close

Вход

Забыли?

вход по аккаунту

?

Estimation of sea surface temperature from passive microwave satellite imagery

код для вставкиСкачать
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI films
the text directly from the original or copy submitted. Thus, some thesis and
dissertation copies are in typewriter face, while others may be from any type of
computer printer.
The quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality illustrations
and photographs, print bleedthrough, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and continuing
from left to right in equal sections with small overlaps.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6” x 9” black and white
photographic prints are available tor any photographs or illustrations appearing
in this copy for an additional charge. Contact UMI directly to order.
ProQuest Information and Learning
300 North Zeeb Road. Ann Arbor. Ml 48106-1346 USA
800-521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ESTIMATION OF SEA SURFACE TEMPERATURE FROM PASSIVE
MICROWAVE SATELLITE IMAGERY
ESTIMATE DE LA TEMPERATURE MARITIME SUPERFICIELLE PAR
IMAGERIE SATELLITE PASSIf A MICRO-ONDE
A Thesis Subm itted
to the Faculty of the Royal M ilitary College of Canada
by
Andrew K. Langille
In P artial Fulfillment of the Requirements for the Degree of
M aster of Science in Physics
February 2002
© T his thesis may be used within the Departm ent of National Defence, but copyright
for open publication remains th e property of the author.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
■♦I
National Library
of Canada
BMiolhtaue nationaie
du Canada
Acquisitions and
Bfetographic Services
Acquisitions at
services btotiographiques
385 WSangton Straat
Ottawa ON K1A0N4
385. ruaWaKnglon
OttaweON K1A0N4
Our mm N aam tM m tet
The author has granted a non­
exclusive licence allowing die
National Library of Canada to
reproduce, loan, distribute or sell
copies o f this diesis in microform,
paper or electronic formats.
L’auteur a accords une licence non
exclusive pennettant &la
Biblioth&que nationaie du Canada de
reproduire, prSter, distribuer ou
vendre des copies de cette these sous
la forme de microfiche/film, de
reproduction sur papier ou sur format
61ectronique.
The author retains ownership of die
copyright in this thesis. Neither die
thesis nor substantial extracts from it
may be printed or otherwise
reproduced without the author’s
permission.
L’auteur conserve la propri&e du
droit d'auteur qui protege cette these.
Ni la these ni des extraits substantiels
de celle-ci ne doivent etre imprimis
ou autrement reproduits sans son
autorisation.
0-612-67080-5
CanadS
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ROYAL MILITARY COLLEGE OF CANADA
COLLEGE MILITAIRE ROYAL DU CANADA
DIVISION OF GRADUATE STUDIES AND RESEARCH
DIVISION DES ETUDES SUPERIEURES ET DE LA RECHERCHE
This is to certify th a t this thesis prepared by / Ceci certifie que la these redigee par
Andrew K. Langille
entitled / intit ulee
ESTIMATION OF SEA SURFACE TEMPERATURE FROM PASSIVE
MICROWAVE SATELLITE IMAGERY
ESTIME DE LA TEMPERATURE MARITIME SUPERFICIELLE PAR
IMAGERIE SATELLITE PASSIF A MICRO-ONDE
complies with the Royal M ilitary College of Canada regulations and th at it meets
the accepted standards of the G raduate School with respect to quality / satisfait
aux reglements du College m ilitaire royal du Canada et qu’elle respecte les normes
acceptees par la Faculte des etudes superieures quant a la qualite,
for the degree of / pour le dipldme de
M aster of Science in Physics / M aitrise en Science, en Physique
Signed by the final examining com m ittee:/
les membres du comite exam inateur de la soutenance de these
Chair / President
External Examiner / Examinateur exteme
Main Supervisor / Directeur de these principal
Approved by the Head of Department : /
Approuve par le Directeur du D e p a rte m e n ^ ^ -£ -5 -_
Date:
To the Librarian: This thesis is not to be regarded as classified./
Au Bibliothecaire: C ette these n ’est pas
c^rr"™* a publication restreinte.
Main Supervisor /
Di
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
de these principal
Acknowledgement
I wish to extend my sincere gratitude to Lt(N) Dewain Emrich for his enthusiasm
towards and assistance throughout this project. Many thanks to Dr. Jean-M arc Noel
for his guidance and support. To my thesis advisor Dr. Joseph Buckley many thanks
for his extensive reviews of this manuscript and his encouragement.
A special thanks to my classmates C apt. Stu MacWilliams, Capt. Dave Weston,
M aj. Tony Masys, 2Lt Jeremy Hansen and 2Lt Charles MacLeod for their assistance
in completing the required course work. Also like to thank Capt. Brendan Cook
for his assistance with
To my officemates, Miss. Rhonda Millikin, C apt. Eric
Travis and M aj. Vivier Lefebvre many thanks for the constant encouragement.
I would also like to extend my thanks to Mr. Michael MacDonald of Satlantic Incorpo­
rated for his assistance in finding an approaprate cloud masking technique. Mr. Mark
Grandmaison of MetOc Centre Halifax for suppling the Sea Surface Analysis D ata.
I would like to thank Ms. Jennifer Mullan for her efficiency in obtaining reference
m aterial.
My sincerest thanks are extended to my family and friends for their continued sup­
port and encouragement.
This project was financially supported by the GEOIDE Network.
iii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Abstract
Langille, Andrew K.. M.Sc. (Physics). Royal M ilitary College of Canada. February
2002. Estim ation of Sea Surface Temperature from Passive Microwave Satellite Im­
agery.
Supervisor: Dr. Joseph R. Buckley.
The measurement of sea surface tem perature (SST) is predominately esti­
mated using infrared radiometers, which produce good results but only for cloud free
viewing of the ocean surface. Thus, the purpose of this research was to evaluate
the potential of SST retrieval using a passive microwave sensor, the Special Sensor
Microwave Imager (SSM /I), which can view the ocean surface without interference
from clouds. The SST results of the Advanced Very High Resolution Radiometer
(AVHRR) were averaged over the field of view (FOV) of SSM /I to act as the calibra­
tion and validation data set for SST. A model developed at the Centre for Research
in Earth and Space Technology which proposed the indirect estim ation of SST via
a non-linear relationship with water vapour was evaluated in the north-west North
A tlantic with unsatisfactory results. A second model was developed using band dif­
ferences, to eliminate the effects of certain geophysical param eters. This new model
produced the large scale SST fields which were shown in the AVHRR SST estimates
and showed th at the presence of cloud liquid water limits the accuracy of SST re­
trieval. Therefore, this model has the potential, with further calibration, to retrieve
SST in near all weather conditions or at least allow for the interpolation of clear SST
estim ations from a well-calibrated sensor across the SSM/I SST gradient. This de­
velopment would increase our knowledge of the global SST distribution on each pass
of the satellite rather than relying on composite images.
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Resume
Langille, Andrew K.. M. Sc. (Physique). College M ilitaire Royal du Canada. Fevrier
2002. Estim ate de la tem perature maritime superficielle par imagerie satellite passif
a micro-onde
Directeur de these: Dr. Joseph R. Buckley.
La mesure de la tem perature de surface des oceans (SST- Sea Surface Tem­
perature) est habituellement estime en utlisant des radiometres infra-rouge passifs.
Ces capteurs produisent des images satisfaisantes lorsqu’il n ’y a pas de nuages dans
I’atmosphere. Le but principal de cette recherche est d ’evaluer la possibility d ’extraire
le SST en utilisant un capteur a micro-onde suraomme SSM /I (Special Sensor Mi­
crowave Sensor). Ce capteur possede la capacite de “voir’ la surface des oceans sans
etre encombre par les nuages. La moyenne des mesures de SST obtenues par le capteur
AVHRR (Advanced Very High Resolution Radiometer) sert comme donnees initiates
afin de calibrer et valider le modele develope par CRESTech (Centre for Research
in E arth and Space Technology) et le notre. Le modele de CRESTech, qui propose
l’estim ation indirecte de SST par la relation non-lineaire de la vapeur d ’eau, a ete
utilisee pour estime les tem peratures de la surface dans le nord-ouest le l’ocean atlantique. Les resultats obtenus par ce modele sont non-satisfaisants. Le deuxieme model
base sur les differences entres les bandes fut devolope par notre groupe de recherche
dans le but d ’eliminer les effets de certains param etres geophysiques. Le nouveau
modele obtient des gradients de tem peratures qui sont semblables a ceux obtenus par
le capteur AVHRR. En plus, notre modele demontre que I’eau liquide contenue dans
les nuages peut jouer un role assez im portant dans la precision des valeurs calculees
pour le SST. Done, le modele a le potentiel d ’estimer le SST dans quasiment toutes les
conditions atmospheriques,ou, du moins, perm ettre ^interpolation de SST sans nuages
avec un capteur qui est bien calibre avec le gradient du capteur SSM /I. Nos connaissances de la distribution globale de SST pourraient etre ameliorer considerablement
en se servant de notre modele au lieu de dependre des images composites.
v
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Contents
A B ST R A C T
r
iv
£ su m £
v
LIST OF FIG U R ES
x
LIST OF TABLES
xi
LIST OF A C R O N Y M S A N D SYM BO LS
x ii
1
IN T R O D U C T IO N
1
2
TH EO R Y
2.1 Emission T h e o r y .......................................................................................
2.2 S S M /I ..........................................................................................................
2.2.1 Geophysical Param eterR e trie v a l.................................................
2.2.2 W ater Vapourand Cloud Liquid W a te r.......................................
2.2.3 W ind Speed ..................................................................................
2.2.4 Rain R a te .........................................................................................
2.2.5 Sea Surface T e m p e ra tu re .............................................................
2.3 A V H R R .......................................................................................................
2.3.1 SST A lgorithm ...............................................................................
2.3.2 Cloud M asking ...............................................................................
5
5
10
13
14
15
17
20
22
25
D a ta C ollection and P rocessing
3.1 D ata C o llectio n ...........................................................................................
3.1.1 Study R e g io n ..................................................................................
3.1.2 Study D u ra tio n ...............................................................................
3.1.3 D ata S o u rc e s ..................................................................................
3.2 D ata Processing...........................................................................................
3.2.1 AVHRR D a ta ..................................................................................
3.2.2 SSM /I D a ta .....................................................................................
31
31
31
33
33
35
35
37
3
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8
vii
3.3
4
5
D ata Preparation and C o-location...........................................................
42
A n alysis and R esu lts
4.1 Construction of Calibration S e t ...............................................................
4.2 CRESTech M o d e l.....................................................................................
4.2.1 Original - csst ..............................................................................
4.2.2 Modified CRESTechModel - n c s s t .............................................
4.3 L S S T ...........................................................................................................
4.3.1 Calibration S e ts ..............................................................................
4.3.2 Validation S e t s ..............................................................................
44
D iscussion and C onclusions
5.1 Future W ork.................................................................................................
93
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
45
46
46
50
54
56
73
L ist o f F igures
1.2
Contributions to the microwave radiation received at the sensor [1]. .
The radiometric sensitivity to various geophysical parameters..............
3
4
2.1
2.2
Change in microwave emmissivity with viewing angle [1]......................
Decision process for cloud masking technique..........................................
27
3.1
Image is a 5.60 day composite AVHRR SST image for June 02, 2001
at 23:24 UTC [2]. The black rectangle outlines the study area.............
A schematic of data processing flow for AVHRR....................................
A schematic of data processing flow for SSM /I......................................
Example of SAA search result map..........................................................
Example of SAA search results..................................................................
32
36
38
39
40
1.1
3.2
3.3
3.4
3.5
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
4.10
4.11
4.12
4.13
4.14
4.15
4.16
4.17
Determination of p c f cut off for calibration set.......................................
caat results for image pair 4........................................................................
AVHRR SST results for image pair 4........................................................
ASST (AVHRR - cast) results for image pair 4.......................................
An example of the caat versus AVHRR SST. Results for image pair 1 .
ncast results for image pair 4......................................................................
ASST (AVHRR - ncaat) results for image pair 4....................................
Comparison plot of L S S T vs. AVHRR SST for the calibration d ata set.
L S S T results for image pair 2....................................................................
AVHRR SST results for image pair 2........................................................
ASST (AVHRR - L S S T ) results for image pair 2..................................
Spatial plot of d ata above and below the calibration cut off for image
pair 2 ..............................................................................................................
Cloud liquid water content for image pair 2.............................................
W ater vapour content for image pair 2......................................................
L S S T results for image pair 3....................................................................
AVHRR SST results for image pair 3........................................................
ASST (AVHRR - L S S T ) results for image pair 3...................................
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8
47
48
49
49
51
53
53
57
58
59
59
60
60
61
62
62
63
ix
4.18 Spatial plot of data above and below the calibration cut off for image
pair 3..............................................................................................................
4.19 Cloud liquid water content for image pair 3.............................................
4.20 W ater vapour content for image pair 3......................................................
4.21 L S S T results for image pair 6 ....................................................................
4.22 AVHRR SST results for image pair 6 ........................................................
4.23 ASST (AVHRR - L S S T ) results for image pair...6 .................................
4.24 Cloud liquid w ater content for image pair 6 .............................................
4.25 W ater vapour content for image pair 6 ......................................................
4.26 L S S T results for image pair 7....................................................................
4.27 AVHRR SST results for image pair 7........................................................
4.28 Cloud liquid water content for image pair 7.............................................
4.29 W ater vapour content for image pair 7......................................................
4.30 L S S T results for image pair 10 ..................................................................
4.31 AVHRR SST results for image pair 10......................................................
4.32 ASST (AVHRR - L S S T ) results for image pair 10 .................................
4.33 Cloud liquid water content for image pair 10 ...........................................
4.34 W ater vapour content for image pair 10....................................................
4.35 L S S T verse AVHRR SST for image pair 9...............................................
4.36 L S S T results for image pair 1....................................................................
4.37 AVHRR SST results for image pair 1........................................................
4.38 Cloud liquid w ater content for image pair 1 .............................................
4.39 W ater vapour content for image pair 1......................................................
4.40 L S S T results for image pair 4....................................................................
4.41 AVHRR SST results for image pair 4........................................................
4.42 ASST (AVHRR - L S S T ) results for image pair...4.................................
4.43 Cloud liquid water content for image pair 4.............................................
4.44 W ater vapour content for image pair 4......................................................
4.45 L S S T results for image pair 5....................................................................
4.46 AVHRR SST results for image pair 5........................................................
4.47 ASST (AVHRR - L S S T ) results for image pair...5.................................
4.48 Cloud liquid w ater content for image pair 5.............................................
4.49 W ater vapour content for image pair 5......................................................
4.50 L S S T results for image pair 8 ....................................................................
4.51 Cloud liquid w ater content for image pair 8 .............................................
4.52 W ater vapour content for image pair 8 ......................................................
4.53 L S S T results for image pair 9....................................................................
4.54 AVHRR SST results for image pair 9........................................................
4.55 Cloud liquid w ater content for image pair 9.............................................
4.56 W ater vapour content for image pair 9......................................................
4.57 L S S T results for image pair 11 ..................................................................
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
63
64
64
65
66
66
67
67
68
69
69
70
71
71
72
72
73
75
76
77
77
78
79
80
80
81
81
82
83
83
84
84
85
86
86
88
88
89
89
90
X
4.58
4.59
4.60
4.61
AVHRR SST results for image pair 11......................................................
ASST (AVHRR - L S S T ) results for image pair 11.................................
Cloud liquid water content for image pair 11 ...........................................
W ater vapour content for image pair 11 ....................................................
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
91
91
92
92
L ist o f T ables
2.1
2.2
2.3
2.4
The spectral channels and EFOV for the S S M /I..............................
9
Spillover and cross-polarization correction factorsfor the SSM /I [3]. . 12
Rain flag for the SSM /I wind speed retrieval......................................
16
The spectral channels and IFOV for AVHRR [4]................................
21
3.1 Surface type codes [3]..............................................................................
34
4.1 A list of the image times for th e validation and calibration image pairs.
4.2 Statistical comparison between csst and AVHRR SST with pixels con­
taining cloud liquid water removed
50
4.3 Statistical comparison between ncsst and AVHRR SST with pixels
containing cloud liquid water removed.............................................
54
4.4 Singular value decomposition weight (W) results............................
55
74
4.5 Statistical comparison between L S S T and AVHRR SST..............
xi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
46
L ist o f A cronym s and Sym bols
a
AOI
AVHRR
c
CF
clw
CRESTech
csst
DMSP
DN
EDR
EFOV
e
FNMOC
h
h-pol
HRPT
IFOV
IR
LSST
A
MM
ncsst
NESDIS
NOAA
pcf
RMC
rr
SAA
Wien’s constant, 2.897 x 10 _3m • K
Area of Interest
Advanced Very High Resolution Radiometer
Velocity of light in a vacuum, 2.9979 x 10®m-s-1
Confidence Factor
cloud liquid water, kg-m” 2
Centre for Research in E arth & Space Technology
CRESTech’s model for estim ating SST with SSM /I, °C
Defense Meteorological Satellite Program
digital numbers
Environmental D ata Records files
Effective Field of View
spectral emissivity
Fleet Numerical Meteorology and Oceanography Center
Planck’s constant, 6.626 x 10-34 W-sec2
Horizontal polarization
High Resolution Picture Transmission
Instantaneous Field of View
Infrared
Langille Sea Surface Temperature model, °C
Wavelength of radiation, m
The wavelength of maximum spectral radiance, m
Main maximum
central wave number, cm -1
modified CRESTech’s model for estim ating SST with SSM /I, °C
National Environmental Satellite, D ata, and Information Service
National Oceanographic and Atmospheric Administration
percent cloud free
Royal M ilitary College of Canada
rain rate, mm-hr_1
Satellite Active Archive
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a
SMMR
SSM /I
SST
SSTA
STD
SVD
S
S'( A)
S( A)
T
ta
TIROS
ws
wv
UTC
v-pol
Stefan-Boltzmann constant, 5.670 x 10-8 W * m ~2 Scanning Multichannel Microwave Radiometer
Special Sensor Microwave Imager
Sea Surface Temperature
Sea Surface Temperature Analysis
Standard deviation
Singular Value Decomposition
Total flux em itted by a blackbody
radiance em itted by the surface, W-m -3
radiance em itted by a blackbody, W-m -3
is the thermodynamic tem perature of the object, K
Antenna Temperature
Television Infrared Observational Satellite
wind speed, m-s-1
w ater vapour, kg-m -2
Universal Coordinated Time
Vertical polarization
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 1
INTRODUCTION
A greater understanding of the spatial distribution of sea surface tem perature (SST)
in the world’s oceans has applications in a wide range of fields including meteorology
and oceanography. Estim ates of SST, turbidity and other oceanographic features
were not available from satellite until the mid-1970s [1], thus the collection of SSTs
was dependent upon ship and buoy measurements. O ther than from moored buoys
concentrated on the continental margins, the repeat tim e for measurements varied
greatly. Spacebome Infrared (IR) sensors increased the spatial coverage and shortened
the repeat tim e even though they were limited to cloud free viewing of the E arth’s
surface. The advent of spacebome microwave sensors increased the spatial coverage
of atmospheric constituents estim ations, surface wind fields, along with the potential
to measure SST. The advantage of using a microwave sensor is th at it provides a near
all-weather observation of the E arth’s solid and liquid surfaces. The wavelengths
measured by a microwave radiom eter are much longer than those seen by an IR
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2
radiometer, therefore microwaves are subject to much less attenuation by atmospheric
effects than is IR radiation. T hat being said, the sources th at affect the radiance
received by a passive microwave sensor viewing the E arth’s surface are similar to
those encountered in infrared radiom etry [5], [1].
Even though the peak of the therm al radiation spectrum for a body at
300K is in the infrared, there is a small but measurable amount of this radiation at
microwave frequencies. Thermal radiation received by microwave radiometers in space
is a combination of (a) radiation em itted by the sea surface, (b) downward emission
of the atmosphere reflected off the sea surface, (c) solar emissions and deep space
radiation reflected by the sea surface, and (d) upward emissions of the atmosphere.
These are illustrated in Figure 1. 1.
There are strong absorption bands scattered throughout the microwave spec­
trum caused by water vapour, oxygen and other atmospheric constituents. Further­
more, the microwave response of the sea surface varies with the salinity, tem perature,
surface roughness and with the presence or absence of foam.
The sea state and dielectric properties play a m ajor role in the effective
retrieval of SST. Figure 1.2 shows the radiometric sensitivity to wind speed (ws),
w ater vapour (wv), cloud liquid w ater (clw), and SST. This figure shows th at the
effect of wind speed and cloud li^ iid w ater will mask out any radiometric changes
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3
Sensor
Ocean
Figure 1 . 1: Contributions to the microwave radiation received at the sensor [1].
due to changes in SST if corrections for these param eters are not made. W ater vapour
is mainly a concern near the water vapour absorption peak at 22.235 Ghz. It has been
found th at the effect of salinity on the received brightness tem perature is negligible
for frequencies greater than 4 GHz [5].
Hollinger and Lo [9], determined th at the direct estim ation of SST using
any linear combination of SSM/I brightness tem perature channels is not possible with
appreciable confidence. The present research aims to show otherwise. In Section 2,
the background theory and geophysical retrieval algorithms for AVHRR and SSM /I
will be discussed. Section 3 will describe th e sensors (AVHRR & SSM /I) and the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4
2
120
19.35 GHz
37J) GHz
i.s
• 100
1
■
00
09
0
•0 9
-1
•
1.9
wv
sst
Figure 1.2: The radiometric sensitivity to various geophysical param eters following
Ulaby et al. [5]. The radiometric sensitivity to clw is plotted using the y-axis on the
right hand side of the plot. The plot was unnormalized using
« 0.1 K*°C_1 at
19 Ghz [6 ],
= 1.5 K mm-1 at 22 Ghz vertical [7],
= 90 K-mm " 1 at 37 Ghz
H-pol [7], and §*£ = 1.06 ± 0.16 K m "l s at 19.34 Ghz H-pol [8 ] at 55°.
procedure th a t will be followed in the rem ainder of this thesis. The presentation
and discussion of the results from the new algorithms will be covered in Section 4.
Conclusions and recommendations for future work will follow in Section 5.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 2
THEORY
2.1
E m ission T h eory
Even though the two sensors being used in this project are looking at different sections
of the electromagnetic (EM) spectrum , the physics behind their measurements is very
similar.
In thermal remote sensing, the sensor measures electromagnetic radiation
em itted by a surface. The spectral distribution of this radiation is directly related to
the thermodynamic tem perature of the surface. This spectral distribution was first
accurately measured by Lummer and Pringsheim in 1899 [10] and was theoretically
described by Planck. The spectral distribution of the radiation from an ideal therm al
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
source, a “blackbody3’ is described by Planck’s law [1], [11],
where,
5(A) = spectral radiant emittance, W /m 3
A = wavelength of radiation, m
h = Planck’s constant, 6.626 x 10-34 W sec2
T — absolute tem perature, K
c = velocity of light in a vacuum, 2.9979 x 10s m sec-1
k = Boltzmann’s constant, 1.38 x 10-23 J-K -1
The total flux em itted, 5 , by a blackbody is defined by the Stefan-Boltzmann’s
Law, which can be found by integrating equation (2 . 1 ) over the entire EM spectrum.
The resulting expression is ([1 ], [11]):
5 = oT 4
(2.2)
where o — 5.670 x 10-8W • m~2 - K~*, is known as the Stefan-Boltzmann constant.
As the tem perature of the body increases, the spectral peak shifts toward
higher frequencies. The wavelength for maximum to tal energy radiated is found using
W ien’s displacement law which reads ([1], [11]):
AmaxT = O
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2.3)
where Amo* is the wavelength of maximum spectral radiance in metres and T is the
thermodynamic tem perature of the object in K and a is the so-called Wien’s constant,
empirically determined to be 2.897 x 10-3 m • K. At 300 K, the blackbody spectrum
peaks at 9.6 fjm where the value of 5(A) is 1.69x 107 W-m”3. At microwave frequencies
of 10 and 100 GHz, 5(A) is 8.6 xlO -6 W-m -3 and 8.5 xlO -2 W-m-3, respectively.
Real surfaces are not true blackbodies and emit a lesser amount of energy,
5 '(A). Spectral emissivity e of a surface is defined as the ratio of the radiance em itted
by the surface 5'(A) to the radiance th a t would be em itted by a blackbody, 5(A), at
the same tem perature. The spectral emissivity is give by,
The emissivity of th e sea surface is about 0.98 for wavelengths ranging between 3 and
14/im (IR) and a represenitive plot of emmissivity for microwave frequencies in give
in Figure 2.1 [1]. Planck’s blackbody spectrum can be rew ritten as brightness per
unit frequency, B f, which is more typical for microwave radiometry, which reads, [1],
[11]
c2 e x p (h f/k T ) —1
(2.5)
Since, h f / k T < 1 for microwave frequencies, e x p (h f/k T ) ~ l+ ( h f/k T ) and equation
(2.5) reduces to,
2k f 2Te{$,<t>)
which is known as the Rayleigh-Jeans approximation (see [1], [11]).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2.6)
8
•a
e
•A
•a
9
Figure 2.1: General shape of the variation of emissivity with the viewing angle for
passive microwave for vertical and horizontial polarization [1].
2 .2
S S M /I
First launched on 19 June 1987, the Special Sensor Microwave Imager (SSM /I) is a
seven channel, linearly polarized, passive microwave radiometer flown onboard the
Defense Meteorological Satellite Program (DMSP) Block 5D-2 spacecraft. These
satellites have a sun-synchronous, near polar orbit at an altitude of around 833 km,
w ith an orbital period of 102 minutes. The 98.8° inclination of the DMSP satellite
orbit means th a t all but a 2.4° area centred at the poles remains unmapped [12].
The SSM /I operates at 19.35, 22.235, 37.0 and 85.5 GHz (hereafter referred
to as 19, 22, 37 and 85 GHz) and is used for a variety of geophysical and atmospheric
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
9
applications. The 19, 37 and 85 GHz channels are dual polarized (i.e., one channel
for each of the vertical and horizontal polarization per frequency) and the 22 GHz
channel is only vertically polarized. Table 2.1 displays the frequency, polarization,
tem poral resolution and effective field of view (EFOV) for these seven channels.
Table 2.1: The temporal and spatial resolution of the seven channels of the SSM/I.
(Note th a t the “V” and “H” following the frequency denotes vertical and horizontal
polarization, respectively [12].)
Frequency
Integration Period
(GHz)
(ms)
Along-track (km)
Cross-track (km)
19.35 V
7.95
69
43
19.35 H
7.95
69
43
22.235 V
7.95
50
40
37.0 V
7.95
37
28
37.0 H
7.95
37
29
85.5 V
3.89
15
13
85.5 H
3.89
15
13
EFOV on the E arth’s Surface
The scanning mechanism for the SSM /I is a broadband, corrugated, offset
parabolic reflector th a t rotates in a plane parallel to the spacecraft body at 31.6
rpm . The 61x66 cm microwave reflector has a 45° down angle (nadir angle), which
m aintains a constant footprint (viewing area) on the E arth’s surface at an incident
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
angle of 53°. For each scan, 128 discrete, uniformly spaced radiometric samples are
taken by the two 85 GHz channels and, on alternate scans, 64 discrete samples are
taken at the remaining five frequency channels [13].
The total radiance received by the antenna may be converted to an apparent
tem perature through the inversion of Planck’s Law (Equation (2.1)). This temper­
ature has not been corrected for antenna properties and for atmospheric effects. It
is known as the ‘antenna tem perature’, T \. Calibration information for the SSM /I
antenna tem peratures is collected on each rotation by passing the feed horn antenna
beneath a hot-load absorber calibrated at a nominal tem perature of 250 K and re­
flecting cosmic background radiation at a nominal tem perature of 3 K into the feed
horn’s field-of view [14]. The SSM /I has an absolute calibration of ±3 K [13].
2 .2 .1
G eop h ysical P aram eter R etriev a l
The retrieval of geophysical param eters using the SSM /I begins by converting ra­
diometric counts to antenna tem peratures and then into calibrated brightness tem­
peratures. As mentioned in the previous section, the radiometers view the E arth’s
surface along with hot and cold references th at are used to calibrate the antenna
tem peratures. These calibrations are performed at the Fleet Numerical Meteorology
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
11
and Oceanography Center (FNMOC) in Monterrey, California, using [15], [16]:
rp
A
[C (T, ~ T bc) + {TbcC h - TwCc))
(Cfc - Cc)
(2'7)
where Tw is the thermodynamic tem perature of the onboard hot body reference, and
C, Ch,and Cc are the radiometer counts when viewing the Garth, hotloadand cold
reflector, respectively. Tbc is the brightness tem perature of cold space and it varies
from 2.7 K at 19 GHz to 3.2 K at 85 GHz [13] [15] .
There are two corrections th at need to be applied to the antenna tempera­
tures before they can be converted to brightness tem peratures. The first removes the
systematic 1 K drop-off near one edge of the scan which is believed to be caused by the
feedhorn partially seeing the cold reflector [15]. And the second is offsets of 2.0, 3.5,
1.3, - 1.6 , and -0.2 K that are applied to the 19V, 19H, 22V, 37V, and 37H channels,
respectively. These offsets are to compensate for absolute errors in the measurements
[15]. The brightness tem peratures can then be found by using,
T b v = G w T av + G hvTAh + G w
(2.8)
Tb „ — GhhTAh + G vhTav + G0h
(2.9)
where, TAv and T \h are the vertical polarization (v-pol) and horizontal (h-pol) cor­
rected antenna tem peratures. T bv and T b h are v-pol and h-pol brightness tempera­
tures. The G ,mare the correction factors for spillover and cross-polarization and their
values are given in Table 2.2 for the corresponding frequencies.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
12
Table 2.2: Spillover and cross-polarization correction factors for the SSM /I [3].
19 GHz
22 GHz
37 GHz
85 GHz
Gw
1.036983
1.01993
1.0368094
1.0263188
Ghv
-0.0039359
0.0
-0.0222607
-0.0143165
Goo
-0.0892273
1.994
-0.0392815
-0.0324062
Ghh.
1.0384912
0.0
1.0421693
1.0321901
Gvh -0.0054442
0.0
-0.02762606
-0.0201877
G0h -0.0892271
0.0
-0.0392816
-0.0324065
In terms of spatial and tem poral variability, the most im portant geophysi­
cal param eters which affect the brightness tem perature are sea surface tem perature
(SST), wind speed (ws) , water vapour (wv) , cloud liquid water (clw) and rain rate
(rr) [5]. Sections 2.2.2, 2.2.3, and 2.2.4 review the formulation for estim ating water
vapour, cloud water, wind speed and rain rate. These formulas were derived using
regression techniques. The following sections will discuss the accuracy of the expected
results and th e importance of specific bands used in th e formulas where applicable.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
13
2 .2 .2
W ater V apour an d C lou d L iquid W ater
W ater vapour and cloud liquid w ater are very im portant in the transportation of
energy between the ocean and atmosphere. The evaporation of ocean waters transfers
heat across the ocean/air interface [17] [18] [19]. The 22 GHz channel is essential for
the estim ation of water vapour since this frequency corresponds to a peak in the water
vapour absorption spectrum. Thus, changes in water vapour content will have the
greatest effect at this frequency (Figure 1.2). It will also affect, to a lesser extent, the
19 and 37 GHz bands because of pressure broadening. Estim ation of total precipitable
water (water vapour) with the SSM /I is calculated using [20]:
WV = Oo + Oi • Tb 19V -I- 02 *Tb32V -+- 03 • TBtiv2 + 04 • 7 s3Tt,
(2.10)
where, o0= 232.89393 kg*m"2, ai = -0.148596 kg m ^ K "1, o 2 = -1.829125
kg-m~2-K_ l, 03 = 6.193 x 10-3 kg-m~2 -K-2, and 04 = —0.36954 kg-m- 2-K_ l. W ater
vapour retrieval has an absolute accuracy of ± 2.0 kg-m“ 2 over a range of 0 to 80
kg m ~2 in increments of 0.10 kg m -2 [20 ].
The evaporated surface w ater will eventually condense into clouds and re­
turn to the earth as precipitation. The estim ation of cloud liquid water content was
found to be [20 ]:
d w = 60 + h - TBv9U + 62 *T b w + 63 • T bz7v + 64 *TB37U
(2-11)
where, 6b = -2.838179 kg m’ 2, 6X= 8.333 x 10~3 kg m ^ K "1, 62 = -7.5959 x 10“ 3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
14
kg-m_ 2*K~l , 63 = 2.0131 x 10"2 kg-m“ 2-K -1, and 64 = -5.3066 x 10“ 3 kg m ^ K "1.
Cloud liquid water is quantized in 0.05 kg-m-2 levels over a range from 0 to 1 kg*m-2
with an absolute accuracy of ± 0.1 kg-m-2 [20 ].
2 .2 .3
W in d S p eed
The emission spectrum from a wind roughened surface is detectable throughout the
microwave spectrum as an increase of 0.5—1.5 K in brightness temperature per m-s-1
increase in wind speed [21]. There is an overall increase in the sensitivity of bright­
ness temperature to wind generated effects as frequency increases, regardless of the
polarization [11]. As the wavelength decreases, the sensitivity of the vertically polar­
ized brightness temperature decreases [14]. This decrease is attributed to an increase
in the calm water emissivity and the resulting compression of the possible range of
emissivity variations as wavelength decreases, and to a lesser extent on the changing
ocean surface wave structure [22]. Thus, the sensitivity of the vertical brightness
temperature to wind speeds depends on the presence of surface foam, whereas the
horizontal sensitivity is a combination of changes in wave structure and surface foam
effects [22 ].
A calm sea surface is characterized by highly polarized emission. As the
surface becomes rough, the emission increases and becomes less polarized (except at
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
15
incident angles above 60° for which the vertically polarized emissions decrease) [15]
(See Figure 2.1). The three mechanisms th at are responsible for the variation in the
surface emissivity due to wind speed are: ( 1) Horizontal and vertical polarization
states and the local angle of incidence are changed by surface waves with wavelength
much greater than the wavelength of radiation; (2) Sea foam increases the emissivity of
both polarizations; and (3) Surface waves th at are short with respect to the microwave
emissions cause diffraction in the microwave radiation [15].
Therefore, wind speed is inferred from passive microwave data using [20],
ws = Co 4- ci *Tb19V 4- C2 -
4 c3 • Tb^7V 4- c4 • TB„ H
(2-12)
where, the constants cq = 147.90 m-s”2, Ci = 1.0969 m s-I K -1, c2 = —0.4555
m-s- 1-K_1, c3 = —1.7600m-s-1-K_1, and c4 = 0.7860
were determined
empirically from measurements of surface wave structure, foam signatures and wind
speed. Wind speed retrievals are valid over a range of 3 to 25 m-s-1, with an absolute
accuracy of ±2 m-s-1.
2 .2 .4
R ain R a te
Rain in the E arth’s atmosphere heavily attenuates microwave radiation at the SSM/I
frequencies, thus the wind speed signature generated through surface roughness and
foam on the ocean surface is masked out, making it necessary to use a rain flag
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
16
to indicate changes in the wind speed retrieval accuracy [20], [23]. The rain flag
suggested in [20] is described in Table 2.3.
Table 2.3: Rain flag for the SSM/I wind speed retrieval.
Rain Flag
Criteria
Accuracy of
Wind Speed
0
T B37V
< 2 m-s-1
- T ssru > 50
and T b 19H < 165
1
T b stv
— T B3?h < 50
2 - 5 m-s_I
or T b ioh > 165
2
t b 3TV
-
37
5 - 1 0 m-s-1
3
t b 37V
- t b3JH < 30
> 10 m-s-1
T b 37H <
Rain is determined to be present in a pixel if the following empirical condi­
tion is met [20 ],
(-11.7939 - 0.02727 • TB„V + .09920 • T b 3TH) > OK
(2.13)
Where Equation (2.13) is true the rain rate (rr) is estimated using Equation (2.14)
[20].
r r = exp (do + di - TBbsv -I- d2 • TBbsh
*Tb?2v + d3 *TBxw) — 8.0
d3 • 7 ^ 7vr
m m /hr
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2-14)
17
where, dQ = 3.06231, dx = -0.0056036 K” 1, d2 = 0.0029478 K-1, d3 = -0.0018119
K-1, d4 = —0.00750 K-1, d$ = 0.0097550 K _1 and T buv and Tbkh are the brightness
temperatures of the 85 GHz channels averaged over the IFOV of the 37 GHz channel.
The rain rate estimates are valid over a range from 0 - 2 5 mm-hr-1 with an absolute
accuracy of ±5 mm-hr-1. If the above condition is not met, there is no rain present
in the pixel. Thus the rain rate is set to 0 mm-hr-1.
2 .2 .5
Sea Surface T em p eratu re
Hollinger and Lo [9] determined the optimal four frequency combination for retrieval
of SST, based on RMS error and confidence factor (CF) values. The 19V, 22V, 85V
and H channels were determined to be the optimal channel combination for SST
retrieval [9]. These results are only true if the 85 GHz data are averaged to match
the resolution of the other channels before retrieval, since this reduces the instrument
noise in these channels. These results agree with a previous study by Lo and Hollinger
referred to in [9]. In these studies they concluded that the SSM/I is not well suited for
the direct retrieval of sea surface temperature. This result agrees with a recent study
by Wentz [7] that the sensitivity of brightness temperature to sea surface temperature
was too small to allow for the retrieval of SST.
Nevertheless, other attem pts have been made to estimate SST from SSM/I
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
18
data. One model th at was tested in this thesis was developed at the Centre for
Research in Earth & Space Technology (CRESTech) and will be referred to in this
thesis as the CRESTech Sea Surface Temperature model or csst model. The csst
model proposes an indirect measure of SST using the SSM/I by modeling SST with
respect to columnar water vapour content (wv2 ) , through the following equation [24],
csst = 14.8 + 14.5773 • Zn(0.571 • wv2)
(2.15)
where:
w v2 = 0.1 • (235.407- (0.129241 t/19c) + v2t+ (-0.377398-t/37c))
g em "2 (2.16)
and
i/21 = (-1.86322) • v22c + (0.0062527 • (v22cf)
(2.17)
v22c
= Tb226 —0.5 *w s
(2.18)
v37c
= T b 37V —0.35 • ws
(2-19)
vl9c
= Tb xw —0.5 - w s
(2.20)
Apparently, Equation (2.15) was successfully validated in the tropics [24].
Since the csst model in this form allows no empirical calibration, a more gen­
eral form of the model, Equation (2.21), was created in order to allow the calibration
of the constant ao, ai, and a2 to our region of interest.
ncsst = a0 + ai • ln(a2 - wv)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2.21)
19
Equation (2.10), was used to estimate water vapour in Equation (2.21) as this is an
updated version of Equation (2.16) [20].
A third model, hereafter referred to as the Langille Sea Surface Temperature
(L S S T ) model, was developed by the author to determine if the SSM/I SST results
could be improved. The L S S T model is based on a linear combination of four band
differences and ws, having the form:
L S S T = go + g \ • t \ -F- g 2 •
t%4- gz *£3 + 9 \ • £4 -F- <75 • w s
(2.22)
£1
= T b xw
—T b , w
(2.23)
£2
= T ^ v
- Tb^ v
(2.24)
£3
= T b S7H
T b 19H
(2.25)
*4
= T Bm v
- T b 72V
(2.26)
where band differences, £1, £2, and £3, were chosen since the resultant values are wind
direction independent. This result is based on the assumption that the wind direction
signature does not vary significantly over the 19 to 37-GHz range [15]. Band difference
£4 ,
was included since this parameter reduced the standard error more than any of
the remaining band differences. It is well known th at the wind speed influences the
emissivity of the ocean surface, therefore a wind speed term was introduced to act as
an initial wind speed adjustment for SST retrieval. In addition, by viewing almost
exclusively with the vertical polarization channels, £3 being the only exception, the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
20
effect of wind speed is reduced since wind speed mostly affects the h-pol channels and
surface temperature mostly affects the v-pol [25].
The results of this model will be presented in Section 4.3.
2.3
AVHRR
The Advanced Very High Resolution Radiometer (AVHRR) is a five channel visible,
near infrared and thermal infrared radiometer flown on board the TIROS/NOAA
series of satellites which are in a near polar, sun synchronous orbit. These satellites
have an orbital period of approximately 102 minutes, which leads to 14.1 orbits per
day. The local hour of latitude crossing remains constant but the non-integer number
of orbits per day means that the ground track is not repeated daily. An area of ap­
proximately 2.4° around the poles remains unmapped due to the 98 degree inclination
of the satellite orbit.
W ith a scan rate of 360 scans per minute, the AVHRR collects 2048 samples
per channel per scan, which spans ±55.4 degrees from nadir [26]. On every scan the
thermal channels view an on-board hot body reference at 290 K, and deep space,
a cold body reference, of 3 K. Four platinum-resistance thermometers measure the
temperature of the hot body reference. The resultant measurements are used in the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
21
calibration of the brightness temperature (Section 2.3.1).
Table 2.4 displays the spectral band widths of each channel, along with the
IFOV. Note that an IFOV of 1.4 milliradians leads to a nadir resolution of 1.1 km for
the normal satellite altitude of 830 km [4].
Table 2.4: The spectral channels and IFOV for AVHRR [4].
Channel
Spectral bands (/zm)
IFOV (mr)
1
0.58 - 0.68
1.39
2
0.725 - 1.10
1.41
3
3.55 - 3.93
1.51
4
10.3 - 11.3
1.41
5
11.5 - 12.5
1.30
The visible channel (0.58-0.68 ^m) is used for daytime cloud and surface
mapping. Channel 2 (near IR) is typically used to delineate land/water boundaries.
The remaining three channels are located in atmospheric windows within the thermal
IR region and are used for day and night SST and cloud mapping, with the exception
of channel 3 which can only be used for night-time images because of contamination
due to reflected sunlight.
Before the raw data can be entered into most image processing software
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
22
it must be converted into a standard NOAA Level-IB format, which requires much
processing. The information received directly from the satellite contains the digital
numbers (DNs) representing the radiance received at each pixel across the scan line
when viewing the Earth, and viewing the hot and cold references. Calibration infor­
mation is also received in the form of the temperature measurements from the four
resistance thermometers measuring the physical temperature of hot load blackbody.
As well, the receiver software records the time of acquisition of the first scan, the
calibrated slope and intercept of the calibration curves (see Section 2.3.1), and the
ephemeris (set of position and velocity vectors over time) used to track the satellite
into the d ata header. The remaining data received are recorded directly into a file,
with the calibration information removed. The file is converted to a Level-IB format
using the Q tolB (Copyright Delta Data Systems Inc.) program which computes the
satellite’s position based on the time of the first received scan and the nominal orbit
as described by the ephemeris. These Earth location data (latitude and longitude) are
appended to the d ata for every 40th pixel on each line starting at pixel 25 (25,65,105,
..., 1945,1985,2025) [4].
2 .3 .1
S S T A lgorith m
In order to calculate SST from AVHRR data, the digital numbers (DN) which rep­
resent the amount of radiation received by the thermal channels must be calibrated.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
23
The thermal channel calibration coefficients (slope and intercept) are extracted from
the Level-IB data header. The slope is scaled to units of mW-m-2 • steradian • cm -1
• count -1 and the intercept to units of mW-m -2 • steradian • cm-1. Using the scaled
slope (Si) and intercept (/*), the DNs ranging from 0 to 1023 counts for Ith channel
are converted to radiant energy Ei by [4]:
Ei = (Si)~ (D N ) + Ii
(2.27)
W ith the calibrated radiance values E for each of the thermal channels, the surface
brightness temperature is calculated using the inverse of Planck’s radiation equation
W,
T{E) -
(2'28)
where /j is the central wave number of the channel (cm-1), and C\ and C 2 are constants
with values 1.1910659 x 10-5 mW m -2 * steradian • cm -4 and 1.43883 cm • K,
respectively.
Note that the central wave numbers referred to in Equation (2.28) are func­
tions of temperature and vary between satellites. These values can be found in [4] for
satellites up to and including NOAA 14.
W ith the calibrated brightness temperatures, the sea surface temperature
can be calculated using one of the following methods:
Split-window
S S T = Oq 4* fliTu -I- 02(^11 —^ 12)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(2.29)
24
Dual-window
S S T = ao + aiTu + a2(T3,7 —Tu )
(2.30)
Triple-window
S S T = ao 4- aiT u + a2(T3.7 — T i 2)
(2-31)
where, T3.7, Tu and T\2 are the brightness temperature values from Equation (2.28)
for channels 3, 4, and 5, respectively. An initial estimate of SST is made from T u.
Since the effects of the atmosphere are different at different wavelengths, spectral
differences are used as an empirical atmospheric correction. The coefficients ao, ai
and a2 are referred to as the McClain coefficients and depend on the satellite and
time of day. The approximate physical meaning for ao is the conversion from Kelvin
to Celsius, ai is an approximation of the surface IR emissivity and a2 acts as an
atmospheric correction. There is a fourth term which can be added to the above
equations, a3(Ti —Tj){sec{<j>) — 1), which accounts for the path length differences,
where <f>is the angle betwen the zenith line (pointing straight up) and the direction
to the satellite (i.e., satellite zenith angle), Ti and 7} are the brightness temperatures
of two of the three thermal channels depending on which method is used (split-, dualor triple-window). This term also accounts for changes in emissivity with respect to
incident angles.
Since the 3.7 fim channel is highly contaminated by reflected sunlight during
the day, only Equation (2.29) can be used during the day. Thus Equation (2.29) was
used for all AVHRR SST calculations and the expected root-mean-square difference
for this method is 0.78°C during the day and 0.63°C at night. The use of the 3.7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
25
fim channel was expected to improve the SST retrieval accuracy as long as the sensor
noise was low, but there is a noise problem in this channel [1]. It should also be noted
th at the radiation received in the 3 and 14 ^m wavelength range comes from the
top 0.1 mm of the water surface, and therefore the SST estimation is for the surface
“skin” only, see [1].
Having completed the calculations, each pixel is assigned a latitude/longitude
coordinate pair. Using the information stored during the scan (51 GCPs per line), the
latitude and longitude of each pixel on the line is determined by a linear interpolation.
For a full discussion see [27].
2 .3 .2
C loud M asking
In order to retrieve reliable sea surface temperatures for AVHRR or other IR sensors,
pixels contaminated by clouds must be detected and removed. Clouds are typically
classified by their high albedo and low temperatures [28]. A pixel that is completely
filled by cloud is easily identified as cloud due to its low temperature. Cloud detection
becomes a problem when a pixel is only partially filled. In these pixels, the apparent
SST is a linear combination of the true SST and the temperature of the cloud tops, and
is somewhat lower than the true SST. Often, this temperature is not an unreasonable
SST, but will be too low in a regional context. This is mainly a problem in areas of
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
cloud over warm water.
There are a wide variety of techniques for detecting clouds, from a simple
threshold to complex solutions using artificial intelligence. A simple threshold that
can be applied is either an albedo (daytime) or a brightness temperature (nighttime).
These conditions can be applied because of the fact th at clouds have high albedo and
low temperatures when compared to the sea surface. The problem with a threshold
mask is that the threshold varies from scene to scene depending on the cloud free
background. The cloud free background varies as a function of wind speed which
affects surface albedo, and atmospheric aerosols in the visible spectrum and is a
function of surface temperature, atmospheric temperature and humidity [29].
A combination of various techniques has been found to be more successful
than an individual cloud removal technique [28]. Two methods were considered in this
thesis. The first method described by Gallegos, et al. [28], combines multiple band
signatures and edge detection using cluster shade texture measurements based on the
gray level co-occurrence matrix. The second, described by Thiermann and Ruprecht
[29], uses a combination of a modified spatial coherence test and an absolute threshold
test. The Thiermann and Ruprecht method [29] was chosen because of its simplicity,
effectiveness and the fact th at the algorithm is the same day or night. This method
will be outlined in the paragraphs that follow and a synopsis is given in Figure 2.2.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
27
AVHRR SST
IMAGE
Automatic Scene
Threshold Determination
Xi, X2, ^3,
or X4 > 0.25
Yes Remove from
calculations
No
No
cloud free
pixel
Yes
histogram
cloudy
pixel
bin SST < -2°C
Yes Remove from
calculations
No
bin freq. < 5%
bin SST < MM
Yes Remove from
calculations
No
cloud free
pixel
No
SST < TH
Yes
cloudy
pixel
determine 95%
Cumulative Freq.
bin SST
Subtract 2°C
scene threshold
(TH)
Figure 2 .2 : Decision process for cloud masking technique.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
28
To classify a pixel with brightness temperature 722 as cloudy, its brightness
temperature is compared to its nearest neighbours.
Tn
T x2 T13
T21 T22 X23
731 T32 X33
In the classical standard deviation test, the standard deviation of all nine
pixels is computed and compared to a threshold. In [29] half of the absolute difference
between T22 and the neighbouring pixels in the north-south, east-west and diagonal
directions is considered:
I I = (|7 i2 — T 2 2 I + IT32 — I 2 2 D/ 2
* 2 = (1 ^ 2 1 — T 22 I + 17 2 3 ~ T 22 D / 2
X3 =
( |T n — 722| + |733 — 7 2 2 D /2
X4 = (|7 i3 — 7221 + iXil ~ ^ 2 2 |)/2
and if any of the four test values, x i to x4, exceed the threshold X c, T22 is classified as
cloudy, otherwise it is clear. In this thesis, the threshold X l was set to 0.25 following
Thiermann and Ruprecht [29]. See the top left side of Figure 2.2. The coherency test
used in [29] differs from the classical standard deviation test since the weight of the
pixel being tested is increased. This increase in weighting raises the coherency test
sensitivity in low cloud conditions [29].
The coherency test works well for detecting cloud edges and areas where
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
29
there are differences in cloud top temperature due to cloud thickness. But, for areas
of homogenous clouds, the coherency test was unable to classify these pixels as clouds
since there is a very small difference in temperature. In regions where the cloud cover
is homogenous, the AVHRR pixels will be completely cloud filled, therefore easily
detectable due to their low temperature. A low temperature cut off was determined
by the following automatic procedure (see right-hand side of Figure 2.2). First, a
coherency test was run with X l = 0.05 [29]. After the coherency test, all the remain­
ing pixels with temperatures less than or equal to -2 .0 ° were eliminated since their
low temperature indicated that they were cloud or ice. A histogram was constructed
from the remaining pixels. Next, temperature bins with a frequency count less than
5% of the remaining pixels and on the low temperature side of the main maximum
(MM) were considered cloudy and eliminated from further analysis. Finally, the cu­
mulative frequency was calculated starting at the warmest temperature bin, and the
bin temperature that corresponded to a 95% cumulative frequency was extracted.
The temperature threshold for the scene was then set equal to the extracted tem­
perature minus 2°C. The automatic scene threshold was applied to the AVHRR SST
image, and all pixels with SST less than the scene threshold were flagged as cloud
and eliminated from further processing (lower left of Figure 2.2).
Thermann and Ruprecht [29] noticed th at their technique was too sensitive
at high look angles, the outer 10 % of the swath. Also, as with all coherency tests,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
30
this technique performs poorly for low warm clouds with temperatures close to that
of the sea surface.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 3
Data Collection and Processing
3.1
3 .1 .1
D a ta C o llectio n
S tu d y R egion
The study area extended from 35°N 75°W to 50°N 40°W. This region was selected
because it is visible by AVHRR sensor while the sensor is in the line of sight of the
RMC ground station. Furthermore, SSTs ranging from near freezing in the northern
regions to mid-20s to the south made this area attractive, since it would allow the
testing of the algorithm over a large range of temperatures and various atmospheric
and surface environments. The large temperature gradient along the north wall of
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
32
Figure 3.1: Image is a 5.60 day composite AVHRR SST image for June 02, 2001 at
23:24 UTC [2]. The black rectangle outlines the study area.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
33
the Gulf Stream (Figure 3.1) and the warm and cold eddies associated with the Gulf
Stream allows for a validation of the algorithm for both absolute surface temperature
and temperature gradients.
3 .1 .2
S tu d y D u ration
Data collection for this study commenced on 20 May 2001, and was prematurely
terminated on 6 June 2001 due to an antenna positioner failure at the RMC AVHRR
ground station. However, in this 17 day period, a sufficient number of clear images
were obtained for the study to proceed.
3.1 .3
D a ta S ou rces
The SSM/I d ata used in this study were acquired from the Satellite Active Archive
(SAA) operated by the National Oceanic and Atmospheric Administration (NOAA).
The selected d ata sets were downloaded from ftp.saaspl.saa.noaa.gov. These data
records contain geolocation information and surface type codes (Table 3.1) along
with antenna temperatures for each channel in the scan line. The data from SAA are
12-bit precision antenna temperatures [30].
The AVHRR data were received directly from the satellite by a ground sta­
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
34
tion operated by the Physics Department at RMC. The system, supplied by Quorum
Communications Inc, consists of a 1.2m parabolic dish, which is controlled by Quo­
rum software on a Pentium PC. Following the process outlined in the upper portion of
Figure 3.2, the raw data were stored directly on the disk system of a Unix workstation
via a computer network connection for further processing.
Table 3.1: Surface type codes [3],
Surface Type Code
Surface Type
0
Land
1
Vegetal - covered land
2
Spare
3
Permanent sea ice
4
Possible sea ice
5
Water
6
Coast
7
Index not available
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
35
3.2
D a ta P ro cessin g
3 .2 .1
A V H R R D a ta
The lower portion of Figure 3.2 gives an overview of the data processing procedure
applied to the AVHRR data, which will be expanded upon in the following para­
graphs. Once the imagery was received, each image in the potential data set was
viewed using image processing software to determine which passes visible from RMC
covered the study area and were sufficiently clear. The majority of the scans, which
were viewing the area of interest (AOI), were taken at night or early dawn. There­
fore, the visual inspection was conducted using the three thermal channels where the
clouds’ colder brightness temperature was used to roughly determine the concentra­
tion and distribution of cloud in the image. Images were eliminated: (1) if they were
predominately cloud, or (2) if excess receiver noise existed in the data. The selected
files were transferred to a PHI 700 MHz PC for further processing.
The Q tolB program (Delta Data Systems) was used to add the geographic
information to the Quorum Level-IB files, converting them into NOAA Level-IB
format. A further IDL program made minor adjustments to the file header so that
ENVI would recognize these d ata as proper NOAA Level-IB files.
In order to determine the appropriate McClain coefficients for the SST com-
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
36
RMC
Ground Station
Transfered to
Stored to disk
8 mm data tapes
Unix Network
drive
Desired
Images list
Visual Inspection
Network drive
Transfer desired
Images to PC
QtolB
HeacLadd
Day/Night
determination
SST image
calculation
Cloud detection
algorithm
SST image
Masked SST Image
(*img)
Cloud Mask
Geocorrected
Masked SST Image
(*-png)
Figure 3.2: A schematic of data processing flow for AVHRR.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
37
putations, an image must be classified as either a daytime or a nighttime pass. Fol­
lowing Landry [26], one thousand pixels were randomly selected and the channel 1
radiance were sampled. The images whose pixel average was greater than or equal
to the threshold set by Landry [26] were classified as daytime passes. All others were
classified as nighttime.
The next stage in the processing was to calibrate the thermal channels and
calculate the SST. The processes involved in the calibration of the thermal channels
and the estimation of SST were discussed in Section 2.3.1. Upon completion of the
SST calculations using the Split Window method, the non-geocorrected SST image
was input into the cloud masking algorithm described in Section 2.3.2. The resulting
mask was a series of Is and Os corresponding to cloud free and cloud contaminated
pixels, respectively. The mask was then multiplied by the SST image producing a
non-geocorrected, masked SST image. This image was used later for comparison with
SSM /I’s results. Another output from the AVHRR processing was a geocorrected,
masked SST *.png file.
3 .2 .2
S S M /I D a ta
Figure 3.3 is summary of the data processing procedure that was applied to the SSM/I
data. The initial selection process of the SSM/I TDR files used in this thesis was
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
38
Sun Workstation
SAA
Search
r
TD R 1
ssmitdrta.c
files
ssmitdrtb.f
Transferee! to PIII"
Average 85 GHz
Channels
Subset data
to AOI
Yes
Data was removed form
further processing.
Surface type No
Data was removed form
further processing.
Test for
rain
1
No
Yes
L
Calculate
Invironmenta] ■
parameters
Unformatted
binary file
F*.dat
Figure 3.3: A schematic of d ata processing flow for SSM/I.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
J
39
Figure 3.4: SAA map displaying the SSM/I search results for May 26, 2001 between
06:26 UTC and 12:26 UTC. Area of interest is outlined by the dashed rectangle and
the solid rectangles corresponds to the area covered by the SSM/I file with the same
color. See figure 3.5
based on scan location and timing. The SSM/I sensor had to pass over the study
region within ±3 hours of the corresponding AVHRR pass. For each search there
were between 3 and 6 scans which met the above criteria. From these scans, the
pass th at was closest to the AVHRR scan temporally and covered a large portion
of the study area was selected. Figures 3.4 and 3.5, are an example of the map
showing the possible passes and the corresponding names of the files meeting the
search criteria, respectively. In this example, orbit 31843 was selected since it was
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
40
31842 2001/05/26 08:19:40 6256 F13 NPR.TDRR.S7D01146JS0819J1003B3184243
31843 2001/05/26 10:01:20 6233 F13 NPR.TDRRS7D01146.S1001 J 2 145*3184344.NS
31844 2001/05/26 11:42:44 6225 FI 3 NPR.TDRRJS72)01146JS1142J2326£3184445.NS
7505 2001/05/26 12:03:31 6241 F15 NPR.TDRR.S9JXJ1146JS1203J1347J30750506.NS
21369
2001/05/26 12:17:59 8405 FI 4 NPR.TDRR.S8D01146.S1217£L438£2136970NS
21370
21369
2001/05/26 12:17:59 11238 F14 NPR.TDRR.S8£ 01146.S 1217£1525£2137071NS
21370
Figure 3.5: SAA SSM/I TDR file search results for May 26, 2001 between 06:26 UTC
and 12:26 UTC.
the closest temporally to the AVHRR pass, 09:26 UTC, and covered a large ocean
surface area within the AOI.
The selected files were downloaded and converted to a more usable format by
two computer programs obtained from the NOAA SAA. The first, ssmitdrta.c, reads
the data and outputs the satellite identification, UTC, revolution number, antenna
temperatures, surface type, latitude and longitude for each pixel. All the latitudes
range from -90 to 90 degrees, with positive latitudes being north and negative indi­
cating southern latitudes, and the longitude ranged from 0 to 360° east. The output
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
41
antenna temperatures were then converted into brightness temperatures using the
ssmitdrtb.f code which is based on a method described by Equations (2.8) and (2.9)
(see Section 2.2.1). The resulting output is identical to the input file, but has the
antenna temperatures replaced by the brightness temperature.
Since there was no partioning into spatial subsets performed at SAA, the
next step in the process was to subset the images to the study area. Next, the reso­
lution of the 85 GHz channels had to be degraded to match the other channels. This
resolution degradation or averaging was conducted by taking all the 85 GHz pixels
for each polarization which were within 0.25 degrees of the central latitude/longitude
of each of the low resolution channels. The distance between each pixel and the lowresolution reference in metres was calculated using the map_2points function in IDL.
The average brightness temperature was set equal to the weighted average of the 9
nearest points with the co-located point receiving a weight of one and the remaining
8 points a weight of 0.5. If there were fewer than 9 points the average was calculated
in the same manner as long as there was at least 4 points. In the unlikely event that
there were less than 4 point, the average was set equal to the brightness temperature
of the co-located point.
Since the aim of this study was to validate/calibrate a model, only th e pixels
containing no rain were used. This condition represents the most ideal situation for
comparison. Furthermore, only pixels with a surface type 5 referring to water (see
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
42
Table 3.1) were used in the calculation of wind speed, water vapour, cloud liquid, and
SST. All other surface types were eliminated.
The geophysical information was output as an unformatted binary file con­
taining the longitude, latitude, surface type, wind speed, water vapour, cloud liquid
water, rain rate, csst, and the seven brightness temperature channels.
3.3
D a ta P rep a ra tio n and C o-location
Once the two sensor files were processed (see Sections 3.2.1 and 3.2.2) the correspond­
ing file pairs were randomly divided into two groups. One of the groups was used for
calibration of the models, while the other was used for validation. In total there were
11 image pairs, 5 were used for the calibration and 6 for the validation. The initial
stage of image processing for all image pairs was the same and will be explained in
the paragraphs that follow.
For calibration and validation of the models, the AVHRR SST estimates
were used as the standard on which the model was compared. To achieve a spatial
scale with AVHRR comparable to th at of SSM/I, all the cloud free SST pixels from
AVHRR were averaged over the IFOV of the 37 GHz channel. Three criteria for a
pixel to be entered into the average AVHRR SST calculations were used. First, the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
43
pixel had to be w ithin 12.5 km of the central latitude/longitude of the SSM /I pixel
under consideration. Next, the AVHRR SST had to be less than 40° C, and not equal
to -2°C (cloud/land mask value). From the pixels th at met these conditions an initial
average was calculated. As a way of eliminating cloud contam inated pixels th at were
missed by the cloud masking technique, any pixel with a SST less than the average
minus 2.5° C was eliminated and the average and standard deviation was calculated
for the remaining pixels. The percent of the IFOV th at was cloud and ice free was
calculated using:
,
slcount , ^
p c f = —jjg — * 100
„.
(3.1)
where, slcount is the number of cloud free pixels used in the calculation of the average
AVHRR SST forthe
FOV,and 443±37.5 is the total number of AVHRRpixels which
could be contained in the FOV.
As each pixel was passed through the processing algorithm those pixels
which met the selection criteria, slcount greater than 22 and a csst between -2°C and
30°C, had their information output to a file. The following information was stored for
each successful pixel grouping: longitude, latitude, surface type, 7 SSM /I brightness
tem peratures, wind speed, wv, clw, rr, csst, mean AVHRR SST, STD of the AVHRR
SST, p c f from AVHRR.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 4
Analysis and Results
In this chapter, the three models introduced in Section 2.2.5 are tested. Model results
are analysed and compared. Five image pairs were randomly selected from Table 4.1
to calibrate the ncsst and L S S T algorithms. The remaining six image pairs were
used for validation. The same six image pairs were used as validation images for
all the algorithms so th a t a fair comparison could be made between models. This
ensured th at the same environmental conditions were in effect for all the results.
The results presented in the following sections are displayed on a 0.25° x 0.25°
grid. These grid points were assigned data values using a ^ weighted average of the
d ata points which were within 0.25° of the grid point. T he statistics were calculated
before and after the re-averaging and there was no significant change in the statistical
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
45
results.
4.1
C o n stru ctio n o f C alib ration S et
The five data sets listed as calibration d ata sets in Table 4.1 were combined into one
calibration file. Any SSM /I pixel which had a corresponding AVHRR result of: (1)
an AVHRR STDEV greater than 1.0°C, (2) AVHRR SST less than 10° and a latitude
less than 40°N, (3) AVHRR SST less than 7°C and a latitude less than 45°, or (4)
AVHRR SST less than 2°C was eliminated. The 40°N line approximates the north
wall of the Gulf Stream during the study period, so th at this criterion effectively
removes from the modeling any cloud contam inated AVHRR pixels th at might have
been missed in the formal procedure.
For the L S S T model, the calibration set was reduced further. The root
mean square difference between the model’s estim ate of SST and AVHRR’s estim ate
of SST was plotted for p c / from 5% to 90%. At each p c f increment, the constant
go through g$ (see Section 2.2.5) were recalculated for the L S S T model and a new
L S S T estim ate was made. The change in slope in Figure 4.1 was used to determine
the p c f cutoff for the calibration set. The resultant cutoff was determined to be a
p c f equal to 50% (see Figure 4.1).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
46
Table 4.1: A list of the image tim es for the validation and calibration image pairs.
AVHRR
Image Pair
SSM/I
Val./Cal.
date
time
date
time
1
May 21
20:14
May 21
21:03
Val.
2
May 25
09:38
May 25
10:14
Cal.
3
May 25
21:05
May 25
21:50
Cal.
4
May 26
09:26
May 26
10:01
Val.
5
May 26
20:53
May 26 21:38
Val.
6
May 27 09:15
May 27 09:48
Cal.
7
May 27
20:42
May 27 21:25
Cal.
8
May 29
08:52
May 29
11:03
Val.
9
May 31
10:06
May 31
10:36
Val.
10
June 1
09:54
June 1
10:23
Cal.
11
June 2
09:42
June 2
10:10
Val.
4.2
C R E ST ech M o d el
4 .2 .1
O riginal —csst
The csst model (see Section 2.2.5) proposes an indirect measure of SST using the
SSM /I by modeling the change in SST with respect to changes in columnar water
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
47
3-0
2.8
„ 2.6
U
c
</> 2.4
2
2.2
2.0
0
20
40
60
pcf
ao
100
Figure 4.1: Determination of p c f cut off for calibration set.
vapour. If the model were to perform well, then the csst SST field would resemble that
shown in Figure 4.3, estim ated from the corresponding AVHRR data set. The spatial
distribution and concentration of SST estim ated using csst model for image pair 4,
Figure 4.2, shows the results of the csst estim ation of SST. The resulting tem perature
field was small, ranging from 8 to 14°C, for th e region. Typically SST values for this
region would range from approximately 2°C in the northern region to the mid-20’s
in the southern region. Also the tem perature gradient (Figure 4.2) appears to be
in a south-east north-west direction rather than the expected north-easterly flowing
structures displayed in Figure 4.3. Also note the much larger tem perature variation
(0 to 26°C) across the study region seen in Figure 4.3. Figures 4.4 is a spatial plot of
th e difference between the SST predicted using AVHRR minus the csst predictions.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
48
In this figure there is a large area having a tem perature difference greater than ±5°C,
w ith a narrow transitional region where the difference changes sign. This result would
be expected even if the model produced a constant tem perature field.
75°W
70°W
65°W
60°W
55°W
50°W
45°W
in
o
la
in
fO
LP
75°W
70°W
65°W
60°W
5Q°W
25
4 5 °w
35
Figure 4.2: csst results for image pair 4.
Similar results were observed in the remaining validation images, and will
not be presented in this thesis. The statistical results for all the validation images are
summarized in Table 4.2. The large mean and standard deviation in the differences
between the AVHRR SST and csst prediction indicates th a t csst model is unable to
estim ate SST in the study region in its present form.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
49
70°W
65°W
60°W
55°W
50°W
7 0 aW
65°W
6 0 °W
55°W
50°W
45°W
3 5 “N
40°N
N097
45°N
75°W
•
5
If
15
2«
25
M
Figure 4.3: AVHRR SST results for image pair 4.
70°W
6 5 °W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
4-5 °W
NoQ£
35°N
40°N
Nodi’
45°N
N .s t’
75°W
<-5 -5 -4 -3 -2 -1 0 1 2 3 4 5 5<
Figure 4.4: ASST (AVHRR - csst) results for image pair 4.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
50
Table 4.2: Statistical comparison between csst and AVHRR SST with pixels contain­
ing cloud liquid w ater removed.
4 .2 .2
Date
Time
number of
data points
May 21
21:30
486
7.46
4.00
May 26
10:01
1290
6.79
4.01
May 26
21:37
1529
4.14
4.79
May 31
10:37
889
6.87
5.17
June 2
10:10
1115
4.31
4.84
Mean Difference STD
(°C)
(°C)
M od ified C R E ST ech M od el —ncsst
In an attem pt to improve the performance of the CRESTech model, two modifica­
tions were made. The method used to estim ate water vapour in the csst model from
Hollinger [13] was replaced with an improved version from Hollinger [20]. Also, when
the csst results were plotted versus the AVHRR SST estim ates (Figure 4.5) the ex­
istence of two populations was noticed. One population corresponded to the water
vapour being greater than 15 kg/m 2 while th e other corresponded to the water vapour
being less than or equal to 15 kg/m 2.
Therefore, the calibration data set (see Section 4.1) was separated into the
two populations from which the constants for the ncsst model were determined. Each
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
51
18
-2 ------------------------------------------------------------------------------------------
AVHRR88T (* f. C)
Figure 4.5: An example of the csst versus AVHRR SST. Results for image pair 1.
population group was entered into a gradient-expansion algorithm to compute a non­
linear least squares fit to equation 2.21 [31], with the initial constants set to those
in Equation (2.15). Iterations are performed until the change between the estim ated
constants changed by less than 0.1. T he difference between the resulting constants in
the first and second iterations where less than 2 x 10-4 therefore, no further iterations
were necessary. The results of the iterative solution for the modified SSM /I SST
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
52
estim ation model are given in the following equations,
wv < 15.0 kg/m 2:
w v > 15.0 kg/m 2:
ncsst = x0 4- xi • ln(x 2 • wv)
ncsst = 20 +
• Jn(x2 • uw)
where, xo, x j, and x2 are equal to 17.07°C,19.76°C and 0.0660 m2 kg_1, and
(4.1)
zq , z \,
and z2 are equal to 16.65°C, 4.40°C, and 0.0739 m2*kg~l, respectively.
Figure 4.6 is an example of the results obtained by the ncsst model. Com­
paring these results w ith those displayed in Figure 4.2, there is a noticeable increase
in tem perature field values Sargasso Sea from 8-14°C in Figure 4.2 to 14-18°C in
Figure 4.6. This increase in tem perature is an improvement over the behaviour of
the csst model as it approaches th e expected values of 18-24°C surface tem peratures
for this region. The region north of the Gulf Stream in Figure 4.6 however, shows
SST higher (8-18°C) than expected (2-12°C). However, these modifications have not
changed the general shape of the SST field. A comparison between the AVHRR re­
sults and ncsst results are displayed in Figure 4.7. Comparing the statistical results
shown in Tables 4.2 and 4.3, the mean difference and standard deviations have been
reduced by the modification to the original model but not to a point where the model
is usable. Therefore, even with these modifications, the csst model was not able to
reproduce the SST field as estim ated by the AVHRR estim ations.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
53
70°W
65°W
60°W
55°W
50°W
45°W
7 5 “W
7Q°W
65°W
60°W
55 W
50°W
45°W
NoQf
35°N
40°N
NoOt>
45°N
N .St’
75°W
25
N
Figure 4.6: ncsst results for image pair 4.
70°W
65°W
6 0 °W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
4 5 °W
35°N
NoQC
40°N
NoOt’
45°N
NoSt'
75°W
<-5 -5 -4 -3 -2 -1 0 1 2 3 4 5 5<
Figure 4.7: ASST (AVHRR - ncsst) results for image pair 4.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
54
Table 4.3: Statistical comparison between ncsst and AVHRR SST with pixels con­
taining cloud liquid water removed.
4 .3
Date
Time
number of
d ata points
Mean Difference
(°C)
STD
(°C)
May 21
21:30
486
3.98
4.32
May 26
10:01
1290
2.05
3.78
May 26
21:37
1529
0.35
4.52
May 31
10:37
889
2.46
5.16
June 2
10:10
1151
0.14
4.65
LSST
Singular value decomposition (SVD) was used to confirm th at the variables chosen for
the L S S T model on a physical bases (see Section 2.2.5) had the greatest influence on
the SST estim ate. A M xN m atrix was created from the calibration data set, where
M was the number of d ata points in the calibration d ata set, and N is the number
of param eters. The 18 param eters consisted of all the individual channel brightness
tem peratures and all possible like-pol differences, as well as the wind speed and a
constant.
In an over-determined system like this, where there are more equations than
unknows, the SVD procedure is equivalent to a linear least squares fitting procedure,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
55
but has the added advantage of producing a diagonal m atrix th a t estimates the rel­
ative contribution of each param eter to the fit [32]. This diagonal or weight m atrix
shows th at there are only 6 param eters th at have a relative contribution of greater
than 10~4, implying th at the influence of the others is negligible (see Table 4.4).
Table 4.4: Singular value decomposition weight (W) results.
Variable
W
W /m ax(W )
Variable
V19-V22
37843.4
1.000
V37-V85
2.54 xlO "2 6.731 xlO -7
V37-V22
1133.69
3.00 xlO "2
H37-H85
4.02 xlO "4
1.06x10"®
H37-H19
352.77
9.32 x l0 “3
V19
8 .8 5 x l0 "5
2.34x10"®
V85-V22
172.00
4.54 xlO "3
H19
6.12x10-®
1.62x10"®
constant
119.61
3.16 xlO "3
V22
4.07x10"®
1.08x10"®
ws
68.66
1.81 xlO "3
V37
3.42x10"®
9.05 xlO “ 10
V19-V37
0.141
3.73 x l0 “6
H37
9.76x10“®
2.58x10"®
V19-V85
1.35 xlO "3
3.58x10“®
V85
3.53x10"®
9.34 xlO” 10
H19-H85
4.54 xlO "4
1.20x10-®
H85
3.74x10"®
9.88 xlO "10
W
W /max(W )
The relative weights in Table 4.4 of the param eters not chosen for the L S S T
model (Section 2.2.5) are approximately equal to or less than the floating point ac­
curacy of the computations, thus are numerically singular and do not contribute any
useful information to the L S S T model. These SVD results confirm th a t the parame­
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
56
ters chosen for the L S S T model on physical principles are those that have numerical
significance.
Singular value decomposition also yields the values of the constants (i.e.,
go, . . -,gs). The M xN m atrix was redefined to contain only the six most significant
param eters. The SVD results from 2796 data points from the calibration data set
yielded:
nsst = go + <71 • t\ 4- g-i - t<i 4- <73 ■£3 + <74 ' ^4
9s ’ ‘ws
(4*2)
where, g0, gi, 02 » 93 , 94, and gs are equal to 97.386°C, 3.314°C-K-1, -2.743 °C-K_1,
-0.862
0.093 °C-K_1, -0.366 °C s-m“ l , respectively. The root mean square
difference between the AVHRR SST and L S S T was 2.55°C for the calibration set.
4 .3 .1
C alib ration S ets
Figure 4.8 shows th at the L S S T results for the calibration set corresponds to the
AVHRR SST data from which it was created. It is also noticeable th at there is a gen­
eral over estim ate of SST for AVHRR SST less than 10°C and a slight under-estimate
of SST at AVHRR SST in the 20°C range. The two aggregations of points appearing
in Figure 4.8 are due to the lack of data points at interm ediate temperatures. The
lack of d ata is due to a strong gradient between warm and cold water in the study
area.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
57
20
cj
-to
0
to
20
30
LSST (C)
Figure 4.8: Comparison plot of L S S T vs. AVHRR SST for the calibration d ata set.
For the remainder of this section, the results from the individual calibration
image pairs will be presented. It should be noted that the data with a p c f less than
the calibration cutoff of 50% are included in the spatial plots.
Im age P air 2: M ay 25 , 2001 (m orning)
The large scale SST features present in the AVHRR SST estim ate (Figure 4.10)
are reproduced by the L S S T estim ate (Figure 4.9). Note, the approximate 12 to
14°C SST (light blue and light green pixels) estim ates in the south west portion of
Figure 4.10, which were not used as d ata for the model creation since this SST does
not make sense oceanographically. Figure 4.12 shows the location of the d ata points
which were above (red) and below (blue) the p c f cut off used to determine which
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
58
AVHRR SST were used in the calibration of the L S S T model. The large positive
differences (red) in Figure 4.11 are located near cloudy regions where the confidence
level th at the AVHRR pixel is cloud free is low (Figure 4.12).
The area in the south west corner of the image was not used for calibration
of the model due to the presence of clw (Figure 4.13). For this area, the L S S T
estim ate was expected to be less than the AVHRR SST but the L S S T model is
actually over-estimating in this region. The water vapour content is greater than 25
kg*m-2, Figure 4.14, which may indicate th a t atmospheric tem perature is affecting
the L S S T results. Also note th at the white areas within the scan are areas containing
no d ata (i.e., rain).
75°W
70°W
65°W
50°W
55°W
50°W
a
2T
(J1
75°W
70°W
65°W
60°W
55°W
50°W
45°W
Figure 4.9: L S S T results for image pair 2.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
75°W
70°W
65°W
•
5
60°W
It
5 5°W
15
»
5 0 °W
25
45°W
3«
Figure 4.10: AVHRR SST results for image pair 2.
70°W
65°W
6 0 °W
55°W
50°W
45°W
75°W
7 0 °W
65°W
60°W
55°W
50°W
45°W
35°N
N-GC
40°N
N„0t-
45°N
N.Gt-
75°W
<-5 -5 -4 -3 -2 -1 0 1 2
Figure 4.11: ASST (AVHRR - L S S T ) results for image pair 2.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
60
75°W
70°W
65°W
60°W
55°W
50°W
45°W
O'
o
O
*<r
04
in
m
LP
75°W
70°W
65°W
60°W
55°W
50°W
45°W
Figure 4.12: Spatial plot of data above and below the calibration cut off for image
pair 2.
75°W
70°W
65°W
60°W
55°W
5Q°W
45°W
r
in
''T
O
«*r
-SbO
i
in
ro
75°W
70°W
65°W
60°W
55°W
50°W
04
<_n
45°W
<0 OlO OH 0 2 03 0.4 0 5 0 6 0.7 0 8 09 LO
Figure 4.13: Cloud liquid water content for image pair 2.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
61
75°W
70°W
65°W
60°W
55°W
50°W
45°W
I—
L
-t.
o
<J1
75°W
70°W
65°W
60°W
55°W
50°W
4 5 °W
TO ' 80
Figure 4.14: W ater vapour content for image pair 2.
Im age P air 3: M ay 25, 2001 (night)
The L S S T model produced reasonable SST estim ates (Figure 4.15) over both the
region of confidence in the AVHRR SST scan as well as over the region of lower
confidence (Figure 4.18). Comparing Figure 4.15 and 4.16 there is a reasonable
agreement in the large scale SST field.
As noticed in the earlier image, 25 May 2001 at 10:14 UTC, the L S S T
model over-estimates in areas were there is clw (Figure 4.19) and wv (Figure 4.20) in
quantities greater than 25 kg-m-2 . The L S S T model under-estim ates the SST (see
Figure 4.17) in the south eastern section where the wv content is less than 25 kg-m-2
and clw is present.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
62
70°W
65°W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
NoSf
35°N
N.OV
40“N
N»Sfr
45°N
75°W
•
5
18
15
28
25
38
Figure 4.15: L S S T results for image pair 3.
70°W
65°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
N.SC
35°N
N.Ofr
40°N
NoGf
45°N
75°W
8
5
60°W
18
5 5°W
15
28
50°W
25
45°W
38
Figure 4.16: AVHRR SST restilts for image pair 3.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
63
75°W
70°W
65°W
60°W
55°W
50°W
45°W
NoSfr
-e.
cn
N»0fr
o
■e.
7
NoSf
1
75°W
; ~ i
70°W
65°W
60°W
55°W
u>
50°W
45°W
< - 5 - 5 - 4 - 3 - 2 - 1 0 1 2 3 4 5 5<
Figure 4.17: ASST (AVHRR - L S S T ) results for image pair 3.
75°W
70°W
65°W
60°W
55°W
50°W
45°w
N„Sfr
-Ca.
NoOt'
o
N„5C
oen
*
7 5 °W
70°W
65°W
6 0 °W
55°W
5 0 °W
4 5 °W
Figure 4.18: Spatial plot of data above and below the calibration cut off for image
pair 3.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
64
7 0 °W
6 5 °W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
35°N
N.SC
40°N
NgOt'
No St’
45°N
75°W
<0 0.0 0.1 0.2 03 0.4 0.5 0.6 0.7 0 8 0.9 1.0
Figure 4.19: Cloud liquid water content for image pair 3.
70°W
6 5 °W
60°W
55°W
50°W
60°W
55°W
50°W
4 5 °w
NoSf
35°N
NoOfr
40°N
NoS*
45*N
75°W
70°W
Figure 4.20: W ater vapour content for image pair 3.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
65
Im age P air 6: M ay 27, 2001 (m orning)
The L S S T estim ates were within the limits of the calibration, ±2.55°C, for the major­
ity of the scanned area th at overlapped (Figure 4.23). The apparent over-estimation
in the south west comer of the scan is due to low AVHRR SST estimates due to cloud
contamination missed by the cloud masking technique (Figure 4.22). The L S S T es­
tim ates for this region are reasonable (Figure 4.21). The area of 10 to 14°C water
appearing as a north-west south-east structure (south of 40°N and east of 50°W) is
lower than expected for this region. As clw is only present along the eastern edge of
this region, there is no explanation for this discrepancy at this time.
75°W
7Q°W
65°W
6 0 °W
55°W
50°W
45°W
cn
in
o
TT
Lm
U*
in
ro
75°W
70°W
65°W
•
5
60°W
II
55°W
IS
21
50°W
IS
45°W
N
Figure 4.21: L S S T results for image pair 6.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
•
5
II
15
»
25
M
Figure 4.22: AVHRR SST results for image pair 6.
70°W
65°W
60°W
55°W
50°W
45°W
55°W
5 0 “*
45°W
45°N
75°W
75°W
70°W
6 5 °W
60°W
NoO
N8QC
35°N
40°N
■
B&J
<-5 -5 -4 -3 -2-1 0 1 2 3 4 5 5<
Figure 4.23: ASST (AVHRR - L S S T ) results for image pair 6.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
67
70°W
65°W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
NuGf
35°N
NoOf
40°N
NoGf
45°N
75°W
<0 0.0 0.1 0 £ 0 3 0.4 0 J 0.6 0.7 0.8 0 5 1.0
Figure 4.24: Cloud liquid w ater content for image pair 6.
70°W
65°W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
35°N
N„GC
40°N
NoOt'
45°N
NoGt-
75°W
Figure 4.25: W ater vapour content for image pair 6.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
68
Im age P air 7: M ay 27, 2001 (night)
The general structure of the L S S T estim ate (Figure 4.26) is comparable to the ex­
pected structure (Figure 4.27). The combined effects of the presence of clw (Fig­
ure 4.28) and wv (Figure 4.29) on the L S S T estim ate of SST is present in the north­
west and south-east section of the scanned area. The region between 38°N and 42°N
extending horn 54°W to the eastern extent of the scanned area in Figure 4.26 shows
lower than expected SST. The discrepancies between the models in this region are
presently unexplainable.
75°W
70°W
65°W
60°W
55°W
50°W
45°W
t
o
0cn
4
75°W
70 °W
65°W
•
5
60°W
It
55°W
15
»
50°W
25
45°W
3t
Figure 4.26: L S S T results for image pair 7.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
69
70°W
65°W
60°W
75°W
70°W
65°W
5 0 °W
55°W
50°W
45°W
N o5C
35»N
NoOf
4 o “N
N .Sfr
4 5 “N
75°W
•
S
M
5 5°W
15
M
50°W
25
4 5 “W
M
Figure 4.27: AVHRR SST results for image pair 7.
70°W
65°W
60°W
55°W
50°W
45°w
75°W
70°W
65°W
60°W
55°W
50°W
45°W
NoSf
35°N
40°N
NoOt-
N.Sfr
45°N
75°W
<0 0.0 (U 0 2 0 3 0.* 0 5 0 6 0.7 0 8 0 9 LO
Figure 4.28: Cloud liquid water content for image pair 7.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
70
75°W
70°W
65°W
60°W
55°W
50°W
45°W
o
o
04
75°W
70°W
65°W
60°W
55°W
50°W
45°W
Figure 4.29: W ater vapour content for image pair 7.
Im age P air 10: June 1, 2001 (m orning)
The SST estim ates using L S S T (Figure 4.30) in overall structure is comparable to
the AVHRR cloud free estimates (Figure 4.31). The areas of under-estim ation (red
pixels in Figure 4.32) by the L S S T model seems to be caused by the presence of
clw (Figure 4.33) and wv content <20 kg*m-2 (Figure 4.34). The white areas in
Figure 4.33 and 4.34 are areas of rain where no geophysical estim ates were calculated
other than the r r . The corresponding areas in Figure 4.30 are the white pixels and
should be ignored.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
71
70°W
65°W
60°W
55°W
50°W
45°W
70°W
65°W
60°W
5 5 °W
5 0 °W
45°W
NoSC
35°N
40°N
NoOt-
45°N
75°W
•
S
1*
15
2*
25
M
Figure 4.30: L S S T results for image pair 10.
65°W
60°W
65°W
60°W
50°W
35°N
NoOfr
40°N
45°N
NoSt-
75°W
75°W
70°W
55°W
5 0 °W
45°W
Figure 4.31: AVHRR SST results for image pair 10.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
75°W
70°W
65°W
60°W
55°W
50°W
4-5 °'#
<-5 -5 - 4 - 3 - 2 - 1 0 1 2 ? ? S <
Figure 4.32: ASST (AVHRR - L S S T ) results for image pair 10.
70°W
65°W
60°W
55°W
50°W
4 5 °w
I
I
!
I
50°W
45°W
NoS*
45°N
75°W
NoOf
N»Sf
35°N
40°N
J
75°W
70°W
65°W
60°W
55°W
< 0 ao 0.1 (12 03 0.4 0 ^ a 6 0.7 0.8 0.9 1.0
Figure 4.33: Cloud liquid water content for image pair 10.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
73
“W
7 0 °W
65°W
60°W
55°W
50°W
45°W
-£w
<_n
in
u
o
o
04
O'
in
75°W
70°W
65°W
60°W
55°W
50°W
45°W
Figure 4.34: Water vapour content for image pair 10.
4.3.2
V alid ation S ets
In all the validation images, the spatial comparison of the L S S T results to corre­
sponding AVHRR SST estimates were very promising. In this section the statistical
results will be presented first, followed by the discussion of the individual validation
images.
Before computing the validation statistic presented in Table 4.5 all points
containing clw were removed. Also, any pixel which met any of the following criteria
were eliminated: 1) an AVHRR SST less than 10°C with a latitude of less than 40°N,
2) AVHRR SST less than 7°C with a latitude less than 45°N, or 3) AVHRR SST less
than 2°C. No p c f criteria was applied to these d ata so that the model was validated
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
74
against real data.
The mean difference and STD are shown in Table 4.5 for the validation
data sets. It was observed that the differences are generally symmetric about the
mean which is less than 1.5°C. These differences are within the calibration limits
of the model. The 0.5° increase in variance over that of the calibration set may be
attributed to AVHRR SST with low p c f being compared with L S S T estimates.
Table 4.5: Statistical comparison between L S S T and AVHRR SST with pixels con­
taining cloud liquid water removed.
Date
Time
number of
data points
Mean Difference
(°C)
STD
(°C)
May 21
21:03
486
2.78
2.90
May 26
10:01
1290
0.98
2.66
May 26
21:37
1529
-0.56
2.60
May 31
10:37
889
0.70
2.97
June 2
10:10
1149
-0.80
2.86
Figure 4.35 is an example of the relationship between L S S T and AVHRR
SST. Note that had the model been perfect the slope and intercept would be equal
to 1 and 0, respectively. The slopes and intercepts for the validation images ranged
from 0.90 to 1.04 and -1.3 to 3.89, respectively, with R2 fits of around 0.7.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
75
For the remainder of this section, the individual image sets will be discussed,
highlighting both the regions where differences occur and the possible reasons they
occur. Also, the general SST structure will be compared between the results of the
two sensors.
• • «*v
,
,• *
r*
••
*• **♦
** *«*•.A
♦ t .♦ • *•
1■ •«
«
0
S
to
LMT(d
IB
20
a
30
Figure 4.35: L S S T verse AVHRR SST for image pair 9.
Im age P air 1: M ay 21, 2001 (night)
Figure 4.37 shows the expected temperature in some of these regions. Unfortunately
there is a significant number of d ata points missing in Figure 4.37 due to the presence
of cloud in the AVHRR FOV thus making if difficult to compare the L S S T results
Figure 4.36 with a known SST value. As expected there are low L S S T estimates
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
76
corresponding to the presence of clw (Figure 4.38) and wv in quantities less than
25 kg-m-2. The other regions of low SST estimates by the L S S T model in the
Sargasso Sea waters is currently unexplainable. There appears to be some atmospheric
effect influencing the L S S T estimates for this particular day (Figure 4.36), since the
expected large scale SST structures are not reproduced by the L S S T model unlike
the structures in other image pairs. Also, image pair 1 is not contained within the
time period covered by the calibration set, therefore leading us to the aforementioned
conclusion.
75°W
70°W
55°W
60°W
55°W
50°W
4 5°w
■B.
□
75°W
70°W
65°W
•
5
60°W
1#
5 5 °W
15
39
5 0 °W
25
4-5 °W
3#
Figure 4.36: L S S T results for image pair 1.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a
LA
LP
77
75°W
70°W
^ .
V4 ft*.
65°W
6 0 °W
55°W
1
i
1
50°W
45°W
P
^
y
/
,
-
C
0 ! > i .
|1
s
r ?
►
!
N0sr
40°N
**•
/v
»
35°N
■
*
*
N .S fr
0
*
Q
MoOt’
45°N
0
75°W
70°W
6 5 °W
•
5
60°W
1*
5 5°W
15
M
5 0 °W
25
45°W
30
Figure 4.37: AVHRR SST results for image pair 1.
70°W
65°W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
NoSf
35°N
40°N
MoOt-
45°N
N„St-
75°W
<0 (U) 0.1 0.2 0 3 0.4 05 0 6 0.7 0.8 0.9 1.0
Figure 4.38: Cloud liquid water content for image pair 1.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
78
75°W
70°W
65°W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
1 'Ji 1 I!
Figure 4.39: Water vapour content for image pair 1.
Im age P air 4: M ay 26, 2001 (m orning)
The large scale structures in the AVHRR SST image (Figure 4.41) can be seen in
the L S S T estimates (Figure 4.40). Following the boundary just above 40°N, notice
th at the L S S T (Figure 4.40) boundary peaks northwards around 60°W and 54°W
the same as in Figure 4.41. Also, note the warmer water in the south- west comer of
the scan (35°N 64°W). Figure 4.42 shows that the L S S T estimate is within ±2°C of
the AVHRR SST for the majority of the co-located scanned region. The large region
of low L S S T estimates in the upper right hand side of Figure 4.40 correlate with
clw greater than zero (Figure 4.43) and w v <20 kg-m~2 (Figure 4.44). Thus, this
shows th at the L S S T algorithm may be highly sensitive to clw content/atmospheric
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
temperature.
75°W
70°W
65°W
60°W
55°W
50°W
45°W
70°W
65°W
60°W
55°W
50°W
45°W
•
5
I#
15
2»
25
3#
Figure 4.40: L S S T results for image pair 4.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
75°W
70°W
65°W
•
5
60°W
!•
55°W
15
2t
50°W
25
45°W
3#
Figure 4.41: AVHRR SST results for image pair 4.
65°W
60°W
55°W
50°W
45°W
55°W
50°W
45°W
No'st-
70°W
45“N
75 °W
35°N
MgSC
NoOf
40°N
P*
75°W
70°W
65°W
60°W
<•5 -5 -4 -3 -2 -1 0 1 2 3 4 5 5<
Figure 4.42: ASST (AVHRR - L S S T ) results for image pair 4.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
81
70°W
65°W
60°W
55°W
5 0 °W
45°W
75 W
70°W
65°W
60°W
55°W
50°W
45°W
NoSV
N.Sfr
75°W
NoOf
NoOfr
No5f
N„Sf
<0 0 0 0 1 0 2 0.3 0.4 Ol5 0.6 0.7 0.8
1.0
Figure 4.43: Cloud liquid water content for image pair 4.
70°W
65°W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
NoSf
N„Sfr
75°W
NoOV
NoOt'
No5C
NoSf
Figure 4.44: W ater vapour content for image pair 4.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
82
Image Pair 5: May 26, 2001 (night)
The L S S T result is in general agreement with the AVHRR SST (Figure 4.46 and
4.45). The differences between the estimates are within the calibration limits of the
L S S T model except along the western edge of the scan (Figure 4.47). This region
contains clw (Figure 4.48, light blue region) and wv in excess of 30 kg*m~2 (Fig­
ure 4.49, dark green and yellow regions). This result was also seen in the calibration
images, thus re-enforcing the idea that the L S S T model maybe sensitive to changes
in atmospheric temperature rather than simply clw content.
75 W
7 0°w
65°W
60°W
55°W
50°W
45°W
n
O
0O4'
iD
rO
75°W
70°W
6 5 °W
60°W
It
55°W
15
5Q°W
2 t 25
^5°W
3«
Figure 4.45: L S S T results for image pair 5.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
i
83
70°^
65°W
60°W
55°W
50°W
45°W
75°W
7Q°W
65°W
6 0 °W
5 5 °W
50°W
45°W
NoQC
35»N
No0t>
4 0 “n
45°N
NoGt'
75°W
•
5
10
15
20
25
30
Figure 4.46: AVHRR SST results for image pair 5.
70°W
65°W
60°W
55°W
50°W
4 5 °W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
35°N
N-5C
NoO
40°N
45°N
NoQt-
75°W
<-5 -5
-3 *2 -1 0 1 2 3 4 5 5<
Figure 4.47: ASST (AVHRR - L S S T ) results for image pair 5.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
84
70°W
65°W
60°W
55°W
50°W
45°W
65°W
60°W
55°W
50°W
45°W
m
75°W
k
i
70°W
NoGf
35°N
40°N
No0t’
45°N
N0G7
75°W
<0 0.0 0.1 0 2 0.3 0.4 0.5 0.6 0 7 0.8 0.9 1.0
35°N
NoG£
40°N
NoOt>
45°N
NoGt’
Figure 4.48: Cloud liquid water content for image pair 5.
75°W
70°W
65°W
60°W
55°W
50°W
45°W
40 ’ 50 ' 60 ' 70 ' 80
Figure 4.49: Water vapour content for image pair 5.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
85
Image Pair 8: May 29, 2001
For this day, there where less than 90 co-located pixels between the two sensors, which
had cloud free AVHRR SST results. The number of pixels was too small to draw
any reliable statistics, also the average p c f was only 18.2±7.1%. Thus, there is low
confidence in the SST estimate by the AVHRR due to possible cloud contamination.
However, the L S S T model was able to produce a reasonable SST structure to what
would be expected in the region with the Gulf Stream breaking away from the coast
(see Figure 4.50). Without SST to compare with the L S S T estimates, few conclusions
can be drawn on the effects of clw (Figure 4.51) and wv (Figure 4.52) have on L S S T
estimates of SST.
75 W
70°W
65°W
60°W
55°W
50°W
45°W
-fit
O
'
&
o
75°W
70°W
65°W
60°W
55°W
50°W
45°W
Figure 4.50: L S S T results for image pair 8.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
86
►
c
55°W
5Q°W
45°W
i
i
i
i
►
►
%
N«SV
^
60°W
35°N
40°N
45°N
»«i. c*
’ *
>
6 5 °W
NoOr
70°W
75°W
70°W
65°W
6 0 °W
55°W
i
i
50°W
45°W
NoSf
75°W
<0 0.0 O l 0 2 0 3 0.4 0 3 0.6 0.7 0 3 0.9 1.0
Figure 4.51: Cloud liquid water content for image pair 8.
65°W
60°W
55°W
50°W
%
A
&
j
45°w
i
NcGt'
70°W
i
NoSf
35°N
N.0fr
40°N
45°N
75°W
7 5 °W
70°W
65°W
60°W
55°W
5Q°W
45»w
Figure 4.52: Water vapour content for image pair 8.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
87
Image Pair 9: May 31, 2001 (morning)
As with the previous data sets, the large scale SST structures in Figure 4.53 and 4.54
are comparable. The presence of clw in both the northern and southern regions of
the study area was not present in previous image pairs (see Figure 4.55). The effect
of clw is consistent in the northern region (39°N 65°W), low (negative) SST estimate
by the L S S T model (Figure 4.53). However, in the region between 35° to 40°N and
60° to 55°W there are areas of reasonable and high estimates of SST by the L S S T
model for approximately the same amount of clw. Once again the presence of wv in
quantities greater than 20 kg-m~2 (Figure 4.56) appears to lessen the effect of clw on
L S S T estimates of SST. The over- and under-estimate of SST by the L S S T model
being spatially correlated to varying amounts of tin; in the presence of clw seem to
re-enforce the idea th at atmospheric temperature has an effect on the L S S T model’s
ability to estimate SST.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
88
70°W
65°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
35°N
N.SC
40°N
N.Ofr
45°N
75°W
•
5
60°W
1«
55°W
15
M
50°W
25
45°W
3t
Figure 4.53: L S S T results for image pair 9.
70°W
6 5 °W
60°W
55°W
75°W
70°W
65°W
60°W
55°W
35°N
NoSC
40°N
NoOfr
45°N
NoGV
75°W
•
5
I#
50°W
45°W
15
Figure 4.54: AVHRR SST results for image pair 9.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
89
70°W
65°W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
NoSf
35°N
N.Ofr
40°N
N»Sfr
45°N
75°W
<0 0.0 a i 0.2 0 3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Figure 4.55: Cloud liquid water content for image pair 9.
65°W
60°W
50°W
45°W
50°W
45°W
No St-
70°W
45°M
75°W
NbO*
40°N
A
75°W
70°W
65°W
6 0 CW
NoQC
35°N
A
55°W
Figure 4.56: Water vapour content for image pair 9.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
90
Image Pair 11: June 2, 2001 (morning)
Areas of low estimate of SST by the L S S T model (Figure 4.57), are well correlated
with regions of clw (Figure 4.60). This correlation is consistent in both the northern
and southern regions of the study area except in areas were the wv is greater than 20
kg-m-2 (Figure 4.61). Figure 4.59 shows that the L S S T estimates are comparable to
the AVHRR SST estimates within the calibration limits (±2.55°C). However, this area
corresponds to AVHRR temperatures which appear low for the region (Figure 4.58).
75°W
70°W
65°W
60°W
55°W
5Q°W
45°W
75°W
70°W
65°W
50 °W
55°W
50°W
45°W
•
5
II
15
at
25
3*
Figure 4.57: L S S T results for image pair 11.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
91
70°W
65°W
6 0 “W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
5 5 3W
50°W
45°W
NoQV
NoSt-
75°W
n , ov
NoOfr
N„Qf
•
5
1«
15
M
25
M
Figure 4.58: AVHRR SST results for image pair 11.
75°W
70°W
*
i
50°W
4-5°W
■=»
>
i
1
l
*
L
■
NoOt'
Be
H
r
►
NoSf
i
75°W
pC
70°W
6 5 “W
60°W
NoSt’
■V '
i
5 5 °W
55°W
i
i
_____________________
50°W
1
45°W
<-5 -5 -4 -3 -2 -1 0 1 2
Figure 4.59: ASST (AVHRR - L S S T ) results for image pair 11.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
NoOt’
No5t,
>
60°W
N„Qf
« * .
r
55°W
92
70°W
65°W
60°W
55°W
50°W
45°W
75°W
70°W
65°W
6 0 °W
55°W
50°W
45°W
NoQ£
35°N
40°N
NoOt'
NoG f
45°N
75°W
<0 a o 0.1 0 2 03 0.4 0 3 0 6 0.7 03 0.9 1 3
Figure 4.60: Cloud liquid water content for image pair 11.
?0°W
65°W
6 0 °W
55°W
50°W
45°W
75°W
70°W
65°W
60°W
55°W
50°W
45°W
35°N
NoQC
40°N
N odi '
45°N
NoSt-
75°W
40
60 ' 70 ' 80
Figure 4.61: W ater vapour content for image pair 11.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chapter 5
Discussion and Conclusions
The retrieval of SST with the SSM/I utilizing a non-linear relationship between SST
and atmospheric water vapour content (csst) proposed by Rubinstein [24] did not
produce resonable results for either the gradient or absolute measure of SST. Modi­
fication of this model by improving water vapour retrieval and by modeling the two
populations individually improved the results slightly but it was still unsuccessful in
estimating the SST comparable to those computed from the AVHRR.
A new band difference model, L S S T , proposed by the author showed a
dramatic improvement over the csst and ncsst results in both spatial distribution
and absolute value of SST. The mean standard deviation of the difference between
the model’s estimate of SST and AVHRR’s estimate of SST for the validation images,
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
94
with pixels containing clw removed, was 4.60°C, 4.46°C and 2.76°C for the csst, ncsst
and L S S T models, respectively. The regions where AVHRR SST - L S S T < -5°C were
concentrated along cloud edges. Because cloud masking techniques are not perfect
and partially cloud/fog filled pixels along the edges of clouds are the most difficult to
detect, the discrepancies in these regions were considered to be mainly due to cloud
contamination in the AVHRR data.
The presence of cloud liquid water and water vapour in quantities less than
approximately 25 kg-m~2 produced under estimates of SST using the L S S T model.
Over estimates of SST were well correlated with the presence of cloud liquid water and
water vapour in quantities greater than 25 kg-m~2. These over and under estimates
of SST using the L S S T model when cloud liquid water is present with different levels
of water vapour content seems to suggest that atmospheric temperature has an effect
on the L S S T model’s ability to estimate SST. Atmospheric temperature is one of
the factors which determines the amount water in vapour form that a given volume
of atmosphere can hold and therefore is considered one of the factors limiting the
models results. When the volume of atmosphere is saturated beyond this point the
water begins to condense. Thus, having the same amount of clw with varying amount
of w v could indicate a change in atmospheric temperature.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
95
5.1
F u tu re W ork
The primary aim of this study was to explore the possibility of using the SSM/I to re­
trieve SST either directly or in combination with another sensor such as AVHRR. The
results presented in Chapter 4 suggest that the potential exists to directly estimate the
SST using the SSM/I. The uncertainty in the present estimates could be improved by
using a more accurate truth measurement, which is not affected by clouds (e.g., buoy
data) and better correlated temporally to adjust the constants of the L S S T model.
The length of the study could be increased so that periodic atmospheric changes can
be taken into account in the model. Using a direct measure of SST, buoy data, future
investigation into the effects of cloud liquid water/atmospheric temperature has on
passive microwave retrieval of SST could be undertaken. Also, since the SST gradi­
ents were comparable to those displayed in the AVHRR SST results, a calibration of
the SSM /I SST field using cloud free AVHRR SST values may yield increased SST
coverage with greater accuracy than directly applying the L S S T model.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Bibliography
[1] Robinson I. S. Satellite Oceanography. Ellis Horwood Limited, Toronto,
Ontario, 1985.
[2] Ocean remote sensing John Hopkins University Applied Physics Laboratory,
h ttp ://fermi.jhuapl.edu/avhrr/SW /averages/01jun/SW_ 01jun02_ 2324. mult.gif,
1998.
[3] Armstrong R. NO A A /N A S A S S M /I pathfinder antenna temperatures,
h ttp :// nsidc.org/N ASA / G U ID E /D O C S /D ataset. documents/ssmi.
pathfinder. land, surface, products, dataset, documents.html, 1996.
[4] Kidwell K. NOAA Polar Orbiter Users Guide. National Oceanic and
Atmospheric Administration, Washington, D.C., June 1995.
[5] Ulaby F. T., R. K. Moore, and A. K. Fung. Microwave Remote Sensing,
volume III. Artech House Inc., Norwood, MA, 1986.
[6] Stogryn A. The apparent temperature of the sea at microwave frequencies.
IEEE Transactions on Antennas and Propagation, AP-15(2):278—286, 1967.
[7] F. J. Wentz. A well-calibrated ocean algorithm for special sensor microwave
imager. Journal o f Geophysical Research, 102(C4):8703-8718, 1997.
[8] Ulaby F. T-, R. K. Moore, and A. K. Fung. Microwave Remote Sensing,
volume II. Addison—Wesley Publishing Company, Don Mills, Ontario, 1982.
[9] Hollinger J. P. and R. C. Lo. S S M /I project summary report. NRL technical
memorandum report 5055, Naval Research Laboratory, Washington, D.C.
20375, April 1983.
[10] Eisberg R. and R. Resnick. Quantum Physics o f atoms, molecules, solids,
nuclei, and particles. John Wiley & Sons, Toronto, 2nd edition, 1985.
97
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
98
[11] Ulaby F. T., R. K. Moore, and A. K. Fung. Microwave Remote Sensing,
volume I. Addison-Wesley Publishing Company, Don Mills, Ontario, 1981.
[12] Case K. Special sensor microwave imager ( s s m / i ) .
http://podacc.jpl.nass.gov.2031/SENSOR_ DOCS/ssmi.html, 1996.
[13] Hollinger J. P. D M SP special sensor microwave/imager calibration/validation.
Final report vol. i, Naval Research Laboratory, Washington, D.C. 20375, July
1989.
[14] Hollinger J., R. Lo, G. Poe, R. Savage, and J. Peirce. Special sensor
microwave/imager user’s guide. Technical report, Naval Research Laboratory,
Washington, D.C. 20375, September 1987.
[15] Wentz F.J. Measurement of oceanic wind vector using satellite microwave
radiometers. IEEE Transactions on Geoscience and Remote Sensing,
30(5):960-972, 1992.
[16] Wentz F.J. personal communication.
[17] Alishouse J. C., S. A. Snyder, J. Vongsathom, and R. R. Ferraro.
Determination of oceanic total precipitable water from the s s m / i . IEEE
Transactions on Geoscience and Remote Sensing, 28(5):811-816, 1990.
[18] Gill A. E. Atmosphere-Ocean Dynamics, volume 30 of International Geophysics
Series. Academic Press, Inc., Toronto, 1982.
[19] Knauss J. A. Introduction to Physical Oceanography. Prentice Hall, Inc,
Englewood Cliffs, New Jersey, 1978.
[20] Hollinger J. P. D M SP special sensor microwave/imager calibration/validation.
Final report vol. II, Naval Research Laboratory, Washington, D.C. 20375, May
1991.
[21] Piepmeier J.R. and A.J. Gasiewski. High-resolution passive polarimetric
microwave mapping of ocean surface wind vector fields. IEEE Transactions on
Geoscience and Remote Sensing, 39(3):606-622, 2001.
[22] Ross D. B. and P. Gloersen. Spectral characteristics of the microwave emission
from a wind-driven foam-covered sea. Journal o f Geophysical Research,
81:3095-3099, 1976.
[23] Lo R. C. A comprehensive description of the mission sensor microwave imager
( s s m / i ) environmental parameter extraction algorithm. NRL memorandum
report 5199, Naval Research Laboratory, Washington, D.C. 20375, September
1983.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
99
[24] Rubinstein I. G. personal communication.
[25] Wentz F .J. A M SR ocean algorithm. RSS tech. report 110398, Remote Sensing
Systems, Santa Rosa, CA, November 1998.
[26] Landry Lt(N). D.J. Determining sea surface temperature structure in the north
west atlantic using high resolution AVHRR imagery. Masters thesis, Royal
Military College of Canada, 1998.
[27] Di L. and D.C. Rundquist. A one-step algorithm for correction and calibration
of AVHRR level lb data. Photogrammetric Engineering & Remote Sensing,
60(2):165—171, 1994.
[28] Gallegos S.C., J.D. Hawkins, and C.F. Cheng. A new automated method of
cloud masking for advanced very high resolution radiometer full-resolution data
over the ocean. Journal o f Geophysical Research, 98(C5):8505-8516, May 1993.
[29] Thiermann V. and E. Ruprecht. A method for the determination of clouds
using AVHRR infrared observations. International Journal o f Remote Sensing,
13(10):1829-1841, 1992.
[30] Satellite active archive help, http://www.saa.noaa.gov/, July 2001.
[31] IDL Reference Guide. Research Systems, Inc., 1998.
[32] Press W. H., B.P. Flannery, S.A. Teukolsky, and W.T. Vetterling. Numerical
Recipes. Cambridge University Press, New York, 1986.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Документ
Категория
Без категории
Просмотров
0
Размер файла
3 688 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа