass estimation were made. In watershed ID and
20B, simulated o°ve* values exceeded the recorded amount of total backscatter in 22%
and 30% of the sample sites, respectively. A similar percentage o f errors can be assumed
to have occurred throughout the study area.
Any error in biomass estimation will
influence the calculated value o°wa. When a biomass overestimate causes o°ve| to exceed
o°t0ui, it is a mathematical impossibility to calculate o°»ii in decibel form (i.e., logarithm
o f a negative number).
Because o f this, instances o f biomass overestimation were
particularly evident in their reduction o f the number o f transect sample points from which
208
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the relationship between soil backscatter and volumetric soil moisture could be
developed.
The final factor causing a weaker correlation between o0*,a and soil moisture is
that plant canopy moisture status is unaccounted for in the radar cloud model. Because
canopy moisture status varies over time directly with available soil water, simulated o°veg
is overestimated when conditions are dry and underestimated when soil water is
abundant. A similar problem likely exists regarding the influence of the litter layer in
unbumed watersheds. Because the biomass comprising the litter layer is already dead, its
moisture content is not tied directly to soil moisture levels. Instead, litter moisture will
vary negatively with time since the last rainfall event.
The litter layer increases the
reflectivity o f unburned watersheds in radar images acquired just after significant
precipitation because the free water trapped in the dead biomass will not have had the
opportunity to evaporate.
The relative contribution o f these last two types o f errors in the radar cloud model
can be evaluated by examining the o°Wirsoil moisture relationship on a day-by-day basis
for each watershed type. The correlation between o°»ii and volumetric soil moisture for
watershed ID are much improved compared to the overall relationship (Figure 5.27).
Two dates, however, stand out from the others. On October 10, the correlation is both the
weakest and in a negative direction. Examination of biomass estimates and volumetric
soil moisture measurements made across the transect on this date shows that the soil
moisture trend follows a different trajectory than would be expected given the
topographic gradient (Table 5.09). The October 10 image was acquired three days after a
209
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
30+ mm rain event.
This large rainfall combined with decreased cooler season
evapotranspiration may be skewing the observed soil moisture values and the resulting
correlation with soil backscatter.
3.0
6.0
r - 0.86
4.0
10
i “
*
- 6.0
a
-8.0
V
r - i] 46
r-0.62
- 10.0
110
14.0
16.0
18.0
0
20
10
30
40
SO
60
V il— Hri» W>tw Ok MM (* )
• 23-JU aOl-Aa
*0W«» aOl-Ort *10011 *03-M«v
Figure 5.27. Regression relationship between soil backscatter (o°Mii) and volumetric soil
moisture on a daily basis for watershed 1D (burned watershed).
210
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Sampling
Point
I
2
3
4
5
6
7
8
9
10
11
ANPP
(g/m2)
184.90
276.52
350.90
356.76
373.85
373.85
317.34
317.34
242.41
126.04
126.96
0v(%)
40.91
40.43
35.84
35.64
39.02
24.57
37.60
34.36
29.23
30.58
35.71
Table S.09. Estimated biomass production and measured volumetric soil moisture (6v)
for watershed ID (burned watershed) on October 10.
The remaining anomalous date is September 5. Though the correlation is strong
at r = 0.62, the data points on this day occupy the extreme left side o f the scatterplot
because it was the driest o f all the days for which radar imagery was acquired. However,
o°soii values are o f intermediate magnitude compared to the other, and wetter, days. It is
evident here that not including a measure o f canopy moisture content has affected soil
backscatter values. Because soil moisture conditions were very dry, the vegetation was
likely experiencing some water stress leading to low canopy moisture content.
In
simulating o°veg, the cloud model does not account for this expected decline in vegetation
moisture and assumes that the canopy is behaving in a constant manner over time. If
canopy moisture content were included in the vegetation backscatter estimate, o°joii
would be lower and more similar to the recorded total backscatter values (Table S. 10).
This reduction in o°»u would place the September S data in a more comparable position
211
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
compared to the remaining image dates. Note also that at several o f the transect sampling
points, biomass was overestimated, leading to a o°veg that exceeded the total backscatter
value.
Sampling Point
I
2
3
4
5
6
7
8
9
10
11
o°totti(dB) o°«m (dB) 0v(%)
-8.26
15.19
-10.05
N/A
19.85
-16.80
N/A
23.29
-11.01
N/A
-14.11
20.52
-7.57
-10.60
14.65
-10.60
-7.57
20.05
-7.84
-1.97
29.16
-7.84
-1.97
22.28
-6.71
-1.74
17.08
-19.59
N/A
10.46
-20.41
N/A
16.20
Table S. 10. Comparison o f total backscatter (o°toui) and soil backscatter (o°»ii) with
volumetric water content data for watershed ID (burned watershed) on September 5,
1996 (DOY 249).
Similar to watershed ID, the correlations between o0»ii and volumetric soil
moisture for watershed 20B were examined on a daily basis (Figure S.28).
Again two
dates stand out: July 23 (DOY 205) and August I (DOY 214). Data from July 23 were
affected by precipitation. The July 23 image was actually acquired at 2342 hours LST on
July 22. As more than 30 mm o f rain fell on July 22 (DOY 204), the litter layer was
clearly wet when the image was acquired. Because the litter layer was wet, its radar
reflectivity increased and masked the moisture differences that existed in the soil (Table
5.11). With the litter layer reflecting near the maximum amount o f radar energy across
212
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the topographic gradient, it dominates the total backscatter value, leading to a flat
relationship between o°soii and soil moisture. Based on the relationship between a°soii and
soil moisture in both watersheds on all days except September 5 (DOY 249), the cloud
model appears to better simulate o°veg when canopy moisture status would be near
maximum levels. Because o f this, a measure o f the litter layer moisture level would
remove the influence o f the wet litter on o°totai and improve the positive correlation.
4.0
10
r - 0 00
0.0
I
r - -0 75
•10
■4.0
1
r - 0 .7 4
-10.0
-110
r - 0.79 •
•14.0
0
10
20
30
40
50
60
V a ta a n r ir tM aar O hmm M(H )
♦ 23-JUI >01-Aug • 27-Aug ♦OS-S* ■ 01-Oct *10-Oct *0S-Nov
Figure 5.28. Regression relationship between soil backscatter (o ^ a) and volumetric soil
moisture on a daily basis for watershed 20B (unbumed watershed).
213
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Sampling Point
1
2
3
4
5
6
7
8
9
10
11
o°to u i (dB)
-4.37
-4.60
-4.60
-6.90
-7.84
-7.27
-7.27
-6.73
-3.14
-3.14
-6.26
(dB) ev (%)
37.48
1.53
38.79
2.17
47.45
2.17
43.30
-0.10
40.68
-1.84
0.62
37.47
0.62
48.22
0.66
44.05
2.63
45.19
2.39
35.48
-0.86
40.49
O°soil
Table 5.11. Comparison o f total and soil backscatter with soil moisture data for
watershed 20B (unburned watershed) on July 23,1996 (DOY 205).
In watershed 20B, the second date that stands apart is August 1 (DOY 214). The
correlation between o°w,i and soil moisture on this day is strongly negative, but was
calculated using only 4 o f 11 possible data points because o f problems with biomass
overestimation.
The general overestimation o f o°veg causes the observed negative
relationship that would otherwise not exist.
Comparing the daily data from each watershed, the difference between the slopes
o f the regression relationship becomes evident (Table 5.12). Overall, the correlation
coefficients for watershed 20B are more consistent and slightly stronger than those found
for watershed ID. However, the slopes o f the regression lines for watershed ID data are
closer to 1, indicating a 1:1 relationship that should exist between backscatter and
volumetric soil moisture. The differences in slope values between watershed types also
provide hints about the performance o f the cloud model over burned and unbumed
214
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
grassland areas. The lower slope values for unburned areas indicate that the simulated
contribution o f the litter layer to o0ve* may have been inconsistent, leading to a non-l:l
slope for the relationship between o0Mii and volumetric soil moisture.
Date
07-23-96
08-01-96
08-27-96
09-05-96
10-01-96
10-10-96
11-05-96
DOY
205
214
240
249
275
284
310
VWatershed ID
Slope Y-Intercept
-25.94
0.50
-11.28
0.16
-37.36
0.72
0.37
-12.08
0.58
-31.06
-0.27
10.31
0.70
-37.31
R2
0.22
0.21
0.73
0.38
0.40
0.15
0.43
Watershed 20B
Slope Y-Intercept
0.00
0.94
-0.20
2.26
0.29
-18.12
0.54
-20.34
0.34
-16.97
0.25
-10.25
0.19
-15.71
R2
0.00
0.56
0.55
0.62
0.42
0.37
0.29
Table 5.12. Slope and y-intercept values for linear regression equations between soil
backscatter and volumetric soil moisture.
Despite the problems encountered with the cloud model, the single date
correlations for watershed ID and 20B are among the highest reported when using radar
satellite data over vegetated study areas. Using ERS-1 imagery, Cognard et al. (1995)
compared o°toai and soil volumetric water content in several agricultural fields in
northwestern France. In fields with vegetation sharing similar structural characteristics
with tallgrass prairie flora, they reported correlations o f r - 0.44 for cereal grams and r =
0.23 for grass pastures. In the U.K., Griffiths and Wooding (1996) also used ERS-1 data
to examine the relationship between o0Mai and volumetric soil moisture for three bare soil
fields and three grass fields. While bare soil correlations were high, ranging from a low
215
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
o f r = 0.76 to a high o f r = 0.99, the relationships between o°t0ui and soil moisture in the
grass fields were referred to only as statistically insignificant.
Most recently, Biftu and Gan (1999) estimated near-surface soil moisture
conditions from RADARSAT imagery on six days for carefully selected agricultural
fields, pastureland, and herbaceous rangeland sites using both linear regression and the
IEM model. They indicated excellent results using linear regression, with correlation
coefficients ranging from r = 0.67 to r = 0.92. In addition, IEM was found to consistently
outperform linear regression. The correlation values reported by Biflu and Gan (1999)
using linear regression are stronger than those reported here, but direct comparison is
difficult because o f the different operating characteristics o f RADARSAT and the ERS
series o f radar satellites.
5.7
Radar-Based Soil Moisture Estimates
Removing the anomalous dates from the data set improves the overall relationship
between o°soii and soil moisture for both the burned and unbumed watersheds (Figure
5.29). The correlation coefficient for the annually burned watershed shows the greatest
increase, moving from r = 0.09 to r = 0.61. The correlation for the unbumed watershed
improves slightly as well, from r - 0.32 to r = 0.38.
216
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
4.0
2. 0
'
0. 0
-
20
-
- .
S
ID Data Excluded;
OS Sep
lOOct
2QB Data Excluded:
22M
01Aug
-4.0'
-O r-0 .3 8
- 8. 0 '
/
2
t
• l G
-
0
12.0
e i d.
dO d^
•
-14.0 ■
r - 0 .6 1
-L6.0 ■
•18.0
0
10
30
20
40
SO
60
VoliMMrir W ear CeaiMM (S )
d WataaM 10 (Bm—Q
mW a*»M X > (TM aaO
Figure S.29. Scatterplot o f soil backscatter (o0»ii) versus volumetric soil moisture by
watershed burning treatment, excluding outlying image dates.
However, to use the relationships in the improved datasets would be to ignore the
limitations shown to exist in the current configuration o f the radar cloud model,
limitations that are exacerbated by accuracy problems with the post-peak biomass
production estimates. Because no parameter accounting for canopy moisture content was
used to generate simulated o0^ values, single linear regression equations for burned and
unburned watersheds are not adequate for predicting soil moisture over the entire study
period. Rather, separate equations by watershed type and image date have to be used and
the results from these computations merged to form a complete picture o f the entire study
area.
217
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
To accomplish this, the linear regression equations shown in Table 5.12 were
inverted and solved for volumetric soil moisture.
Figure 5.30 shows estimated soil
moisture conditions for Konza Prairie Biological Station on August 27 (DOY 240) and
September 5 (DOY 249).
Differences in the soil moisture content o f burned and
unbumed watersheds are clearly evident as are the variations in moisture based on
topographic position with lowland areas having generally higher soil moisture levels than
upland sites. Using image differencing techniques, changes between dates also can be
examined (Figure 5.31).
Figure 5.30. Estimated near surface volumetric soil moisture conditions for KPBS on
August 27 (DOY 240) and September 5 (DOY 249) after application o f a 3 x 3 low pass
filter.
218
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 5.31. Changes in estimated near surface volumetric soil moisture conditions for
KPBS from August 27 (DOY 240) to September 5 (DOY 249) after application o f a 3 x
low pass filter.
219
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6
CONCLUSION
Data acquired by radar sensors are increasingly being used to measure the
biophysical characteristics o f the Earth's surface. Radar, itself, is by no means a new
technology and has been used in a number of military and civilian applications since its
inception in 1922. But it was not until 1978 and the short-lived SEASAT mission that
the potential of spacebome synthetic aperture radar for terrestrial scientific investigations
was realized. Research in the 1970s and 1980s, using ground-based and airborne radar
sensors, documented in great detail the complex relationship that exists between the radar
sensor design, the geometric and electrical properties o f an illuminated surface, and the
magnitude of reflected radar energy, or backscatter.
The foundation o f radar-based soil moisture investigations is the linear relationship
between backscatter and volumetric soil moisture in the near surface soil profile (Dobson
and Ulaby 1986, Pultz et al. 1990, Lin et al. 1994, Engman and Chauhan 199S).
However, because of the impact o f geometric and electrical controls over the backscatter
coefficient, a simple inversion procedure does not always result in accurate soil moisture
estimates, especially over vegetated surfaces. The successes reported when using the
relatively fine resolution ground-based and airborne radar systems have not translated
well when using data from spacebome systems. It is only relatively recently that an
emphasis has been placed on research using data acquired by radar satellites (e.g.,
Cognard et al. 1995, Griffiths and Wooding 1996, Henebry and Knapp 1996, Moran et
al. 1997, Sano et al. 1998, Weimann et al. 1998, Biflu and Gan 1999, Tansey et al. 1999.
220
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Much of the research using radar satellite imagery has not developed past the point
o f simple correlation between total backscatter and volumetric soil moisture. Depending
on the nature of the study area surfaces (e.g., bare soil or vegetated), these correlations
could be good or very poor. Increasingly, the impact o f surface features on active
microwave energy is being quantified through modeling approaches. This is especially
true for vegetated areas, where the structure o f various vegetation types is idealized and
the resulting scattering behavior is quantified mathematically. One o f the simplest of
these modeling approaches is the cloud model used here.
The goal o f this research was to produce an accurate and spatially distributed
estimate of soil moisture for Konza Prairie Biological Station using active microwave
imagery acquired by an orbiting SAR sensor. In support of this goal were five objectives:
(1) to measure soil moisture conditions in the field concurrently with satellite overpass,
(2) to quantify and minimize the influence o f topography on radar backscatter values, (3)
to estimate the contribution of vegetation backscatter
(o ° v c |)
to the total backscatter
coefficient (0°tMii) using a backscatter simulation model and, in doing so, quantify the
amount of backscatter contributed by the soil surface (o0*,ii), (4) to generate a spatial
estimate of biomass throughout the study period using AVHRR-modified LANDSAT
TM NDVI, and (5) to determine the linear relationship between o°«mj, as the dependent
variable, and volumetric soil moisture, as the independent variable, that could then be
inverted and solved for volumetric soil moisture to quantify near surface soil moisture
conditions across the study area and over tim e.
221
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Overall, the backscatter modeling effort did not perform as well as expected. The
correlation between o°joii and volumetric soil water content was weak and o f the same
value found for a0tom and volumetric soil moisture (r = 0.21). In addition, the correlation
between o°»ji and volumetric soil moisture in the burned and unburned resulted in very
different values, r = 0.09 in the burned watershed versus r = 0.32 in the unbumed
watershed. The modeling process failed to account for a key factor or factors causing
significant differences in the backscattering behavior o f the selected watershed types,
preventing the development o f a single inversion equation to estimate soil moisture over
the entire study area.
Examination of the relationship between 0°Wii and volumetric soil moisture by
date helped to decipher the poor explanatory power o f the combined backscatter values.
When plotted by burning treatment, daily correlation values were strong when compared
to results o f similar studies found in the literature. In the annually burned watershed
(ID), all but the two driest dates fell along a line indicating an approximate 1:1
relationship between o°»ii and volumetric soil moisture.
O f these image dates,
correlation coefficients ranged between a low o f r - 0.46 to a high o f r = 0.86. Outlying
dry image dates weakened the correlation o f the time series as a whole.
The relationship between o0*,!! and volumetric soil moisture on September S
(Figure 5.26) illustrates the weakness o f the cloud model with regard to the burned
watershed.
Soil conditions on this date were very dry, leading to the reasonable
conclusion that the grass canopy was under water stress. The magnitude o f the o0»ii
values, however, suggest wetter conditions. The higher than expected soil backscatter
222
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
values resulted from the overestimation o f vegetation backscatter at a time when the grass
canopy was least active in attenuating the radar signal. If a method o f accounting for
canopy moisture had been included in the cloud model process, a°soii values for
September 5 would have declined, placing them in better agreement with the values
estimated for days when soil water content was higher and canopy moisture content near
its maximum.
A similar flaw in the cloud model was detected when plotting o°soii versus
volumetric soil moisture on a daily basis for the unbumed watershed (20B). With the
exception o f one date, the relationship between o°soii and volumetric soil moisture is very
strong (Figure 5.27), with correlation coefficients even higher than those calculated for
watershed ID. In this unbumed watershed, however, a neat dichotomy does not exist
between wet and dry days. Unlike the burned watershed, o°»u values for September 5 are
in their proper position relative to other dates. Also unlike watershed ID, the lines o f
best fit for the wetter days have clearly different y-intercepts and are "stacked" on top o f
each other, from top to bottom, in the order expected if the separation criteria were
increasing time since the last significant rainfall event. This suggests that the moisture
content o f the litter layer may be exerting the strongest, yet unaccounted for influence on
estimated o°wgvalues in the unbumed watersheds.
The lack o f canopy and litter layer moisture information severely impacted the
performance o f the cloud model and the resulting attempt to derive a single inversion
equation to estimate soil moisture.
An additional source o f error was inaccuracy in
biomass estimation. On several dates for each watershed, biomass was overestimated to
223
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
such an extent that the resulting o°vtg exceeded the total backscatter coefficient,
preventing the estimation of o°wa. While known instances o f biomass overestimation
were encountered, it is likely that biomass also was underestimated in other areas. No
attempt was made in this study to verify the accuracy o f the NDVI-derived biomass
production values.
While biomass estimates were based on indices derived from optical imagery,
another source of error results from the nature of radar imagery itself. Though the rms
errors associated with image registration were very good, resampling and scaling issues
are important considerations when trying to match point estimates o f a variable such as
soil moisture to a composite backscatter value resulting from the averaged radar
reflectance o f a 900 m2 area. It is not known how well the measured gravimetric soil
samples represented actual mean values for the resampled 30 m x 30 m pixels of the radar
images used in the study. This kind o f mixed pixel problem can be even more severe for
active microwave imagery than for optical images such as LANDSAT because o f the
random noise added by radar speckle. In addition, very small sub-pixel sized elements
can actually dominate measured radar backscatter values through comer reflection and
resonance effects.
The final concern over the accuracy of the derived soil moisture estimates conies
from a lack of detailed soil bulk density information. The bulk density data from the
SSURGO digital soil layer used in this study represents only the characteristics o f the
general soil types or slope classes and may not reflect well the variation in values
expected for watersheds spanning a topographic gradient. In watershed ID, onfy a single
224
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
bulk density value was used to transform all o f the gravimetric soil moisture samples into
volumetric form. In watershed 20B, two values were used, one each for the upland and
lowland areas.
Though not shown previously, the correlation between o°»a and
volumetric soil moisture was better compared to that calculated when using mass
wetness. It is expected that more detailed bulk density information for the transect soils
would result in an even stronger relationship between backscatter and soil moisture.
Despite the several problem areas noted above, many encouraging results were
obtained. First, the AVHRR-modified TM NDVI values used to develop the NDVIbiomass regression relationship was a useful and cost-effective alternative to using a time
series o f LANDSAT TM images to estimate aboveground primary production.
An
assessment o f the accuracy o f this method would be an excellent candidate for future
study, but fell outside o f the intent and purpose o f this research. In this project, the
multisensor NDVI technique was simply an attempt to obtain input values for the cloud
model that better reflected the actual cycle of vegetation phenological development than
would a single and static estimate o f net primary production.
Also, the topographic correction applied to the radar imagery during the pre­
processing phase performed as expected. Relationships between the selected topographic
variables and total backscatter values, as well as the final reduction in image variance,
were comparable to similar efforts reported in the literature.
Enhancements to the
topographic correction procedure offer the potential to greatly improve the sensitivity o f
radar backscatter to soil moisture.
The use o f finer resolution digital terrain data
products, or then development through radar-based mterferometric techniques, would
225
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
allow for a more detailed characterization o f the land surface and, hopefully, further
reductions in image variance.
Many studies have questioned the sensitivity o f C-band radars, operating at
moderate incidence angles, to near surface soil moisture conditions.
Evidence from
Figures S.20 and S.21, however, clearly shows that ERS-2 total backscatter is capable of
monitoring general soil moisture conditions over even dense natural vegetation
characteristic o f tallgrass prairie. On a daily basis, that capability is further supported by
the strong correlations between o°»ii and soil moisture indicated by Figures 5.27 and
5.28. This is significant in that the operating characteristics o f the ERS-2 sensor appear
to be more than sufficient for estimating and monitoring soil moisture conditions over
virtually any grassland area. Though weaknesses in the cloud model design prevented
the development o f one inversion equation to estimate soil moisture, the single date
correlations for both the burned and unbumed watershed are among the highest yet
reported using radar satellite data. Therefore, optimism is justified for making routine
spatially distributed estimates o f near surface soil moisture conditions over large
grassland regions on a single-date basis.
Future research will be oriented along two major themes. One will further our
understanding o f the impact o f topography and vegetation on the radar backscatter
coefficient.
Promising techniques include using principal components analysis to
separate soil moisture information from other physical and biological factors that can
dominate the backscatter o f radar imagery (e.g., Henebry 1997, Verhoest et al. 1998).
Incorporating remotely-sensed landcover information, determined using radar (e.g., Hill
226
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
et al. 1999) or optical sensors, with backscatter models capable o f recognizing key
structural differences in landcover types, will expand the number o f regions that can be
studied. The second, and more urgent, o f the future research themes should deal with
specifying model parameters.
The ultimate success of radar-based soil moisture
estimates will be measured not only in by their accuracy, but also in the reduction and
simplification o f the input requirements for backscatter models.
227
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7
REFERENCES
Altese, E., O. Bolognani, M. Mancini, and P.A. Tock. 1996. Retrieving soil moisture over
bare soil from ERS 1 synthetic aperture radar data: Sensitivity analysis based on a
theoretical surface scattering model and field data. Water Resources Research 32(3):653661.
Anderson, R.C. 1990. The Historic Role o f Fire in the North American Grassland. Pp.
8-18 in Fire in North American Tallgrass Prairies. S.L. Collins and L.L. Wallace, eds.
University of Oklahoma Press: Norman, Oklahoma.
Attema, E.P.W. and F.T. Ulaby. 1978. Vegetation modeled as a cloud model. Radio
Science 13:357-364.
Axelrod, D.I. 1985. Rise o f the grassland biome, central North America. Botanical
Review 51:163-201.
Bailey, R. G. 1983. Delineation of ecosystem regions. Environmental Management 7:
365-373.
Bailey, R. G.; Avers, P. E.; King, T.; McNab, W. H., eds. 1994. Ecoregions and
subregions of the United States (map). Washington, DC: USDA Forest Service.
1:7,500,000. With supplementary table o f map unit descriptions, compiled and edited by
W. H. McNab and R. G. Bailey.
Balling, R.C. Jr. 1985. Warm season nocturnal precipitation in the Great Plains o f the
United States. Journal o f Climate and Applied Meteorology 24:1383-1387.
Bark,D. 1987. Konza Prairie Research Natural Area, Kansas. The Climates o f the
Long-Term Ecological Research Sites, ed. D. Greenland. Institute o f Arctic ami Alpine
Research: University o f Colorado.
Bauer, B.O., T.T. Veblen, and J.A. Winkler. 1999. Old methodological sneakers:
Fashion and function in a cross-training era. Annals o f the Association ofAmerican
Geographers 89(4):679-686.
Bayer, T., R. Winter, and G. Schreier. 1991. Terrain influences in SAR backscatter and
attempts to their correction. IEEE Transactions on Geoscience and Remote Sensing
29(3):451-462.
228
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Benallegue, M., O. Taconet, D. Vidal-Madjar, and M. Normand. 199S. The use o f radar
backscattering signals for measuring soil moisture and surface roughness. Remote
Sensing o f Environment 53:61-68.
Biflu, G.F. and T.Y. Gan. 1999. Retrieving near-surface soil moisture from Radarsat
SAR data. Water Resources Research 35(5): 1569-1579.
Bindlish, R. and A.P. Barros. 2000. Multifrequency soil moisture inversion from SAR
measurements with the use o f IEM. Remote Sensing o f Environment 71:67-88.
Blanchard, A.J., R.W. Newton, L. Tsang, and B.R. Jean. 1982. Volumetric effects in
cross polarized airborne radar data. IEEE Transactions on Geoscience and Remote
Sensing GE-20:36-41.
Borchert, J.R. 1950. The climate o f the central North American grassland. Annals o f the
Association o f American Geographers 40:1-39.
Box, E.O., B.N. Holben, and V. Kalb. 1989. Accuracy o f the AVHRR vegetation index
as a predictor o f biomass, primary production and net C02 flux Vegetatio 80(2):7l-89.
Briggs, J.M. and M.D. Nellis. 1989. Landsat thematic mapper digital data for predicting
aboveground biomass in a tallgrass prairie ecosystem. Pages 53-55 in Proceedings o f the
Eleventh North American Prairie Conference. T.B. Bragg and J. Stubbendieck, eds.
Lincoln, NE: University o f Nebraska Press.
Briggs, J.M., D.R. Rieck, C.L. Turner, G.M. Henebry, D.G. Goodin, and M.D. Nellis.
1997. Spatial and temporal patterns o f vegetation in the Flint Hills. Transactions o f the
Kansas Academy o f Science 100(l-2):10-20.
Briggs, J.M., M.D. Nellis, C.L. Turner, G.M. Henebry, and H. Su. 1998. A Landscape
Perspective o f Patterns and Processes in Tallgrass Prairie. Chapter 15 in Grassland
Dynamics: Long-Term Ecological Research in Tallgrass Prairie. A.K. Knapp, J.M.
Briggs, D.C. Hartnett, and S.L. Collins, eds. New York: Oxford University Press.
Callahan, J.T. 1984. Long-term ecological research. BioScience 34:363-367.
Campbell, J.B. 1987. Introduction to Remote Sensing. New York: Guilford.
Chauhan, N., D. LeVine, and R. Lang. 1994. Use of discrete scatter model to predict
active and passive microwave sensor response to com: Comparison o f theory and data.
IEEE Transactions on Geoscience and Remote Sensing 32:416-426.
Chen, K.S., S.K. Yen, and W.P. Huang. 1995. A simple model for retrieving bare soil
moisture from radar-scattering coefficients. Remote Sensing o f Environment 54:121-126.
229
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Cognard, A.L, C. Loumagne, M. Normand, P. Olivier, C. Ottle, D. Vidal-Madjar, S.
Louahala, and A. Vidal. 1995. Evaluation o f the ERS 1/ synthetic aperture radar
capacity to estimate surface soil moisture: Two-year results over the Naizin watershed.
Water Resources Research 3 1(4):975-982.
Corcoran, W.T. 1982. Moisture stress, mid-tropospheric pressure patterns, and the
forest/grassland transition in the south central states. Physical Geography 3(2): 148-159.
Devore, J. and R. Peck. 1986. Statistics: The Exploration and Analysis o f Data. New
York: West Publishing.
Dobson, M.C., F.T. Ulaby, T. LeToan, A. Beaudoin, E.S. Kasischke, and N. Christensen.
1992a. Dependence o f radar backscatter on coniferous forest biomass. IEEE
Transactions on Geoscience and Remote Sensing 30(2):412-415.
Dobson, M.C., L. Pierce, K. Sarabandi, F.T. Ulaby, and T. Sharik. 1992b. Preliminary
analysis o f ERS-1 SAR for forest ecosystem studies. IEEE Transactions on Geoscience
and Remote Sensing 30(2):203-211.
Dobson, M.C., F.T. Ulaby, and L.E. Pierce. 1995. Land-cover classification and
estimation o f terrain attributes using synthetic aperture radar. Remote Sensing o f
Environment 51:199-214.
Dobson, M.C. and F.T. Ulaby. 1986. Active microwave soil moisture research. IEEE
Transactions on Geoscience and Remote Sensing GE-24(l):23-36.
Dubois, P.C., J. van Zyl, and T. Engman. 1995a. Measuring soil moisture with imaging
radars. IEEE Transactions on Geoscience and Remote Sensing 33(4):9l5-926.
Dubois, P.C., J. van Zyl, and T. Engman. 1995b. Corrections to "measuring soil
moisture with imaging radars." IEEE Transactions on Geoscience and Remote Sensing
33(6): 1340.
Duell, A.B. 1990. Effects o f burning on infiltration, overland flow, and sediment loss on
tallgrass prairie. M.S. Thesis. Kansas State University, Manhattan, Kansas, USA.
Durden, S.L., L.A. Morrissey, and G.P. Livingston. 1995. Microwave backscatter and
attenuation dependence on leaf area index for flooded rice fields. IEEE Transactions o f
Geoscience and Remote Sensing 33(3):807-810.
Eidenshenk, J.C. 1992. The 1990 conterminous US AVHRR data set. Photogrammetric
Engineering and Remote Sensing 58(6):809-913.
230
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Elachi, C. 1987. Introduction to the Physics and Techniques o f Remote Sensing. Wiley:
New York.
Engheta, N. and C. Elachi. 1982. Radar scattering from a diffuse vegetation layer over a
smooth surface. IEEE Transactions on Geoscience and Remote Sensing GE-20(2):212216.
Engman, E.T. and N. Chauhan. 1995. Status o f microwave soil moisture measurements
with remote sensing. Remote Sensing o f the Environment 51:189-198.
ERDAS, Inc. 1994. ERDAS Field Guide. Atlanta, GA: ERDAS.
Flora, S.D. 1948. Climate o f Kansas. Topeka: Kansas State Board o f Agriculture
Reports, Vol. 67.
Franklin, S.E., M.B. Lavigne, J.R. Hunt ER, B.A. Wilson, D.R. Peddle, G.J. McDermid,
and P.T. Giles. 1995. Topographic dependence o f synthetic aperture radar imagery.
Computers and Geosciences 21(4):521-532.
Freeman, C.C. and L.C. Hulbert. 1985. An annotated list o f the vascular flora of
Konza Prairie Research Natural Area, Kansas. Transactions o f the Kansas
Academy o f Science 88:84-115.
Freeman C.C. 1998. The Flora o f Konza Prairie: A Historical Review and
Contemporary Patterns. Chapter 5 in Grassland Dynamics: Long-Term Ecological
Research in Tallgrass Prairie. A.K. Knapp, J.M. Briggs, D.C. Hartnett, and S.L. Collins,
eds. Oxford: New York.
Fung, A.K. and F.T. Ulaby. 1983. Matter-Energy Interaction in the Microwave Region.
In Manual o f Remote Sensing Vol. I: Theory, Instruments, and Techniques. R.N.
Colwell, D.S. Simonett and F.T. Ulaby, eds. Falls Church, VA: American Society o f
Photogrammetry.
Fung, A.K., and H J. Eom. 1985. A comparison between active and passive sensing o f
soil moisture from vegetated terrains. IEEE Transactions on Geoscience and Remote
Sensing GE-23(5):768-775.
Fung, A.K. and K.S. Chen. 1992. Dependence o f backscattering coefficients on
frequency, roughness, and polarization states. International Journal o f Remote Sensing
13(9):1663-1680.
Fung, A.K., Z. Lee, and K.S. Chen. 1992. Backscattering from a randomly rough
dielectric surface. IEEE Transactions on Geoscience and Remote Sensing 30(2):356-369.
231
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Fussell, J., D. Rundquist, and J.A. Harrington, Jr. 198S. On defining remote sensing.
Photogrammetric Engineering & Remote Sensing 52(9): 1507-1511.
Gogineni, S., J. Ampe, and A. Budihardjo. 1991. Radar estimates o f soil moisture over
the Konza Prairie. International Journal o f Remote Sensing 12(11):2425-2432.
Goward, S.N., C.J. Tucker, and D.G. Dye. 1985. North American vegetation patterns
observed with the NOAA-7 Advanced Very High Resolution Radiometer. Vegetatio
64(1):3-14.
Goyal, S.K., M.S. Seyfried, and P.E. O'Neill. 1998. Effect of digital elevation model
resolution on topographic correction of airborne SAR. International Journal o f Remote
Sensing 19(16):3075-3096.
Gray, L.J., G.L. Macpherson, J.K. Koelliker, and W.K. Dodds. 1998. Hydrology and
Aquatic Chemistry. Chapter 10 in Grassland Dynamics: Long-Term Ecological
Research in Tallgrass Prairie. A.K. Knapp, J.M. Briggs, D.C. Hartnett, and S.L. Collins,
eds. Oxford: New York.
Greenland, D. Use o f satellite-based sensing in land surface climatology. Progress in
Physical Geography I8(l):l-15.
Griffiths, G.H. and M.G. Wooding. 1996. Temporal monitoring o f soil moisture using
ERS-1 SAR data. Hydrological Processes 10:1127-1138.
Harrington, J.A., Jr. and J.R. Harman. 1991. Climate and vegetation in central North
America. Great Plains Quarterly 11:103-112.
Hayden, B. P. 1998. Regional Climate and the Distribution of Tallgrass Prairie. Chapter
2 in Grassland Dynamics: Long-Term Ecological Research in Tallgrass Prairie. A.K.
Knapp, J.M. Briggs, D.C. Hartnett, and S.L. Collins, eds. Oxford: New York.
Henderson-Sellers, A. 1996. Soil moisture: A critical focus for global change studies.
Global and Planetary Change 13:3-9.
Henebry, G.M. and A.K. Knapp. 1996. Spatio-temporal dynamics o f soil moisture in
tallgrass prairie using ERS-l SAR image time series: First year results. Proceedings o f
IGARSS '96:166-168.
Henebry, G.M. 1997. Advantages o f principal components analysis for land cover
segmentation from SAR image series. Paper presented at the 3"1ERS Symposium (ESA),
March 18-21, Florence, Italy. Noordwijk, Netherlands: European Space Agency
Publications Division.
232
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Hill, M.J., G.E. Donald, and P.J. Vickery. 1999. Relating radar backscatter to
biophysical properties o f temperate perennial grassland. Remote Sensing o f Environment
67:15-31.
Hillel, D. 1982. introduction to Soil Physics. San Diego: Academic Press.
Hirschboeck, K.K. 1999. A room with a view: Some geographic perspectives on
dilettantism, cross-training, and scale in hydrology. Annals o f the Association o f
American Geographers 89(4):696-706.
Hoffman, B. 1999. Personal communication.
Hussin, Y.A., R.M. Reich, and R.M Hoffer. 1991. Estimating slash pine biomass using
radar backscatter. IEEE Transactions on Geoscience and Remote Sensing 29(3):427-431.
Hutchinson, J.M.S. 1998. Extracting Soil Moisture Information from ERS-1 SAR
Imagery. Applied Geographic Studies 2(3):177-194.
Hutchinson, J.M.S. and B.C. Rundquist. 1999. Rural Landcover Classification using
Multi-Temporal SAR Imagery. Papers and Proceedings o f the Applied Geography
Conference 22:332-340.
Jantz, D.R., R.F. Hamer, H.T. Rowland, and D.A. Grier. 1975. Soil survey o f Riley
County and part o f Geary County, Kansas. Soil Conservation Service, United States
Department o f Agriculture, Washington, D.C.
Jensen, J.R. 1983. Biophysical remote sensing. Annals o f the Association o f American
Geographers 73(1):111-132.
Karam, M.A., A.K. Fung, R.H. Lang, and N.S. Chauhan. 1992. A microwave scattering
model for layered vegetation. IEEE Transactions on Geoscience and Remote Sensing
30(4):767-784.
Kasischke, E.S. and L.L. Bourgeau-Chavez. 1997. Monitoring south Florida wetlands
using ERS-1 SAR imagery. Photogrammetric Engineering and Remote Sensing
63(3):281-291.
Kazmaier, R. 1993. Checklist o f the plants o f the Konza Prairie Research Natural Area.
Manhattan, Kansas: Konza Prairie Office.
Knapp, A.K., J.T. Fahnestock, S.P. Hamburg, L.B. Statland, T.R. Seastedt, and D.S.
Schimel. 1993. Landscape patterns in soil-plant water relations and primary production
ui tallgrass prairie. Ecology 74(2):549-560.
233
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Knapp, A.K. and T.R. Seastedt. 1998. Introduction: Grasslands, Konza Prairie, and
Long-Term Ecological Research. Chapter 1 in Grassland Dynamics: Long-Term
Ecological Research in Tallgrass Prairie. A.K. Knapp, J.M. Briggs, D.C. Hartnett, and
S.L. Collins, eds. Oxford: New York.
Knapp, A.K. J.M. Briggs, J.M. Blair, and C.L. Turner. 1998. Patterns and Controls o f
Aboveground Net Primary Production in Tallgrass Prairie. Chapter 12 in Grassland
Dynamics: Long-Term Ecological Research in Tallgrass Prairie. A.K. Knapp, J.M.
Briggs, D.C. Hartnett, and S.L. Collins, eds. New York: Oxford University Press.
Kollmorgen, W.M. and D. Simonett. 1965. Grazing operations in the Flint Hills
Bluestem pastures o f Chase County, Kansas. Annals o f the Association ofAmerican
Geographers 55:260-290.
Kramer, H.J. 1996. Observation o f the Earth and its Environment. New York: Springer.
Kuchler, A.W. 1974a. Potential natural vegetation: Revised edition map. American
Geographical Society, United State Geological Survey, Washington D.C., USA.
Kuchler, A.W. 1974b. A new vegetation map of Kansas. Ecology 55:586-604.
Lamm, F. R., D. A. Pacey, and H. L. Manges. 1987. Spreadsheet template for the
calculation of penman reference evapotranspiration. American Society o f Agricultural
Engineers Paper No. MCR 87-106.
Laur, H., P. Bally, P. Meadows, J. Sanchez, B. Schaettler, E. Lopinto, and D. Esteban.
1998. Derivation o f the backscatter coefficient sigma-nought in ESA ERS SAR PRI
products. ES-TN-RS-PM-HL09, Issue 2, Revision Sb (September 7, / 998).
Lewis, S.L. 1996. Development o f Water-Usage Coefficientsfo r the Tallgrass Prairie.
M.S. Thesis. Kansas State University, Manhattan, Kansas, USA.
Lillesand, T.M. and R.W. Kiefer. 1994. Remote Sensing and Image Interpretation. New
York: John Wiley and Sons, Inc.
Lin, D.-S., E.F. Wood, P.A. Troch, M. Mancini, and T.J. Jackson. 1994. Comparisons of
remotely sensed and model-simulated soil moisture over a heterogeneous watershed.
Remote Sensing o f Environment 48:159-171.
Maidment, D.R. 1993. GIS and hydrologic modeling. In Environmental Modeling with
GIS, M. Goodchild, ed. New York: Oxford University Press.
Martin, R.D. Jr., G. Asrar, and E.T. Kanemasu. 1989. C-band scatterometer
measurements o f a tallgrass prairie. Remote Sensing o f Environment 29:281-292.
234
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Moran, M.S., A. Vidal, D. Troufleau, J. Qi, T.R. Clarke, P.J. Pinter, Jr., T.A. Mitchell, Y.
Inoue, and C.M.U. Neale. 1997. Combining multifrequency microwave and optical data
for crop management. Remote Sensing o f Environment 61:96-109.
Narayanan, R.M., R. Pardipuram, and D.C. Rundquist. 1994. Statistical characteristics
of simulated radar imagery from bare soil surfaces: Effects of surface roughness and soil
moisture variability. IEEE Transactions on Geoscience and Remote Sensing 32(l):l59167.
Oh, Y., K. Sarabandi, and F.T. Ulaby. 1992. An empirical model and an inversion
technique for radar scattering from bare soil surfaces. IEEE Transactions on Geoscience
and Remote Sensing 30(2):370-381.
ONeill, P.E., T J. Jackson, N.S. Chauhan, and M.S. Seyfried. 1993. Microwave soil
moisture estimation in humid and semi-humid watersheds. Advances in Space Research
13(5):115-118.
ONeill, P.E., N.S. Chauhan, and T.J. Jackson. 1996. Use of active and passive
microwave remote sensing for soil moisture estimation through com. International
Journal o f Remote Sensing 17(10):1851-1865.
Oviatt, C.G. 1998. Geomorphology of Konza Prairie. Chapter 3 in Grassland
Dynamics: Long-Term Ecological Research in Tallgrass Prairie. A.K. Knapp, J.M.
Briggs, D.C. Hartnett, and S.L. Collins, eds. Oxford: New York.
Pultz, T.J., R. Leconte, R.J. Brown, and B. Brisco. 1990. Quantitative soil moisture
extraction from airborne SAR data. Canadian Journal o f Remote Sensing 16(3): 5662.
Ransom, M.D., C.W. Rice, T.C. Todd, and W.A. Wehmueller. 1998. Soils and Soil
Biota. Chapter 4 in Grassland Dynamics: Long-Term Ecological Research in Tallgrass
Prairie. A.K. Knapp, J.M. Briggs, D.C. Hartnett, and S.L. Collins, eds. Oxford: New
York.
Raunkiaer, C. 1934. The Life Forms o f Plants and Statistical Plant Geography.
Clarendon Press: Oxford, UK.
National Research Council (NRC), 1997. Geography’s contributions to scientific
understanding. In Rediscovering Geography: New Relevance fo r Science and Society.
Rediscovering Geography Committee, Board o f Earth Sciences and Resources,
Commission on Geosciences, Environment, and Resources, National Research Council
Washington, D.C.: National Academy Press.
235
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ridley, J., F. Strawbridge, R. Card, and H. Phillips. 1996. Radar backscatter
characteristics o f a desert surface. Remote Sensing o f Environment 57:63-78.
Rosenberg, N.J. 1986. Climate o f the Great Plains region o f the United States. Great
Plains Quarterly 7:22-12.
Saatchi, S.S., D.M. Le Vine, and R.H. Lang. 1994. Microwave backscattering and
emission model for grass canopies. IEEE Transactions on Geoscience and Remote
Sensing 32(1):177-186
Saatchi, S.S. and J. van Zyl. 1995. Estimation o f canopy water content in Konza Prairie
grasslands using synthetic aperture radar measurements during FIFE. Journal o f
Geophysical Research 100(D12):25,481-25,496.
Samson, F. and F. Knopf. 1994. Prairie conservation in North America. BioScience
44:418-421.
Sano, E.E., M.S. Moran, A.R. Huete, and T. Miura. 1998. C- and multiangle Ku-band
synthetic aperture radar data for bare soil moisture estimation in agricultural areas.
Remote Sensing o f Environment 64:77-90.
Schmugge, T.J.. 1983. Remote sensing o f soil moisture: Recent advances. IEEE
Transactions on Geoscience and Remote Sensing GE-21(3):336-344.
Seastedt, T.R., and A.K. Knapp (1993) Consequences o f non-equilibrium resource
availability across multiple time scales: the transient maxima hypothesis. American
Naturalist 141:621-633.
Sellers, P.J., F.G. Hall, G. Asrar, D.E. Strebel, and R.E. Murphy. 1992. An overview of
the First International Satellite Land Surface Climatology Project (ISLSCP) Field
Experiment (FIFE). Journal o f Geophysical Research 97(D17): 18345-18371.
Skolnik, M.I. 1980. Introduction to Radar Systems. New York: McGraw-Hill.
Smith, A.M., D.J. Major, M.J. HOI, W.D. Williams, B. Brisco, C.W. Lindwall, and R J.
Brown. 1994. Airborne synthetic aperture radar analysis o f rangeland revegetation o f a
mixed prairie. Journal o f Range Management 47:385-391.
Smith, R.L. 1996. Ecology and Field Biology (5th Edition). HarperCollins: New York.
Steyaert, L.T. 1993. A perspective on the state of environmental simulation
modeling. In Environmental Modeling with GIS, M. Goodchild, ed. New York:
Oxford University Press.
236
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Suits, G.H. 1983. The Nature o f Electromagnetic Radiation. In Manual o f Remote
Sensing Vol. J: Theory, Instruments, and Techniques. R.N. Colwell, D.S. Simonett and
F.T. Ulaby, eds. Falls Church, VA: American Society o f Photogrammetry.
Taconet, O., M. Benallegue, D. Vidal-Madjar, L. Prevot, M. Dechambre, and M.
Normand. 1994. Estimation o f soil and crop parameters for wheat from airborne radar
backscattering data in C and X bands. Remote Sensing o f Environment 50:287-294.
Taconet, O., D. Vidal-Madjar, Ch. Emblanch, and M. Normand. 1996. Taking into
account vegetation effects to estimate soil moisture from C-band radar measurements.
Remote Sensing o f Environment 56:52-56.
Tansey, K.J., A.C. Millington, A.M. Battikhi, and K.H. White. 1999. Monitoring soil
moisture dynamics using satellite imaging radar in northeastern Jordan. Applied
Geography 19:325-344.
Tomassini, C. 1999. Personal communication (http://earthl .esrin.esa.it/I2/10/eeo 10).
Ulaby, F.T., R.K. Moore, and A.K. Fung. 1981. Microwave Remote Sensing: Active
and Passive (Vol. I): Microwave Remote Sensing Fundamentals and Radiometry.
Reading, MA: Addison-Wesley.
Ulaby, F.T., R.K. Moore, and A.K. Fung. 1982. Microwave Remote Sensing: Active
and Passive (Vol. II): Radar Remote Sensing and Surface Scattering and Emission
Theory. Reading, MA: Addison-Wesley.
Ulaby, F.T., R.K. Moore, and A.K. Fung. 1986. Microwave Remote Sensing: Active
and Passive (Vol. Ill): From Theory to Applications. Norwood, MA: Artech House.
Ulaby, F.T., C.T. Allen, and G. Eger III. 1984. Relating the microwave backscattering
coefficient to leaf area index. Remote Sensing o f Environment 14:113-133.
Ulaby, F.T. and R.P Jedlicka. 1984. Microwave dielectric properties o f plant materials.
IEEE Transactions on Geoscience and Remote Sensing GE-22:406-4l4.
Ulaby, F.T., P.C. Dubois, and J. van Zyl. 1996. Radar mapping o f surface soil moisture.
Journal o f Hydrology 184:57-84.
van Oevelen, P.J. and D.H. Hoekman. 1999. Radar backscatter inversion techniques for
estimation of surface soil moisture: EFEDA-Spain and HAPEX-Sahel case studies.
IEEE Transactions on Geoscience and Remote Sensing 37(1): 113-123.
237
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Verhoest, N.E.C., P.A. Troch, C. Paniconi, and F.P. De Troch. 1998. Mapping basin
scale variable source areas from multitemporal remotely sensed observations o f soil
moisture behavior. Water Resources Research 34(2):3235-3244.
Wang, J.R., S.P. Gogineni, and J. Ampe. 1992. Active and passive microwave
measurements o f soil moisture in FIFE. Journal o f Geophysical Research 97(D17):
18,979-18,985.
Wang, Y., E.S. Kasischke, J.M. Melack, F.W. Davis, and N.L. Christensen, Jr. 1994.
The effects o f changes in loblolly pine biomass and soil moisture on ERS-l SAR
backscatter. Remote Sensing o f Environment 49:25-31.
Weast, R.C., ed. 1984. CRC Handbook o f Chemistry and Physics. M.J. Astle and W.H.
Beyer, assoc, eds. Boca Raton, FL: CRC Press.
Wehmueller, W.A. 1996. Genesis and morphology o f soils on the Konza Prairie
Research Natural Area, Riley and Geary Counties, Kansas. M.S. Thesis, Kansas State
University, Manhattan, Kansas, U.S.A.
Weimann, A., M. von Schonermark, A. Schumann, P. Jom, and R Gunther. 1998. Soil
moisture estimation with ERS-l SAR data in the East-German loess soil area.
International Journal o f Remote Sensing 19(2):237-243.
Winokur, R.S. 1996. Operational Use O f Civil Space-Based Synthetic Aperture Radar
(SAR). Interagency Ad Hoc Working Group on SAR, National Oceanic and Atmospheric
Administration (NOAA).
Wood, E.F., D.S. Lin, M. Mancini, D. Thongs, P.A. Troch, T.J. Jackson, J.S. Famiglietti,
and E.T. Engman. 1993. Intercomparison between passive and active microwave remote
sensing, and hydrological modeling for soil moisture. Advances in Space Research
13(5): 167-176.
Xu, H., M.D. Steven, and K.W. Jaggard. 1996. Monitoring leaf area index o f sugar beet
using ERS-l SAR data. International Journal o f Remote Sensing 17(17):340l-3410.
Zerdev, N.G. and G.P. Kulemin. 1993. Determination o f sofl moisture content by
multichannel radar methods. Soviet Journal o f Remote Sensing 11(1):139-152.
238
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
8
APPENDIX A: SOIL MOISTURE DATA (MEASURED)
Watershed
Date
Sample Point
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
07-23-96
1
2
3
4
5
6
7
8
9
10
11
1
2
3
4
5
6
7
8
9
10
11
1
2
3
4
5
6
7
8
9
10
11
08-01-96
08-27-96
Bulk
Volumetric
Gravimetric
Water Content Density Water Content
Pb (g/cm3)
w (g/g)
9v(%)
1.35
52.1
0.39
1.35
44.0
0.33
1.35
46.4
0.34
42.4
1.35
0.31
1.35
41.5
0.31
47.9
1.35
0.36
0.34
1.35
45.9
1.35
0.33
44.5
1.35
0.33
44.5
1.35
39.0
0.29
1.35
39.9
0.30
1.35
0.20
27.3
1.35
35.9
0.27
0.32
1.35
42.9
0.30
1.35
41.0
0.24
1.35
31.9
1.35
0.26
34.8
1.35
42.5
0.32
1.35
38.7
0.29
0.29
1.35
39.3
0.20
1.35
27.5
0.20
1.35
27.3
1.35
44.6
0.33
1.35
56.7
0.42
50.4
0.37
1.35
48.4
1.35
0.36
1.35
50.5
0.37
1.35
49.5
0.37
48.8
1.35
0.36
1.35
47.5
0.35
1.35
43.6
0.32
36.9
1.35
0.27
40.8
0.30
1.35
239
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Watershed
Date
Sample Point
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
09-05-96
1
2
3
4
5
6
7
8
9
10
11
1
2
3
4
5
6
7
8
9
10
11
1
2
3
4
5
6
7
8
9
10
11
10-01-96
10-10-96
Volumetric
Gravimetric
Bulk
Water Content Density Water Content
w(g/g)
Pb (g/cm3)
0V(%)
15.2
0.11
1.35
19.8
0.15
1.35
23.3
0.17
1.35
20.5
0.15
1.35
14.6
0.11
1.35
20.0
1.35
0.15
29.2
1.35
0.22
22.3
0.17
1.35
17.1
1.35
0.13
10.5
0.08
1.35
16.2
1.35
0.12
41.2
0.31
1.35
0.30
40.3
1.35
46.8
0.35
1.35
42.1
1.35
0.31
43.5
0.32
1.35
38.1
0.28
1.35
39.2
0.29
1.35
41.4
1.35
0.31
32.9
0.24
1.35
28.5
0.21
1.35
30.4
1.35
0.23
40.9
0.30
1.35
40.4
0.30
1.35
35.8
0.27
1.35
35.6
0.26
1.35
39.0
0.29
1.35
24.6
1.35
0.18
37.6
0.28
1.35
34.4
0.26
1.35
29.2
1.35
0.22
30.6
0.23
1.35
35.7
0.27
1.35
240
Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
Watershed
Date
Sample Point
ID
11-05-96
I
2
3
4
5
6
7
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
8
9
10
11
07-23-96
I
2
3
4
5
6
7
8
9
08-01-96
10
11
1
2
3
4
5
6
7
8
9
10
11
Bulk
Volumetric
Gravimetric
Water Content Density Water Content
pb (g/cm3)
w(g/g)
9 v (% )
49.1
1.35
0.36
41.0
1.35
0.30
48.3
1.35
0.36
46.2
1.35
0.34
40.0
1.35
0.30
43.4
1.35
0.32
1.35
43.1
0.32
41.6
1.35
0.31
39.0
1.35
0.29
36.5
1.35
0.27
38.7
0.29
1.35
37.5
0.28
1.35
38.8
0.29
1.35
0.35
1.35
47.5
1.35
43.3
0.32
1.40
40.7
0.29
1.40
37.5
0.27
1.40
48.2
0.34
1.40
44.1
0.32
1.40
45.2
0.32
1.35
35.5
0.26
40.5
1.35
0.30
31.7
1.35
0.24
1.35
29.8
0.22
37.7
1.35
0.28
1.35
36.2
0.27
1.40
40.3
0.29
1.40
48.3
0.35
46.6
1.40
0.33
1.40
49.5
0.35
46.0
1.40
0.33
40.3
1.35
0.30
24.1
1.35
0.18
241
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Watershed
Date
Sample Point
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
08-27-96
I
2
3
4
5
6
7
8
9
10
11
I
2
3
4
5
6
7
8
9
10
11
I
2
3
4
5
6
7
8
9
10
11
09-05-96
10-01-96
Bulk
Gravimetric
Water Content Density
Pb (g/cm3)
w te/g)
0.24
1.35
0.30
1.35
1.35
0.33
0.34
1.35
1.40
0.38
1.40
0.36
0.39
1.40
1.40
0.28
0.36
1.40
0.32
1.35
1.35
0.25
0.14
1.35
0.13
1.35
0.18
1.35
0.21
1.35
1.40
0.22
1.40
0.20
1.40
0.25
1.40
0.19
1.40
0.21
1.35
0.15
0.10
1.35
0.27
1.35
0.24
1.35
1.35
0.31
0.30
1.35
1.40
0.33
1.40
0.37
1.40
0.35
0.32
1.40
0.29
1.40
1.35
0.25
1.35
0.25
Volumetric
Water Content
0V(%)
32.6
40.1
43.9
45.2
53.7
49.8
54.6
39.3
49.6
42.6
33.4
18.6
17.7
24.6
27.7
30.2
27.8
34.4
26.0
29.3
20.2
13.8
35.7
31.9
41.5
40.2
46.8
51.3
48.9
44.5
40.4
33.5
33.1
242
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Watershed
Date
Sample Point
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
10-10-96
I
2
3
4
5
6
7
8
9
10
11
1
2
3
4
5
6
7
8
9
10
11
11-05-96
Gravimetric
Volumetric
Bulk
Water Content Density Water Content
w(g/g)
Pb (g/cm3)
9v(%)
35.0
0.26
1.35
0.24
32.6
1.35
35.9
0.27
1.35
38.2
0.28
1.35
46.4
0.33
1.40
51.7
0.37
1.40
0.34
1.40
47.2
1.40
43.6
0.31
49.4
1.40
0.35
37.3
0.28
1.35
27.4
0.20
1.35
41.3
0.31
1.35
0.31
1.35
41.3
42.2
1.35
0.31
47.9
0.36
1.35
1.40
52.6
0.38
1.40
52.9
0.38
1.40
56.6
0.41
50.2
1.40
0.36
49.9
0.36
1.40
39.3
0.29
1.35
32.8
0.24
1.35
243
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
9
APPENDIX B: TRANSECT SAMPLE POINT LOCATIONS
Watershed Sample Point
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
ID
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
20B
1
2
3
4
5
6
7
8
9
10
11
I
2
3
4
5
6
7
8
9
10
11
Elevation
Latitude
Longitude
(UTM, NAD27) (UTM, NAD27) (m ASL)
4328242.3
710811.7
440.5
710864.9
442.0
4328254.4
434.2
4328248.2
710911.5
710931.1
425.3
4328253.5
710948.5
432.5
4328248.2
425.9
710970.5
4328250.5
710992.3
428.3
4328261.7
435.4
711007.4
4328263.4
434.4
711033.5
4328264.6
4328265.9
711056.1
442.1
4328258.4
711089.8
445.8
709461.9
437.3
4327416.5
709480.4
433.9
4327416.7
427.1
4327418.1
709498.2
709532.7
422.8
4327414.0
424.2
709551.9
4327417.5
709599.4
418.8
4327411.1
421.9
709625.6
4327412.3
423.4
709661.0
4327407.9
709687.6
421.5
4327411.4
425.9
4327416.9
709716.3
709734.6
431.2
4327413.7
Note: GPS-derived location and elevation data.
244
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
APPENDIX C: KONZA PRAIRIE BURN HISTORY
Wat ershed
Bum Date
(1995)
Computer Code
Map Code
099A
02-21
WA
02-21
WB
099B
04-03
HOC
001A
04-06
1A
001B
04-06
IB
04-06
1C
001C
04-08
2Co
002C
04-13
NIB
N01B
04-13
SpB
04-19
001D
ID
04-19
C01B
C1B
04-19
c
o
tc
C1C
04-21
KOIB
K1B
04-24
HQA
04-24
HQB
04-24
N01A
N1A
04-24
K1A
K01A
SpA
04-25
04-27
Belowground
04-27
Hulbert
04-27
Nature
04-28
TxHog
11-07
FA
OOFA
11-07
FB
OOFB
Notes
Source: Konza Prairie Biological Station LTER Web Site (climate.konza.ksu.edu)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Wat ershed
Bum Date
(1996)
Map Code
Computer Code
WA
099A
02-09
WB
099B
02-09
1C
001C
02-25
00 ID
ID
02-25
2De
002D
02-25
C4A
C04A
02-25
CIB
C01B
02-25
CIC
COIC
02-25
NIB
N01B
02-25
K1B
K01B
02-25
K2A
K02A
02-25
S3A
S03A
02-25
S03B
S3B
02-25
S3C
S03C
02-25
WP
099P
02-25
4F
004F
02-25
SuB
02-25
N20B
N20B
03-26
K4B
K04B
03-26
K1A
KOIA
04-05
C4D
C04D
04-09
1A
001A
04-16
N2A
N02A
04-20
HQC
04-20
OOFB
IB
001B
04-23
SuA
04-23
04-26
2Ao
002A
NIA
04-30
N01A
N2B
04-30
N02B
05-11
Belowground
2Be
05-13
002B
SpB
05-13
05-13
10B
OlOB
11-21
FA
OOFa
FB
11-21
OOFb
11-22
Fall Hulbert
Notes
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
Wildfire
East half burned in 02-25 wildfire
East half burned in 02-25 wildfire
Wildfire
West third burned in 02-25 wildfire
Source: Konza Prairie Biological Station LTER Web Site (climate.konza.ksu.edu)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Документ
Категория
Без категории
Просмотров
0
Размер файла
9 328 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа