close

Вход

Забыли?

вход по аккаунту

?

Disaggregation Of Passive Microwave Soil Moisture For Use InWatershed Hydrology Applications

код для вставкиСкачать
DISAGGREGATION OF PASSIVE MICROWAVE SOIL MOISTURE FOR USE IN
WATERSHED HYDROLOGY APPLICATIONS
by
Bin Fang
Bachelor of Engineering
Nanjing Forestry University, 2005
Master of Engineering
Chinese Academy of Sciences, 2008
Submitted in Partial Fulfillment of the Requirements
For the Degree of Doctor of Philosophy in
Geological Sciences
College of Arts and Sciences
University of South Carolina
2015
Accepted by:
Venkat Lakshmi, Major Professor
Subrahmanyam Bulusu, Committee Member
Camelia Knapp, Committee Member
Thomas Jackson, Committee Member
Lacy Ford, Vice Provost and Dean of Graduate Studies
UMI Number: 3704331
All rights reserved
INFORMATION TO ALL USERS
The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
UMI 3704331
Published by ProQuest LLC (2015). Copyright in the Dissertation held by the Author.
Microform Edition © ProQuest LLC.
All rights reserved. This work is protected against
unauthorized copying under Title 17, United States Code
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
© Copyright by Bin Fang, 2015
All Rights Reserved.
ii
ACKNOWLEDGEMENTS
Many people have been supporting and encouraging me during my doctoral
studies. Firstly I gratefully want to thank my advisor Dr. Venkat Lakshmi, for advising
and providing me the freedom to accomplish research projects in the past five years. I
would deeply appreciate his inspiration, guidance and faith which will strive me for
pursuing my academic and career goals in the future. I sincerely thank my dissertation
committee members Dr. Thomas Jackson, Dr. Subrahmanyam Bulusu and Dr. Camelia
Knapp for their valuable efforts, suggestions and feedbacks. I would like also to thank the
help from Dr. Rajat Bindlish and Dr. Michael Cosh from USDA, Dr. Jeffrey Basara from
University of Oklahoma and Dr. Andreas Colliander from JPL. This dissertation is
funded by NASA Terrestrial Hydrology Program and I would acknowledge the funding
source supporting me to complete my Ph.D. work.
I would like also to thank all the other faculty, staff and students in our
department, specifically the help coming from Ms. Stephanie Bradley and my labmates
Jessica Price Sutton, Reyadh Albatakat and Matthew Balint.
Particularly and most importantly, I would be genuinely grateful to my parents for
their understanding and support through my life.
iii
ABSTRACT
In recent years the passive microwave remote sensing has been providing soil
moisture products using instruments on board satellite/airborne platforms. Spatial
resolution has been restricted by the diameter of antenna which is inversely proportional
to resolution. As a result, typical products have a spatial resolution of tens of kilometers,
which is not compatible for some hydrological research applications. For this reason, the
dissertation explores three disaggregation algorithms that estimate L-band passive
microwave soil moisture at the subpixel level by using high spatial resolution remote
sensing products from other optical and radar instruments were proposed and
implemented in this investigation. The first technique utilized a thermal inertia theory to
establish a relationship between daily temperature change and average soil moisture
modulated by the vegetation condition was developed by using NLDAS, AVHRR, SPOT
and MODIS data were applied to disaggregate the 25 km AMSR-E soil moisture to 1 km
in Oklahoma. The second algorithm was built on semi empirical physical models (NP89
and LP92) derived from numerical experiments between soil evaporation efficiency and
soil moisture over the surface skin sensing depth (a few millimeters) by using simulated
soil temperature derived from MODIS and NLDAS as well as AMSR-E soil moisture at
25 km to disaggregate the coarse resolution soil moisture to 1 km in Oklahoma. The third
algorithm modeled the relationship between the change in co-polarized radar backscatter
and the remotely sensed microwave change in soil moisture retrievals and assumed that
iv
change in soil moisture was a function of only the canopy opacity. The change detection
algorithm was implemented using aircraft based the remote sensing data from PALS and
UAVSAR that were collected in SMPAVEX12 in southern Manitoba, Canada. The
PALS L-band h-polarization radiometer soil moisture retrievals were disaggregated by
combining them with the PALS and UAVSAR L-band hh-polarization radar spatial
resolutions of 1500 m and 5 m/800 m, respectively. All three algorithms were validated
using ground measurements from network in situ stations or handheld hydra probes. The
validation results demonstrate the practicability on coarse resolution passive microwave
soil moisture products.
v
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ............................................................................................... iii
ABSTRACT ....................................................................................................................... iv
LIST OF TABLES ........................................................................................................... viii
LIST OF FIGURES ............................................................................................................ xi
LIST OF ABBREVIATIONS .......................................................................................... xvi
CHAPTER I INTRODUCTION ..........................................................................................1
1.1 MICROWAVE SOIL MOISTURE ......................................................................1
1.2 SOIL MOISTURE RETRIEVAL ALGORITHM ................................................2
1.3 PASSIVE/ACTIVE SOIL MOISTURE DISAGGREGATION ...........................5
1.4 DISSERTATION STRUCTURE .........................................................................8
CHAPTER II THERMAL INERTIA CONCEPT BASED PASSIVE MICROWAVE
SOIL MOISTURE DOWNSCALING USING VEGETATION INDEX AND SKIN
SURFACE TEMPERATURE............................................................................................11
2.1 ABSTRACT .......................................................................................................11
2.2 INTRODUCTION ..............................................................................................12
2.3 DATA SOURCES ..............................................................................................16
2.4 METHODOLOGY .............................................................................................20
2.5 RESULTS ...........................................................................................................25
2.6 CONCLUSIONS AND FUTURE WORK .........................................................32
2.7 TABLES AND FIGURES ..................................................................................36
vi
CHAPTER III AMSR-E SOIL MOISTURE DISAGGREGATION USING SOIL
EVAPORATION EFFICIENCY MODELS ......................................................................60
3.1 ABSTRACT .......................................................................................................60
3.2 INTRODUCTION ..............................................................................................61
3.3 DATA .................................................................................................................64
3.4 METHODOLOGY .............................................................................................66
3.5 RESULTS AND DISCUSSIONS .......................................................................72
3.6 CONCLUSIONS ................................................................................................77
3.7 TABLES AND FIGURES ..................................................................................80
CHAPTER IV PASSIVE/ACTIVE MICROWAVE SOIL MOISTURE RETRIEVAL
AND DISAGGREGATION USING SMAPVEX12 DATA ............................................108
4.1 ABSTRACT .....................................................................................................108
4.2 INTRODUCTION ............................................................................................109
4.3 STUDY AREA AND DATA ............................................................................111
4.4 METHODOLOGY ...........................................................................................114
4.5 RESULTS .........................................................................................................119
4.6 CONCLUSIONS AND DISCUSSIONS ..........................................................125
4.7 ACKNOWLEDGEMENT ................................................................................127
4.8 TABLES AND FIGURES ................................................................................128
CHAPTER V CONCLUSIONS AND FUTURE WORK ................................................147
5.1 CONCLUSIONS ..............................................................................................147
5.2 FUTURE WORK..............................................................................................152
REFERENCES ................................................................................................................154
APPENDIX A – REPRINT PERMISSIONS ...................................................................165
vii
LIST OF TABLES
Table 2.1. Studies on downscaling soil moisture using various remote sensing and
modeling techniques. .........................................................................................................36
Table 2.2. Sources of land surface data used in the downscaling of soil moisture and their
spatial resolution and temporal repeat. ..............................................................................37
Table 2.3a. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Oklahoma Mesonet in May: RMSE, unbiased RMSE, bias
and standard deviation are in m3/m3. .................................................................................38
Table 2.3b. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Oklahoma Mesonet in July: RMSE, unbiased RMSE, bias
and standard deviation are in m3/m3. .................................................................................40
Table 2.3c. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Oklahoma Mesonet in August: RMSE, unbiased RMSE,
bias and standard deviation are in m3/m3. .........................................................................42
Table 2.4. Monthly generalized comparison statistics between the 1 km downscaled,
NLDAS and AMSR-E soil moisture compared to the Oklahoma Mesonet for three
months: RMSE, unbiased RMSE, bias and standard deviation are in m3/m3. ...................43
Table 2.5a. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Little Washita soil moisture observations in May. RMSE,
unbiased RMSE, bias and standard deviation are in m3/ m3 ("-" is for the value less than
0.01). ..................................................................................................................................44
Table 2.5b. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Little Washita soil moisture observations in July. RMSE,
unbiased RMSE, bias and standard deviation are in m3/ m3 ("-" is for the value less than
0.01). ..................................................................................................................................46
Table 2.6. Monthly generalized comparison statistics between the 1 km downscaled,
NLDAS and AMSR-E soil moisture compared to the Little Washita soil moisture
observations for three months. RMSE, unbiased RMSE, bias and standard deviation are
in m3/ m3. ...........................................................................................................................48
viii
Table 3.1. Soil moisture validation variables: slope, RMSE, unbiased RMSE and spatial
standard deviation of downscaled soil derived from NP89 and LP92 models, AMSR-E
and NLDAS comparing with Little Washita watershed Micronet on May 2004. .............80
Table 3.2. Soil moisture validation variables: slope, RMSE, unbiased RMSE and spatial
standard deviation of downscaled soil derived from NP89 and LP92 models, AMSR-E
and NLDAS comparing with Little Washita watershed Micronet on June 2005. .............82
Table 3.3. Soil moisture validation variables: slope, RMSE, unbiased RMSE and spatial
standard deviation of downscaled soil derived from NP89 and LP92 models, AMSR-E
and NLDAS comparing with Little Washita watershed Micronet on July 2005. ..............84
Table 3.4. Soil moisture validation variables: slope, RMSE, unbiased RMSE and spatial
standard deviation of downscaled soil derived from NP89 and LP92 models, AMSR-E
and NLDAS comparing with Little Washita watershed Micronet on August 2005. .........85
Table 3.5. Overall soil moisture validation variables of the four months: slope, RMSE,
unbiased RMSE and spatial standard deviation of downscaled soil derived from NP89
and LP92 models, AMSR-E and NLDAS comparing with Little Washita watershed
Micronet. ............................................................................................................................86
Table 3.6. Studies of disaggregating microwave soil moisture using fine scale remote
sensing imagery and model data. ......................................................................................87
Table 4.1. Remote sensing and in situ data sets used in soil moisture retrieval and
disaggregation. .................................................................................................................128
Table 4.2. PALS Soil moisture retrievals comparing with in situ soil moisture
measurements from June 25, July 3, July 10, July 13 and July 14. .................................129
Table 4.3. Disaggregated PALS change in soil moisture at 1500 m resolution validated
by change in situ soil moisture measurements from the days of June 15-June 12, June 17June 15, June 25-June 22, July 5-July 3 and July 17-July 14. .........................................129
Table 4.4. Disaggregated UAVSAR change in soil moisture at 800 m resolution
validated by change in situ soil moisture measurements from the days of June 25-June 22,
July 5-July 3, July 8-July 5, July 10-July 8, Jul 13-July 10. ...........................................130
Table 4.5. PALS change in radar backscatter compares with aggregated UAVSAR
change in radar backscatter at 1500 m resolution from the days of July 5-July 3, July 8July 5, July 10-July 8, Jul 13-July 10. .............................................................................131
Table 4.6. Validation results of disaggregated UAVSAR change in soil moisture at 5 m
resolution by change in situ soil moisture measurements from the days of July 5-July 3,
July 8-July 5, July 10-July 8, Jul 13-July 10. ..................................................................131
ix
Table 4.7. Spatial standard deviation of disaggregated UAVSAR change in soil moisture
at 5 m resolution comparing with change in situ soil moisture measurements by crop
fields from the days of July 5-July 3, July 8-July 5, July 10-July 8, Jul 13-July 10. ......132
x
LIST OF FIGURES
Figure 2.1. Imagery Maps of study region of Oklahoma and the Little Washita
Watershed and the locations of the Mesonet Stations are denoted in open yellow circles
and the soil moisture sites for Little Washita are noted in red dots. .................................49
Figure 2.2. Maps of Variables used in the soil moisture downscaling algorithm from July
21, 2005, over Oklahoma (a) MODIS Aqua 1 km land surface temperature during the
day; (b) MODIS Aqua 1 km land surface temperature at night; (c) 1/4o spatial resolution
AMSR-E soil moisture; (d) 1/8o spatial resolution NLDAS soil moisture; (e) MODIS
Aqua 1 km NDVI. .............................................................................................................50
Figure 2.3. (a) Top shows the various elements in the disaggregation procedure (b)
Bottom shows construction of the lines corresponding to constant NDVI between average
soil moisture and change in surface temperature. .............................................................51
Figure 2.4. Data flow of the soil moisture downscaling algorithm. .................................52
Figure 2.5. Daily temperature difference versus daily average soil moisture
corresponding to (Latitude: 101.875oW~102oW; Longitude: 35.125oN ~35.625oN) and
different NDVI values for May, August and July. .............................................................53
Figure 2.6. Maps of the NLDAS, AMSR-E and 1 km soil moisture (m3/m3) from May
22, 2004, July 17, 2005 and August 9, 2005 in Oklahoma. ..............................................54
Figure 2.7. Maps of the NLDAS, AMSR-E and 1 km soil moisture (m3/m3) from May
22, 2004, July 17, 2005 and August 2, 2005 in Little Washita. ........................................55
Figure 2.8. Time-series maps of 1 km soil moisture (m3/m3) of five days on July 2005
show the dry-down tendency in Little Washita region. ....................................................56
Figure 2.9a. Overall scatter plots of NLDAS, AMSR-E and 1 km soil moisture versus the
Little Washita Micronet soil moisture observations for all the months. ............................57
Figure 2.9b. Overall scatter plots of NLDAS, AMSR-E and 1 km detrended soil moisture
versus the Little Washita Micronet soil moisture observations for all the months. ...........58
Figure 2.10. Overall scatter plots of spatial standard deviation of NLDAS, AMSR-E and
1 km soil moisture versus the Oklahoma Mesonet and Little Washita Micronet soil
moisture observations for all the months. .........................................................................59
xi
Figure 3.1. Maps of Oklahoma and Little Washita Watershed boundary and the Micronet
soil moisture stations are shown in green dots. .................................................................89
Figure 3.2. Averaged and standard deviation values of soil moisture observations from
the Micronet in May 2004, June 2005, July 2005 and August 2005 and the gaps represent
the days of missing soil moisture measurements. .............................................................90
Figure 3.3. Maps of variables used in building the soil moisture downscaling algorithm
over Little Washita watershed region from July 22, 2005 (a) Daytime 1 km resolution
MODIS land surface temperature; (b) 1 km resolution MODIS NDVI; (c) NLDAS 1/8o
resolution soil temperature at 0-10 cm depth; (d) NLDAS derived field capacity; (e)
NLDAS 1/8o resolution soil moisture at 0-10 cm depth; (f) AMSR-E 1/4o resolution soil
moisture. ............................................................................................................................91
Figure 3.4. Correlation of measured soil evaporation efficiency and soil moisture (C =
0.35). NP89 is the model proposed by Noilhan and Planton (1989), while LP92 is the
model proposed by Lee and Pielke (1992). ......................................................................92
Figure 3.5a. Bivariate correlation analysis among surface soil temperature (0-10cm),
surface temperature and fractional vegetation cover from May 20, 2004, June 26, 2005,
July 22, 2005 and August 9, 2005, at 12.5 km spatial resolution. ....................................93
Figure 3.5b. Correlation analysis between surface soil temperature (0-10cm) and
fractional vegetation cover from May 20, 2004, June 26, 2005, July 22, 2005 and August
9, 2005, at 12.5 km spatial resolution. ..............................................................................94
Figure 3.5c. Correlation analysis between surface soil temperature (0-10cm) and surface
temperature from May 20, 2004, June 26, 2005, July 22, 2005 and August 9, 2005, at
12.5 km spatial resolution. ................................................................................................95
Figure 3.6a. Maps of 1 km soil evaporation efficiency, 1 km disaggregated soil
temperature and NLDAS soil temperature in Little Washita watershed region from May
20, 2004 and June 26, 2005. .............................................................................................96
Figure 3.6b. Maps of 1 km soil evaporation efficiency, 1 km disaggregated soil
temperature and NLDAS soil temperature in Little Washita watershed region from July
22, 2005 and August 9, 2005. ...........................................................................................97
Figure 3.7a. Maps of 1 km disaggregated soil moisture derived from NP89 and LP92
models, AMSR-E and NLDAS soil moisture in Little Washita watershed region from
May 20, 2004 and June 26, 2005. .....................................................................................98
Figure 3.7b. Maps of 1 km disaggregated soil moisture derived from NP89 and LP92
models, AMSR-E and NLDAS soil moisture in Little Washita watershed region from
July 22, 2005 and August 9, 2005. ....................................................................................99
xii
Figure 3.8a. Bar plots of soil moisture validation variables: slope, RMSE, unbiased
RMSE and spatial standard deviation of downscaled soil derived from NP89 and LP92
models, AMSR-E and NLDAS comparing with Little Washita watershed Micronet on
May 2004. .......................................................................................................................100
Figure 3.8b. Bar plots of soil moisture validation variables: slope, RMSE, unbiased
RMSE and spatial standard deviation of downscaled soil derived from NP89 and LP92
models, AMSR-E and NLDAS comparing with Little Washita watershed Micronet on
June 2005. .......................................................................................................................101
Figure 3.8c. Bar plots of soil moisture validation variables: slope, RMSE, unbiased
RMSE and spatial standard deviation of downscaled soil derived from NP89 and LP92
models, AMSR-E and NLDAS comparing with Little Washita watershed Micronet on
July 2005. ........................................................................................................................102
Figure 3.8d. Bar plots of soil moisture validation variables: slope, RMSE, unbiased
RMSE and spatial standard deviation of downscaled soil derived from NP89 and LP92
models, AMSR-E and NLDAS comparing with Little Washita watershed Micronet on
August 2005. ...................................................................................................................103
Figure 3.8e. Bar plots of overall soil moisture validation variables: slope, RMSE,
unbiased RMSE and spatial standard deviation of downscaled soil derived from NP89
and LP92 models, AMSR-E and NLDAS comparing with Little Washita watershed
Micronet. .........................................................................................................................104
Figure 3.9. Overall scatter plots of 1 km disaggregated soil moisture derived from NP89
and LP92 models, AMSR-E and NLDAS soil moisture comparing with Little Washita
Micronet soil moisture observations on May 2004, June 2005, July 2005 and August
2005. ................................................................................................................................105
Figure 3.10. Overall scatter plots of detrended 1 km disaggregated soil moisture derived
from NP89 and LP92 models, detrended AMSR-E and NLDAS soil moisture comparing
with Little Washita Micronet soil moisture observations on May 2004, June 2005, July
2005 and August 2005. ...................................................................................................106
Figure 3.11. Overall scatter plots of spatial standard deviation of 1 km disaggregated soil
moisture derived from NP89 and LP92 models, AMSR-E and NLDAS soil moisture
comparing with Little Washita Micronet soil moisture observations on May 2004, June
2005, July 2005 and August 2005. ..................................................................................107
Figure 4.1. Overview of the SMAPVEX12 study area (above), with a box showing the
scanning swath of the flight. Different colors indicate the types of sampling fields:
forestry site, USDA agricultural field and permanent AAFC (Agriculture and Agri-Food
Canada) field. Close view of on crop field (below) site-55. The ground soil moisture was
sampled from 16 points of two parallel rows, showing in red. .......................................133
xiii
Figure 4.2. PALS radiometer (high altitude) brightness temperature at L-band (6 GHz),
ground soil moisture measurements compares with daily accumulated precipitation from
June 7- July 19, 2012. The precipitation comes from of rain gauges of Canada
WeatherFarm weather station network. ..........................................................................134
Figure 4.3. Data flow of microwave soil moisture retrieval, disaggregation and validation
using PALS/ UAVSAR radar data and in situ soil moisture measurements. .................135
Figure 4.4. Data sets used for building soil moisture disaggregation model from July 13 July 10: (i) PALS change in soil moisture retrieval at 1500 m resolution (m3/m3); (ii)
PALS change in radar backscatter at 1500 m resolution (dB); (iii) UAVSAR change in
radar backscatter at 800 m resolution; (iv) Vegetation Water Content (VWC) at 800 m
resolution (kg/m2). ..........................................................................................................136
Figure 4.5. PALS soil moisture retrievals (m3/m3) at 1500 m resolution from five days:
June 25, July 3, July 10, July 13 and July 14. .................................................................137
Figure 4.6. PALS soil moisture retrievals at 1500 m resolution validated using in situ
measurements from June 25, July 3, July 10, July 13 and July 14. The validation results
are classified into four VWC groups. .............................................................................138
Figure 4.7. Aggregated PALS at 4500 m resolution and UAVSAR radar backscatter at
2400 m resolution compare with PALS soil moisture retrievals from July 5- July 3, July
13- July 10 and July 17- July 14. ....................................................................................139
Figure 4.8. Disaggregated PALS change in soil moisture (m3/m3) at 1500 m resolution
from June 25-June 22, July 5-July 3, July 10-July 8, July 13-July 10 and July 17-July 14.
..........................................................................................................................................140
Figure 4.9. Validation of disaggregated PALS change in soil moisture at 1500 m
resolution from June 15-June 12, June 17-June 15, June 25-June 22, July 5-July 3 and
July 17-July 14, being divided into four WVC groups. ..................................................141
Figure 4.10. Disaggregated UAVSAR change in soil moisture (m3/m3) at 800 m
resolution from June 25-June 22, July 5-July 3, July 10-July 8, July 13-July 10, Jul 17July 14. ............................................................................................................................142
Figure 4.11. Validation of Disaggregated UAVSAR change in soil moisture at 800 m
resolution from June 25-June 22, July 5-July 3, July 8-July 5, July 10-July 8, Jul 13-July
10, corresponding to four VWC groups. .........................................................................143
Figure 4.12. Comparison between PALS radar backscatter change and aggregated
UAVSAR radar backscatter change at 1500 m from July 5-July 3, July 8-July 5, July 10July 8, Jul 13-July 10, corresponding to four WVC groups. ..........................................144
xiv
Figure 4.13. Spatial variation of disaggregated UAVSAR change in soil moisture
(m3/m3) at 5 m resolution from July 3 to July 13 at the sites of three typical crop types
canola, soybeans and corn: (i): site-62, (ii):site-112, (iii): site-71. ..................................145
Figure 4.14. Validation of disaggregated UAVSAR change in soil moisture at 5 m
resolution from the days of July 5-July 3, July 8-July 5, July 10-July 8, Jul 13-July 10, at
six sites representing different crop types. ......................................................................146
xv
LIST OF ABBREVIATIONS
AAFC ............................................................................ Agriculture and Agri-Food Canada
AMSR-E .......................................... Advanced Microwave Scanning Radiometer for EOS
AMSR2 ........................................................Advanced Microwave Scanning Radiometer 2
ASTER .................... Advanced Spaceborne Thermal Emission and Reflection Radiometer
AVHRR......................................................... Advanced Very High Resolution Radiometer
EF ......................................................................................................... Evaporative Fraction
ESA ................................................................................................ European Space Agency
GCOM-W1 .............................. Global Change Observation Mission - Water "SHIZUKU"
JAXA ........................................................................ Japan Aerospace Exploration Agency
JPL ............................................................................................... Jet Propulsion Laboratory
LST ............................................................................................. Land Surface Temperature
MESONET.............................................................................................Mesoscale Network
MICRONET .......................................................................................... Microscale Network
MODIS ....................................................Moderate-Resolution Imaging Spectroradiometer
NASA........................................................ National Aeronautics and Space Administration
NCEP ..........................................................National Centers for Environmental Prediction
NDVI.................................................................... Normalized Difference Vegetation Index
NLDAS ................................................... North American Land Data Assimilation System
NOAA .................................................. National Oceanic and Atmospheric Administration
PALS .......................................................................... Passive Active L- and S-band Sensor
RMSE............................................................................................. Root Mean Square Error
SCA ............................................................................................. Single Channel Algorithm
SEE ........................................................................................... Soil Evaporation Efficiency
SGP99 ................................................................ Southern Great Plains Experiment in 1999
xvi
SMAP........................................................................ Soil Moisture Active Passive Mission
SMAPVEX12 .......................................................... SMAP Validation Experiment in 2012
SMOS.................................................................. Soil Moisture and Ocean Salinity mission
SPOT .......................................................................Satellite Pour l’Observation de la Terre
UAVSAR ........................................ Uninhabited Aerial Vehicle Synthetic Aperture Radar
USDA................................................................... United States Department of Agriculture
VWC ............................................................................................Vegetation Water Content
xvii
CHAPTER I INTRODUCTION
1.1 MICROWAVE SOIL MOISTURE
Soil moisture plays an important role in hydrological processes and air-land interactions.
The soil moisture in the near surface layer determines the proportion of latent of energy
fluxes as well as infiltration and runoff from precipitation. A previous study showed that
the accuracy of modeled precipitation primarily rely on the soil moisture conditions
(Betts et al., 1996). The microwave remote sensing studies have shown the advantages of
this approach over other optical bands to estimate water content in the top soil layer
(Jackson, 1993): (1) The microwave signal is not influenced by the atmospheric
conditions so the measurements are available for all weather coverage; (2) Microwave
frequency bands are independent to soil illumination so the measurements can be
acquired in day or night time; (3) The microwave signal has the ability to penetrate the
vegetation cover and monitor the underlying surfaces, where the microwave
measurements are highly related to the target dielectric properties, which makes it
possible to retrieve the water present in the soil layer.
The remotely sensed soil moisture data has been providing by microwave sensors
radiometer/radar over the past two decades from various platforms. It is quite useful in
regions with limited in situ observations (Jackson and Schmugge. 1989; Jackson et al.,
1999; Schmugge et al., 2002; Lakshmi, 2004, 2013; Schmugge et al., 1974; Schmugge
and Jackson, 1994; Njoku et al., 1999). For some geophysical applications, such as
1
agriculture, hydrology, extreme weather prediction and monitoring, soil moisture
products at high spatial resolution are required. However, these have been limited by the
antenna diameter, which is inversely proportional to the spatial resolution of microwave
sensors and directly related to the altitude of the platforms what carry the sensors.
Currently available microwave soil moisture data has a spatial resolution on the order of
tens of kilometers. Satellite based radars have much higher spatial resolution, but their
temporal and spatial coverage are limited. In addition, soil moisture retrieved from radar
sensors is more complicated than from radiometers. In the past few years, two satellites
carrying passive microwave sensors have been launched with a goal of solving the large
antenna limitation issue. These satellite include the Soil Moisture and Ocean Salinity
mission (SMOS) launched by ESA(European Space Agency) in 2009, and the recently
launched (January, 2015) satellite SMAP (Soil Moisture Active Passive Mission) by
NASA. The SMAP program will be providing soil moisture products at three resolutions
of 3km 9 km and 36 km.
1.2 SOIL MOISTURE RETRIEVAL ALGORITHM
A commonly used passive microwave soil moisture retrieval algorithm, called the
Single Channel Algorithm (SCA) employs single wavelength (L-Band) radiometer
observations as well as other ancillary data sets, including temperature, vegetation and
soil properties.
A first step in retrieval is to flag pixels with the land cover types that are not
suitable for retrieving soil moisture. Then a correction for the atmospheric moisture is
applied to TB .
2
For a soil surface with vegetation cover, the radiometer brightness temperature
can be expressed as a combination of radiation emitted by vegetation as well as radiation
emitted by the underlying soil which is affected by vegetation (Ulaby et al., 1986).
Assuming the vegetation layer is uniform, the radiative transfer model can be expressed
as
TB = [1 + (1 − esur )Γveg (1 − Γveg )(1 − α)Tveg + (esur Γveg Tsoil )
Equation 1.1: Radiative transfer model
Where, esur is the soil surface emissivity. α is single scattering albedo of the vegetation.
Γveg is tranmissivity of vegetation. Tveg and Tsoil are temperatures of the vegetation and
soil surface, respectively.
At microwave wavelengths, the vegetation layer has an attenuation effect on the
background emission rather than completely masking it. The single scattering albedo α at
the same microwave wavelengths is close to zero. Then if the vegetation temperature and
soil temperature can be considered the same and the Equation 1.1 can be simplified as
e=
TB
= 1 − (1 − esur )Γ 2
Tveg
Equation 1.2: Soil surface emissivity
Relating this back to the retrieval algorithm, the second step is to calculate the soil
surface emissivity e which is the brightness temperature TB divided by the temperature of
the soil surface Tsoil,
3
Next, the influences of vegetation and surface roughness factors are accounted for.
For the vegetation, the transmissivity of the vegetation canopy Γ is determined from the
vegetation optical depth τ and sensor look angle θ, which can be expressed as
Γ = exp⁡[−τ(secθ)]
Equation 1.3: Vegetation canopy transmissivity
The vegetation optical depth τ is a function of vegetation structure and vegetation
water content, expressed as
τ = bwc /cosθ
Equation 1.4: Vegetation optical depth
Where b is a parameter that depends on vegetation type and is proportional to frequency
(Jackson and Schmugge, 1991). cosθ is the slant path from the sensor look angle. wc is
vegetation water content.
A model was proposed by Choudhury et al (1979) to correct for the soil roughness
effects by using an effective roughness number h to calculate the soil emissivity esoil.
The equation is given as
esoil = 1 + (esur − 1)exp⁡(hcos2 θ)
Equation 1.5: Soil emissivity
Where, h is related to the standard deviation of surface heights.
4
Using soil emissivity esoil , the soil dielectric constant εeff can be calculated from
the viewing angle θ and soil surface reflectivity R⁡(R = 1 − esoil) by the inversion of the
Fresnel equation
εeff = ±sin2 α + cos 2 α(
√R + 1 2
)
√R − 1
Equation 1.6: Soil dielectric constant
Several mixing models can be used to estimate the soil dielectric properties by
describing the partitioning of dielectric properties of different soil components as a
function of soil water content. The model proposed by Wang and Schumugge (1980)
simplified the relationship between soil moisture and dielectric constant by using εeff
from remote sensing data. Dobson et al. (1985) used soil texture fractions and bulk
density from in situ measurements.
1.3 PASSIVE/ACTIVE SOIL MOISTURE DISAGGREGATION
There have been previous numerical studies directed at improving the coarse
spatial resolution of passive microwave soil moisture products. These can be calculated
as two types of disaggregation methodologies.
The first category is to integrate radar backscatter observations of high spatial
resolution with soil moisture retrievals based on the linear relationships between remote
sensing soil moisture and radiometer/radar backscatter measurements (Njoku et al., 1999).
As the soil moisture retrieved from active microwave observations depends on the radar
sensitivity related parameters including vegetation, topography and soil properties, by
using the statistical methods the complicated formulae between active microwave remote
5
sensing observation and remotely sensed soil moisture can be simplified. The high spatial
resolution radar data can be integrated with passive microwave soil moisture retrievals to
estimate disaggregated high resolution soil moisture change. Previously, the soil moisture
was estimated using Microwave/Imager satellite data on the Red River Basin by Lakshmi
et al., 1997. A integrated model by combining forward modelling, regression retrievals
and observations was developed and applied to disaggregate soil moisture using the data
from Southern Great Plains experiment in 1999 (Bolten et al., 2003). An algorithm
synthesized PALS soil moisture retrievals with high spatial resolution radar data from
Airsar acquired in SMEX02 experiments in Iowa. The soil moisture was disaggregated
from 400 m to 100 m resolution (Narayan et al., 2006). A continuing study on soil
moisture disaggregation by fusing soil moisture estimated from AMSR-E at 50 km
resolution and TRMM-PR radar backscatter at 5 km resolution in Little Washita
watershed, Oklahoma was carried out as well as analyses on subpixel variability of
disaggregated soil moisture (Narayan and Lakshmi, 2008). A change detection algorithm
was proposed to estimate high resolution soil moisture from SMAP satellite by using 3
km radar backscatter and 36 km radiometer brightness temperature from an observation
system simulation experiment (OSSE) to generate 10 km product (Piles et al., 2009). The
spatial variability of soil moisture from The Advanced Synthetic Aperture Radar (ASAR)
Global Monitoring (GM) was evaluated by a temporal change detection methodology
applied in southeastern Australia by using data from National Airborne Field Experiment
2005 (Mladenova et al. 2010). A statistical based algorithm was developed by merging
SMAP radiometer and radar data to obtain high resolution soil moisture data at 9 km
resolution from SMEX02 and OSSE data (Das et al., 2011).
6
As the soil moisture is directly related to many land surface variables including
land surface temperature (LST), vegetation cover as well as evapotranspiration (ET), the
products of relatively higher spatial resolution from other types satellite platforms in
visible/infrared bands can also be used to model such relationships. The soil evaporation
efficiency, which can be described as ratio of the soil evaporation rate between soil
surface and water surface, was applied on soil moisture disaggregation. The relationship
between soil evaporation efficiency and soil moisture was examined and several models
was used to calculate high spatial resolution of soil temperature data. This methodology
was applied on watershed study in southeastern Australia (Merlin et al., 2010). Similar to
this study, two indices derived from soil evaporation efficiency: evaporation fraction (EF)
and actual evaporation fraction (AEF) were used to build the disaggregation algorithm
and applied in southeastern Arizona (Merlin et al., 2008). A sequential model by combing
MODIS and the very high spatial resolution remote sensing data ASTER (Advanced
Spaceborne Thermal Emission and Reflection Radiometer) was developed to find out the
optimal resolution for disaggregating soil moisture (Merlin et al., 2009). An algorithm
which is called Physical And Theoretical scale Change (DisPATCh) modelled the
relationship between soil temperature derived from MODIS data and soil moisture and
the algorithm was applied to disaggregate SMOS soil moisture product in southeastern
Australia (Merlin et al., 2012). Based on the triangular relationship between land surface
temperature, NDVI and soil moisture, several algorithms were proposed to disaggregate
soil moisture. The triangular relationship was studied by comparing land surface
temperature with vegetation index (TvX) from Aqua MODIS data (Goetz, 1997; Sandholt
et al., 2002; Mallick et al., 2009). The soil wetness index (SWI) was proposed to compare
7
with NDVI and LST for disaggregating soil moisture. The related studies include: An
integration method based on thermal inertia theory to use the airborne thermal infrared
observations to estimate soil moisture on bare soil field (Minacapilli et al., 2009); The
relationship between fractional vegetation cover and soil moisture distribution pattern
derived from AVHRR data was studied and applied in a region of England, UK. The
enhanced vegetation index (EFI) was integrated with land surface temperature from
MODIS to disaggregate AMSR-E soil moisture in California and compared with some
others algorithms (Kim and Hogue, 2012); The other supportive studies on the
correlations between surface temperature and soil moisture were published by Lakshmi et
al., 2001; Lakshmi and Susskind, 2000. Very recently the data assimilation technique was
also explored to build the disaggregation models. These studies include (Parada and
Liang, 2004; Zhou et al., 2008; Pan et al., 2009; Sahoo et al., 2012).
1.4 DISSERTATION STRUCTURE
The overall object of this dissertation is to improve the spatial resolution of
aircraft/satellite derived passive microwave soil moisture retrievals by using other
sources of high spatial resolution remotely sensed data from visible/infrared bands as
well as radar backscatter at L band. The first section will introduce a soil moisture
downscaling algorithm for AMSR-E (Advanced Microwave Scanning Radiometer for
EOS) that uses vegetation and surface temperature data derived from satellite imagery
and North American Land Data Assimilation System (NLDAS). This algorithm is based
on thermal inertia theory, i.e., the lower values of daily average soil moisture will
correspond to higher value of daily temperature difference and vice versa. With this
theory, the soil moisture relationship curves corresponding to various vegetation
8
conditions of each NLDAS grid box will be extracted and applied to MODIS surface
temperature to retrieve the 1 km soil moisture, which will then be used to adjust AMSR-E
soil moisture. The ground observation data that include the Oklahoma Mesonet and Little
Washita watershed will be used to evaluate the accuracy of this downscaling algorithm.
The second section will focus on another soil moisture downscaling method based
on the relationship between soil evaporative efficiency and surface volumetric soil
moisture. The soil evaporative efficiency is defined as the ratio of soil evaporation rate of
a soil surface and a watered surface and is also related to humidity, soil type and wind
speed. In past few years, there have been several studies on the relationship between soil
evaporative efficiency and surface volumetric soil moisture and semi-empirical
relationships equations have been formulated (Noilhan and Planton, 1989; Lee and Pielke,
1992; Komatsu et al., 2003). In addition, other studies have shown that the soil
evaporative efficiency can be retrieved from the satellite derived surface soil temperature.
Therefore, in this study, the disaggregated surface soil temperature will be used to
compute the soil evaporative efficiency. The next step in this approach is the retrieval of
surface volumetric soil moisture using a semi-empirical soil evaporative efficiency model.
The high spatial resolution soil moisture will be applied to downscale soil moisture
products from AMSR-E. Several in situ dataset will then be compared with the results for
evaluation of the accuracy.
The third section will implement a change detection algorithm to disaggregate the
change in coarse resolution passive microwave soil moisture which is retrieved from
PALS L-band sensor at h polarization from SMAPVEX12 campaign (SMAP validation
experiment in 2012), by combining with the high spatial resolution radar backscatter data
9
from PALS (Passive Active L- and S-band Sensor) and UAVSAR (Uninhabited Aerial
Vehicle Synthetic Aperture Radar) sensors at L-band hh polarization. This algorithm was
derived by making two assumptions, that (a) there is a linear relationship between change
in soil moisture and change in radar backscatter; (b) the change of vegetation canopy in a
short period is insignificant comparing to the copolarized radar backscatter retrieved soil
moisture, and the microwave soil moisture is a function of the canopy opacity only. The
PALS soil moisture was applied to disaggregated PALS radar data at 1500 m as well as
UAVSAR radar data at 5 m/800 m and validated using ground soil moisture
measurements from coincident days.
10
CHAPTER II THERMAL INERTIA CONCEPT BASED PASSIVE
MICROWAVE SOIL MOISTURE DOWNSCALING USING
VEGETATION INDEX AND SKIN SURFACE TEMPERATURE 1
2.1 ABSTRACT
Soil moisture satellite estimates are available from a variety of passive
microwave satellite sensors, but their spatial resolution is frequently too coarse for use
by land managers and other decision makers. In this paper, a soil moisture downscaling
algorithm based on a regression relationship between daily temperature changes and
daily average soil moisture is developed and presented to produce an enhanced spatial
resolution soil moisture product. The algorithm was developed based on the thermal
inertial relationship between daily temperature changes and averaged soil moisture
under different vegetation conditions, using 1/8o spatial resolution North American
Land Data Assimilation System (NLDAS) surface temperature and soil moisture data,
as well as 5 km Advanced Very High Resolution Radiometer (AVHRR) (1981-2000)
and 1 km Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized
Difference Vegetation Index (NDVI) and surface temperature (2002-present) to build
the look-up table at 1/8o resolution. This algorithm was applied to the 1 km MODIS
land surface temperature to obtain the downscaled soil moisture estimates and then
1
Fang, B., Lakshmi, V., Bindlish, R., Jackson, T. J., Cosh, M., & Basara, J. (2013)
Passive microwave soil moisture downscaling using vegetation index and skin surface
temperature. Vadose Zone Journal, 12(3).
11
used to correct the soil moisture products from Advanced Microwave Scanning
Radiometer – EOS (AMSR-E). The 1 km downscaled soil moisture maps display
greater details on the spatial pattern of soil moisture distribution. Two sets of groundbased measurements, the Oklahoma Mesonet and the Little Washita Micronet were
used to validate the algorithm. The overall averaged slope for 1 km downscaled results
versus Mesonet data is 0.219, which is better than AMSR-E and NLDAS, while the
spatial standard deviation (0.054) and unbiased RMSE (0.042) of 1 km downscaled
results are similar to the other two datasets. The overall slope and spatial standard
deviation for 1 km downscaled results versus Micronet data (0.242 m3/m3 and 0.021
m3/m3, respectively) are significantly better than AMSR-E and NLDAS, while the
unbiased RMSE (0.026) is better than NLDAS and further than AMSR-E. In addition,
Mesonet comparisons of all three soil moisture datasets demonstrate a stronger
statistical significance than Micronet comparisons and the p-value of 1 km downscaled
is generally better than the other two soil moisture datasets. The results demonstrate
that the AMSR-E soil moisture was successfully disaggregated to 1 km. The enhanced
spatial heterogeneity as well as the accuracy of the soil moisture estimates are superior
than the AMSR-E and NLDAS estimates, when compared with in situ observations.
Keywords: Soil Moisture, downscaling algorithm, NLDAS, MODIS, AMSR-E
2.2 INTRODUCTION
Soil moisture remote sensing has a long history with microwave radiometry. The spatial
resolution obtained by a microwave radiometer (hence forth referred to as radiometer) is
inversely proportional to the diameter of the antenna and directly proportional to the
height of the satellite platform for a given frequency. Higher spatial resolution is desired
12
by a diverse set of fields of application such as agriculture, monitoring and prediction of
weather, droughts and floods. Consequently, to obtain high spatial resolution data one
would need a large aperture antenna or a low orbit. Lowering the altitude is undesirable
because it reduces temporal frequency and decreases the lifetime of the mission.
Solutions to the large antenna limitations are currently being investigated using the Soil
Moisture and Ocean Salinity mission (SMOS) by ESA and Soil Moisture Active Passive
Mission (SMAP) by NASA but it remains a problem for many operational systems.
Recognizing the need for improved spatial resolution and the limitations
described above, other solutions to improve resolution of soil moisture monitoring
should be explored. One example is SMAP mission (Entekhabi et al., 2010), set for
launch in 2014. SMAP will utilize a very large antenna and combined radiometer/radar
measurements to provide soil moisture at higher resolutions than radiometers alone can
currently achieve. SMAP (Entekhabi et al., 2010) consists of both passive and active
microwave sensors. The radiometer will have a nominal spatial resolution of 36 km and
the active radar will have a resolution of 1 km. The active microwave remote sensing
data can provide a higher spatial resolution observation of backscatter than those
obtained from a radiometer (order of magnitude: radiometer - 40 km and radar - 1 km or
better). However, radar data are more strongly affected by local roughness, micro-scale
topography and vegetation than a radiometer, implying that it is difficult to invert
backscatter to soil moisture accurately. Therefore, the current radar alone algorithms
cannot meet the accuracy requirements of the soil moisture mission (0.04 m3/m3).
SMAP will use high-resolution radar observations to disaggregate coarse resolution
radiometer observations to produce a soil moisture product at 9 km resolution. On the
13
other hand, soil moisture has been retrieved from radiometer data successfully using
various sensors and platforms and these retrieval algorithms have an established heritage
(Njoku and Entekhabi, 1996; Schmugge et al., 1974).
In this paper we implemented an alternative approach to derive higher resolution
soil moisture that would complement the SMAP approach and at the same time has the
potential to be used immediately with available satellite systems as well as downscaling
historical satellite products. This method will help to establish a long-term record of high
spatial resolution soil moisture, beginning with the AMSR (Advanced Microwave
Scanning Radiometer) instrument, which is on board NASA’s Aqua satellite and
launched in 2002, and continuing with AMSR2 (Advanced Microwave Scanning
Radiometer 2), which is on board JAXA’s GCOM-W1 satellite and launched in 2012
(Imaoka et al., 2010). In this paper, we propose the use of land surface temperature and
vegetation index data derived from two NASA sensors – AVHRR (Advanced Very High
Resolution Radiometer) since the late 1970s, and MODIS (Moderate Resolution
Imaging Spectroradiometer) as well as AMSR-E (Advanced Microwave Scanning
Radiometer - Earth Observing System sensor) on board the Aqua (2002-2011) spacecraft.
Over the past few years various methods have integrated the use of active
sensors with a higher spatial resolution to downscale passive microwave soil moisture
retrievals (Narayan et al., 2006; Narayan and Lakshmi, 2008; Das et al., 2010). Recently,
numerous studies have addressed the soil moisture downscaling problem using MODISsensor derived temperature, vegetation, and other land surface variables. The major
publications in this area of study include: A method based on a “universal triangle”
concept was used to retrieve soil moisture from Normalized Difference Vegetation
14
Index (NDVI) and Land Surface Temperature (LST) data (Piles et al., 2011). A
relationship between surface soil moisture and soil evaporative efficiency was explored
for catchment studies in Southeastern Australia (Merlin, et al., 2010). A method to
downscale soil moisture by using two soil moisture indices (Evaporative Fraction (EF)
and Actual EF (AEF)) was developed and applied in southeastern Arizona (Merlin et al.,
2008). A sequential model using MODIS as well as ASTER (Advanced Spaceborne
Thermal Emission and Reflection Radiometer) data was proposed for downscaling soil
moisture (Merlin et al., 2009). In addition, Merlin et al. (2012) used the algorithm of
Physical And Theoretical scale Change (DisPATCh) to convert high spatial resolution
soil temperature from MODIS into soil moisture. They applied this method to
disaggregate SMOS soil moisture in Southeastern Australia (Merlin et al., 2012). Kim
and Hogue (2012) developed an integrated algorithm using enhanced vegetation index
and surface temperature derived from MODIS to downscale AMSR-E soil moisture in
California and compared their results with several previous downscaling methods (Kim
and Hogue, 2012). Table 2.1 lists these studies, the methods, and significant results of
the soil moisture downscaling. Furthermore, quite a few of studies have reported using
dynamic and three dimensional data assimilation techniques to develop soil moisture
downscaling algorithms (Parada and Liang, 2004; Pan et al., 2009; Sahoo et al., 2012;
Zhou et al., 2008).
In our proposed downscaling method, we use the relationship between soil
moisture and surface temperature modulated by vegetation. This approach has a
background with past studies of Mallick et al. (2009), who used the triangular
relationship (Goetz, 1997; Sandholt et al., 2002) between surface temperature and the
15
vegetation index (TvX) derived from the MODIS Aqua sensor data. From this, they
derived the soil wetness index that was converted to soil moisture at a 1 km scale.
Minacapilli et al. (2009) used thermal infrared observations from an airborne platform
to estimate soil moisture using the thermal inertia principle for a bare soil field. They
found that the estimated soil moisture correlated very well with in situ observations.
Gillies and Carlson (1995) devised a method that derived the fractional vegetation and
spatial patterns of soil moisture using the AVHRR data set and demonstrated this
method in a region of England. In our method, the Diurnal Temperature Range (DTR)
was used (Karl et al., 1984), which was affected by vegetation (Collatz et al., 2000), soil
moisture and clouds (Dai et al., 1999).
2.3 DATA SOURCES
The State of Oklahoma was selected as the study area due to its long history of
soil moisture research. The Oklahoma Mesonet and Little Washita River Experimental
Watershed Micronet are two long-term in-situ soil moisture networks providing a solid
foundation for soil moisture remote sensing research (shown in Figure 2.1). The Little
Washita has been the location for various soil moisture field experiments including
Southern Great Plains (SGP) SGP97, SGP99, and SMEX03 (Jackson et al., 1999;
Jackson et al., 2002) and has been a key element in satellite validation studies (Jackson
et al., 2010; Mladenova et al., 2011). In addition to the ground resources, a variety of
spaceborne sensors also contributed to this study. Descriptions and maps of the datasets
used in this article are shown in Table 2.2 and Figure 2.2. Table 2.2 lists the spatial
resolution and temporal repeat of these sensors and their data products.
16
2.3.1 NLDAS DATA
The NLDAS (North American Land Data Assimilation System
http://ldas.gsfc.nasa.gov/nldas/) phase 2 hourly mosaic data is used in this study. NLDAS
is run hourly on a geographical grid with a spatial resolution of 1/8o (12.5 km). The
NLDAS data output includes various surface variables, such as radiation flux, surface
runoff, surface temperature, vegetation indices and soil moisture (Mitchell et al., 2004).
Soil, vegetation and elevation are parameterized using high resolution datasets (1 km
satellite data in the case of vegetation). The forcing data (Cosgrove et al., 2003; Luo, et
al. 2003) and outputs have been extensively validated (Lohmann et al., 2004; Robock et
al., 2003; Schaake et al., 2004). The soil moisture downscaling model included the use
of two variables: surface skin temperature and soil moisture at 0-10 cm depth. The data
used in this study correspond to the closest local overpass times of Aqua satellite for the
Oklahoma region, which are approximately 08:00 and 20:00 in UTC time.
2.3.2 AMSR-E DATA
The Advanced Scanning Microwave Radiometer on board the EOS Aqua platform
(AMSR-E) collected microwave observations at 6, 10, 19, 37 and 85 GHz frequencies
from 2002- 2011 (Njoku et al., 2003). The AMSR-E instrument provided global passive
microwave measurements of terrestrial, oceanic, and atmospheric parameters for
hydrological studies from 2002-2011 (Njoku et al., 2003, Njoku and Chan, 2006). The
soil moisture retrievals from AMSR-E are posted on 1/4o (25 km) spatial resolution. The
estimate of AMSR-E soil moisture accuracy is approximately 0.1 m3/m3 and cannot be
estimated in areas where vegetation biomass exceeds 1.5 kg/m2 (Njoku and Li, 1999).
The AMSR-E soil moisture was estimated using the single channel algorithm (SCA)
17
(Jackson, 1993; Jackson et al., 2010). The single channel algorithm uses the X-band
observations at H-polarization (the most sensitive channel) for estimation of soil
moisture. The C-band observations cannot be used for land surface applications because
they are significantly affected by radio frequency interference (RFI). The land surface
temperature was estimated using the 37 GHz v-pol observations. AVHRR derived
climatological dataset was used to account for vegetation impact on microwave
radiations emitted from the soil surface. For matching with other geo-referenced
datasets, a drop-in-the-bucket method was applied to the AMSR-E data and it was
gridded to a 25×25 km EASE (Equal-Area Scalable Earth) grid cell size. This method
averaged all the AMSR-E points by determining if their center coordinates were within
the border of a particular EASE grid cell.
2.3.3 MODIS DATA
Surface temperature data corresponded to the Oklahoma local times of 01:30
and 13:30, as well as the NDVI from MODIS/Aqua. MODIS has 36 spectral bands,
including visible, near infrared and thermal infrared spectrum and provides 44 global
data products (Justice et al., 2002). The algorithms to derive the MODIS products are
well established and have been extensively evaluated, including NDVI (Tucker, 1979;
Myneni et al., 1995), leaf area index (LAI; Myneni et al., 2002), Land cover
classification (DeFries et al., 1998; Friedl et al., 2002) and surface temperature (Wan
and Li, 1997). In the current study, surface temperature and NDVI products at two
different spatial resolutions were used for downscaling soil moisture. The datasets
included 1 km daily surface temperature (MYD11A1), 1 km biweekly NDVI
(MYD13A2), and 500 m biweekly Climate Modeling Grid (CMG) NDVI (MYD13C1).
18
The dry down lines of soil moisture during May 2004, July 2005 and August 2005 in
Oklahoma were examined. During these three months, clear days (due to the
requirement of surface temperature in our algorithm) were selected for the
downscaling algorithm application.
2.3.4 AVHRR DATA
Before the launch of the Aqua satellite and the availability of MODIS data, the 5
km CMG daily NDVI data from AVHRR sensor (AVH13C1) was used. The AVHRR
sensor is on-board the NOAA satellites, including N07, N09, N11 and N14, and
provides global and long-term surface ground measurements. Daily AVHRR NDVI data
are available between 1981~1999 (http://ltdr.nascom.nasa.gov/ltdr/ltdr.html). Because
the N14 orbit drifted greatly and degraded the data quality, AVHRR NDVI data after
year 2000 was not used in this soil moisture downscaling curve fitting.
2.3.5 OKLAHOMA MESONET
The Oklahoma Mesonet is a network of 120 automated environmental
monitoring stations with at least one site in each of the 77 counties in Oklahoma
(McPherson et al., 2007). Environmental variables are obtained at intervals spanning
every 5 to 30 minutes, depending on the variable. The data quality is verified by a
series of automated and manual checks, performed by the Oklahoma Climatological
Survey (Illston et al., 2008). In this investigation, 5 cm soil moisture content
measurements from 116 stations were extracted and geo-located for comparison with
the 1 km downscaled, AMSR-E and NLDAS soil moisture values. The locations of
the Oklahoma Mesonet stations are denoted by open yellow circles in Figure 2.1.
19
2.3.6 LITTLE WASHITA WATERSHED MICRONET
The Little Washita Watershed is located in the southwestern portion of
Oklahoma and over 20 stations within a 25×25 km region referred to as the Little
Washita Micronet. The point watershed soil moisture observations from 9 stations with
the closest time to the Aqua overpass times: 1:30 and 13:30 were extracted and then
averaged to match up with the estimated average daily soil moisture (Cosh et al., 2004;
Jackson et al., 2010). The locations of these stations are denoted by red dots in Figure
2.1.
2.4 METHODOLOGY
2.4.1 DAILY NDVI INTERPOLATION
AVHRR and MODIS NDVI products at spatial resolutions of 5 km and 1 km
were used for model building and implementation. The biweekly 5 km MODIS NDVI,
between 2003 and 2011, was aggregated to 12.5 km and geo-located to NLDAS
pixels for gap filling the AVHRR NDVI data. The biweekly 1 km MODIS NDVI
between 2003 and 2011 was input to the model for retrieving 1 km soil moisture. To
provide 5 km and 1 km resolution NDVI estimates on a daily basis, all the NDVI
records through each year of 2003-2011 were fitted using the sinusoidal method as
NDVId = a0 sin(a1 ∗ D + a2 ) + a3
Equation 2.1: MODIS daily NDVI interpolation
Where, a0, a1, a2, a3, are the regression coefficients, NDVId is the daily NDVI value, D is
the day of year. This equation was applied to the biweekly NDVI data, for all years, to
obtain daily NDVI values. This method assumes a single crop cycle and has to be
modified for locations with multiple crop cycles. So the daily NDVI varies in a near
20
sinusoidal fashion through all the days every year (Zhang et al. 2003; Knight et al.,
2006). Almost the entire study domain is dominated by forest, rangeland or winter wheat
(single crop cycle).
2.4.2 THERMAL INERTIA THEORY
Thermal inertia is the resistance of a material to temperature change, which is
indicated by time dependent variations in temperature during a full heating/cooling
cycle. It is defined as the square root of the product of the material's bulk thermal
conductivity (k) and volumetric heat capacity, where the latter is the product of density
() and specific heat capacity (c):
I = √kρc
Equation 2.2: Thermal inertia
An approximation to thermal inertia can be obtained from the amplitude of the
diurnal temperature curve. The temperature of a material with low thermal inertia will
change more during the day, than a material with high thermal inertia. Many attempts
have been made since the HCMM (Heat Capacity Mapping Mission), the first of a
series of Applications Explorer Missions (AEM) (Heliman and Moore, 1982), to
capture the thermal characteristics of the earth surface. The objective of the HCMM
was to provide comprehensive, accurate, high-spatial-resolution thermal surveys of the
surface of the earth to determine thermal inertia.
The heat capacity of water is greater than dry soil. Therefore, soil with higher
moisture content corresponds to smaller temperature changes. Because soil volumetric
heat capacity increases with higher soil moisture, it is our assertion that lower daily
average soil moisture (θav) will correspond to higher daily temperature differences (ΔTs)
21
and vice-versa. Higher soil moisture also corresponds to higher evapotranspiration,
cooling the soil surface. ΔTs can be described as
ΔTs = Tmax − Tmin
Equation 2.3: Daily temperature difference on soil surface
Where, Tmax, Tmin are the daily highest and lowest temperatures, respectively. The two
local overpass times of MODIS/Aqua approximately correspond to the highest and
lowest temperatures.
2.4.3 CONSTRUCTION OF THE DOWNSCALING MODEL
The MODIS sensor provides two very important products: NDVI and surface
temperature Ts. In this study, these two variables were extracted for each 1 km MODIS
pixel (i, j) in the 1/4o gridded AMSR-E radiometer data. We denote these variables by
NDVI (i, j) and Ts (i, j), respectively. The AMSR-E derived soil moisture,
corresponding to the 01:30 (descending orbit) overpass, is denoted as Θa while 13:30
(ascending orbit) overpass is denoted as Θp for the entire 1/4o pixel. Soil moisture
values for AM and the PM overpass for each of the MODIS pixels are referred as θa (i, j)
and θp (i, j), respectively. The time-average value of the pixel soil moisture is denoted as
θav (i, j), which refers to the predicted arithmetic mean of soil moisture for the 1 km
MODIS pixel of the AM and PM overpass (see Figure 2.3). The MODIS sensor on
Aqua was used because it matched the time of AMSR-E soil moisture estimates.
Three principles motivate the pixel-based downscaling algorithm. First, we must
consider that the soil moisture history of each pixel is unique with regard to
precipitation, evapotranspiration and runoff and can be summarized by the average soil
moisture θav (i, j). Second, based on the thermal inertia theory, the thermal inertia and
22
soil moisture depend on soil thermal conductivity, which for a wet pixel will show a
smaller change while a dry pixel will show larger change in surface temperature
(Minacapilli et al., 2009) due to modulation by evapotranspiration. Wetter pixels have
larger ET and lower surface temperature change and vice-versa (Kurc and Small, 2004).
Third, vegetation biomass within each pixel will vary and can also modulate the change
of surface temperature, which is represented by ∆Ts (Merlin et al., 2010; Lakshmi et al.,
2011). The comparison of the pixel sizes between the three datasets and the regression
curve building method is shown in Figure 2.3.
The key to the proposed disaggregation procedure is establishing the relationship
between the change in surface temperature and the average soil moisture for the 1 km
pixel. To construct the regression relationship, we plotted separately the daily NLDAS
from all the years of a particular month of the study period (i.e., the July plot will have
data for the surface temperature change and the average NLDAS soil moisture for all
years from 1981 to 2011). The data for equal NDVI lines at increments of 0.3 in NDVI
were subsequently organized. For example, during some months (e.g., January),
vegetation growth was limited and few NDVI lines in the Upper Midwest might be
constructed. On the other hand, during other periods, such as July in the Upper Midwest,
rapid changes in NDVI due to crop growth could occur and many NDVI lines would be
created. The daily NLDAS lines based on temperature difference and averaged soil
moisture at 01:30 and 13:30 of all the days from 1981 to 2011 (except 2000-2002) were
then fitted.
For simplicity, a linear regression model between the daily average soil
moisture θav(s, t) and daily temperature difference ΔTs(s, t) at NLDAS scale for each
23
month was developed as follows
θav (s, t) = a0 + a1 ∆Ts (s, t)
Equation 2.4: Relationship between daily soil moisture average and temperature
difference
Where s and t represent the NLDAS pixel location, a0 and a1 are the regression model
coefficients that correspond to several different NDVI intervals. The growing season
between May and September was examined in this study and the NDVI was
subdivided into three intervals: 0-0.3, 0.3-0.6, 0.6-1. For each month, three NLDASbased regression lines of each pixel, corresponding to the three NDVI intervals, were
built and the regression coefficients were obtained.
2.4.4 CORRECTION OF 1 KM DOWNSCALED SOIL MOISTURE
We presumed that the soil moisture variation within each NLDAS pixel could be
ignored and the downscaling model at 1/8o could be applied to the 1 km MODIS surface
temperature. On a daily basis, we used the lines corresponding to the NLDAS data
closest to the MODIS pixel to calculate the 1 km averaged soil moisture θav (i, j) from
the ΔTs (i, j) of each 1 km MODIS pixel by Eq. 4. We then averaged θav (i, j) from all the
1 km MODIS pixels and compared the values to daily average AMSR-E soil moisture
(Θa + Θp)/2, and then corrected each θav (i, j) with the difference between (Θa + Θp)/2 and
averaged θav (i, j) within the AMSR-E pixel. The cloud-covered or data gap pixels of
MODIS and AMSR-E were not used in the calculation. The corrected soil moisture θavc
(i, j) is given by
θ
avc (i,
j) = θ
av (i,
Θa + Θp
1
j) + [(
) − ∑ θav (i, j)]
2
N
i,j
24
Equation 2.5: AMSR-E soil moisture correction
Where N is the number of 1 km θav (i, j) within the AMSR-E pixel. We subsequently
generated daily values of θavc (i,j) at 1 km. This satisfied the following conditions:
(a) The average of the disaggregated soil moistures over the AMSR-E pixel is the
same as that recorded by AMSR-E; (b) The MODIS 1 km vegetation modulates the
distribution of the disaggregated soil moisture through its relationship with the daily
change in the MODIS 1 km surface temperature; and, (c) The 1 km scale changes in
surface temperature is reflected in the soil moisture distribution as evidenced in the
disaggregated soil moisture. The limitation of this methodology is that it can only be
applied over areas with no cloud cover. The data flow diagram for this algorithm is
shown in Figure 2.4.
2.5 RESULTS
2.5.1 ΘAV-ΔTS REGRESSION LINES
Figure 2.5 shows the regression fit results between NLDAS derived daily
temperature difference and daily average soil moisture of a pixel (Latitude:
101.875oW~102oW; Longitude: 35.125oN~35.625oN) for the growing months of May,
July and August. Notice that the daily average soil moisture values for all the months
generally range between 0.05~0.3 m3/m3. The points that correspond to each NDVI
interval (0~0.3; 0.3~0.6 and 0.6~1.0) yield nearly parallel lines (R2 values for July are
0.54, 0.56 and 0.40, respectively for each NDVI interval). The relationship between
daily temperature difference (ΔTs) and daily average soil moisture (θav) of the particular
NLDAS pixel (Figure 2.5) of different NDVI intervals was also examined by Student’s ttest. The results show that the relationships for May 2004, July 2005 or August 2005 at α =
25
0.05 level are statistically significant through all three NDVI intervals. Further, the daily
average soil moisture has a negative relationship with the daily temperature change,
which is consistent with the assumptions that (a) the temperature change between
morning and night is determined by pixel wetness, and (b) vegetation modulates the
change of surface temperature and the pixel with higher vegetation is less sensitive to
the temperature change.
2.5.2 1 KM DOWNSCALED SOIL MOISTURE ANALYSIS
Examples of maps of daily 1 km downscaled soil moisture are shown for May 22,
2004 and July 17, 2005 and August 9, 2005 (Figure 2.6). The 1 km downscaled soil
moistures in the lower Mideast part of Oklahoma (Figures 2.6 (iii) and (vi) and (ix)) were
missing because precipitation and heavy cloud cover dominated this area resulting in
missing MODIS surface temperature data. This area also corresponded to a gap between
AMSR-E sensor swaths. These downscaled maps illustrate the pattern of soil moisture
distribution where soil moisture content gradually increases from west to east, which
roughly corresponds to the NDVI variation in Oklahoma. In addition, the 1 km
downscaled soil moisture maps also exhibit similar spatial patterns as those of AMSR-E
and NLDAS.
The NLDAS soil moisture (Figures 2.6(i), (iv), and (vii)), which is obtained
from models, always has complete coverage because it is not impacted by cloud cover
or missing data due to gaps in swath coverage. On May 22, 2004 (Figures 2.6(i – iii)), a
wet area in the northeast corner of Oklahoma was not clearly shown by either the
AMSR-E or the 1 km downscaled soil moisture. In general the spatial patterns of the
three estimates for the May 22, 2004 case resembles each other. On July 17, 2005, the
26
western half of Oklahoma was very dry with soil moisture close to 0.02 m3/m3 and with
larger values in the east. The spatial structure shown by the 1 km soil moisture
demonstrated variability in the dry western part of the state, which was not observable
using the 1/4o AMSR-E estimates alone. In addition, the 1 km soil moisture captured
the wet area in the east central part of the state. A similar west to east dry to wet pattern
could be observed in all the estimates of the soil moisture for August 9, 2005.
Focusing on the Little Washita region (Figures 2.7), the spatial distribution of
soil moisture exhibited greater heterogeneity in 1 km downscaled map, depicting far
more dry 1 km pixels than shown in the AMSR-E and NLDAS maps, easily explainable
by single AMSR-E pixel that covers the entire watershed and the few (~5) NLDAS
pixels. The ability to show heterogeneity in soil moisture at the catchment scale is one
of the strong points of the 1 km downscaled soil moisture product. This is seen
especially in a time series of soil moisture changes in 1 km downscaled map of the
Little Washita watershed during July 2005 (Figure 2.8). Here, the soil moisture dry
down can be clearly observed through these days, particularly in the westernmost
portion of the watershed, as well as a smaller sub-catchment near the middle-east
portion of the larger watershed.
2.5.3 VALIDATION BY OKLAHOMA MESONET SOIL MOISTURE DATA
Validating the disaggregation algorithm was done by comparing two sets of
ground observations: Oklahoma Mesonet and Little Washita watershed soil moisture
values with the three gridded soil moisture datasets: NLDAS, AMSR-E and 1 km
disaggregated soil moisture. The ground observation soil moisture points were compared
with closest pixel of the three gridded soil moisture datasets. We set a threshold for the
27
minimum number of ground observation points to compare with the three gridded soil
moisture datasets from the Mesonet and Little Washita watershed (20 points and 5
points, respectively). Because the NLDAS uses Oklahoma Mesonet soil moisture (Luo
et al., 2003) and has been scaled to Mesonet data, the NLDAS soil moisture should
perform better than the AMSR-E and 1 km estimates.
The statistical variables being validated include: slope, RMSE (Root Mean Square
Error), unbiased RMSE and spatial standard deviation. The equations are as follows:
m=
θ̂i
θi
Equation 2.6: Slope of linear regression between estimated θ̂ and in-situ θ
∑n (θ̂i − θi )2
RMSE = √ i=1
n
Equation 2.7: RMSE between estimated θ̂ and in-situ θ
∗
RMSE = √
∗
∑ni=1(θ̂i − θi )2
n
Equation 2.8: Unbiased RMSE between predicted θ̂* from linear regression and in-situ θ
σ=√
∑ni=1(θ̂i − θ̅)2
n
Equation 2.9: Spatial standard deviation between estimated θ̂ and in-situ θ
Where m is the slope of linear regression between estimated soil moisture θ̂ (1 km,
AMSR-E and NLDAS) and in-situ soil moisture θ (Mesonet and Micronet). RMSE* is the
∗
unbiased RMSE. θ̂i is the predicted soil moisture value from linear regression between
estimated soil moisture θ̂ and in-situ soil moisture θ, σ is the spatial standard deviation of
28
each soil moisture data, including estimated soil moisture and in-situ soil moisture, n is
the number of data points over the study area (Mesonet or Micronet) collected in a single
day. The RMSE and RMSE* are used to characterize the uncertainty of estimated soil
moisture θ̂ to in-situ soil moisture θ. The spatial standard deviation σ is used to represent
the spatial variation of the study area. The Chi-squared test was also applied for both
Mesonet and Micronet comparisons for examining the goodness of fit, which is specified
at α = 0.05 level.
Table 2.3 shows the statistical values of comparisons with Mesonet data for single
days during May 2004, July 2005 and August 2005, while Table 2.4 shows the monthly
overall and total averaged values of the three months. From Table 2.3, we note that the
slope of 1 km downscaled comparison is generally better than NLDAS and AMSR-E,
while the unbiased RMSE and spatial standard deviation of some days are better as well.
The p-value of Chi-squared test (at α = 0.05 level) of 1 km downscaled soil moisture is
statistically better than NLDAS and AMSR-E. Table 2.3 also shows the day-to-day
variability of the performance of the 1 km soil moisture estimates as compared to AMSRE and NLDAS as well as the changes in the spatial standard deviation as compared to the
Mesonet estimates.
From Table 2.4, we observed that the slope of 1 km downscaled comparison for
two months (July 2005 and August 2005), which are 0.078 m3/m3 and 0.2 m3/m3,
respectively, is better than NLDAS and AMSR-E, and the total averaged slope of 1 km
downscaled comparison as well. Although the RMSE of 1 km downscaled comparison is
worse than NLDAS and AMSR-E, the unbiased RMSE of 1 km downscaled comparison
for all three months and overall is very similar to NLDAS and AMSR-E (0.042 m3/m3
29
versus 0.04 and 0.042 m3/m3, respectively). In addition, the spatial standard deviation of
1 km downscaled results on May 2004 and August 2005, which is 0.058 m3/m3 and 0.044
m3/m3, respectively, are closer to Mesonet than NLDAS, which is 0.066 m3/m3 and 0.047
m3/m3 and a little further than AMSR-E, which is 0.056 m3/m3 and 0.047 m3/m3. These
results demonstrate that the disaggregated soil moisture provides improvements in both
accuracy and spatial resolution.
We also note that the soil moisture values for all three datasets are systematically
lower than the in situ Mesonet observation values. This could be attributed to several
factors. First, the accuracy of AMSR-E soil moisture is limited. This methodology is
based on preserving the 25 km mean soil moisture from the AMSR-E platform. So, any
overall day-to-day bias presented in the AMSR-E soil moisture retrievals will appear in
the disaggregated 1 km estimates. Second, the MODIS retrieved daytime surface
temperature is higher than the NLDAS land surface model output, particularly during the
growing season. This may cause the daily temperature difference to be greater than
NLDAS and consequently the downscaled soil moisture would be lower than NLDAS.
Third, the Oklahoma Mesonet is equipped with sensors to measure soil water potential
(model 229-L, Campbell Scientific, Inc., Logan, UT), which is then converted to
volumetric soil moisture. Biases in this conversion may be present in gravimetric and
neutron probe samples, of which RMSE is between 0.006~0.052 m3/m3 (Illston et al.,
2008). From the matric potential values, soil water content is calculated by the van
Genuchten equation (1980), using coefficients based on soil parameters collected at the
time of installation. In addition, Minet (2012) used GPR (Ground penetrating radar)
derived soil moisture to study the uncertainties of field scale variability of surface soil
30
moisture and concluded that these arise from the errors introduced in mapping and
interpolation, and model inadequacies.
It has been observed in the data series that this methodology can result in a
biased high soil moisture estimate in the Mesonet, compared to other sampling methods.
Mesoscale models typically also provide high soil moisture estimates due to their
shortcomings in modeling soil water infiltration. Because of these independent
deficiencies, both the Mesonet soil moisture and NLDAS soil moistures are biased wet.
The soil moisture estimates were re-scaled to remove the bias to mitigate the calibration
bias between different soil moisture datasets.
2.5.4 VALIDATION USING LITTLE WASHITA WATERSHED SOIL MOISTURE
DATA
Figure 2.9(a-b) shows the overall soil moisture and detrended soil moisture
comparisons of NLDAS, AMSR-E, and 1 km downscaled data, with Little Washita
Micronet. Table 2.5 shows the statistical values comparing Micronet data for single days
on May 2004 and July 2005, while Table 2.6 shows the monthly overall and total
averaged results of the two months (August 2005 did not have enough valid values so it
was dropped). From Table 2.5, the slope of the 1 km downscaled results is obviously
better than NLDAS and AMSR-E, while the unbiased RMSE of some days is better than
either NLDAS or AMSR-E. In addition, when comparing the Mesonet data, the spatial
standard deviation for 1 km is significantly improved and much closer to Micronet than
NLDAS and AMSR-E. The results of Chi-squared test (at α = 0.05) of all three dataset are
not as good as the Mesonet comparison. However, improvement of the p-value of 1 km
downscaled results can be noticed comparing with the other two soil moisture datasets.
31
From Table 2.6, the overall slope and spatial standard deviation for 1 km
downscaled result are 0.242 m3/m3 and 0.021 m3/m3, respectively, which shows
significant advantages compared to NLDAS (0.096 m3/m3 and 0.007 m3/m3, respectively)
and AMSR-E (0.076 m3/m3 and 0.005 m3/m3, respectively). The overall spatial standard
deviation of Micronet observations was found to be 0.028 and the 1 km spatial standard
deviation is 0.021 – much closer to the Micronet observations as compared to AMSR-E
and NLDAS – 0.005 and 0.007, respectively. This will be particularly important in small
watershed studies when one pixel of NLDAS might cover an entire catchment and not
provide information on spatial variability. In addition, the unbiased RMSE of 1 km
downscaled result is also better than NLDAS. Results show that RMSE 0.024 versus
0.025 for NLDAS on May 2004, 0.027 versus 0.031 for NLDAS on July 2005, and 0.026
versus 0.028 for NLDAS of total average. By analyzing the comparison plots of spatial
standard variation between estimated soil moisture and Mesonet in-situ data (Figure 2.10),
we show that the spatial standard variation of 1 km results is more systematically
distributed than NLDAS and AMSR-E, and also closer to the in-situ soil moisture.
2.6 CONCLUSIONS AND FUTURE WORK
In this study, a soil moisture downscaling algorithm based on NLDAS derived
regression relationship related daily surface temperature changes and average daily soil
moisture was developed. This algorithm was applied using MODIS products of clear days
during crop growing seasons (May, July and August of 2004~2005) in Oklahoma. We
used two sets of validation data, Oklahoma Mesonet and Little Washita Micronet soil
moisture observations to compare with the three estimates: 1 km downscaled soil
moisture values, AMSR-E soil moisture values and NLDAS soil moisture values.
32
Statistical analysis was used to study the accuracy of the downscaling algorithm.
Several observations can be made using these results. First, the regression relationship
supports our assumption that the surface temperature change depends on the wetness of
the land surface and that the vegetation modulates this relationship. Second, the 1 km
downscaled maps provide details on the soil moisture spatial distribution patterns in
Oklahoma that are not available using AMSR-E product. The results also compare well
with Oklahoma Mesonet soil moisture values. Third, considering Mesonet is biased wet
(Illston et al., 2008), as is NLDAS (Mo et al., 2012), and that our system is based on
NLDAS, we also have a slight bias; however, we are biased drier compared to Mesonet
and NLDAS. This feature can be observed in Figure 2.9 (a) by comparing the distribution
pattern among NLDAS, AMSR-E and 1 km results, where more points of 1 km
comparison plot are below the diagonal line than the other two comparison plots. So the
downscaled results are quite close to true values. Fourth, validation results of the three
estimated soil moisture products against field observations from the Oklahoma Mesonet
show that the slope for 1 km downscaled soil moisture are generally better and the spatial
standard deviation is partially better than NLDAS and AMSR-E products. The overall
spatial standard deviation of 1 km on August 2005 is closer to Mesonet, while on July
2005 and August 2005 is closer to either NLDAS or AMSR-E. Although the RMSE of 1
km downscaled soil moisture is poorer than the other soil moisture datasets, the unbiased
RMSE of 1 km is better for some months. The p-value of the Chi-squared goodness of fit
test also shows that the comparison of field data with the 1 km downscaled map is
statistically better than the NLDAS and AMSR-E products. Another advantage of the 1
km downscaled soil moisture result can be observed from comparisons with Little
33
Washita Micronet data. The overall slope and spatial standard deviation using 1 km
downscaled results in Table 2.6 are definitely better than the other two datasets- AMSR-E
and NLDAS. In addition, the overall unbiased RMSE using 1 km downscaled is always
better than NLDAS, while better than AMSR-E on July 2005. Although the results of
Chi-squared test using data from the Micronet site and the three soil moisture datasets are
poor, we did observe an improvement in the comparison of the results.
By comparing the scatter plots of spatial standard deviation, 1 km downscaled soil
moisture demonstrates a better correlation with in-situ soil moisture than NLDAS and
AMSR-E soil moisture. Such trends are also noted when comparing with Micronet
observations, where the NLDAS and AMSR-E comparisons are biased. Considering the
slope and spatial standard deviation are two important variables that indicate similarity
between observed and predicted measurements, this downscaling methodology not only
increases resolution and adds information about the spatial variability of the soil moisture
within the AMSR-E estimates, but also preserves the average soil moisture. Any bias
between the in situ observations and AMSR-E soil moisture will also be reflected in the
disaggregated soil moisture estimates. We reported the unbiased RMSE estimates for
different soil moisture products; however, the RMSE does not reflect the added value that
the downscaled soil moisture provides about the spatial heterogeneity of soil moisture.
The validation results proved that the soil moisture downscaling algorithm is applicable.
In addition, if we compare our results with those reported in the literature and presented
in Section 1 and Table 2.1, we note that this analysis included a large area (the entire
state of Oklahoma) and a longer period of time compared to some previous studies, which
only included shorter-term field experiments or smaller catchments and our results show
34
much lower RMSE. These works by Rodríguez-Iturbe et al. (1995), Mohanty et al.
(2000), Famiglietti et al. (2008) and Minet et al. (2012) on spatial variability and
uncertainty of soil moisture across scales as they relate to our research. In particular,
research reported by Mohanty et al. (2000) and Famiglietti et al. (2008) was based on
observations made over the same study domain. These papers clearly demonstrate the
scaling properties of soil moisture and the ability to observe soil moisture patterns across
different scales. Remote sensing observations provide a spatially average estimate of soil
moisture over the entire footprint.
Regarding these probable sources of uncertainty, several limitations still exist in
this algorithm: (a) the MODIS temperature and NDVI products are often influenced by
cloud coverage; therefore, this method for downscaling is not appropriate for all weather
conditions; (b) the NDVI data comes from two sensors (AVHRR and MODIS) that were
available for different time periods; (c) the accuracy of NLDAS and AMSR-E soil
moisture determines the accuracy of the 1 km downscaled soil moisture; (d) only
vegetation and temperature were used to develop this downscaling algorithm, and high
spatial resolution data of these variables would be required for broader applications. This
methodology is based on preserving the 25 km mean soil moisture, as is done for the
AMSR-E soil moisture estimates. So, any overall day-to-day bias present in the AMSR-E
soil moisture retrievals also will be present in the disaggregated 1 km estimates. Future
work will combine this approach with our previous active-passive downscaling approach
(Narayan et al., 2006), yielding an advantage that it can be applied in cloud-free as well
as cloudy areas.
35
2.7 TABLES AND FIGURES
Table 2.1. Studies on downscaling soil moisture using various remote sensing and
modeling techniques.
Author
Methodology
Time and Region
Results
Based on the relationship
between soil evaporative
efficiency and soil moisture
NAFE 2006 (OctNov), Yanco,
Southeastern
Australia
Mean correlation
slope between
simulated and
measured data is
0.94, the most
accuracy with an
error of 0.012
m3/m3
Piles et al.
(2011)
Build model between LST,
NDVI and soil moisture
Jan-Feb 2010,
Murrumbidgee
catchment, Yanco,
Southeastern
Australia
R2 is between
0.14~0.21 and
RMSE is between
0.9~0.17 m3/m3
Merlin et
al. (2008)
Downscaling algorithm is
derived from MODIS and
physical based soil evaporative
efficiency model
NAFE 2006 (OctNov),
Murrumbidgee
catchment, Yanco,
Southeastern
Australia
Overall RMSE is
between
1.4%~1.8% vol.
Based on two soil moisture
indices EF and AEF
June and August
1990(Monsoon'90
experiment),
USDA-ARS
WGEW in
southeastern
Arizona
Total accuracy is
3% vol. for EF and
2% vol. for AEF,
and correlation
coefficient is
0.66~0.79 for EF
and 0.71~0.81 for
AEF
Sequential model
NAFE 2006 (OctNov),Yanco,
southeastern
Australia
RMSE is -0.062
m3/m3 and the bias
is 0.045 m3/m3
Merlin et
al. (2010)
Merlin et
al. (2008)
Merlin et
al. (2009)
36
Jan, Feb and Sept
2010,
Murrumbidgee
catchment, Yanco,
Southeastern
The correlation
coefficient between
disaggregated and
in situ soil
moisture between
0.70~ 0.85 in
summer
Merlin
et al.
(2012)
Physical And Theoretical scale
Change (DisPATCh) method
Kim
and
Hogue
(2012)
SMEX04 field
Enhanced vegetation index and
Spatial correlation
measurement from
surface temperature derived
are generally from
the San Pedro
from MODIS
-0.08 to 0.34
River Basin
Table 2.2. Sources of land surface data used in the downscaling of soil moisture and
their spatial resolution and temporal repeat.
Source
Data
Spatial
Resolution
Temporal
Repeat
Soil Moisture Content (0-10 cm
layer, kg/m2 )
1/8 degree
(12.5 km)
Hourly
NLDAS
1/8 degree
Surface Skin Temperature (K)
Hourly
(12.5 km)
Normalized Difference
Vegetation Index (NDVI)
5 km
Daily
Normalized Difference
Vegetation Index (NDVI)
5 km
Biweekly
Land Surface Temperature (K)
1 km
Daily
AMSR-E
Soil Moisture Content (m3/m3)
1/4 degree
(25 km)
Daily
Mesonet
Surface Soil Moisture Content
(0-5 cm layer, m3/m3)
116~117
stations
5 minutes
Micronet
Surface Soil Moisture Content
(0-5 cm layer, m3/m3)
9 stations
Hourly
AVHRR
MODIS
37
Table 2.3a. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Oklahoma Mesonet in May: RMSE, unbiased RMSE, bias
and standard deviation are in m3/m3.
Day
May 2,
2004
May 4,
2004
May 6,
2004
May 7,
2004
May 8,
2004
Dataset
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
Spatial
Standard
Deviation
(m3/m3)
pvalue
0.404
0.136
0.063
0.074
0.019
0.442
0.132
0.061
0.07
0.024
0.552
0.101
0.056
0.071
0.067
0.132
0.268
0.144
0.057
0.069
0.009
0.35
0.141
0.055
0.065
0.009
0.497
0.108
0.052
0.069
0.058
0.1
-0.285
0.138
0.032
0.031
0.005
-0.341
0.135
0.033
0.026
0.006
-0.579
0.108
0.033
0.027
0.034
0.063
0.37
0.151
0.053
0.056
0.006
0.395
0.146
0.051
0.052
0.01
0.423
0.118
0.048
0.073
0.055
0.073
0.241
0.108
0.032
0.057
0.167
0.263
0.109
0.032
0.052
0.165
0.703
0.086
0.031
0.046
0.033
0.378
38
Number
of
Points
81
81
87
35
45
May 9,
2004
May 20,
2004
May 22,
2004
May 23,
2004
May 30,
2004
May 31,
2004
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
1 km
Downsca
led
AMSRE
NLDAS
Mesonet
0.544
0.119
0.050
0.041
0.176
0.72
0.116
0.049
0.041
0.192
1.071
0.096
0.043
0.072
0.052
0.322
0.462
0.105
0.064
0.062
0.146
0.453
0.101
0.061
0.06
0.148
0.717
0.109
0.06
0.08
0.072
0.103
0.884
0.128
0.025
0.1
0.027
0.993
0.122
0.028
0.113
0.041
1.501
0.111
0.016
0.101
0.062
0.086
1.04
0.11
0.054
0.063
0.038
0.496
0.11
0.052
0.056
0.028
-0.01
0.107
0.049
0.075
0.06
0.048
-0.012
0.118
0.023
0.043
0.107
0.112
0.114
0.022
0.043
0.109
0.14
0.122
0.018
0.056
0.029
0.049
0.25
0.105
0.041
0.043
0.073
0.254
0.099
0.039
0.039
0.095
0.488
0.106
0.037
0.053
0.042
0.086
39
24
57
34
60
37
65
Table 2.3b. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Oklahoma Mesonet in July: RMSE, unbiased RMSE, bias
and standard deviation are in m3/m3.
Day
July 3,
2005
July 8,
2005
July
10,
2005
July
12,
2005
July
17,
2005
July
21,
2005
Dataset
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
Spatial
Standard
Deviation
(m3/m3)
pvalue
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
0.237
0.166
0.059
0.055
0.002
0.039
0.095
0.161
0.119
0.064
0.063
0.034
0.025
0.064
0.007
0.117
0.224
0.177
0.047
0.061
0.002
0.08
0.168
0.169
0.106
0.048
0.047
0.052
0.045
0.048
0.005
0.224
-0.043
0.174
0.053
0.06
0.001
-0.005
0.213
0.162
0.113
0.053
0.051
0.052
0.048
0.053
0.004
0.127
0.032
0.162
0.047
0.059
0.003
-0.007
-0.013
0.154
0.091
0.047
0.047
0.052
0.046
0.047
0.008
0.361
0.176
0.168
0.043
0.057
-
0.228
0.16
0.165
0.104
0.043
0.043
0.052
0.045
0.043
0.145
-0.262
0.13
0.031
0.047
0.009
-0.141
-0.106
0.125
0.09
0.031
0.031
0.05
0.047
0.031
0.014
0.215
40
Number
of
Points
85
85
83
90
95
99
July
22,
2005
July
28,
2005
July
29,
2005
July
31,
2005
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
1 km
Downscale
d
AMSR-E
NLDAS
Mesonet
-0.146
0.16
0.035
0.051
-
-0.027
-0.108
0.153
0.112
0.035
0.035
0.05
0.046
0.035
0.059
0.345
0.146
0.043
0.076
0.01
0.305
0.217
0.137
0.098
0.044
0.044
0.058
0.047
0.045
0.017
0.209
0.295
0.166
0.039
0.073
-
0.16
0.098
0.159
0.119
0.039
0.039
0.047
0.034
0.039
0.031
-0.077
0.155
0.032
0.051
-
-0.067
-0.255
0.145
0.111
0.032
0.032
0.056
0.048
0.032
0.036
41
102
103
79
106
Table 2.3c. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Oklahoma Mesonet in August: RMSE, unbiased RMSE, bias
and standard deviation are in m3/m3.
Day
August 2,
2005
August 4,
2005
August 9,
2005
August
24, 2005
August
25, 2005
Dataset
1 km
Downsc
aled
AMSRE
NLDAS
Mesonet
1 km
Downsc
aled
AMSRE
NLDAS
Mesonet
1 km
Downsc
aled
AMSRE
NLDAS
Mesonet
1 km
Downsc
aled
AMSRE
NLDAS
Mesonet
1 km
Downsc
aled
AMSRE
NLDAS
Mesonet
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
Spatial
Standard
Deviation
(m3/m3)
pvalue
0.056
0.164
0.036
0.036
-
0.077
0.16
0.036
0.051
-
-0.307
0.124
0.035
0.051
0.053
0.015
-0.346
0.143
0.025
0.026
0.045
-0.46
0.138
0.025
0.033
0.067
-1.17
0.106
0.023
0.035
0.059
0.427
0.119
0.167
0.041
0.041
-
0.043
0.162
0.041
0.049
-
0.051
0.113
0.041
0.051
0.046
0.086
0.393
0.147
0.047
0.055
0.073
0.315
0.144
0.047
0.05
0.085
0.316
0.064
0.047
0.047
0.034
0.751
0.377
0.118
0.05
0.065
0.235
0.366
0.122
0.05
0.049
0.207
0.121
0.062
0.054
0.049
0.046
0.823
42
Numbe
r of
Points
80
26
51
34
27
August
30, 2005
1 km
Downsc
aled
AMSRE
NLDAS
Mesonet
0.601
0.18
0.024
0.039
0.003
0.38
0.173
0.025
0.052
0.003
0.666
0.105
0.024
0.05
0.051
0.367
44
Table 2.4. Monthly generalized comparison statistics between the 1 km downscaled,
NLDAS and AMSR-E soil moisture compared to the Oklahoma Mesonet for three months:
RMSE, unbiased RMSE, bias and standard deviation are in m3/m3.
Day
May 2004
Dataset
1 km
Downscaled
AMSR-E
NLDAS
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
Spatial
Standard
Deviation
(m3/m3)
0.379
0.124
0.045
0.058
0.376
0.500
0.120
0.107
0.044
0.040
0.056
0.066
Mesonet
July 2005
August
2005
Total
1 km
Downscaled
AMSR-E
NLDAS
Mesonet
1 km
Downscaled
AMSR-E
NLDAS
Mesonet
1 km
Downscaled
AMSR-E
NLDAS
Mesonet
Number
of Points
606
0.051
0.078
0.160
0.043
0.059
0.057
0.047
0.153
0.106
0.044
0.043
0.050
0.043
0.044
0.200
0.153
0.037
0.044
0.120
-0.054
0.150
0.096
0.037
0.037
0.047
0.047
0.045
0.219
0.146
0.042
0.054
0.184
0.164
0.141
0.103
0.042
0.040
0.051
0.052
0.047
43
927
262
1795
Table 2.5a. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Little Washita soil moisture observations in May. RMSE,
unbiased RMSE, bias and standard deviation are in m3/ m3 ("-" is for the value less than
0.01).
Day
May 4,
2004
May 6,
2004
May 8,
2004
May
15,
2004
May
22,
2004
May
23,
2004
Dataset
1 km
Downsca
led
AMSR-E
NLDAS
Micronet
1 km
Downsca
led
AMSR-E
NLDAS
Micronet
1 km
Downsca
led
AMSR-E
NLDAS
Micronet
1 km
Downsca
led
AMSR-E
NLDAS
Micronet
1 km
Downsca
led
AMSR-E
NLDAS
Micronet
1 km
Downsca
led
AMSR-E
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
Spatial
Standard
Deviation
(m3/m3)
pvalue
0.083
0.044
0.043
0.015
0.882
0.083
0.158
0.041
0.048
0.031
0.04
0.005
0.016
0.044
0.891
0.835
0.1
0.052
0.041
0.015
0.785
0.1
0.173
0.041
0.058
0.026
0.036
0.006
0.014
0.042
0.863
0.729
0.346
0.077
0.017
0.012
0.662
0.165
0.145
0.069
0.078
0.009
0.022
0.004
0.011
0.023
0.686
0.660
0.149
0.037
0.017
0.01
0.889
0.065
0.05
0.035
0.036
0.033
0.004
0.01
0.036
0.810
0.912
0.526
0.059
0.018
0.021
0.719
0.034
0.284
0.051
0.039
0.014
0.014
0.001
0.007
0.021
0.746
0.794
0.459
0.068
0.018
0.019
0.614
Number
of
Points
9
9
9
9
6
8
0.084
0.055
0.011
44
0.002
0.673
May
24,
2004
May
30,
2004
NLDAS
Micronet
1 km
Downsca
led
AMSR-E
NLDAS
Micronet
1 km
Downsca
led
AMSR-E
NLDAS
0.16
0.046
0.017
0.006
0.02
0.711
0.812
0.075
0.013
0.02
0.577
0.013
0.179
0.063
0.045
0.008
0.015
0.002
0.006
0.017
0.626
0.702
0.438
0.092
-
0.009
0.730
0.002
0.072
0.023
-
0.001
0.770
-
0.842
0.043
Micronet
0.012
45
9
5
Table 2.5b. Comparison statistics between the 1 km downscaled, NLDAS and AMSR-E
soil moisture compared to the Little Washita soil moisture observations in July. RMSE,
unbiased RMSE, bias and standard deviation are in m3/ m3 ("-" is for the value less than
0.01).
Day
Dataset
1 km
Downsca
led
July 3,
AMSR-E
2005
NLDAS
Micronet
1 km
Downsca
led
July 8, AMSR-E
2005
NLDAS
Micronet
1 km
Downsca
led
July 12, AMSR-E
2005
NLDAS
Micronet
1 km
Downsca
led
July 17, AMSR-E
2005
NLDAS
Micronet
1 km
Downsca
led
July 22, AMSR-E
2005
NLDAS
Micronet
1 km
Downsca
July 28,
led
2005
AMSR-E
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
Spatial
Standard
Deviation
(m3/m3)
pvalue
0.237
0.066
0.064
0.028
0.747
0.039
0.095
0.049
0.061
0.058
0.05
0.005
0.004
0.066
0.775
0.832
0.224
0.065
0.048
0.030
0.813
0.08
0.168
0.039
0.021
0.039
0.044
0.004
0.010
0.050
0.669
0.810
0.032
0.049
0.029
0.025
0.864
-0.007
-0.013
0.02
0.033
0.023
0.03
0.003
0.005
0.032
0.840
0.785
0.176
0.05
-
0.026
0.825
0.228
0.033
-
0.002
0.871
0.16
0.057
-
0.002
0.806
Numbe
r of
Points
9
9
9
6
0.011
-0.146
0.055
0.009
0.013
0.759
-0.027
-0.108
0.029
0.051
0.015
0.017
0.005
0.001
0.009
0.813
0.686
0.345
0.084
-
0.030
0.725
0.305
0.06
-
0.003
0.750
46
9
5
NLDAS
Micronet
1 km
Downsca
led
July 29, AMSR-E
2005
NLDAS
Micronet
1 km
Downsca
led
July 30, AMSR-E
2005
NLDAS
Micronet
0.217
0.085
-
0.004
0.020
0.683
0.295
0.048
0.012
0.027
0.787
0.16
0.098
0.042
0.07
0.012
0.014
0.015
0.002
0.005
0.759
0.621
-0.204
0.08
0.002
0.034
0.036
-0.210
-0.238
0.053
0.056
-
0.007
0.002
0.016
0.036
0.036
47
9
5
Table 2.6. Monthly generalized comparison statistics between the 1 km downscaled,
NLDAS and AMSR-E soil moisture compared to the Little Washita soil moisture
observations for three months. RMSE, unbiased RMSE, bias and standard deviation are in
m3/ m3.
Day
May
2004
July
2005
Total
Dataset
1 km
Downscaled
AMSR-E
NLDAS
Micronet
1 km
Downscaled
AMSR-E
NLDAS
Micronet
1 km
Downscaled
AMSR-E
NLDAS
Micronet
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
Spatial
Standard
Deviation
(m3/m3)
0.364
0.063
0.024
0.015
0.080
0.145
0.055
0.049
0.020
0.025
0.003
0.010
0.027
0.120
0.062
0.027
0.027
0.071
0.047
0.041
0.054
0.029
0.031
0.006
0.004
0.028
0.242
0.063
0.026
0.021
0.076
0.096
0.048
0.052
0.025
0.028
0.005
0.007
0.028
48
Number of
Points
64
61
125
Figure 2.1. Imagery Maps of study region of Oklahoma and the Little Washita
Watershed and the locations of the Mesonet Stations are denoted in open yellow
circles and the soil moisture sites for Little Washita are noted in red dots.
49
Figure 2.2. Maps of Variables used in the soil moisture downscaling algorithm from July
21, 2005, over Oklahoma (a) MODIS Aqua 1 km land surface temperature during the day;
(b) MODIS Aqua 1 km land surface temperature at night; (c) 1/4o spatial resolution
AMSR-E soil moisture; (d) 1/8o spatial resolution NLDAS soil moisture; (e) MODIS
Aqua 1 km NDVI.
50
Figure 2.3. (a) Top shows the various elements in the disaggregation procedure (b)
Bottom shows construction of the lines corresponding to constant NDVI between
average soil moisture and change in surface temperature.
51
Figure 2.4. Data flow of the soil moisture downscaling algorithm.
52
Figure 2.5. Daily temperature difference versus daily average soil moisture
corresponding to (Latitude: 101.875oW~102oW; Longitude: 35.125oN ~35.625oN)
and different NDVI values for May, August and July.
53
Figure 2.6. Maps of the NLDAS, AMSR-E and 1 km soil moisture (m3/m3) from May
22, 2004, July 17, 2005 and August 9, 2005 in Oklahoma.
54
Figure 2.7. Maps of the NLDAS, AMSR-E and 1 km soil moisture (m3/m3) from May
22, 2004, July 17, 2005 and August 2, 2005 in Little Washita.
55
Figure 2.8. Time-series maps of 1 km soil moisture (m3/m3) of five days on July 2005
show the dry-down tendency in Little Washita region.
56
Figure 2.9a. Overall scatter plots of NLDAS, AMSR-E and 1 km soil moisture versus
the Little Washita Micronet soil moisture observations for all the months.
57
Figure 2.9b. Overall scatter plots of NLDAS, AMSR-E and 1 km detrended soil
moisture versus the Little Washita Micronet soil moisture observations for all the months.
58
Figure 2.10. Overall scatter plots of spatial standard deviation of NLDAS, AMSR-E and
1 km soil moisture versus the Oklahoma Mesonet and Little Washita Micronet soil
moisture observations for all the months.
59
CHAPTER III AMSR-E SOIL MOISTURE DISAGGREGATION USING
SOIL EVAPORATION EFFICIENCY MODELS2
3.1 ABSTRACT
High spatial resolution soil moisture is necessary for hydrological, agricultural and
environmental related research. In this study, a soil moisture retrieval algorithm based on
soil evaporation efficiency models was applied for disaggregating Advanced Microwave
Scanning Radiometer (AMSR-E) soil moisture. Moderate Resolution Imaging Spectroradiometer (MODIS) products at 1 km resolution, including Land Surface Temperature
(LST) and Normalized Differential Vegetation Index (NDVI), as well as North American
Land Data Assimilation Systems (NLDAS) phase 2 data at 1/8o spatial resolution for soil
temperature of 0-10 cm soil layer were used to develop a bivariate regression analysis for
disaggregating the soil temperature to 1 km. The disaggregated soil moisture was derived
from two soil evaporation efficiency models NP89 and LP92, based on relationships
among soil evaporation efficiency, soil moisture and field capacity and used along with
the 1/4o AMSR-E soil moisture. Four months of growing season between 2004-2005
were used to test this algorithm, and four clear days for each month were picked out and
mapped for comparing with corresponding AMSR-E and NLDAS soil moisture. The
disaggregated soil moisture and the AMSR-E and NLDAS soil moisture were validated
2
Fang, B., & Lakshmi, V. (2014) AMSR-E Soil Moisture Disaggregation Using MODIS
and NLDAS Data. Remote Sensing of the Terrestrial Water Cycle, 277-304, Published by
John Wiley.
60
using ground observations from 9 stations of Little Washita watershed region, located in
southwest Oklahoma, covered by cropland and grassland. The overall results of NP89
and LP92 show the slope of disaggregated soil moisture ranges -0.042~0.056 m3/m3,
RMSE of disaggregated soil moisture ranges 0.042~0.069 m3/m3, unbiased RMSE ranges
0.031~0.055 m3/m3, and spatial standard deviation ranges 0.013~0.017 m3/m3. For the
disaggregated soil moisture, the RMSE and unbiased RMSE are better than NLDAS,
while the RMSE, unbiased RMSE and slope of two out of four months are better than
that for AMSR-E. The spatial standard deviation of disaggregated soil moisture is better
compared to all estimates. This proves that the disaggregated soil moisture maintains the
accuracy as well as demonstrates spatial soil moisture variability in more detail,
especially in the months of July and August.
Keywords: Soil Moisture disaggregation, Soil Evaporation Efficiency Model,
MODIS, NLDAS, AMSR-E
3.2 INTRODUCTION
The land surface is spatially heterogeneous - soils, vegetation and topography and
as a result, the hydrological responses are represented by soil moisture, surface
temperature and evapotranspiration are also spatially variable. A combination of
hydrological models, ground observations have been used for capturing hydrological
variability in both space and time (Jackson et al., 1989; Schmugge et al., 1994). Remote
sensing has a rich heritage of soil moisture mapping (Schmugge et al., 2002; Njoku et al.,
2003). Passive microwave remote sensing is able to obtain soil moisture in near real-time
by simple retrieval algorithms (Njoku et al., 1996; Njoku et al., 1999), especially with the
61
high sensitivity of L-band to soil moisture. However, due to the limitation of satellite
wavelength of L-band and antenna diameter, the spatial resolution of L-band microwave
remote sensing imagery is restricted to tens of kilometer scales (Schmugge et al., 1974;
Schmugge et al., 1994).
The Soil Moisture and Ocean Salinity (SMOS) and Soil Moisture Passive/Active
(SMAP) are the first two satellites dedicated for soil moisture monitoring using L-band
radiometers. The SMOS mission satellite, launched by European Space Agency (ESA) in
2009, provides soil moisture observations of 40 km on average at L-band (Kerr et al.,
2012). The NASA SMAP mission satellite is designed and to be launched in 2014 and
will provide the microwave radiometer for soil moisture product at 36 km and radar soil
moisture at 3 km (Entekhabi et al., 2010). The Advanced Microwave Scanning
Radiometer (AMSR2) onboard Global Change Observation Mission (GCOM)-W1
measures microwave energy from land surface and provides high accurate soil moisture
product of two spatial resolutions 0.1/0.25 degrees (10/25km), with six frequency bands
between 7-89 GHz and revisits time of 1-2 days (Imaoka, Kachi et al., 2010;
http://www.jaxa.jp/projects/sat/gcom_w/index_e.html). The coarse soil moisture can be
disaggregated by combining radiometer and radar data and/or land surface model outputs,
based on the relationships among vegetation cover, soil moisture and surface temperature
(Lakshmi et al., 1997; Narayan et al., 2004; Carlson et al., 2007; Narayan and Lakshmi,
2008; Das et al., 2010; Piles et al., 2011; Merlin et al., 2013; Fang et al., 2013). In order
to improve the soil moisture downscaling algorithms derived from the relationships
between soil moisture, land surface temperature variation and vegetation, there are
several issues which should be considered. Soil property data sets are seldom available at
62
fine spatial scales and detailed vegetation information can be time consuming to generate
at appropriate resolutions. Consequently, the land-atmosphere interactions related
variables, such as evaporation can be used to improve the soil moisture downscaling
algorithms.
Previous studies have tried to estimate soil evaporation using both mechanical and
physical methods (Mahfouf and Noilhan, 1991; Chanzy and Bruckler, 1993; Yamanaka
et al., 1998; Brunsell et al., 2011). Other studies have suggested methods to retrieve soil
evaporation by using remote sensed soil moisture (Nishida et al., 2003; Zhang et al.,
2003). The surface skin temperature has been used to estimate soil evaporation (Nishida
et al., 2003; Kustas et al., 1993); Merlin et al. applied soil efficiency models to SMOS
data for retrieving and then downscaling soil moisture products derived from microwave
radiometry (Merlin et al., 2010a).
The mechanical based methods work generally better than the physical methods,
since soil evaporation can be expressed using simple formulations. Two types of
mechanical methods for estimating soil evaporation have been developed. The first
method utilizes the resistance of vapor diffusion rss , which uses soil surface temperature
to estimate soil evaporation (Monteith, 1981; Camillo and Gurney, 1986; Passerat de
Silans, 1986; Kondo et al., 1990; Sellers et al., 1992; Daamen and Simmonds, 1996;
Merlin et al., 2010b). However, this method is only applicable when water flow is
determined by vapor transport diffusion. The second method is using the soil evaporation
efficiency to estimate the soil evaporation (Merlin et al., 2010a).
In this study we aim to (1) disaggregate the NLDAS derived soil temperature by
the relationships among soil temperature (0-10 cm layer), land surface temperature and
63
fractional vegetation cover; (2) derive the soil evaporation efficiency variable of Merlin
et al., 2010a using the disaggregated soil temperature as well as calculate the soil
moisture from two soil evaporation models (Noilhan and Planton, 1989; Lee and Pielke,
1992) to disaggregate AMSR-E soil moisture.
3.3 DATA
3.3.1 LITTLE WASHITA RIVER WATERSHED MICRONET
Little Washita River Watershed Micronet is a valuable resource for soil moisture
validation. The Little Washita River Watershed is located in southwestern Oklahoma,
which is composed of 20 Micronet stations within a 625 km2 region (Basara et al., 2009;
McPherson et al., 2007; Cosh et al., 2004) and measures soil moisture at the near surface,
providing an accurate estimate of the 0-5 cm soil layer. The locations of Micronet
stations are shown in Figure 3.1.
For the length of this study, there were 9 stations of Micronet soil moisture
observations consistently available to provide the large scale surface estimate. The
nearest network average available at 1:30 PM of Little Washita local time was used for
comparison. The reliable stations were geolocated for comparison with the disaggregated
as well as AMSR-E, NLDAS soil moisture data. Figure 3.2 shows the averaged and
standard deviation values of the soil moisture from the 9 Micronet stations in Little
Washita of the four months in this study. It is observed that the averaged soil moisture
has a similar trend as the standard deviation of soil moisture.
3.3.2 NLDAS DATA
NLDAS (North American Land Data Assimilation System,
64
http://ldas.gsfc.nasa.gov/nldas/) phase-2 hourly forcing data derived from NOAA’s
(National Oceanic and Atmospheric Administration) Noah land surface model, which
was developed for the NCEP (National Centers for Environmental Prediction)
mesoscale Eta Model and soil hydraulic properties dataset were used. NLDAS provides
real-time land surface models output data at 1/8o (12.5 km) spatial resolution for
variables such as surface temperature, solar radiation, vegetation indices and soil
wetness. In this study, three variables were used to retrieve 1 km resolution soil moisture.
These included skin surface temperature, which refers to the averaged top surface
temperature of vegetation, bare soil and snow, soil temperature at 0-10 cm depth, and
field capacity. The data was processed at Oklahoma local time 1:30 PM for matching up
with Aqua satellite overpass. The NLDAS dataset has been extensively validated
(Lohmann et al., 2004; Robock et al., 2003; Schaake et al., 2004).
3.3.3 MODIS DATA
MODIS on-board Aqua satellite has 36 spectral bands covering visible and
infrared bands and provides 44 products. In this study, MODIS products, including
daytime (local time 1:30 PM) daily LST (MYD11A1) and biweekly NDVI (MYD13A2)
of the most consecutive cloud-free days on May 2004 and June-August 2005 were
selected. For acquiring daily NDVI, the biweekly NDVI imageries of these months were
collected and fitted using sinusoidal regression among each MODIS pixel to interpolate
for the daily NDVI.
3.3.4 AMSR-E DATA
Soil moisture is retrieved from L-band microwave observations using the
Advanced Microwave Scanning Radiometer on board the Aqua satellite (AMSR-E) using
65
the Single Channel Algorithm (SCA) (Jackson et al., 1993; Jackson et al., 1999; Jackson
et al., 2010; Njoku et al., 2003). The AMSR-E soil moisture is produced at 1/4o resolution
and covers the years from 2002 to 2011. Previous research for AMSR-E soil moisture
validation has shown that the error of retrieved soil moisture is less than 0.1 m3/m3 in low
vegetation areas. (Njoku and Li, 1999; Njoku et al., 2006; Mladenova et al., 2011).
In this study, the footprint based AMSR-E level-2 soil moisture retrieval was geolocated and gridded at 1/4o spatial resolution by averaging all the soil moisture values
within each AMSR-E pixel.
The variables used in this study, including MODIS 1 km LST, 1 km NDVI,
NLDAS 1/8o soil temperature, soil moisture and field capacity, AMSR-E 1/4o soil
moisture for July 22, 2005 are shown for the Little Washita watershed in Figure 3.3.
3.4 METHODOLOGY
Several assumptions were made in this study: (a) The land surface temperature is
a linear combination of soil temperature and vegetation temperature. These are
proportional to fractional photo-synthetic vegetation cover (Gutman and Ignatov, 1998;
Merlin et al., 2010b). (b) The top layer soil moisture is a function of soil evaporation
efficiency and field capacity. Previous experiments have shown similar non-linear
relationships under different soil depths and soil types, and the models K89 and LP92 are
applicable for the soil moisture disaggregation model building (Komatsu et al., 2003; Lee
and Pielke, 1992; Noilhan and Planton, 1989) (c) We assume that the field capacity
among each NLDAS scale pixel is homogeneous and variation at the 1 km scale is not
accounted for in this study.
66
The methodology will be presented in three subsections: (a) The method to
disaggregate NLDAS soil temperature Tsoil to 1 km by using MODIS LST and NDVI (b)
The 1 km resolution θ will be calculated from the two soil evaporation efficiency models
K03 and LP92 using the disaggregated Tsoil. (c) The disaggregated 1 km θ will be used to
correct AMSR-E soil moisture retrievals.
3.4.1 SOIL TEMPERATURE DISAGGREGATION
The land surface temperature from satellite Trad can be expressed as linear
combination of soil temperature Tsoil ⁡and vegetation temperature Tveg , as
Trad = (1 − fG98 )Tsoil + fG98 Tveg
Equation 3.1: The combination of land surface temperature
Where, fG98 is fractional vegetation cover using Gutman and Ignatov, 1998 method
calculated using MODIS derived NDVI as
fG98 =
NDVI − NDVIS
NDVIv − NDVIS
Equation 3.2: Fractional vegetation cover calculation
Where, NDVIv and NDVIs ⁡are NDVI values corresponding to fully vegetated and bare
soil, respectively. In this study we denoted the minimum NDVI value of imagery as
NDVIs and the maximum NDVI value of imagery as NDVIv .
High spatial resolution vegetation temperature is usually difficult to obtain.
Therefore a LST disaggregation method using fractional vegetation cover and surface
temperature at high resolution (Merlin et el., 2010) was applied to NLDAS soil
67
temperature in this study. Bivariate regression analysis among three parameters Trad , Tsoil
and fG98 is expressed as
Tsoil = a × fG98 + b × Trad + c
Equation 3.3: Regression fitting for soil temperature
Where, a, b and c are the fitting coefficients.
The variation of Tsoil of the MODIS pixels among each NLDAS pixel can be
adjusted by the variations of fG98 and Trad , as
∆Tsoil =
∂Tsoil
∂Tsoil
∆fG98 +
∆T
∂fG98
∂Trad rad
Equation 3.4: ∆Tsoil calculation using ∆fG98 and ∆Trad
Where, ∆Tsoil, ∆fG98 and ∆Trad are the differences of Tsoil , fG98 and Trad between
MODIS and disaggregated NLDAS pixels,
∂Tsoil
∂fG98
∂T
and ∂Tsoil are the partial derivatives of
rad
Tsoil to fG98 and Trad .
This regression model was derived at 1/8o NLDAS scale and applied to fG98 and
Trad at 1 km to calculate the total derivative ∆Tsoil . ∆Tsoil , that was then added up to each
MODIS pixel to obtain corrected 1 km Tsoil.
3.4.2 SOIL EVAPORATION EFFICIENCY MODEL
Soil evaporation efficiency β can be defined as a ratio of soil evaporation rate
between soil surface and water surface. The standard equation of soil evaporation
efficiency (Komatsu et al 2003) is expressed as
68
E = βρa Δq/ra
Equation 3.5: Standard equation of soil evaporation efficiency
Where, E is the evaporation rate of water from soil surface, Δq is the specific humidity
difference between air and soil surface, ρa is the mass density of air and ra is evaporative
resistance.
The soil evaporation efficiency β at 1 km is estimated using disaggregated soil
temperature T1km by
β=
Tmax − T1km
Tmax − Tmin
Equation 3.6: Soil evaporation efficiency expressed by soil temperature
Where Tmax and Tmin are the maximum and minimum soil temperatures of the imagery,
respectively.
Practically these parameters are difficult to be measured and collected on a largescale region. Consequently, simplified soil evaporation efficiency model was developed
as (Komatsu et al 2003)
β = (1 − exp(−θmm /θc )
Equation 3.7: Simplified soil evaporation efficiency by Komatsu et al 2003
Where, θmm is experimental soil moisture data at 1~3 mm and θc is a semi-empirical
parameter field capacity.
The θc is determined by humidity, soil type and wind speed, defined as
69
ref
rah
θc = θc0 (1 +
)
rah
Equation 3.8: Semi-empirical parameter field capacity
ref
Where, θc0 is a soil dependent parameter ranging between 0.01~0.04 vol/vol, rah
is the
reference aerodynamic resistance, while rah is the aerodynamic resistance (Komatsu et al.,
2003).
However, this model is only applicable for the soil layer depth of several
millimeters. In recent years several semi-empirical soil evaporation efficiency models
have been developed for soil layer of centimeter scale, which are consistent with the
sensing depth of NLDAS as:
(a) NP89 method (Noilhan and Planton, 1989)
βmod,NP89 = (0.5 − 0.5cos(πθ/θC,NP89 )
Equation 3.9: Soil evaporation efficiency model by Noilhan and Planton, 1989
Where θC,NP89 is the soil moisture field capacity of NP89 experiment and θ is the
measured soil moisture. This model was derived from the comparison of relative
humidity, soil moisture and field capacity.
(b) LP92 method (Lee and Pielke, 1992)
βmod,LP92 = [0.5 − 0.5cos⁡(πθ/θC,LP92 )]2
Equation 3.10: Soil evaporation efficiency model by Lee and Pielke, 1992
70
Where θC,LP92 is the soil moisture field capacity of LP92 experiment and θ is measured
soil moisture. This model was derived from several numerical atmospheric models and is
particularly suitable for sand or silt loam soil.
Both of the soil evaporation efficiency models correspond to various soil depth
and types, including agricultural soil, sand and cornstarch. The β-θ curves under different
soil types demonstrate similar patterns. Figure 3.4 shows the β-θ curves of these two
models when θC equals to 0.35.
To determine the field capacity θC , NLDAS provides gridded θC at 1/8 degree
covering the whole Oklahoma. The NLDAS θC was resampled to 1 km scale by
comparing the coordinates of NLDAS and MODIS pixels to match up with 1 km soil
temperature pixels.
3.4.3 AMSR-E SOIL MOISTURE CORRECTION
The AMSR-E soil moisture was disaggregated using the calculated 1 km daily
soil moisture θ from model NP89 and LP92. Among each 1/4o AMSR-E pixel, the
difference between AMSR-E soil moisture Θ and averaged 1 km disaggregated soil
moisture θ within this AMSR-R pixel boundary was added to each 1 km pixel. These
cloud contaminated or no-data pixels were discarded.
The AMSR-E soil moisture correction equation is given by
θc (i, j) = θ(i, j) + [Θ −
1
∑ θ(i, j)]
N i,j
Equation 3.11: AMSR-E Soil moisture correction
71
Where θc (i, j) is 1 km corrected soil moisture, θ(i, j) is soil evaporation efficiency model
output 1 km soil moisture, Θ is AMSR-E descending orbit soil moisture and N is the
number of 1 km pixels among each AMSR-E pixel.
3.5. RESULTS AND DISCUSSIONS
3.5.1 CORRELATION ANALYSIS
Figure 3.5 (a) shows the bivariate correlation between the MODIS daily 12.5 km
land surface temperature and fractional vegetation cover as well as aggregated 12.5 km
NLDAS soil temperature, of four days from May 20, 2004, June 26, 2005, July 22, 2005
and August 9, 2005 covering Oklahoma, while Figure 3.5 (b) displays the relationship
between surface soil temperature and fractional vegetation cover, and Figure 3.5 (c)
shows the relationship between surface soil temperature and surface temperature. All the
scatter plots show good correlation between the three variables and inverse relationships
of surface soil temperature-surface temperature and surface-temperature-fractional
vegetation cover. The R2 of fG98 - Tsoil and fG98 - Trad for May 20, 2004 are 0.73 and 0.82,
for June 26, 2005 are 0.38 and 0.669, for July 22, 2005 are 0.318 and 0.357 and August 9,
2005 are 0.268 and 0.398, respectively. The bivariate correlation coefficients for the other
days in the study period show similar values.
3.5.2 SOIL VARIABLES MAPPING
In Figure 3.6 (a-b), soil variables, including soil evaporation efficiency,
disaggregated 1 km soil temperature and NLDAS soil moisture of four days with low
cloud cover were chosen to examine soil moisture dry-down patterns. The west most and
northeast parts of the watershed correspond to high soil temperature and low soil
72
evaporation efficiency, while the center part of the basin corresponds to lower soil
temperature and higher soil evaporation efficiency. The regional variations in
temperature responses to seasonal change are better displayed in the 1 km than AMSR-E
soil moisture.
Corresponding to these days, the soil moisture at different resolutions: 1 km
disaggregated soil moisture output from NP89 and LP92 models, AMSR-E and NLDAS
were mapped (Figure 3.7, a-b). Compared with the coarser spatial resolution soil
moisture from AMSR-E and NLDAS, which contain only 3-7 pixels, the disaggregated
results demonstrate soil moisture spatial distribution pattern with far more detail and have
around 625 pixels of 1km. However, the difference of soil moisture ranges amongst the
three scales must be noted. NLDAS soil moisture values are higher than the
disaggregated and AMSR-E soil moisture. A major reason is the mismatch of soil sensing
depth between microwave remote sensing imagery and land surface model output. The
microwave satellite sensor can only penetrate soil layer of 0-2 cm depth, matching the
sensing depths of the two experimental soil evaporation efficiency models NP89 and
LP92, while the NLDAS soil moisture is at 0-10 cm recording the soil water content
corresponding to a deeper layer. Moreover, an unnatural transition of soil moisture
distribution pattern in southeast portion of the watershed could be observed, due to the
coarse resolution of NLDAS derived field capacity. It also can be observed that the soil
moisture of July 22, 2005 shows the least spatial variation compared to the other days,
while June 26, 2005 shows the most variability. If comparing the days shown in Figure
3.6 and 7 with soil moisture dry down curves in Figure 3.2, soil moisture of May 20,
2004, June 26, 2005 and August 9, 2005 was showing apparent drying trend, while the
73
soil moisture of July 22, 2005 was temporally invariant. For June 26, 2005, the soil
moisture curve showed a very rapid drying trend.
3.5.3 VALIDATION WITH MICRONET DATA
Table 3.1-3.4 display the statistics of 1 km disaggregated soil moisture derived
from NP89 and LP92, AMSR-E and NLDAS validated with Little Washita Micronet
observations and includes slope, RMSE, unbiased RMSE and spatial standard deviation.
Table 3.5 shows the overall averaged statistical variables of each month. We set the
threshold of the valid number of point pair as 5 and the days with a lesser number of
point pairs were dropped. The bar plots of the statistical values of Table 3.1-3.5 are also
shown in Figure 3.8 (a-e).
The statistical variables are listed as below.
m=
∆θ̂
∆θ
Equation 3.12: Slope of linear regression between estimated θ̂ and in-situ θ
∑n (θ̂i − θi )2
RMSE = √ i=1
n
Equation 3.13: RMSE between estimated θ̂ and in-situ θ
RMSE = √
∗
∗
∑ni=1(θ̂i − θi )2
n
Equation 3.14: Unbiased RMSE between predicted θ̂* from linear regression and in-situ
θ
σ=√
∑ni=1(θ̂i − θ̅)2
n
74
Equation 3.15: Spatial standard deviation between estimated θ̂ and in-situ θ
Where, m is the slope, which is the ratio of estimated soil moisture θ̂ (1 km disaggregated,
AMSR-E and NLDAS) to Micronet moisture observations θ (Micronet). RMSE* is
∗
unbiased RMSE and θ̂i is the predicted soil moisture derived from linear regression
fitting between θ̂ and θ. σ is spatial standard deviation of each soil moisture dataset.
From Table 3.1-3.5 and Figure 3.8 (a-e) we can summarize as follows: (a) The
slopes of all four soil moisture sources are biased – i.e. the slopes are different from 1
when compared with the Micronet soil moisture. The slope of the disaggregated soil
moisture ranges -0.042~0.056 m3/m3. Specifically, for July 2005 and August 2005, slopes
of NP89 and LP92 are better than NLDAS and closer to AMSR-E. (b) The RMSE and
unbiased RMSE of disaggregated soil moisture range 0.042~0.069 m3/m3 and
0.031~0.055 m3/m3, respectively. For May 2004 and June 2005, RMSE and unbiased
RMSE for disaggregated soil moisture of NP89 and LP92 are better than NLDAS and
slightly worse than AMSR-E. For July 2005 and August 2005, RMSE for the two model
disaggregated soil moistures are better than both AMSR-E and NLDAS. (c) The spatial
standard deviation of disaggregated soil moisture ranges 0.013~0.017. Comparing the
spatial standard deviation of each soil moisture dataset with ground observations - NP89
and LP92 methods disaggregated soil moisture are closer to Micronet soil moisture
measurements than AMSR-E and NLDAS. (d) Comparing the accuracy and spatial
variation between the two evaporation efficiency models, LP92 has better RMSE but a
worse unbiased RMSE than NP89, while the spatial standard deviation is always better
than NP89.
75
Figure 3.9 shows that NP89 and LP92 have much more variance than AMSR-E
and NLDAS. AMSR-E and NLDAS comparison results appear tightly grouped but are
obviously biased. In Figure 3.10 the correlation of all the detrended soil moisture
comparisons with Micronet are greatly enhanced (as compared to the raw data in Figure
3.9). It should be pointed out that the NLDAS forcing data has been scaled to ground
observations (Luo et al., 2003), so the scatter plot of its correlation with Micronet soil
moisture in Figure 3.9-3.10 tends to be more convergent than the disaggregated and
AMSR-E soil moisture, but biased wet. This explains the reason as to the unbiased
RMSE of NLDAS being significantly smaller and closer to NP89 and NP92. The Figure
3.11 compares the spatial standard deviation between Micronet and the four soil moisture
datasets. It can be noted that all the soil moistures are biased low compared to the Little
Washita Micronet. However, the comparisons of NP89 and LP92 spatial standard
deviation are definitely better than AMSR-E and NLDAS as they are closer to the
corresponding Micronet spatial standard deviation.
Two factors significantly influence soil moisture validation: sensing depth and
scale. (a) The mismatch of the sensing depth influences the soil temperature
disaggregation. The 0-10 cm soil layer NLDAS soil temperature was downscaled and
corrected using MODIS LST and NDVI, as well as AMSR-E soil moisture with sensing
depth of a few of millimeters. Another sensing depth inconsistency comes from soil
moisture validation. Micronet soil moisture observations are at 0-5 cm depth, while
microwave sensor retrieved AMSR-E soil moistures corresponds to less than 1 cm depth
and the disaggregated 1 km soil temperature is at 0-10 cm. (b) The mismatch of soil
moisture spatial scales also needs to be discussed. The Little Washita Micronet records
76
soil moisture observations at a point scale, while the disaggregated, remotely sensed and
land surface model output soil moisture corresponds to kilometer scale. The bias of soil
moisture comparison as well as the spatial standard deviation comparison between
Micronet observations and estimated soil moistures can be noted (Figure 3.9, 3.11).
The validation results prove that the soil evaporation efficiency disaggregated soil
moisture is a compromise between accuracy and spatial variation. It retains the accuracy
of satellite soil moisture, but also significantly improves the variability from the other
sources.
3.6. CONCLUSIONS
In this study the AMSR-E soil moisture was disaggregated through three steps.
Firstly the 1/8o spatial resolution NLDAS soil temperature of top soil layer was
disaggregated to 1 km by using a bivariate regression analysis among top layer soil
temperature, land surface temperature and fractional vegetation cover and the total
derivative of land surface temperature and fractional vegetation cover to soil temperature
was used for correcting the disaggregated soil temperature. The 1 km soil temperature
was then used for estimating soil evaporation efficiency and retrieving 1km soil moisture
from two experimental soil evaporation efficiency models. Lastly in order to eliminate
the inconsistency among different sensors, including AMSR-E and MODIS as well as
land surface models output, the disaggregated 1 km soil moisture was bias-corrected
using AMSR-E soil moisture. Table 3.6 compares this study to some soil moisture
disaggregation studies. It can be concluded that the RMSE and unbiased RMSE obtained
in this study are as good as most of the previous studies and better than some of them,
77
such as Piles et al., (2011) and Fang et al., (2013). An obvious improvement of this study
is it estimates instantaneous soil moisture rather than using averaged value of two instant
times as daily average from Fang et al., (2013). Another advantage is this study improved
the soil moisture to 1 km scale as well as a daily temporal coverage and this has not been
achieved in most other studies.
The study results demonstrate that (1) Strong correlations among top layer soil
temperature, land surface temperature and fractional vegetation cover and soil
temperature variation were discovered and can be used to disaggregate soil temperature
to a 1km scale. (2) The disaggregated 1 km soil moisture maps definitely captures the
spatial distribution pattern and variation in greater detail than AMSR-E and NLDAS. (3)
NLDAS soil moisture was found apparently wet biased (Mo et al., 2012), mostly due to
the mismatch of the soil layer depth. (4) Validation using the Little Washita River
Watershed Micronet soil moisture, it can be seen that the NP89 and LP92 1 km soil
moistures sustain accuracy as well as improves spatial variation. Although the slope of
disaggregated soil moisture of two months are better than either AMSR-E or NLDAS, the
spatial standard deviation of disaggregated soil moisture output from NP89 and LP92 is
always better than AMSR-E and NLDAS, while the RMSE is better than NLDAS and is
close to AMSR-E.
However this study has various limitations that need to be addressed. Firstly, our
study is based on a couple of assumptions, so mismatches and uncertainties exist during
the soil temperature and soil moisture disaggregation. (a) The dataset for this algorithm
come from the sources with different sensing depths. MODIS and AMSR-E respond to
soil depths of a couple of millimeters, while NLDAS data is at 0-10 cm and Micronet
78
observation is at 0-5 cm. (b) Vegetation temperature at fine spatial scales is difficult to
obtain, the bivariate regression analysis was used and we also presumed straight-forward
linear relationship between top layer soil temperature (0-10 cm), land surface temperature
and fractional vegetation cover. (c) Soil evaporation efficiency was estimated by an
empirical equation using disaggregated soil temperature (d) Field capacity data used in
this study is at 1/8o and the spatial variation between 1/8o and 1 km has been ignored. (e)
The inconsistencies of imagery scales among MODIS, AMSR-E and NLDAS would
cause mismatch of pixels between different spatial resolutions. These issues will be
addressed in future work that will focus on solving inconsistencies and uncertainties on
scale and sensing depth, as well as consider using other newer sources of remotely sensed
soil moisture, such as SMOS and SMAP.
79
3.7 TABLES AND FIGURES
Table 3.1. Soil moisture validation variables: slope, RMSE, unbiased RMSE and spatial
standard deviation of downscaled soil derived from NP89 and LP92 models, AMSR-E
and NLDAS comparing with Little Washita watershed Micronet on May 2004.
Day
May 4,
2004
May 5,
2004
May 6,
2004
May 15,
2004
May 20,
2004
May 23,
2004
May 24,
Dataset
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
0.019
0.016
0.068
0.097
0.045
0.045
0.045
0.108
0.043
0.043
0.044
0.107
0.051
0.05
0.098
0.106
0.044
0.044
0.039
0.123
0.042
0.042
0.039
0.123
0.056
0.052
0.108
0.120
0.049
0.049
0.038
0.136
0.041
0.041
0.030
0.136
0.04
0.044
-0.003
0.048
0.04
0.041
0.057
0.072
0.036
0.036
0.037
0.036
-0.069
-0.129
0.05
0.15
0.056
0.057
0.037
0.111
0.03
0.03
0.03
0.026
0.049
0.006
0.077
0.221
0.065
0.065
0.044
0.119
0.022
0.022
0.022
0.019
0.247
0.079
0.017
80
Spatial
Standard
Deviation
(m3/m3)
0.013
0.013
0.005
0.01
0.043
0.014
0.014
0.006
0.01
0.042
0.015
0.016
0.006
0.01
0.041
0.013
0.014
0.003
0.009
0.037
0.019
0.023
0.002
0.009
0.03
0.015
0.016
0.002
0.009
0.022
0.014
Number
of Points
8
8
8
5
8
7
7
2004
LP92
AMSR-E
NLDAS
Micronet
0.282
-0.013
0.234
0.077
0.06
0.12
81
0.017
0.01
0.017
0.015
0.002
0.009
0.018
Table 3.2. Soil moisture validation variables: slope, RMSE, unbiased RMSE and spatial
standard deviation of downscaled soil derived from NP89 and LP92 models, AMSR-E
and NLDAS comparing with Little Washita watershed Micronet on June 2005.
Day
June 2,
2005
June 4,
2005
June 6,
2005
June 13,
2005
June 15,
2005
June 18,
2005
June 19,
2005
Unbiased
RMSE
RMSE
(m3/m3)
(m3/m3)
Dataset
Slope
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
0.012
-0.001
0.088
0.001
0.052
0.053
0.052
0.104
0.052
0.052
0.041
0.052
0.156
0.185
0.11
0.059
0.043
0.042
0.05
0.113
0.036
0.037
0.025
0.026
-0.044
-0.055
0.013
0.013
0.062
0.063
0.063
0.078
0.057
0.055
0.059
0.058
-0.103
-0.1
-0.072
0.002
0.084
0.081
0.075
0.14
0.057
0.058
0.058
0.14
-0.047
-0.05
-0.028
0.017
0.087
0.085
0.075
0.16
0.057
0.057
0.059
0.061
-0.14
-0.145
-0.077
-
0.066
0.067
0.071
0.075
0.024
0.026
0.031
0.055
-0.096
-0.104
-0.019
-0.002
0.062
0.064
0.069
0.086
0.038
0.04
0.032
0.052
82
Spatial
Standard
Deviation
(m3/m3)
0.016
0.017
0.008
0.01
0.052
0.011
0.014
0.006
0.003
0.05
0.012
0.012
0.011
0.007
0.059
0.025
0.027
0.023
0.013
0.059
0.011
0.011
0.009
0.009
0.062
0.012
0.013
0.01
0.01
0.055
0.013
0.014
0.005
0.01
Number
of Points
7
5
6
5
6
5
5
June 20,
2005
June 22,
2005
June 24,
2005
June 26,
2005
June 27,
2005
June 29,
2005
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
-0.081
-0.103
0.008
0.003
0.051
0.052
0.049
0.113
0.039
0.039
0.045
0.046
-0.093
-0.111
-0.003
0.002
0.045
0.045
0.038
0.138
0.033
0.034
0.037
0.037
0.046
0.051
0.073
0.002
0.047
0.045
0.035
0.147
0.033
0.033
0.026
0.033
-0.061
-0.079
0.067
0.004
0.052
0.051
0.037
0.143
0.03
0.03
0.023
0.03
-0.064
-0.072
0.024
0.005
0.056
0.054
0.039
0.144
0.026
0.027
0.016
0.028
0.036
0.044
0.072
0.003
0.06
0.058
0.044
0.14
0.026
0.027
0.015
0.027
83
0.052
0.012
0.013
0.003
0.01
0.047
0.013
0.014
0.004
0.011
0.037
0.014
0.015
0.004
0.012
0.033
0.016
0.018
0.003
0.013
0.03
0.011
0.012
0.001
0.012
0.028
0.011
0.011
0.002
0.012
0.027
6
6
6
5
6
6
Table 3.3. Soil moisture validation variables: slope, RMSE, unbiased RMSE and spatial
standard deviation of downscaled soil derived from NP89 and LP92 models, AMSR-E
and NLDAS comparing with Little Washita watershed Micronet on July 2005.
Day
July 3,
2005
July 8,
2005
July 10,
2005
July 11,
2005
July 12,
2005
July 20,
2005
July 22,
2005
Dataset
Slope
RMSE
(m3/m3)
Unbiased
RMSE
(m3/m3)
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
0.035
0.035
0.021
0.056
0.064
0.064
0.074
0.099
0.038
0.049
0.059
0.048
0.002
-0.007
0.042
0.057
0.054
0.055
0.066
0.086
0.048
0.048
0.037
0.041
0.258
0.331
0.078
0.076
0.027
0.026
0.035
0.11
0.024
0.024
0.026
0.027
0.035
0.011
0.048
0.037
0.035
0.036
0.039
0.115
0.035
0.036
0.039
0.035
0.075
0.085
0.073
0.048
0.033
0.032
0.034
0.114
0.031
0.031
0.033
0.031
0.085
0.072
0.165
-0.125
0.048
0.046
0.031
0.125
0.02
0.02
0.014
0.02
-0.143
-0.184
0.108
-0.153
0.038
0.037
0.022
0.123
0.019
0.019
0.014
0.019
84
Spatial
Standard
Deviation
(m3/m3)
0.012
0.012
0.004
0.005
0.065
0.012
0.013
0.004
0.007
0.048
0.016
0.02
0.004
0.006
0.031
0.014
0.014
0.004
0.008
0.036
0.012
0.013
0.005
0.008
0.033
0.013
0.013
0.005
0.009
0.021
0.012
0.013
0.003
0.009
Number
of Points
6
7
6
5
7
7
6
Micronet
0.021
Table 3.4. Soil moisture validation variables: slope, RMSE, unbiased RMSE and spatial
standard deviation of downscaled soil derived from NP89 and LP92 models, AMSR-E
and NLDAS comparing with Little Washita watershed Micronet on August 2005.
Day
August
2, 2005
August
9, 2005
August
18, 2005
August
25, 2005
August
29, 2005
August
30, 2005
Unbiased
RMSE
(m3/m3)
Dataset
Slope
RMSE
(m3/m3)
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
0.321
0.336
0.322
-0.081
0.038
0.036
0.019
0.129
0.016
0.017
0.018
0.017
-0.013
-0.031
0.025
0.005
0.086
0.088
0.097
0.092
0.074
0.073
0.068
0.074
0.027
0.009
0.07
0.007
0.053
0.055
0.053
0.091
0.053
0.054
0.044
0.055
-0.077
-0.069
-0.113
-0.009
0.081
0.081
0.099
0.104
0.057
0.058
0.054
0.06
0.01
0.015
-0.006
-0.031
0.076
0.077
0.094
0.097
0.064
0.064
0.064
0.061
0.015
0.022
-0.002
-0.037
0.077
0.078
0.095
0.103
0.061
0.061
0.061
0.057
85
Spatial
Standard
Deviation
(m3/m3)
0.02
0.022
0.008
0.006
0.018
0.014
0.015
0.006
0.005
0.074
0.01
0.01
0.006
0.009
0.055
0.02
0.022
0.016
0.01
0.061
0.015
0.016
0.006
0.01
0.064
0.017
0.018
0.007
0.011
0.061
Number
of Points
6
5
6
6
6
6
Table 3.5. Overall soil moisture validation variables of the four months: slope, RMSE,
unbiased RMSE and spatial standard deviation of downscaled soil derived from NP89
and LP92 models, AMSR-E and NLDAS comparing with Little Washita watershed
Micronet.
Day
May
2004
June
2005
July
2005
August
2005
Unbiased
RMSE
(m3/m3)
Dataset
Slope
RMSE
(m3/m3)
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
NP89
LP92
AMSR-E
NLDAS
Micronet
0.056
0.046
0.055
0.14
0.054
0.054
0.046
0.113
0.033
0.033
0.03
0.066
-0.037
-0.042
0.02
0.009
0.059
0.058
0.054
0.122
0.039
0.04
0.036
0.05
0.05
0.049
0.076
-0.001
0.043
0.042
0.043
0.11
0.031
0.032
0.032
0.032
0.047
0.047
0.049
-0.025
0.069
0.069
0.076
0.103
0.054
0.055
0.052
0.054
86
Spatial
Standard
Deviation
(m3/m3)
0.015
0.016
0.004
0.009
0.033
0.014
0.015
0.007
0.01
0.046
0.013
0.014
0.004
0.007
0.036
0.016
0.017
0.008
0.008
0.056
Number of
Points
51
74
44
35
Table 3.6. Studies of disaggregating microwave soil moisture using fine scale remote
sensing imagery and model data.
Author
Methodology
Time and Region
Narayan
and
Lakshmi
(2008)
Fused AMSR-E
and TMI soil
moisture with
TRMM-PR
backscatter data
Entire year of
2003, Little
Washita Micronet
Merlin
et al
(2008)
Merlin
et al
(2010)
Das
et al
(2011)
Piles et
al
(2011)
Disaggregated
LPRM soil
moisture using
MODIS data by
soil evaporation
efficiency model
Disaggregated
LPRM soil
moisture using
MODIS data by
combined soil
evaporation
efficiency model
Merged L-band
SMAP soil
moisture and
radar data using
a linear
relationship
model
Disaggregated
SMOS data
using MODIS
and multiple
regression model
Result
The RMSE ranges
0.01~0.36
comparing with
AMSR-E
validation of
0.102~0.239
NAFE 2006
(OctoberNovember) in
Murrumbidgee,
Australia
RMSE ranges
1.4%~1.8%
NAFE 2006
(OctoberNovember) in
Murrumbidgee,
Australia
Average slope is
0.94 and the most
accuracy with an
error of 0.012
PALS SMEX02
in Iowa, USA,
2002
The RMSE ranges
0.015~0.02 cm3/
cm3
Jan-Feb 2010,
Murrumbidgee,
Australia
R2 ranges
0.14~0.21 and
RMSE ranges
0.09~0.17 m3/m3
Kim
and
Hogue
(2012)
Improved
AMSR-E soil
moisture using
MODIS products
SMEX 04 in San
Pedro River
Basin, 2004
Spatial correlation
ranges -0.08~0.34
Fang et
al
(2013)
Disaggregated
AMSR-E soil
moisture using
MODIS and
NLDAS data
Oklahoma
Mesonet and
Little Washita
Micronet, 20042005
Overall spatial
slope is 0.208,
RMSE is 0.059,
unbiased RMSE
is 0.027 m3/m3
87
Fang et
al.
(2013)
Disaggregated
AMSR-E soil
moisture using
MODIS and
NLDAS data by
soil evaporation
efficiency model
Little Washita
Micronet, 20042005
RMSE ranges
0.042~0.069
m3/m3, while
unbiased RMSE
ranges
0.031~0.055
m3/m3 and spatial
standard deviation
ranges
0.013~0.017
Merlin
(2012)
Disaggregated
SMOS soil
moisture based
on Physical and
Theoretical scale
change
(DisPATCh)
method
AACES
experiments in
Southeastern
Australia, 2010
R2 ranges
0.7~0.85 during
summer AACES
Merlin
(2013)
Disaggregated
SMOS soil
moisture using
the DisPATCh
method as well
as calibration
procedure
Apr-Oct 2011
Campaign in
Catalunya, Spain
R2 ranges 0.2~0.5
88
Figure 3.1. Maps of Oklahoma and Little Washita Watershed boundary and the Micronet
soil moisture stations are shown in green dots.
89
Figure 3.2. Averaged and standard deviation values of soil moisture observations from
the Micronet in May 2004, June 2005, July 2005 and August 2005 and the gaps represent
the days of missing soil moisture measurements.
90
Figure 3.3. Maps of variables used in building the soil moisture downscaling algorithm
over Little Washita watershed region from July 22, 2005 (a) Daytime 1 km resolution
MODIS land surface temperature; (b) 1 km resolution MODIS NDVI; (c) NLDAS 1/8o
resolution soil temperature at 0-10 cm depth; (d) NLDAS derived field capacity; (e)
NLDAS 1/8o resolution soil moisture at 0-10 cm depth; (f) AMSR-E 1/4o resolution soil
moisture.
91
Figure 3.4. Correlation of measured soil evaporation efficiency and soil moisture (C =
0.35). NP89 is the model proposed by Noilhan and Planton (1989), while LP92 is the
model proposed by Lee and Pielke (1992).
92
Figure 3.5a. Bivariate correlation analysis among surface soil temperature (0-10cm),
surface temperature and fractional vegetation cover from May 20, 2004, June 26, 2005,
July 22, 2005 and August 9, 2005, at 12.5 km spatial resolution.
93
Figure 3.5b. Correlation analysis between surface soil temperature (0-10cm) and
fractional vegetation cover from May 20, 2004, June 26, 2005, July 22, 2005 and August
9, 2005, at 12.5 km spatial resolution.
94
Figure 3.5c. Correlation analysis between surface soil temperature (0-10cm) and surface
temperature from May 20, 2004, June 26, 2005, July 22, 2005 and August 9, 2005, at
12.5 km spatial resolution.
95
Figure 3.6a. Maps of 1 km soil evaporation efficiency, 1 km disaggregated soil
temperature and NLDAS soil temperature in Little Washita watershed region from May
20, 2004 and June 26, 2005.
96
Figure 3.6b. Maps of 1 km soil evaporation efficiency, 1 km disaggregated soil
temperature and NLDAS soil temperature in Little Washita watershed region from July
22, 2005 and August 9, 2005.
97
Figure 3.7a. Maps of 1 km disaggregated soil moisture derived from NP89 and LP92
models, AMSR-E and NLDAS soil moisture in Little Washita watershed region from
May 20, 2004 and June 26, 2005.
98
Figure 3.7b. Maps of 1 km disaggregated soil moisture derived from NP89 and LP92
models, AMSR-E and NLDAS soil moisture in Little Washita watershed region from
July 22, 2005 and August 9, 2005.
99
Figure 3.8a. Bar plots of soil moisture validation variables: slope, RMSE, unbiased
RMSE and spatial standard deviation of downscaled soil derived from NP89 and LP92
models, AMSR-E and NLDAS comparing with Little Washita watershed Micronet on
May 2004.
100
Figure 3.8b. Bar plots of soil moisture validation variables: slope, RMSE, unbiased
RMSE and spatial standard deviation of downscaled soil derived from NP89 and LP92
models, AMSR-E and NLDAS comparing with Little Washita watershed Micronet on
June 2005.
101
Figure 3.8c. Bar plots of soil moisture validation variables: slope, RMSE, unbiased
RMSE and spatial standard deviation of downscaled soil derived from NP89 and LP92
models, AMSR-E and NLDAS comparing with Little Washita watershed Micronet on
July 2005.
102
Figure 3.8d. Bar plots of soil moisture validation variables: slope, RMSE, unbiased
RMSE and spatial standard deviation of downscaled soil derived from NP89 and LP92
models, AMSR-E and NLDAS comparing with Little Washita watershed Micronet on
August 2005.
103
Figure 3.8e. Bar plots of overall soil moisture validation variables: slope, RMSE,
unbiased RMSE and spatial standard deviation of downscaled soil derived from NP89
and LP92 models, AMSR-E and NLDAS comparing with Little Washita watershed
Micronet.
104
Figure 3.9. Overall scatter plots of 1 km disaggregated soil moisture derived from NP89
and LP92 models, AMSR-E and NLDAS soil moisture comparing with Little Washita
Micronet soil moisture observations on May 2004, June 2005, July 2005 and August
2005.
105
Figure 3.10. Overall scatter plots of detrended 1 km disaggregated soil moisture derived
from NP89 and LP92 models, detrended AMSR-E and NLDAS soil moisture comparing
with Little Washita Micronet soil moisture observations on May 2004, June 2005, July
2005 and August 2005.
106
Figure 3.11. Overall scatter plots of spatial standard deviation of 1 km disaggregated soil
moisture derived from NP89 and LP92 models, AMSR-E and NLDAS soil moisture
comparing with Little Washita Micronet soil moisture observations on May 2004, June
2005, July 2005 and August 2005.
107
CHAPTER IV PASSIVE/ACTIVE MICROWAVE SOIL MOISTURE
RETRIEVAL AND DISAGGREGATION USING SMAPVEX12 DATA3
4.1 ABSTRACT
The SMAPVEX12 experiment was held during June-July 2012 in Manitoba,
Canada with the goal of collecting remote sensing data and ground measurements for the
development and testing of soil moisture retrieval algorithms under different vegetation
and soil conditions for the SMAP (Soil Moisture Active Passive) satellite which was
launched in January 2015. The aircraft based soil moisture data provided by the
passive/active microwave sensor PALS has a spatial resolution of 1500 m. Here, a
disaggregation method called the change detection algorithm that combines passive
microwave soil moisture retrievals at coarse resolution with radar backscatter coefficients
at fine resolution was proposed and implemented with aircraft data sets providing passive
and active observations. The accuracy of the disaggregated change in soil moisture was
examined using ground soil moisture measurements. Results show that R2 for the
disaggregation products improved when compared with the original soil moisture
retrievals. The highest resolution disaggregated product at 5 m shows soil moisture
heterogeneity in much detail and also represent different distribution patterns of crops.
The difference of spatial standard deviation between disaggregated and in situ ranging
3
Fang, B., Lakshmi, V., Bindlish, R., Jackson, T. (2015) High Spatial Resolution
Microwave Soil Moisture Estimation and Downscaling Using SMAPVEX12 Data. To be
submitted to Transactions on Geoscience and Remote Sensing.
108
from -0.007-0.084 also proves the capability of the change detection algorithm at 5 m
scale.
Keywords: SMAP Microwave soil moisture, radar backscatter, disaggregation
change detection algorithm.
4.2 INTRODUCTION
As an important parameter in the land surface processes, soil moisture is related to
the studies such as agriculture, land-air interactions, hydrology and ecology (Lakshmi,
2013). The microwave remote sensing technology has been providing soil moisture
retrievals with improving since the late 1970s. There have been numerical studies on soil
moisture retrieval with passive microwave sensors, including AMSR-E (Advanced
Microwave Scanning Radiometer for the Earth Observing System), AMSR2 (Advanced
Microwave Scanning Radiometer 2) and SMOS (Soil Moisture and Ocean Salinity)
(Jackson and Schmugge. 1989; Jackson et al., 1999; Schmugge et al., 2002; Lakshmi,
2004, 2013; Schmugge et al., 1974; Schmugge and Jackson, 1994; Njoku et al., 1999).
The spatial resolution of these passive products is restricted by the antenna so the spatial
resolution of passive microwave soil moisture is typically at tens of kilometers. As a
result these products cannot meet the requirements for research and applications that
demand a fine scale. The SMAP (Soil Moisture Active Passive) program developed by
NASA was launched in January 2015. The SMAP satellite will provide soil moisture
products with multiple spatial resolutions of 3 km, 9 km and 36 km. In order to support
evaluating and testing of candidate SMAP soil moisture retrieval algorithms, a pre-launch
field campaign was carried out from June 6 - July 19, 2012 in southern Manitoba, Canada
109
that provided remotely sensed data from two aircraft systems PALS (Passive and Active
L and S band System) and UAVSAR (The Unmanned Air Vehicle Synthetic Aperture
Radar), as well as in-situ soil moisture measurements from field sampling and permanent
soil moisture monitoring network stations.
Radars, especially synthetic aperture satellite systems can provide much higher
spatial resolution information than the radiometers. However, it is more difficult to
retrieve soil moisture from radar backscatter coefficients because interaction with the
target is more complicated, which is related to vegetation canopy structure and soil
surface roughness. There have been numerical previous investigation on the retrieval of
soil moisture from active/passive microwave observations (Schmugge et al., 1974; Njoku
and Entekhabi, 1996; Lakshmi, 1997). Furthermore, attempts have been made at
disaggregating coarse spatial resolution passive microwave soil moisture retrievals by
combining them with high resolution active microwave signals, or the satellite products
from visible/infrared bands, such as land surface temperature and vegetation indices
(Bolten et al., 2003; Narayan et al., 2006; Narayan and Lakshmi, 2008; Piles et al., 2009;
Fang et al.,2013a, b.)
Soil moisture variations over time can be monitored by change detection
technologies. In the past, some studies have suggested and applied the change detection
method as an approach determining the change in soil moisture (Engman and Chauhan,
1995; Njoku, et al, 2002; Villasenor et al, 1993). This methodology assumes that the
change in radiometer brightness temperature or radar backscatter recorded by the sensor
only depends on the change in soil moisture of the target. The previous studies have
shown that a linear relationship between the change in brightness temperature/ radar
110
backscatter and the change in soil moisture and the radar sensitivity is a dependent on soil
moisture. However, the study usually requires soil moisture ground measurements which
are typically difficult to obtain for the high vegetation cover area. Therefore the soil
moisture retrieved from radiometer can be applied to the radar backscatter to estimate the
radar sensitivity by considering the factors of vegetation on radar sensitivity and spatial
heterogeneity.
In this investigation, the change detection algorithm was implemented to
disaggregate the PALS passive based soil moisture retrievals by combining them with the
PALS radar backscatter at 1500 m resolution and UAVSAR radar backscatter at 5 m and
aggregated resolution at 800 m, which corresponds to the crop field size, without using
any ground soil moisture measurements. This study presumed that the influence of the
spatial heterogeneity of soil properties, such as soil type and soil roughness on radar
sensitivity is not significant and consequently the variability of radar backscatter within
coarse spatial resolution radiometer pixel boundary only depends on soil moisture and
vegetation water content.
4.3 STUDY AREA AND DATA
The SMAPVEX12 campaign was held in the southern portion of Manitoba
province, Canada, centered on the town of Elm Creek (98o0’23”W, 49o 40’48”N),
covering a rectangular region of 12.8 km wide by 70 km long (Figure 4.1). The soil
texture type is dominated by sand in the west and loam and silt in the east. The major
crop types in the study area include soybeans, wheat, corn, canola and pasture, as well as
some wetland and forest land. In this study, remote sensing data and in situ measurements
111
were obtained over the 55 USDA agricultural fields. The data sets used for retrieving and
disaggregating microwave soil moisture are listed in Table 1.
4.3.1 PALS RADIOMETER/RADAR DATA
The PALS was designated as a simulator of the SMAP for the SMAPVEX12
campaign. This instrument was mounted at the tail of the aircraft with a 40 degree
viewing angle. During the SMAPVEX12 experiment, the PALS provided two sets of
observations, L-band (1.143 GHz) radiometer brightness temperature at h, v polarizations
and L-band (1.26 GHz) radar backscatter coefficient at vv, hh, vh and hv polarizations.
The data was acquired at two elevations (low and high) with an along track spatial
resolutions of around 650 m and 1600 m, respectively. The daily accumulated
precipitation from Canada WeatherFarm weather station network were compared with the
average PALS radiometer and radar observations at h and v polarizations in Figure 4.2.
Heavy rainfall events that occurred in mid-June and on July 19 can be detected. The
PALS brightness temperature is inversely related to the in situ soil moisture, while the
radar backscatter varied in its corresponding with in situ soil moisture.
4.3.2 UAVSAR DATA
The UAVSAR is a fully polarimetric and interferometric L-band radar. It operates
at approximately the same frequency as PALS (1.26GHz), and has spatial resolution of
1.66 m in range scale and 3 m in multi-looked angle. The incidence angle of UAVSAR is
between 35-45 degrees, which corresponding to the incidence angle of SMAP of 40
degrees and the nominal swath of UAVSAR is 21 km. In this study, the orthorectified
UAVSAR radar backscatter observations of hh polarization resampled to 5 m was used
and aggregated to 800 m which was at the size as the sampling fields.
112
4.3.3 VEGETATION WATER CONTENT
The Vegetation Water Content (VWC) data was derived from the Normalized
Difference Water Index (NDWI) using 5 m spatial resolution satellite imagery (SPOT
and Rapideye), covering the SMAPVEX12 field sampling and flight days from June
through July 2012. The VWC product was aggregated to 800 m and 1500 m for matching
with PALS and UAVSAR pixels (McNairn et al., 2013). In this investigation, we
assumed that the variation of VWC over time within each pixel at different spatial
resolutions could be ignored.
4.3.4 NLDAS (NORTH AMERICA LAND DATA ASSIMILATION SYSTEM)
The NLDAS has been providing land surface variables in near real-time on 1/8o
(12.5 km) spatial resolution over North America since January, 1979 by integrating
various land surface models (LSM) (Cosgrove et al., 2003; Mitchell et al., 2004). In this
study, the NLDAS soil temperature of the 0-10 cm soil layer at 1/8o (12.5 km) spatial
resolution was downloaded from the website (http://ldas.gsfc.nasa.gov/nldas/) and
resampled to the PALS radiometer spatial resolution (1500 m) as a key input for soil
surface reflectivity calculation.
4.3.5 GROUND MEASUREMENT
The soil moisture is spatially heterogeneous. In order to capture a representative
value for the SMAPVEX fields, the volumetric soil moisture was collected using
handheld hydro probes. The soil moisture was measured at 16 sampling points along two
transects of 8 points that were parallel to the crop row direction, with three replicate
measurements at each of these sample points.
113
4.4 METHODOLOGY
The data flow for retrieving and disaggregating passive microwave soil moisture
is shown in Figure 4.3 and described in more detail below.
4.4.1 MICROWAVE SOIL MOISTURE RETRIEVAL
The upwelling microwave radiation from the land surface is observed above the
canopy. The forward model considers a uniform layer of vegetation overlying the soil
surface. The radiative brightness temperature is described by the radiative transfer
equation, given by Equation 1.1 and the optical thickness  depends on vegetation water
content  and sensor viewing angle , in an approximately linear relationship, by
Equation 1.4.
4.4.2 CHANGE DETECTION ALGORITHM
Both brightness temperature and radar backscatter have a nearly linear
relationship to surface soil moisture, given uniform vegetation and land surface
characteristics (Narayan et al., 2006). The soil moisture can be retrieved from PALS
radiometer brightness temperature observations by combining it with a few ancillary
parameters that include surface soil temperature and soil properties. The passive
microwave radiometer of PALS can provide the soil moisture product at relatively low
spatial resolution (1500 m) and calculating the absolute soil moisture from the active
radar backscattering is quite complicated. The active microwave soil moisture retrieval
requires modelling a complex signal target interaction and inputting many other
parameters related to vegetation and soil structure to classify the target area into many
uniform subclasses.
114
The dependence of the radar backscatter on the change of soil moisture can be
simplified as being linear. Njoku et al (2002) modeled the relationship between the PALS
radiometer/ radar data acquired from SGP99 campaign, which are at same size of
footprint and in situ soil moisture measurements by classifying them into 3 subclasses
based on the corresponding vegetation water content. The equations are
TBP = A + Bmv
Equation 4.1: Relationship between radiometer brightness temperature and soil moisture
σoPP = C + Dmv
Equation 4.2: Relationship between radar backscatter and soil moisture
Where the regression parameters A, B, C and D correspond to each PALS radiometer or
radar pixel. They were assumed to be functions of surface vegetation and roughness. For
the equation 4.2, the parameters C and D can be calibrated by the soil moisture change on
consecutive days. In this study, the purpose is to estimate the soil moisture change at very
high spatial resolution by combining PALS radiometer soil moisture with the radar
backscatter coefficient information. However, as there is inconsistency in the spatial
resolution between the PALS soil moisture (1500 m) and the UAVSAR (800 m/5 m), it is
our assumption that over a few days, the change in vegetation canopy parameters has an
insignificant contribute on the change in copolarized radar backscatter resulting for the
change in soil moisture. Both the PALS and UAVSAR observations were obtained at a
high frequent revisit over the study region. The linear relationship between the change in
soil moisture and the change in radar backscatter can be expressed as
∆σoPP = DΔθ
115
Equation 4.3: Relationship between change in radar backscatter and soil moisture
Where, ∆σoPP is the copolarized change in radar backscatter and Δθ is the change
in soil moisture. The parameter D is the slope of the linear regression equation between
∆σoPP and Δθ and depends on the attenuation properties of vegetation canopy and soil
surface roughness. Past research (Du et al., 2000) demonstrated that the relative
sensitivity of radar backscatter mostly depends on the vegetation canopy opacity which
can be described as the ratio of the radar sensitivity in the presence of vegetation canopy
(D) and bare soil (D0 ) as
D
= f(τ)
D0
Equation 4.4: Radar sensitivity expressed by vegetation canopy and bare soil
So by combining the Equations 4.3 and 4.4, we have
∆σoPP = f(τ) ∗ D0 ∗ Δθ
Equation 4.5: Change in radar backscatter expressed by radar sensitivity, bare soil and
change in soil moisture
The radar sensitivity of bare soil D0 is determined by the soil roughness
variability for a certain sensor configuration: look angle, frequency and polarization. In
the study of Fung et al (1992), the behavior of the copolarized radar backscatter versus
soil moisture under different root mean square soil roughness from s=0.4~2.4 cm was
plotted and the result demonstrated that the correlation lines are parallel to each other
with the slope but have different intercepts. Therefore, it can be concluded that the radar
116
sensitivity of soil moisture is insignificantly affected by the surface roughness variability.
The Equation 4.5 can be simplified as
Δθ =
Δσ0
S0
Equation 4.6: Change in soil moisture expressed by change in radar backscatter
Where, the ratio S0 = f(τ) ∗ D0 . Microwave soil moisture is available at the
coarser spatial resolution of X and the radar backscatter is at a finer scale of x. To
downscale the change in soil moisture Δθx to a finer resolution, the soil moisture
difference was calculated from two PALS soil moisture images on two days t 0 to t 0 + Δt.
The change in soil moisture at coarse spatial resolution was then aggregated to 3×3 pixels.
The PALS radar backscatter data was aggregated to 4500 m and the UAVSAR radar
backscatter data was aggregated to 2400 m in order to compare it with the PALS soil
moisture, which was aggregated at the same spatial resolution,
1
ΔθX = ( ) ∑ Δθx
N
Equation 4.7: Summation of change in soil moisture to coarse spatial resolution
Where, ΔθX is the PALS radiometer soil moisture at the coarse spatial resolution,
which is based on "N" fine resolution radar pixels, while Δθx is the radar backscatter
downscaled soil moisture to the fine spatial resolution. Combining the Equation 4.6 with
Equation 4.7 we obtain
1
Δσ0x
ΔθX = ( ) ∑ [
]
N
S0
117
Equation 4.8: Change in soil moisture at coarse resolution expressed by radar
backscatter
The average is over all the “N” smaller radar footprints σ0x within the larger
radiometer footprint X, This is the change in soil moisture as measured by the radiometer
at the lower spatial resolution, is the change in soil moisture as measured by the radar at
the higher spatial resolution.
The change in consecutive values of radar backscatter acquired over an area
would be related to change in soil moisture, given by
Δσ0PP = f(τ) ∗ D0 ∗ Δmv =S0 ∗ Δθ
Equation 4.9: Change in radar backscatter expressed only by change in soil moisture
The PALS/UAVSAR mainly recorded data over crop fields. We made the
assumption that the slope S0 would be the same for all the radar pixels falling in one
radiometer pixel and the vegetation characteristics were uniform. In addition, during the
SMAPVEX12 campaign, the incidence angle of UAVSAR ranging 35-45 degrees was
another assumption made that the incidence angle variation of the radar data recorded
from different days is not significant as compared to the change in radar backscatter
within an aggregated PALS change in soil moisture pixel boundary. The Equation 4.9 can
be transformed as
1 ΣΔσ0x
S0 = ( )
N Δθ
Equation 4.10: Ratio between summarized change in radar backscatter and coarse
resolution microwave soil moisture
118
And the change in soil moisture at the radar backscatter resolution can be
calculated from
Δθ =
Δσ0x
S0
Equation 4.11: Change in soil moisture expressed by summarized change in radar
backscatter and ratio
4.5 RESULTS
4.5.1 PALS SOIL MOISTURE RETRIEVALS
Figure 4.4 shows the difference maps of the variables used in soil moisture
downscaling for two days: July 10-July 13. The soil moisture (Figure 4.2-i) varied more
in the southern potion of the flight path than the north. A clear wetting trend of soil
moisture around 0.02 m3/m3 can be noticed at the center of the flight path and a soil
moisture dry-down occurred in the southern part which corresponds to the crop fields.
The spatial distribution pattern of the UAVSAR radar backscatter (Figure 4.2-iii) is
highly correlated to the pattern of the PALS radar backscatter (Figure 4.2-ii) but has
much more detail. Both the UAVSAR and PALS changes in radar backscatter are
positively related to changes in soil moisture. The vegetation water content (Figure 4.2-iv)
composited from multiple flight days shows a region of high VWC in the north of the
flight path that corresponds to the forest region. In the southern part of the flight path the
VWC ranges from 0-4 kg/m2 and shows less spatial variation. In order to validate the
retrieved soil moisture from PALS at 1500 m, as well as the downscaled soil moisture
using PALS radar backscatter at 1500m and UAVSAR radar backscatter at 5 m and 800
m, the soil moisture ground measurements collected during the campaign were used. The
119
ground sampled soil moisture was gridded to the PALS and UAVSAR pixels at different
spatial resolutions by geo-locating and averaging all the point soil moisture
measurements within the corresponded pixel boundaries. In addition, the vegetation water
content data aggregated to the PALS and UAVSAR pixel size was divided into four
classes in order to study the sensitivity of statistical variables under different VWC
conditions. In addition, for analyzing the influence of vegetation water content on the
accuracy of soil moisture retrieval and disaggregation validations, the VWC values
corresponding to PALS/UAVSAR pixels were classified into four groups of
approximately equal numbers.
The ground based soil moisture was gridded for the pixel boundaries at different
scales for validating the soil moisture estimates from PALS and UAVSAR. In this study,
the statistical variables being calculated include R2, slope, RMSE, bias and spatial
standard deviation. The formulae are given by Equation 3.12 - Equation 3.15 and the bias
is calculated using:
n
1
Bias = ∑(θi − θ′i )
n
i=1
Equation 4.12: Bias calculation
The PALS radiometer soil moisture retrievals at a 1500 m spatial resolution for
six sampling days: June 25, July 3, July 10, July 13 and July 14 are mapped in Figure 4.5.
The soil moisture is approximately 0.2 m3/m3 higher in the northern part of the flight area
than the south for all the days. The southern portion has more heterogeneity and a faster
variation trend of soil moisture.
120
From Table 4.2, it can be observed that the R2 values of the PALS soil moisture
retrieval validation for the two VWC intervals of 1.5-2 kg/m2 and 2-2.5 kg/m2 that are
equal to 0.677 and 0.685, respectively, are better than the other two classes. The slopes of
the first three VWC classes are 0.986, 0.984 and 1.032, respectively, which are very close
to the diagonal lines. The RMSE ranges from 0.077-0.109 m3/m3 and bias ranges from 0.034-0.024 m3/m3. These two variables do not vary much through for the different VWC
classes. From the corresponding validation plots of the four intervals in Figure 4.6, we
observe that the two plots of VWC classes of 1.5-2 kg/m2 and 2-2.5 kg/m2 are relatively
concentrated. Obvious wrong prediction of soil moisture retrievals can be noted from the
plot of VWC interval which is over 2.5 kg/m2. We believe that the accuracy of the soil
moisture retrieval is greatly influenced when VWC is over 2.5 kg/m2.
4.5.2 COMPARISON BETWEEN PALS SOIL MOISTURE RETRIEVALS AND
PALS/UAVSAR RADAR BACKSCATTERS
Figure 4.7 shows the scatter plots of the PALS change in soil moisture versus the
aggregated UAVSAR change in radar backscatter at 2400 m and the PALS change in
radar backscatter at 4500 m for three pairs of days: July 5-July 3, July 13-July 10 and
July 17-July 14 are compared. The R2 of the PALS comparisons ranged from 0.623-0.779
and were mostly better than the R2 of the UAVSAR comparisons ranged from 0.4450.609. The PALS radar backscatter should have a better agreement with the PALS
radiometer retrieved soil moisture than the UAVSAR radar backscatter. Another reason
could be the inconsistency in the spatial resolution between the aggregated PALS and the
aggregated UAVSAR. The UAVSAR is at finer resolution so more point pairs were used
for the correlation analysis and consequently more uncertainties were brought in. In
121
addition, it also can be noted that there is a clear difference in the slope 0 between
UAVSAR and PALS comparisons each day. The 0 of the PALS comparison, ranging
14.65-16.7, is always greater than the corresponded UAVSAR comparison 7.805-11.72.
4.5.3 DISAGGREGATED PALS SOIL MOISTURE (1500 M) and UAVSAR SOIL
MOISTURE (800 M)
We selected five pairs of days to represent the rapid variation of soil moisture
during the entire period of the campaign: June 25-June 22, July 5- July 3, July 10- July 8,
July 13- July 10, July 17- July 14. The disaggregated PALS change in soil moisture at
1500 m is shown in Figure 4.8, as well as the disaggregated UAVSAR change in soil
moisture at 800 m is shown in Figure 4.10. The good agreement of the soil moisture
distribution pattern can be seen by comparing the results from the two sensors. Rapid
changes of soil moisture, dry-down and wetting can be noticed in the southern part of the
flight as opposed to in the north, where the variation is normally shown in light blue and
ranges from -0.2-0.2 m3/m3. In the southeast part of the flight path, the soil moisture
varied in response to the rainfall events. Soil type is one reason that might explain such a
change trend. The northern portion is dominated by sandy soil while the south is
dominated by loam soil and silt soil that are more suitable for crops growing. The
evapotranspiration of the crops contributed to the fast transition of soil moisture.
The PALS downscaled soil moisture validation results are shown in Figure 4.9
and Table 4.3 and demonstrate that the R2 values (0.648-0.824) are better than the
original PALS soil moisture validation results for all the VWC classes. A tendency can
also be noted that the PALS downscaled soil moisture is less correlated to in situ soil
moisture as VWC increasing. The bias values, which range from -0.004-0.011 m3/m3
122
indicate that the downscaled soil moisture values are mostly overestimated. This trend
can also be observed from the distribution pattern of the data points in Figure 4.9, as well
as the overall slope values (1.065-1.617). Figure 4.9 also shows that for the first three
VWC classes, most of the data points fall into the northeast and southwest quadrants,
which indicates that the change in soil moisture is correctly predicted, as oppose to the
VWC class over 2.5 kg/m2 that have the most data points falling into the other two
quadrants. The RMSE (0.044-0.072 m3/m3) is always better than the PALS soil moisture
validation of the corresponding VWC class.
By examining the validation results of the UAVSAR downscaled soil moisture at
800 m in Figure 4.11 and Table 4.4, we can assess that the estimation errors, RMSE
(0.06-0.069 m3/m3) as well as bias (-0.003-0.025 m3/m3) indicate apparent decreasing
tendencies as the VWC increases. This could be associated with the increasing influence
of vegetation on the accuracy of soil moisture disaggregation. The overestimation
characteristics of the disaggregated soil moisture can be observed from the slope which
ranges from 1.065-1.244 as well as the distribution pattern in Figure 4.11. For the
UAVSAR validation plots it should be noted that more points are found in the northwest
and southeast quadrants than the PALS validation plots, which indicates that more soil
moisture points are incorrectly predicted.
4.5.4 COMPARISON BETWEEN UAVSAR VS PALS RADAR PIXELS
The comparison between the aggregated UAVSAR and PALS change in radar
backscatters at 1500 m is plotted in Figure 4.12 and statistical variables are shown in
Table 4.5. Four coincident days from July were selected and classified based on their
VWC values. The R2 (0.474-0.777) demonstrates that the two radar backscatter data sets
123
have a good correlation under low VWC conditions and become less correlated while
VWC increases. The other variables, slope (0.636-0.731), RMSE (0.639-1.007 m3/m3)
and bias (-0.179-0.033 m3/m3) are generally worse than either the PALS or UAVSAR
disaggregation validation results.
4.5.5 DISAGGREGATED UAVSAR SOIL MOISTURE AT 5 M RESOLUTION
In Figure 4.13, the UAVSAR disaggregated soil moisture at 5 m from three sites
representing to three typical crop types: canola, soybeans and corn were mapped to show
the spatial and temporal variability of soil moisture at a fine scale. Comparing these in
Figure 4.2, from July 3-July 5, which was just after a rainfall event occurred, a wetting
trend can be viewed for all three sites, with a soil moisture change of -0.1-0.03 m3/m3.
The canola has a larger variation of soil moisture than the other two crops. From July 8July 10 a soil moisture dry-down trend occurred and was followed by a minor wetting
event on July 10-July 13. It also can be observed that the crop canopy pattern for
soybeans fields is reflected by the disaggregated change in soil moisture, with parallel
bright strips indicating the attenuation of radar signals on the crop canopy. Such
characteristics were not found for either canola or corn field. However, the spatial
heterogeneity of soil moisture can be noticed. From the canola maps, soil moisture varied
much less from southeast to north.
By analyzing the validation results for the UAVSAR disaggregated soil moisture
at 5 m from Tables 4.6-4.7. It can be concluded that the canola crop, with R2 ranging
0.746 and 0.769, are relatively better correlated than the other two crop types. From the
results shown in Table 4.7, the difference in the spatial standard deviation σ between the
UAVSAR disaggregated soil moisture and the in situ measurements ranged between 124
0.007-0.084 m3/m3. The soybeans sites 111 and 112, with differences of 0.008 and -0.007
m3/m3, respectively, have the best agreement with the in situ data. The σ of the
disaggregated results are mostly greater than in situ measurements, except for site 112,
indicating an overestimation of soil moisture change. The overestimation trend can also
be observed from the Figure 4.14, where the slopes are mostly over 1.
4.6 CONCLUSIONS AND DISCUSSIONS
In this paper a change detection algorithm for disaggregating microwave soil
moisture data retrieved aircraft observations was developed and implemented on the Lband hh polarization radar backscatter coefficients data at 1500 m and at 5 m/800 m
obtained from the SMAPVEX12 campaign. The algorithm was developed by assuming
that the change in the microwave radiometer and radar observations are linearly
correlated to the change in soil moisture within the satellite data pixel boundary so the
change in microwave soil moisture can be simplified as a function of vegetation opacity
only. Another assumption was made that the change in the vegetation canopy parameter
is insignificant compared to the change in the co-polarized radar backscatter signal
retrieved change in soil moisture. The PALS soil moisture retrievals and
PALS/UAVSAR radar data for some sampling days in June-July, 2012 were applied to
implement the disaggregation algorithm at different spatial resolutions and validated
using ground soil moisture measurements. The results show that the disaggregated
change in soil moisture for both PALS and UAVSAR are significantly better correlated
to the in situ soil moisture measurements, with R2 ranging 0.648-0.824 and RMSE
ranging 0.044-0.072 m3/m3 for PALS validation and R2 ranging 0.651-0.737 and RMSE
ranging 0.06-0.079 m3/m3 for UAVSAR validation, than the original PALS soil moisture
125
validation with R2 ranging 0.367-0.685 and RMSE ranging 0.077-0.109 m3/m3. The
statistical variables for both the disaggregated PALS and UAVSAR change in soil
moisture tend to be less correlated to the in situ soil moisture data as the VWC increases,
which is consistent to the assumption that the vegetation opacity dominates the
microwave soil moisture. In addition, both the PALS and UAVSAR disaggregated
change in soil moisture are overestimated for all WVC groups when compared with the in
situ data. It is also noted that the PALS radar backscatter and aggregated UAVSAR radar
backscatter have a good correlation when VWC <6 kg/m2, with R2 ranging 0.642-0.777
m3/m3. This suggests that the change detection algorithm can disaggregate PALS change
in soil moisture by combining with PALS/UAVSAR change in radar backscatter. The
disaggregation method was also applied with 5 m resolution UAVSAR radar data to
study the change in soil moisture at crop field scale. The change in the soil moisture maps
shows the spatial heterogeneity of soil moisture in much more detail. The overestimation
tendency is also observed in the statistical variables of the validation results, with the
difference of spatial standard deviation ranging from -0.007-0.084 m3/m3. In addition, the
overall R2 (0.374-0.769) does not show a significant difference between different crop
types.
The limitations of study include (1) the disaggregation model was developed
based on the assumption that the influence of change in vegetation canopy on microwave
change in soil moisture over a short period can be ignored. The change in soil moisture
disaggregation can be accurately implemented within 2-3 days. (2) The disaggregated soil
moisture as well as radar backscatter had poor bad correlation for high VWC values. A
126
future study should apply the change detection algorithm method to estimate change in
soil moisture from the SMAP satellite product once available.
4.7 ACKNOWLEDGEMENT
We would like to thank Andreas Colliander from JPL (Jet Propulsion Laboratory)
for providing the daily soil moisture retrievals from PALS radiometer observations used
in this study. And, the SMAPVEX12 UAVSAR and in situ data are provided by National
Snow & Ice Data Center (NSIDC) at http://nsidc.org/data.
127
4.8 TABLES AND FIGURES
Table 4.1. Remote sensing and in situ data sets used in soil moisture retrieval and
disaggregation.
Dataset
Spatial Resolution
Temporal
Resolution
PALS L-band (h-pol) Radiometer
Brightness Temperature (K)
655/1592 m (Along
Track)
Daily
PALS L-band (hh-pol) Radar
Backscatter (dB)
655/1592 m (Along
Track)
Daily
UAVSAR Radar Backscatter (dB)
5 m/800 m
Daily
NLDAS (North America Land Data
Assimilation System) Soil
Temperature (K)
1/8 degree (12.5 km)
Hourly
Vegetation Water Content Map
(kg/m2)
5m
Daily
NSDB (National Soil Database) Soil
Properties
1:1 million
-
Probe-based In situ Soil Moisture
(m3/m3)
800 m
Daily
128
Table 4.2. PALS Soil moisture retrievals comparing with in situ soil moisture
measurements from June 25, July 3, July 10, July 13 and July 14.
VWC
Number
of Points
R2
Slope
RMSE
(m3/m3)
Bias
(m3/m3)
0<VWC<1.5
46
0.449
0.986
0.109
0.017
1.5<VWC<2
64
0.677
0.984
0.086
0.024
2<VWC<2.5
56
0.685
1.032
0.077
0.022
VWC>2.5
45
0.367
0.636
0.096
-0.034
Table 4.3. Disaggregated PALS change in soil moisture at 1500 m resolution validated
by change in situ soil moisture measurements from the days of June 15-June 12, June 17June 15, June 25-June 22, July 5-July 3 and July 17-July 14.
VWC
Number
of Points
R2
Slope
RMSE
(m3/m3)
Bias
(m3/m3)
0<VWC<1.5
43
0.824
1.617
0.071
0.011
1.5<VWC<2
62
0.700
1.197
0.072
0.012
2<VWC<2.5
54
0.720
1.244
0.062
0.014
VWC>2.5
48
0.648
1.065
0.044
-0.004
129
Table 4.4. Disaggregated UAVSAR change in soil moisture at 800 m resolution
validated by change in situ soil moisture measurements from the days of June 25-June 22,
July 5-July 3, July 8-July 5, July 10-July 8, Jul 13-July 10.
VWC
Number
of Points
R2
Slope
RMSE
(m3/m3)
Bias
(m3/m3)
0<VWC<1.3
36
0.651
1.452
0.060
-0.003
1.3<VWC<1.8
41
0.694
1.053
0.066
0.021
1.8<VWC<2.2
39
0.680
1.409
0.079
0.025
VWC>2.2
28
0.737
1.605
0.069
0.021
130
Table 4.5. PALS change in radar backscatter compares with aggregated UAVSAR
change in radar backscatter at 1500 m resolution from the days of July 5-July 3, July 8July 5, July 10-July 8, Jul 13-July 10.
VWC
Number
of Points
R2
Slope
RMSE
(m3/m3)
Bias
(m3/m3)
0<VWC<1.9
463
0.720
0.689
1.007
-0.179
1.9<VWC<2.8
510
0.777
0.728
0.921
-0.128
2.8<VWC<6
495
0.642
0.731
0.858
-0.078
VWC>6
522
0.474
0.636
0.639
0.033
Table 4.6. Validation results of disaggregated UAVSAR change in soil moisture at 5 m
resolution by change in situ soil moisture measurements from the days of July 5-July 3,
July 8-July 5, July 10-July 8, Jul 13-July 10.
Site
Type
Number
R2
Slope
RMSE
(m3/m3)
Bias
(m3/m3)
Site-14
Soybeans
22
0.374
1.214
0.150
-0.132
Site-61
Canola
35
0.769
1.407
0.067
0.043
Site-62
Canola
31
0.746
1.925
0.099
-0.070
Site-63
Soybeans
32
0.677
1.460
0.079
-0.063
Site-71
Corn
51
0.456
1.970
0.099
0.006
Site-111
Soybeans
64
0.657
0.878
0.072
-0.024
Site-112
Soybeans
32
0.736
0.715
0.037
-0.015
131
Table 4.7. Spatial standard deviation of disaggregated UAVSAR change in soil moisture
at 5 m resolution comparing with change in situ soil moisture measurements by crop
fields from the days of July 5-July 3, July 8-July 5, July 10-July 8, Jul 13-July 10.
Site
UAVSAR
(m3/m3)
In Situ
(m3/m3)
Difference
(m3/m3)
Site-14
0.087
0.049
0.038
Site-61
0.094
0.060
0.034
Site-62
0.108
0.052
0.056
Site-64
0.069
0.045
0.024
Site-71
0.122
0.038
0.084
Site-111
0.114
0.106
0.008
Site-112
0.055
0.062
-0.007
132
Figure 4.1. Overview of the SMAPVEX12 study area (above), with a box showing the
scanning swath of the flight. Different colors indicate the types of sampling fields:
forestry site, USDA agricultural field and permanent AAFC (Agriculture and Agri-Food
Canada) field. Close view of on crop field (below) site-55. The ground soil moisture was
sampled from 16 points of two parallel rows, showing in red.
133
Figure 4.2. PALS radiometer (high altitude) brightness temperature at L-band (6 GHz),
ground soil moisture measurements compares with daily accumulated precipitation from
June 7- July 19, 2012. The precipitation comes from of rain gauges of Canada
WeatherFarm weather station network.
134
Figure 4.3. Data flow of microwave soil moisture retrieval, disaggregation and validation
using PALS/ UAVSAR radar data and in situ soil moisture measurements.
135
Figure 4.4. Data sets used for building soil moisture disaggregation model from July 13 July 10: (i) PALS change in soil moisture retrieval at 1500 m resolution (m3/m3); (ii)
PALS change in radar backscatter at 1500 m resolution (dB); (iii) UAVSAR change in
radar backscatter at 800 m resolution; (iv) Vegetation Water Content (VWC) at 800 m
resolution (kg/m2).
136
Figure 4.5. PALS soil moisture retrievals (m3/m3) at 1500 m resolution from five days:
June 25, July 3, July 10, July 13 and July 14.
137
Figure 4.6. PALS soil moisture retrievals at 1500 m resolution validated using in situ
measurements from June 25, July 3, July 10, July 13 and July 14. The validation results
are classified into four VWC groups.
138
Figure 4.7. Aggregated PALS at 4500 m resolution and UAVSAR radar backscatter at
2400 m resolution compare with PALS soil moisture retrievals from July 5- July 3, July
13- July 10 and July 17- July 14.
139
Figure 4.8. Disaggregated PALS change in soil moisture (m3/m3) at 1500 m resolution
from June 25-June 22, July 5-July 3, July 10-July 8, July 13-July 10 and July 17-July 14.
140
Figure 4.9. Validation of disaggregated PALS change in soil moisture at 1500 m
resolution from June 15-June 12, June 17-June 15, June 25-June 22, July 5-July 3 and
July 17-July 14, being divided into four WVC groups.
141
Figure 4.10. Disaggregated UAVSAR change in soil moisture (m3/m3) at 800 m
resolution from June 25-June 22, July 5-July 3, July 10-July 8, July 13-July 10, Jul 17July 14.
142
Figure 4.11. Validation of Disaggregated UAVSAR change in soil moisture at 800 m
resolution from June 25-June 22, July 5-July 3, July 8-July 5, July 10-July 8, Jul 13-July
10, corresponding to four VWC groups.
143
Figure 4.12. Comparison between PALS radar backscatter change and aggregated
UAVSAR radar backscatter change at 1500 m from July 5-July 3, July 8-July 5, July 10July 8, Jul 13-July 10, corresponding to four WVC groups.
144
Figure 4.13. Spatial variation of disaggregated UAVSAR change in soil moisture (m3/m3)
at 5 m resolution from July 3 to July 13 at the sites of three typical crop types canola,
soybeans and corn: (i): site-62, (ii):site-112, (iii): site-71.
145
Figure 4.14. Validation of disaggregated UAVSAR change in soil moisture at 5 m
resolution from the days of July 5-July 3, July 8-July 5, July 10-July 8, Jul 13-July 10, at
six sites representing different crop types.
146
CHAPTER V CONCLUSIONS AND FUTURE WORK
5.1 CONCLUSIONS
Soil moisture disaggregation is important in the improvement spatial resolution of
current passive microwave soil moisture products, which are not suitable for hydrological
study at watershed scale. In this dissertation, three soil moisture disaggregation
algorithms were developed and implemented to improve the spatial resolution of soil
moisture retrievals from satellite/aircraft platforms and were validated using ground
measurements.
In Chapter 2, the thermal inertia concept - relationship between daily average soil
moisture and temperature difference modulated by vegetation condition was formulated
using linear regression model. This model was built by integrating NLDAS soil moisture
at 0-10 cm layer and surface skin temperature, as well as NDVI derived from satellites MODIS, AVHRR and SPOT at 1/8 degree resolution and implemented on microwave
soil moisture retrievals from AMSR-E and SMOS to obtain 1 km daily soil moisture in
Oklahoma between May-August, 2004 and 2005. The disaggregated soil moisture was
validated using ground measurements from Oklahoma Mesonet and Little Washita
watershed Micronet. The disaggregated soil moisture has an overall slope of 0.219
validated by Mesonet data and 0.242 validated by Micronet, which are significantly better
than coarse resolution soil moisture AMSR-E and NLDAS. Particularly improvement
147
was most evident when using Micronet data, with overall RMSE of 0.021 m3/m3 and
unbiased RMSE of 0.026 m3/m3. The validation results are consistent with the
assumption that the daily soil moisture is negatively related to daily temperature change
modified by NDVI. The results demonstrate soil moisture distribution pattern as well as
change tendency in much more detail. In addition, it is observed that the sub-pixel
heterogeneity does not greatly affect the quality of disaggregated soil moisture when
applying the model at NLDAS resolution to 1 km data.
Chapter 3 introduces a physical model based algorithm to disaggregate the
microwave soil moisture using two soil evaporation efficiency models NP89 and LP92
which quantify the relationship between soil moisture, soil capacity and soil evaporation
efficiency. A bivariate regression analysis was applied on three variables: top soil
temperature, land surface temperature, fractional vegetation cover derived from NLDAS
and remote sensing data to disaggregate soil temperature to 1 km resolution.
Subsequently the soil evaporation efficiency and soil moisture at 1 km in Little Washita
watershed, Oklahoma between 2004-2005 was calculated from the soil temperature and
used to disaggregate AMSR-E soil moisture. From the disaggregated soil moisture it was
found that the overall RMSE ranges 0.042-0.069 m3/m3 and unbiased RMSE ranges
0.031-0.055 m3/m3 and these are generally better than either AMSR-E or NLDAS. The
spatial standard deviation which ranges 0.013-0.017 m3/m3 also indicates the spatial
variability of disaggregated soil moisture is closer to in situ measurement.
Chapter 4 introduces a change detection algorithm which integrated microwave
radiometer soil moisture retrievals and radar backscatter coefficients based on the linear
correlations between remotely sensed soil moisture and radiometer brightness
148
temperature/ radar backscatter, by assuming that the variability of vegetation structure
within a short period (1-3 days) could be ignored, as well as the change in microwave soil
moisture could be simplified as a function of vegetation opacity only. This algorithm was
applied on L-band hh polarization radar backscatter data from PALS at 1500 m resolution
and from UAVSAR at 5 m and 800 m resolution which both were acquired from a prelaunch field experiment SMAPVEX12, carried out in June-July 2012 to disaggregate the
change in soil moisture derived from PALS L-band h polarization radiometer brightness
temperature at 1500 m resolution. For the PALS validation, the R2 ranges 0.648-0.824
and RMSE ranges 0.044-0.072 m3/m3. For the UAVSAR validation, the R2 ranges 0.6510.737 and RMSE ranges 0.06-0.079 m3/m3 at 800 m resolution and the R2 ranges 0.3740.769 and RMSE ranges 0.037-0.15 m3/m3 from the selected crop fields at 5 m resolution.
Both of PALS and UAVSAR disaggregated change in soil moistures show a better
agreement with the in situ measurements than the original PALS soil moisture. Other
findings include the accuracy of disaggregated soil moisture decreases as the VWC
increases; the change detection algorithm may not be effective in some crop types with
complicated structures, such as corn.
Limitations exist in each of the disaggregation algorithm. For the thermal inertia
based algorithm, data sets are derived from different sources and at different resolutions:
remote sensing and land surface model output were used and consequently the
uncertainties were brought into the disaggregation model: (1) Mismatch of sensing depth.
The visible/infrared bands remotely sensed observations represent the soil layer at a few
millimeter depth, as oppose to the sensing depths of microwave remote sensing and land
surface model are at centimeters depth; (2) Mismatch of pixel size. The disaggregation
149
algorithm was built at NLDAS of 1/8o resolution and applied on 1 km MODIS LST data
and corrected microwave soil moisture retrievals at 25 km. The assumption was made
that the fine resolution pixels within each coarse resolution pixel have the same thermal
inertia relationship between daily average soil moisture and temperature change.
For the soil evaporation efficiency based disaggregation algorithm, beside the two
limitations existed in the first algorithm, which are mismatch of sensing depth and pixel
size, other limitations should be noted that: (1) The soil evaporation efficiency models
were derived from experiments and expressed by semi-empirical formulae. However, the
soil moisture in the study region may be influenced by other factors; (2) The soil
evaporation efficiency and fractional vegetation cover were also computed by empirical
equations using remote sensing data; (3) The soil temperature was disaggregated through
bivariate regression analysis by land surface temperature and fractional vegetation cover.
The accuracy of disaggregated soil temperature greatly depends on the correlation
coefficients between the three parameters.
For the change detection algorithm that integrates passive/microwave
observations, the limitations mainly come from the radar sensitivity. The active
microwave soil moisture is determined by a couple of factors, including sensor viewing
angle, vegetation canopy structure, topographic properties, soil dielectric properties, etc.
The vegetation and soil parameters are often difficult to obtain. Therefore, the
relationship between remotely sensed soil moisture and radar backscatter has to been
summarized that it only depends on the vegetation optical depth and the disaggregation
model accuracy is affected by the vegetation parameters. In addition, we assumed that
during a short period the influence of change in vegetation canopy structure is not as
150
important as the influence of change in copolarized radar backscatter on the change in
soil moisture. So only the remote sensing data of adjacent flight days can be used for
implementing the algorithm. And again, the mismatch of the pixel size also restricts the
accuracy of disaggregated change in soil moisture. We assumed that the sub-pixel
variability of the Δθ-Δσ ratio at the fine resolution radar backscatter pixels within each of
the coarse resolution soil moisture pixel could be ignored.
As to the achievements and limitations of all three disaggregation algorithms, we
consider the first algorithm based on thermal inertia theory is more practical than the
other two algorithms for future uses. Firstly, unlike the soil evaporation based algorithm,
which requires to obtain many variables, the thermal inertia based disaggregation model
uses only three variables land surface temperature, soil moisture and NDVI so less
uncertain involved in this algorithm. Secondly, as the input variables of the thermal
inertia based algorithm can be acquired from land surface models and remotely sensed
data sets which have high spatial coverage (NLDAS covers North America, while the
satellite data used in this dissertation cover worldwide), it is possible to generate
disaggregated soil moisture products covering large areas. In contrast, the microwave
radar sensors usually have high spatial resolutions and relatively smaller spatial coverage
than either passive microwave or visible/infrared sensors. Thirdly, concerning the data
availability, the change detection algorithm requires high spatial resolution radar data,
which has lower revisit frequency and with high expenditure, as oppose to the change
detection algorithm, which produces daily temporal resolution data (the NLDAS data is
at hourly resolution and the satellite data are at daily resolutions) and with no cost.
Finally, by comparing the validation results between the three algorithms, the RMSE
151
from thermal inertia based algorithm demonstrates a better accuracy of prediction, with
overall RMSE of 0.021 m3/m3 and unbiased RMSE of 0.026 m3/m3 than the other two
algorithms, with overall RMSE ranging 0.042-0.069 m3/m3 and unbiased RMSE ranging
0.031-0.055 m3/m3 for soil evaporation efficiency based algorithm, with RMSE ranging
0.044-0.072 m3/m3 for change detection algorithm using PALS data, and with RMSE
ranging 0.06-0.079 m3/m3 and 0.037-0.15 m3/m3 using UAVSAR data at two resolutions.
5.2 FUTURE WORK
As all three algorithms were built on regression analyses modeling the relationships
between soil moisture and other land surface variables, the accuracy of disaggregated
results rely on the correlational analysis. However, bad correlations were noticed,
specifically the relationship of θ-ΔΤ during non-growing seasons and the relationship
between fraction vegetation cover and surface soil temperature and the relationship
between soil moisture and radar backscatter at high VWC class. Toward this problem, the
author would suggest to use other mathematical models to improve the correlation.
Currently the statistically based algorithms have been well developed and several
algorithms among them could be effective applied in my dissertation studies, such as
artificial neural network, decision tree learning, support vector machines, etc.
On the other hand, regarding the change detection algorithm, the author will also
look for ways to better parameterization of change in radar backscatter with vegetation
optical depth - specifically the spatial heterogeneity of vegetation structure parameter at
sub-pixel level. The VWC derived from high-resolution optical satellite sensors can be
utilized to calculate the vegetation optical depth. In addition, at fine resolution, the
152
relationship between vegetation structure parameter and radar backscatter retrieved soil
moisture over different crop types can be studied for improving the soil moisture
disaggregation accuracy.
For all three algorithms, the study regions were restricted at watershed and
statewide scales, as the in situ measurements are available at these places for validating
the algorithm accuracy. The next step the author may consider extending the algorithms
to global scales and also applying to other microwave soil moisture data. Currently there
are many satellite sensors with high spatial and temporal resolutions providing land
surface parameters from visible/infrared bands for building soil moisture disaggregation
algorithms based on thermal inertia concept and soil evaporation efficiency, as well as
microwave soil moisture products, especially the newly launched satellite SMAP which
will provide very high resolution radiometer and radar observations to implement the
disaggregation algorithm on a global scale.
153
REFERENCES
Basara, J. B., Illston, B. G., Fiebrich, C. A., McPherson, R. A., Bostic, J. P., Browder, P.,
... & Kesler, K. (2009, January). An Overview of the Oklahoma City Micronet. In Eighth
Symposium on the Urban Environment.
Bindlish, R., Barros, A.P., (2002). Subpixel variability of remotely sensed soil moisture:
an inter-comparison study of SAR and ESTAR. IEEE Trans. Geosci. Remote Sens.40 (2),
326–337.
Bolten, J.D., Lakshmi, V., Njoku, E.G., (2003). Soil moisture retrieval using the
passive/active L-and S-band radar/radiometer. IEEE Trans. Geosci. Remote Sens. 41
(12), 2792–2801.
Brown, M. E., Pinzón, J. E., Didan, K., Morisette, J. T., & Tucker, C. J. (2006).
Evaluation of the consistency of long-term NDVI time series derived from AVHRR,
SPOT-Vegetation, SeaWiFS, MODIS, and Landsat ETM+ sensors. Geoscience and
Remote Sensing, IEEE Transactions on, 44(7), 1787-1793.
Brunsell, N. A., & Gillies, R. R. (2001). The effect of emissivity on evaporation, IAHS
PUBLICATION, 276-280.
Camillo, P. J., & Gurney, R. J. (1986). A resistance parameter for bare-soil evaporation
models. Soil Science, 141(2), 95-105.
Carlson, T. N., Gillies, R. R., & Schmugge, T. J. (1995). An interpretation of
methodologies for indirect measurement of soil water content. Agricultural and forest
meteorology, 77(3), 191-205.
Carlson, T. (2007), An overview of the “triangle method” for estimating surface evapotranspiration and soil moisture from satellite imagery, Sensors, 7, 1612–1629.
Chanzy, A., & Bruckler, L. (1993). Significance of soil surface moisture with respect to
daily bare soil evaporation. Water Resources Research, 29(4), 1113-1125.
Choi, M., & Hur, Y. (2012). A microwave-optical/infrared disaggregation for improving
spatial representation of soil moisture using AMSR-E and MODIS products. Remote
Sensing of Environment, 124, 259-269.
154
Choudhury, B. J., Schmugge, T. J., Chang, A., & Newton, R. W. (1979). Effect of surface
roughness on the microwave emission from soils. Journal of Geophysical Research:
Oceans (1978–2012), 84(C9), 5699-5706.
Collatz, G. J., Bounoua, L., Los, S. O., Randall, D. A., Fung, I. Y., & Sellers, P. J. (2000).
A mechanism for the influence of vegetation on the response of the diurnal temperature
range to changing climate. Geophysical Research Letters, 27(20), 3381-3384.
Cosgrove, B. A., Lohmann, D., Mitchell, K. E., Houser, P. R., Wood, E. F., Schaake, J.
C., ... & Meng, J. (2003). Real‐time and retrospective forcing in the North American
Land Data Assimilation System (NLDAS) project. Journal of Geophysical Research:
Atmospheres (1984–2012), 108(D22).
Cosh, M. H., Jackson, T. J., Bindlish, R., & Prueger, J. H. (2004). Watershed scale
temporal and spatial stability of soil moisture and its role in validating satellite estimates.
Remote sensing of Environment, 92(4), 427-435.
Daamen, C. C., & Simmonds, L. P. (1996). Measurement of evaporation from bare soil
and its estimation using surface resistance. Water Resources Research, 32(5), 1393-1402.
Dai, A., & Deser, C. (1999). Diurnal and semidiurnal variations in global surface wind
and divergence fields. Journal of Geophysical Research: Atmospheres (1984–2012),
104(D24), 31109-31125.
Das, N. N., Entekhabi, D., & Njoku, E. G. (2011). An algorithm for merging SMAP
radiometer and radar data for high-resolution soil-moisture retrieval. Geoscience and
Remote Sensing, IEEE Transactions on, 49(5), 1504-1512.
De Fries, R. S., Hansen, M., Townshend, J. R. G., & Sohlberg, R. (1998). Global land
cover classifications at 8 km spatial resolution: the use of training data derived from
Landsat imagery in decision tree classifiers. International Journal of Remote Sensing,
19(16), 3141-3168.
Du, Y., Ulaby, F. T., & Dobson, M. C. (2000). Sensitivity to soil moisture by active and
passive microwave sensors. Geoscience and Remote Sensing, IEEE Transactions on,
38(1), 105-114.
Entekhabi, D., Njoku, E. G., O'Neill, P. E., Kellogg, K. H., Crow, W. T., Edelstein, W.
N., ... & Van Zyl, J. (2010). The soil moisture active passive (SMAP) mission.
Proceedings of the IEEE, 98(5), 704-716.
Famiglietti, J. S., Ryu, D., Berg, A. A., Rodell, M., & Jackson, T. J. (2008). Field
observations of soil moisture variability across scales. Water Resources Research, 44(1).
Fang, B., Lakshmi, V., Bindlish, R., Jackson, T. J., Cosh, M., & Basara, J. (2013). Passive
microwave soil moisture downscaling using vegetation index and skin surface temperature.
155
Vadose Zone Journal, 12(3).
Fang, B., & Lakshmi, V. (2014). Soil moisture at watershed scale: Remote sensing
techniques. Journal of Hydrology, 516, 258-272.
Fang, B., & Lakshmi, V. (2014). AMSR‐E Soil Moisture Disaggregation Using MODIS
and NLDAS Data. Remote Sensing of the Terrestrial Water Cycle, 277-304, Published by
John Wiley.
Friedl, M. A., McIver, D. K., Hodges, J. C., Zhang, X. Y., Muchoney, D., Strahler, A.
H., ... & Schaaf, C. (2002). Global land cover mapping from MODIS: algorithms and
early results. Remote Sensing of Environment, 83(1), 287-302.
Fung, A. K., Li, Z., & Chen, K. S. (1992). Backscattering from a randomly rough
dielectric surface. Geoscience and Remote Sensing, IEEE Transactions on, 30(2), 356-369.
Gillies, R. R., & Carlson, T. N. (1995). Thermal remote sensing of surface soil water
content with partial vegetation cover for incorporation into climate models. Journal of
Applied Meteorology, 34(4), 745-756.
Goddard Earth Sciences Data and Information Services Center (GES DISC). HDISC:
Hydrology Data and Information Services Center.
http://disc.sci.gsfc.nasa.gov/hydrology/. (Accessed in 2011).
Goetz, S. J. (1997). Multi-sensor analysis of NDVI, surface temperature and biophysical
variables at a mixed grassland site. International Journal of Remote Sensing, 18(1), 71-94.
Gutman, G., & Ignatov, A. (1998). The derivation of the green vegetation fraction from
NOAA/AVHRR data for use in numerical weather prediction models. International
Journal of remote sensing, 19(8), 1533-1543.
Heathman, G. C., Cosh, M. H., Han, E., Jackson, T. J., McKee, L., & McAfee, S. (2012).
Field scale spatiotemporal analysis of surface soil moisture for evaluating point-scale in
situ networks. Geoderma, 170, 195-205.
Heilman, J. L., & Moore, D. G. (1982). Evaluating near-surface soil moisture using heat
capacity mapping mission data. Remote Sensing of Environment, 12(2), 117-121.
Illston, B. G., Basara, J. B., Fiebrich, C. A., Crawford, K. C., Hunt, E., Fisher, D. K., ... &
Humes, K. (2008). Mesoscale monitoring of soil moisture across a statewide network.
Journal of Atmospheric and Oceanic Technology, 25(2), 167-182.
Imaoka, K., Kachi, M., Fujii, H., Murakami, H., Hori, M., Ono, A., ... & Shimoda, H. (2010).
Global Change Observation Mission (GCOM) for monitoring carbon, water cycles, and
climate change. Proceedings of the IEEE, 98(5), 717-734.
156
Jackson, T. J., & Schmugge, T. J. (1989). Passive microwave remote sensing system for
soil moisture: Some supporting research. Geoscience and Remote Sensing, IEEE
Transactions on, 27(2), 225-235.
Jackson, T. J. (1993). III. Measuring surface soil moisture using passive microwave
remote sensing. Hydrological processes, 7(2), 139-152.
Jackson, T. J., Le Vine, D. M., Hsu, A. Y., Oldak, A., Starks, P. J., Swift, C. T., ... &
Haken, M. (1999). Soil moisture mapping at regional scales using microwave
radiometry: The Southern Great Plains Hydrology Experiment. Geoscience and Remote
Sensing, IEEE Transactions on, 37(5), 2136-2151.
Jackson, T. J., Gasiewski, A. J., Oldak, A., Klein, M., Njoku, E. G., Yevgrafov, A., ...
& Bindlish, R. (2002). Soil moisture retrieval using the C-band polarimetric scanning
radiometer during the Southern Great Plains 1999 Experiment. Geoscience and Remote
Sensing, IEEE Transactions on, 40(10), 2151-2161.
Jackson, T. J., Cosh, M. H., Bindlish, R., Starks, P. J., Bosch, D. D., Seyfried, M., ... &
Du, J. (2010). Validation of advanced microwave scanning radiometer soil moisture
products. Geoscience and Remote Sensing, IEEE Transactions on, 48(12), 4256-4272.
Jackson, T. J., Bindlish, R., Cosh, M. H., Zhao, T., Starks, P. J., Bosch, D. D., ... &
Leroux, D. (2012). Validation of soil moisture and ocean salinity (SMOS) soil moisture
over watershed networks in the US. Geoscience and Remote Sensing, IEEE
Transactions on, 50(5), 1530-1543.
Karl, T. R., Kukla, G., & Gavin, J. (1984). Decreasing diurnal temperature range in the
United States and Canada from 1941 through 1980. Journal of climate and applied
meteorology, 23(11), 1489-1504.
Justice, C. O., Townshend, J. R. G., Vermote, E. F., Masuoka, E., Wolfe, R. E., Saleous,
N., ... & Morisette, J. T. (2002). An overview of MODIS Land data processing and
product status. Remote sensing of Environment, 83(1), 3-15.
Karl, T. R., Kukla, G., & Gavin, J. (1984). Decreasing diurnal temperature range in the
United States and Canada from 1941 through 1980. Journal of climate and applied
meteorology, 23(11), 1489-1504.
Kerr, Y. H., Waldteufel, P., Richaume, P., Davenport, I., Ferrazzoli, P., & Wigneron, J. P.
(2007). SMOS Level 2 Processor for Soil Moisture, Algorithm Theoretical Based
Document (ATBD). CESBIO, IPSL-Service d’Aéronomie, INRAEPHYSE, Reading
University, Tor Vergata University. SO-TN-ESL-SM-GS-0001 Issue 3. a, 05/12/2007.
Kerr, Y. H., Waldteufel, P., Richaume, P., Wigneron, J. P., Ferrazzoli, P., Mahmoodi,
A., ... & Delwart, S. (2012). The SMOS soil moisture retrieval algorithm. Geoscience and
Remote Sensing, IEEE Transactions on, 50(5), 1384-1403.
157
Kim, J., & Hogue, T. S. (2012). Improving spatial soil moisture representation through
integration of AMSR-E and MODIS products. Geoscience and Remote Sensing, IEEE
Transactions on, 50(2), 446-460.
Knight, J. F., Lunetta, R. S., Ediriwickrema, J., & Khorram, S. (2006). Regional scale
land cover characterization using MODIS-NDVI 250 m multi-temporal imagery: A
phenology-based approach. GIScience & Remote Sensing, 43(1), 1-23.
Komatsu, T. S. (2003). Toward a robust phenomenological expression of evaporation
efficiency for unsaturated soil surfaces. Journal of Applied Meteorology, 42(9), 13301334.
Kondo, J., Saigusa, N., & Sato, T. (1992). A model and experimental study of
evaporation from bare-soil surfaces. Journal of applied meteorology, 31, 304-312.
Kurc, S. A., & Small, E. E. (2004). Dynamics of evapotranspiration in semiarid grassland
and shrubland ecosystems during the summer monsoon season, central New Mexico.
Water Resources Research, 40(9).
Kustas, W. P., & Daughtry, C. S. (1990). Estimation of the soil heat flux/net radiation
ratio from spectral data. Agricultural and Forest Meteorology, 49(3), 205-223.
Lakshmi, V., Wood, E. F., & Choudhury, B. J. (1997). Evaluation of Special Sensor
Microwave/Imager satellite data for regional soil moisture estimation over the Red River
Basin. Journal of Applied Meteorology, 36(10), 1309-1328.
Lakshmi, V., & Susskind, J. (2000). Validation of TOVS land surface parameters using
ground observations. J. Geophys. Res, 105, 2179-2190.
Lakshmi, V., Czajkowski, K., Dubayah, R., & Susskind, J. (2001). Land surface air
temperature mapping using TOVS and AVHRR. International Journal of Remote
Sensing, 22(4), 643-662.
Lakshmi, V. (2004). The role of satellite remote sensing in the prediction of ungauged
basins. Hydrological processes, 18(5), 1029-1034.
Lakshmi, V., Hong, S., Small, E. E., & Chen, F. (2011). The influence of the land surface
on hydrometeorology and ecology: new advances from modeling and satellite remote
sensing. Hydrology Research, 42(2-3), 95-112.
Lakshmi, V. (2013). Remote sensing of soil moisture. ISRN Soil Science, 2013.
Land Long Term Data Record (LTDR). AVHRR data. http://ltdr.nascom.nasa.gov/
(Accessed in 2013).
158
Lee, T. J., & Pielke, R. A. (1992). Estimating the soil surface specific
humidity. Journal of Applied Meteorology, 31(5), 480-484.
Lohmann, D., Mitchell, K. E., Houser, P. R., Wood, E. F., Schaake, J. C., Robock, A., ... &
Tarpley, J. D. (2004). Streamflow and water balance intercomparisons of four land surface
models in the North American Land Data Assimilation System project. Journal of
Geophysical Research: Atmospheres (1984–2012), 109(D7).
Luo, L., Robock, A., Mitchell, K. E., Houser, P. R., Wood, E. F., Schaake, J. C., ... &
Tarpley, J. D. (2003). Validation of the North American Land Data Assimilation System
(NLDAS) retrospective forcing over the southern Great Plains. Journal of Geophysical
Research: Atmospheres (1984–2012), 108(D22).
Luo, Y., Berbery, E. H., Mitchell, K. E., & Betts, A. K. (2007). Relationships between land
surface and near-surface atmospheric variables in the NCEP North American Regional
Reanalysis. Journal of Hydrometeorology, 8(6), 1184-1203.
Mahfouf, J. F., & Noilhan, J. (1991). Comparative study of various formulations of
evaporations from bare soil using in situ data. Journal of Applied Meteorology, 30(9),
1354-1365.
Mallick, K., Bhattacharya, B. K., & Patel, N. K. (2009). Estimating volumetric surface
moisture content for cropped soils using a soil wetness index based on surface
temperature and NDVI. Agricultural and Forest Meteorology, 149(8), 1327-1342.
McPherson, R. A., Fiebrich, C. A., Crawford, K. C., Kilby, J. R., Grimsley, D. L., Martinez,
J. E., ... & Sutherland, A. J. (2007). Statewide monitoring of the mesoscale environment: A
technical update on the Oklahoma Mesonet. Journal of Atmospheric and Oceanic
Technology, 24(3), 301-321.
Merlin, O., Chehbouni, A., Walker, J. P., Panciera, R., & Kerr, Y. H. (2008). A
simple method to disaggregate passive microwave-based soil moisture. Geoscience
and Remote Sensing, IEEE Transactions on, 46(3), 786-796.
Merlin, O., Walker, J. P., Chehbouni, A., & Kerr, Y. (2008). Towards deterministic
downscaling of SMOS soil moisture using MODIS derived soil evaporative efficiency.
Remote Sensing of Environment, 112(10), 3935-3946.
Merlin, O., Al Bitar, A., Walker, J. P., & Kerr, Y. (2009). A sequential model for
disaggregating near-surface soil moisture observations using multi-resolution thermal
sensors. Remote Sensing of Environment, 113(10), 2275-2284.
Merlin, O., Al Bitar, A., Walker, J. P., & Kerr, Y. (2010). An improved algorithm for
disaggregating microwave-derived soil moisture based on red, near-infrared and thermalinfrared data. Remote Sensing of Environment, 114(10), 2305-2316.
Merlin, O., Duchemin, B., Hagolle, O., Jacob, F., Coudert, B., Chehbouni, G., & Kerr, Y.
159
(2010). Disaggregation of MODIS surface temperature over an agricultural area using a
time series of Formosat-2 images. Remote Sensing of Environment, 114(11), 2500-2512.
Merlin, O., Al Bitar, A., Rivalland, V., Béziat, P., Ceschia, E., & Dedieu, G. (2011). An
analytical model of evaporation efficiency for unsaturated soil surfaces with an arbitrary
thickness. Journal of Applied Meteorology and Climatology, 50(2), 457-471.
Merlin, O., Rudiger, C., Al Bitar, A., Richaume, P., Walker, J. P., & Kerr, Y. H.
(2012). Disaggregation of SMOS soil moisture in Southeastern Australia. Geoscience
and Remote Sensing, IEEE Transactions on, 50(5), 1556-1571.
Merlin, O., Escorihuela, M. J., Mayoral, M. A., Hagolle, O., Al Bitar, A., & Kerr, Y.
(2013). Self-calibrated evaporation-based disaggregation of SMOS soil moisture: An
evaluation study at 3km and 100m resolution in Catalunya, Spain. Remote Sensing of
Environment, 130, 25-38.
Minacapilli, M., Iovino, M., & Blanda, F. (2009). High resolution remote estimation of
soil surface water content by a thermal inertia approach. Journal of hydrology, 379(3),
229-238.
Minet, J., Bogaert, P., Vanclooster, M., & Lambot, S. (2012). Validation of ground
penetrating radar full-waveform inversion for field scale soil moisture mapping. Journal
of Hydrology, 424, 112-123.
Mitchell, K. E., Lohmann, D., Houser, P. R., Wood, E. F., Schaake, J. C., Robock, A., ...
& Bailey, A. A. (2004). The multi‐institution North American Land Data Assimilation
System (NLDAS): Utilizing multiple GCIP products and partners in a continental
distributed hydrological modeling system. Journal of Geophysical Research:
Atmospheres (1984–2012), 109(D7).
Mladenova, I., Lakshmi, V., Walker, J. P., Panciera, R., Wagner, W., & Doubkova, M.
(2010). Validation of the ASAR global monitoring mode soil moisture product using the
NAFE'05 data set. Geoscience and Remote Sensing, IEEE Transactions on, 48(6), 24982508.
Mladenova, I., Lakshmi, V., Jackson, T. J., Walker, J. P., Merlin, O., & de Jeu, R. A.
(2011). Validation of AMSR-E soil moisture using L-band airborne radiometer data from
National Airborne Field Experiment 2006. Remote Sensing of Environment, 115(8),
2096-2103.
Mo, K. C., Chen, L. C., Shukla, S., Bohn, T. J., & Lettenmaier, D. P. (2012).
Uncertainties in North American land data assimilation systems over the contiguous
United States. Journal of Hydrometeorology, 13(3), 996-1009.
Mohanty, B. P., Famiglietti, J. S., & Skaggs, T. H. (2000). Evolution of soil moisture
spatial structure in a mixed vegetation pixel during the Southern Great Plains 1997
(SGP97) Hydrology Experiment. Water Resources Research, 36(12), 3675-3686.
160
Monteith, J. L. (1981). Evaporation and surface temperature. Quarterly Journal of the
Royal Meteorological Society, 107(451), 1-27.
Myneni, R. B., Hall, F. G., Sellers, P. J., & Marshak, A. L. (1995). The interpretation of
spectral vegetation indexes. Geoscience and Remote Sensing, IEEE Transactions on,
33(2), 481-486.
Myneni, R. B., Hoffman, S., Knyazikhin, Y., Privette, J. L., Glassy, J., Tian, Y., ... &
Running, S. W. (2002). Global products of vegetation leaf area and fraction absorbed
PAR from year one of MODIS data. Remote sensing of environment, 83(1), 214-231.
Narayan, U., Lakshmi, V., & Njoku, E. G. (2004, September). A simple algorithm for
spatial disaggregation of radiometer derived soil moisture using higher resolution
radar observations. In Geoscience and Remote Sensing Symposium, 2004. IGARSS'04.
Proceedings. 2004 IEEE International (Vol. 3, pp. 1877-1879). IEEE.
Narayan, U., & Lakshmi, V. (2008). Characterizing subpixel variability of low
resolution radiometer derived soil moisture using high resolution radar data. Water
resources research, 44(6).
National Aeronautics and Space Administration (NASA). National Space Science
Data Center. NSSIDC ID: 1978-041A.
http://nssdc.gsfc.nasa.gov/nmc/spacecraftDisplay.do?id=1978-041A(Accessed 2011).
Nishida, K., Nemani, R. R., Running, S. W., & Glassy, J. M. (2003). An operational
remote sensing algorithm of land surface evaporation. Journal of Geophysical Research:
Atmospheres (1984–2012), 108(D9).
Njoku, E. G., & Entekhabi, D. (1996). Passive microwave remote sensing of soil
moisture. Journal of hydrology, 184(1), 101-129.
Njoku, E. G., & Li, L. (1999). Retrieval of land surface parameters using passive
microwave measurements at 6-18 GHz. Geoscience and Remote Sensing, IEEE
Transactions on, 37(1), 79-93.
Njoku, E. G., Jackson, T. J., Lakshmi, V., Chan, T. K., & Nghiem, S. V. (2003). Soil
moisture retrieval from AMSR-E. Geoscience and Remote Sensing, IEEE Transactions
on, 41(2), 215-229.
Njoku, E. G., & Chan, S. K. (2006). Vegetation and surface roughness effects on
AMSR-E land observations. Remote Sensing of environment, 100(2), 190-199.
Noilhan, J., & Planton, S. (1989). A simple parameterization of land surface processes for
meteorological models. Monthly Weather Review, 117(3), 536-549.
161
North American Land Data Assimilation Systems (NLDAS). NLDAS Data.
http://ldas.gsfc.nasa.gov/. (Accessed in 2011).
Pan, M., Wood, E. F., McLaughlin, D. B., Entekhabi, D., & Luo, L. (2009). A multiscale
ensemble filtering system for hydrologic data assimilation. Part I: Implementation and
synthetic experiment. Journal of Hydrometeorology, 10(3), 794-806.
Parada, L. M., & Liang, X. (2004). Optimal multiscale Kalman filter for assimilation of
near‐surface soil moisture into land surface models. Journal of Geophysical Research:
Atmospheres (1984–2012), 109(D24).
Passerat de Silans, A., Bruckler, L., Thony, J. L., & Vauclin, M. (1989). Numerical
modeling of coupled heat and water flows during drying in a stratified bare soil—
comparison with field observations. Journal of Hydrology, 105(1), 109-138.
Peischl, S., Walker, J. P., Allahmoradi, M., Barrett, D., Gurney, R., Kerr, Y., ... & Ye, N.
(2009, July). Towards validation of SMOS using airborne and ground data over the
Murrumbidgee Catchment. In Proc. MODSIM (pp. 3733-3739).
Piles, M., Entekhabi, D., & Camps, A. (2009). A change detection algorithm for
retrieving high-resolution soil moisture from SMAP radar and radiometer observations.
Geoscience and Remote Sensing, IEEE Transactions on, 47(12), 4125-4131.
Piles, M., Camps, A., Vall-Llossera, M., Corbella, I., Panciera, R., Rudiger, C., ... &
Walker, J. (2011). Downscaling SMOS-derived soil moisture using MODIS
visible/infrared data. Geoscience and Remote Sensing, IEEE Transactions on, 49(9),
3156-3166.
Rodrí
guez-Iturbe, I., G. K. Vogel, R. Rigon, D. Entekhabi, F. Castelli, and A.Rinaldo
(1995), On the Spatial Organization of Soil Moisture Fields, Geophysical Research
Letters, 22(20), 2757-2760.
Robock, A., Luo, L., Wood, E. F., Wen, F., Mitchell, K. E., Houser, P. R., ... & Crawford,
K. C. (2003). Evaluation of the North American Land Data Assimilation System over the
southern Great Plains during the warm season. Journal of Geophysical Research:
Atmospheres (1984–2012), 108(D22).
Sánchez-Ruiz, S., Piles, M., Sánchez, N., Martínez-Fernández, J., Vall-llossera, M., &
Camps, A. (2014). Combining SMOS with visible and near/shortwave/thermal infrared
satellite data for high resolution soil moisture estimates. Journal of Hydrology, 516, 273283.
Sahoo, A. K., De Lannoy, G. J., Reichle, R. H., & Houser, P. R. (2013). Assimilation and
downscaling of satellite observed soil moisture over the Little River Experimental
Watershed in Georgia, USA. Advances in Water Resources, 52, 19-33.
162
Sandholt, I., Rasmussen, K., & Andersen, J. (2002). A simple interpretation of the surface
temperature/vegetation index space for assessment of surface moisture status. Remote
Sensing of environment, 79(2), 213-224.
Schaake, J. C., Duan, Q., Koren, V., Mitchell, K. E., Houser, P. R., Wood, E. F., ... &
Tarpley, J. D. (2004). An intercomparison of soil moisture fields in the North American
Land Data Assimilation System (NLDAS). Journal of Geophysical Research:
Atmospheres (1984–2012), 109(D1).
Schmugge, T., Gloersen, P., Wilheit, T., & Geiger, F. (1974). Remote sensing of soil
moisture with microwave radiometers. Journal of Geophysical Research, 79(2), 317-323.
Schmugge, T., & Jackson, T. J. (1994). Mapping surface soil moisture with microwave
radiometers. Meteorology and Atmospheric Physics, 54(1-4), 213-223.
Schmugge, T. J., Kustas, W. P., Ritchie, J. C., Jackson, T. J., & Rango, A. (2002).
Remote sensing in hydrology. Advances in water resources, 25(8), 1367-1385.
Sellers, P. J., Heiser, M. D., & Hall, F. G. (1992). Relations between surface conductance
and spectral vegetation indices at intermediate (100 m2 to 15 km2) length scales. Journal
of Geophysical Research: Atmospheres (1984–2012), 97(D17), 19033-19059.
Steven, M. D., Malthus, T. J., Baret, F., Xu, H., & Chopping, M. J. (2003).
Intercalibration of vegetation indices from different sensor systems. Remote Sensing of
Environment, 88(4), 412-422.
Tarnavsky, E., Garrigues, S., & Brown, M. E. (2008). Multiscale geostatistical analysis of
AVHRR, SPOT-VGT, and MODIS global NDVI products. Remote Sensing of
Environment, 112(2), 535-549.
Tucker, C. J. (1979). Red and photographic infrared linear combinations for monitoring
vegetation. Remote sensing of Environment, 8(2), 127-150.
Tucker, C. J., Pinzon, J. E., Brown, M. E., Slayback, D. A., Pak, E. W., Mahoney, R., ...
& El Saleous, N. (2005). An extended AVHRR 8-km NDVI dataset compatible with
MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing,
26(20), 4485-4498.
Wan, Z., & Li, Z. L. (1997). A physics-based algorithm for retrieving land-surface
emissivity and temperature from EOS/MODIS data. Geoscience and Remote Sensing,
IEEE Transactions on, 35(4), 980-996.
Wigneron, J. P., Chanzy, A., Calvet, J. C., & Bruguier, N. (1995). A simple algorithm to
retrieve soil moisture and vegetation biomass using passive microwave measurements
over crop fields. Remote Sensing of Environment, 51(3), 331-341.
163
Xu, Xiaolan. (2014). SMAPVEX12 UAVSAR Incidence-Angle Normalized Backscatter
Data. Boulder, Colorado USA: NASA DAAC at the National Snow and Ice Data Center.
http://dx.doi.org/10.5067/1A09OKL9G4QW.
Yamanaka, T., Takeda, A., & Shimada, J. (1998). Evaporation beneath the soil surface:
some observational evidence and numerical experiments. Hydrological processes, 12(13‐
14), 2193-2203.
Zhang, R., Sun, X., Zhu, Z., Su, H., & Tang, X. (2003). A remote sensing model for
monitoring soil evaporation based on differential thermal inertia and its validation.
Science in China Series D: Earth Sciences, 46(4), 342-355.
Zhang, X., Friedl, M. A., Schaaf, C. B., Strahler, A. H., Hodges, J. C., Gao, F., ... &
Huete, A. (2003). Monitoring vegetation phenology using MODIS. Remote sensing of
environment, 84(3), 471-475.
Zhao, W., & Li, A. (2013). A Downscaling Method for Improving the Spatial Resolution
of AMSR-E Derived Soil Moisture Product Based on MSG-SEVIRI Data. Remote
Sensing, 5(12), 6790-6811.
Zhou, Y., McLaughlin, D., Entekhabi, D., & Ng, G. H. C. (2008). An ensemble
multiscale filter for large nonlinear data assimilation problems. Monthly Weather Review,
136(2), 678-698.
164
APPENDIX A – REPRINT PERMISSIONS
165
166
167
168
Документ
Категория
Без категории
Просмотров
0
Размер файла
5 163 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа