47 département de géomatique appliquée université … · 47 département de géomatique...
Post on 12-Sep-2018
217 Views
Preview:
TRANSCRIPT
47
Département de Géomatique appliquée
faculté des Lettres et sciences humaines
Université de Sherbrooke
Explorîng Snow Information Content of Interferometric SAR Data
Exploration du contenu en information de l’interférométrie RSO lié à la neige
Ah Esmaeily Gazkoliaui
Thesis submitted in fulfihiment of the requirements for the degree of Doctor of
Phîlosophy (Ph.D.) in Remote Sensing
Thèse présentée pour l’obtention du grade de Philosophiae Doctor (Ph.D.) en
Télédétection
© Ail Esmaeily Gazkohani, 200$
fN
I
Composition du jury
Explorïng Snow lnformatïon Content of Interferometric SAR Data
Exploration du contenu en information de l’interférométrie RSO lié à la neige
Ail Esmaeïiy Gazkohani
Members of jury:
Cette thèse a été évaluée par un jury composé des personnes suivantes
Hardy B. Granberg
Directeur de recherche : (Département de Géomatique appliquée, Université de Sherbrooke)
Q. Hugh J. Gwyn
Codirecteur de recherche : (Département de Géornatique appliquée, Université de Sherbrooke)
Dong-Chen 11e
Membres du jury: (Département de Géomatique appliquée, Université de Sherbrooke)
Matti Leppranta
Membres du jury: (Departrnent ofPhysics, University of Helsinki, Finland)
Bruce G. Thom
Membres du jury: (Geosciences Faculty, University of Sidney, Australia)
Abstract
The objective of this research is to explore the information content of repeat-pass cross-track
Interferornetric SAR (InSAR) with regard to snow, in particular Snow Water Equivalent
(SWE) and snow depth. The study is an outgrowth of earher snow cover modeling and radar
interferometry experiments at Schefferville, Quebec, Canada and elsewhere which has shown
that for reasons of loss of coherence repeat-pass InSAR is not useful for the purpose of snow
cover mapping, even when used in differential InSAR mode. Repeat-pass cross-track InSAR
would overcome this problem.
As at radar wavelengths dry snow is transparent, the main reftection is at the snow/ground
interface. The high refractive index of ice creates a phase delay which is linearly reÏated to the
water equivalent of the snow pack. When wet, the snow surface is the main reflector, aid this
enables measurement of snow depth. Algorithms are elaborated accordingly.
Field experiments were conducted at two sites and employ two different types of digital
elevation models (DEM) produced by means of cross track InSAR. One was from the Shuttie
Radar Topography Mission digital elevation model (SRTM DEM), ftown in February 2000. It
was cornpared to the photogrammetrically produced Canadian Digital Elevation Model
(CDEM) to examine snow-related effects at a site near Schefferville, where snow conditions
are well known from haif a century of snow and permafrost research. The second type of
DEM was produced by means of airborne cross track InSAR (TOPSAR). Several missions
were flown for this purpose in both summer and winter conditions during NA$A’s CoÏd Land
Processes Experiment (CLPX) in Colorado, USA. Differences between these DEM’s were
compared to snow conditions that were well documented during the CLPX field carnpaigns.
The resuits are not straightforward. As a resuit of automated correction routines employed in
both SRTM and ALRSAR DEM extraction, the snow cover signal is contaminated. Fitting
InSAR DEM’s to known topography distorts the snow information, just as the snow cover
distorts the topographic information. The analysis is therefore rnostly qualitative, focusing on
particular terrain situations.
At ScheffervilÏe, where the SRTM was adjusted to Iuown lake levels, the expected dry-snow
signal is seen near such lakes. Mine pits and waste dumps flot included in the CDEM are
depicted and there is also a strong signal related to the spatial variations in SVŒ produced by
wind redistribution of snow near lakes and on the alpine tundra.
In Colorado, cross-sections across ploughed roads support the hypothesis that in dry snow the
SWE is measurable by differential InSAR. They also support the hypothesis that snow depth
may be measured when the snow cover is wet. Difference maps were also extracted for a 1
km2 Intensive Study Area (ISA) for which intensive ground truth was available. Initial
comparison between estimated and observed snow properties yielded low coneÏations which
improved after stratification ofthe data set.
In conclusion, the study shows that snow-related signals are measurable. For operational
applications satellite-borne cross-track InSAR would be necessary. The processing needs to be
snow-specific with appropriate filtering routines to account for influences by terrain factors
other than snow.
Key words: Interferometry, InSAR, Snow Water Equivalent, SWE, Snow depth, Dry Snow,
Wet Snow, AIRSAR, CLPX, SRTM, Schefferville.
Résumé
L’objectif de cette recherche consiste à explorer le contenu en information de l’interférornétrie
radar (InSAR) à passage multiple avec deux antennes, lié à la neige, en particulier l’équivalent
en eau et sa profondeur. L’étude fait suite aux expériences de modélisation de couvert nival et
d’interférométrie radar à Schefferville, Québec, Canada et d’autre endroit. Ces dernières
montrent qu’à cause de perte de cohérence, l’interférométrie à passage multiple n’est pas
convenable pour la détection de la neige, même dans le mode de InSAR Différentielle. InSAR
à passage multiple avec deux antennes devrait résoudre le problème.
Puisque la neige sèche est transparente pour le radar, la rétrodiffusion principale se produit à
l’interface de neige/sol. Le haut indice de réfraction de la glace génère un déphasage qui est
une approximation à l’équivalent en eau du couvert neigeux. Quand la neige est humide, sa
surface est le réflecteur principal et ceci permet la mesure de profondeur de la neige. En
conséquence, des algorithmes sont développés.
Les travaux expérimentaux ont été appliqués pour deux sites en utilisant deux types de modèle
numérique d’altitude (MNA) générés par InSAR avec deux antennes. Le premier venait de
données du Shuttie Radar Topography Mission (SRTM) qui a fait le tour de Globe en février
2000. Ceci a été comparé avec le MNA canadien généré en moyen photogrammétrie pour
examiner les effets de la neige sur le radar dans un site près de Schefferville, où les conditions
de neige est bien connus d’un demie centenaire de recherche sur la neige et pergélisol. Le
deuxième type de MNA a été généré en utilisant les données InSAR aéroportées avec deux
antennes (TOPSAR). Pour cette fin, plusieurs missions ont été planifiées en été et en hiver
dans le cadre de projet Cold Land Processes Experiment (CLPX) au Colorado (ÉU). Les
différences entre ces MNA ont été comparées avec les conditions de la neige qui ont été bien
documentées pendant la campagne du terrain de projet CLPX.
Les résultats n’ont pas permis de valider la méthode de manière définitive. En raison des
routines de corrections automatisées utilisées dans l’extraction de DEM des données SRTM et
ATRSAR, le signal lié à la neige est bruité. L’ajustement de DEM d’InSAR à la topographie
connue affecte l’information de neige, de la même manière que la neige affecte l’information
topographique. L’analyse est donc qualitative, se concentrant sur des situations particulières de
terrain.
À Schefferville, où les MNA de SRTM ont été ajustés sur les niveaux connus des lacs, le
signal prévu de la neige sèche est observé près de tels lacs. Des puits de mine et les décharges
de rebut qui ne sont pas inclus dans le DEM Canadien sont détectés et il y a également un
signal fort lié aux variations spatiales de l’équivalent en eau produit par la redistribution de la
neige par le vent sur la toundra.
Au Colorado, les profils transversaux des routes déneigées soutiennent l’hypothèse que
l’équivalent en eau pour la neige sèche est mesurable par InSAR différentiel. Elles soutiennent
également l’hypothèse que la profondeur de neige peut être mesurée quand la neige est
humide. Des cartes de différence ont été également extraites pour un site d’étude I km2
(Intensive Study Area - ISA) pour lequel la vérité du terrain était disponible de façon très
détaillée. La première comparaison directe entre les propriétés de neige mesurée et estimée a
rapporté de faibles corrélations qui ont été améliorés après la stratification des données.
En conclusion, l’étude prouve que les signaux liés à la neige sont mesurables. Pour des
applications opérationnelles l’approche de l’interférométrie radar avec deux antennes par
moyen satellitaire est nécessaire. Le traitement doit être conçu pour la neige avec des routines
de filtrage appropriées pour expliquer des influences par des facteurs de terrain autres que la
neige.
Mots clefs Interférométrie, kSAR, Équivalent en eau, Profondeur de neige, Neige sèche,
Neige humide, AIRSAR, CLPX
TABLE 0F CONTENTS
Table of contents I
List offigures iii
List of tables vii
Acronyms vIIi
Acknowledgments ix
1 INTRODUCTION1
1.1 Background to the study 1
1.2 Objectives5
1.2.1 General objective5
1.2.2 Specific objectives6
1.3 Hypotheses6
1.4 Thesis outiine6
1.5 Claim to originality7
2 SNOW PROPERTIES AND CHARACTERISTICS 8
2.1 Snowcover$
2.2 Snow spatial variability and distribution 10
2.2.1 Snow accumulation 10
2.2.2 Effect ot topography on snow cover distribution 10
2.2.3 Effect of vegetation 11
2.2.4 Effectofwind12
2.3 Mictowave interactions and dielectric properties ofsnow 12
3 INTERFEROMETRIC SAR 19
3.1 Theory19
3.1.1 Differential lnterferometry 22
3.1.2 Errors and limitations ofInSAR 24
3.2 The influence of wet and dry snow on InSAR DEM’s 27
3.2.1 Dry condition27
3.2.2 Wet condition31
3.3 Theoretical analysis 32
3.4 Interpretation and discussion oftheoretical research 36
11
3.5 Conclusion of theoretical research.3$
4 METHODOLOGY39
4.1 Introduction39
4.2 Schefferville study area and data 39
4.3 Data41
4.4 Colorado study areas and data 41
4.4.1 Characteristics of Meso-ceIl Study Areas (MSA) 43
4.4.2 Characteristics of Intensive Study Areas (ISA) 49
4.4.3 AIRSAR data50
4.4.4 Ground data52
4.5 Method of analysis56
4.5.1 SRTM InSAR data56
4.5.2 CLPX InSAR data56
4.5.3 Elevation correction5$
4.5.4 Interactive image georeferencing 5$
4.5.5 Misfit between TOPSAR DEM’s 60
4.5.6 Snow spatial variability63
4.5.7 SWE and snow depth estimation 63
4.5.8 Data ntegration63
5 RESULTS AND DISCUSSION 65
5.1 Results at Schefferville 65
5.2 Results from Colorado 6$
5.3 Integration of remote sensing and field data 76
5.4 SWE and depth estimation 77
5.5 The nature of the snow depth and water equivalent signais 81
5.5.1 Snow water equivalent$2
5.5.2 Snow depth85
6 CONCLUSIONS90
7 REFERENCES92
8 APPENDIX1102
9 APPENDIX2112
lu
List of Figures
Figure 1: Three examples of coherence images of repeat-pass interferornetry with 3 days
interval, calculated from ERS-1 parent images, (a) January 9 and 12, (b) January 12 and 15,
(e) February 14 and 17 (Images provided by Vachon et aÏ., 1995) 3
Figure 2: (a) RADARSAT coherence image from pair image ofDecernber 1 and 25, 1996, (b)
RADARSAT coherence image from pair image ofFebruary 2$ and Mardi 24, 1997, (e) Part
of permafrost map produced by 10CC in 1972-74 (modified after Granberg and Vachon,
199$)4
figure 3: The penetration depth of snow decreases rapidly with increasing liquid water
content, especiaÏly in the immediate vicinity ofLw = O (rnodified after Ulaby et aL, 1986).... 15
figure 4: Real part of tic dielectric constant of dry snow as a ftmction ofthe density of dry
snow relative to water (modified after Tiuri et aÏ., 1984) 15
Figure 5: The unit ceil of a grain cluster showing a liquid vein at the three-grain junction and
three liquid fluets at grain boundaries (after CoÏbeck, 1980) 16
Figure 6: a- The spectral variation of the permittivity ofwet snow with snow wetness as a
parameter; b - The spectral variation ofthe loss factor ofwet snow with snow wetness as a
parameter. The curves are based on the rnodifiecÏ Debye-Ïike model (afler Haïlikainen et aï.,
1986)17
Figure 7: Geometry ofrepeat-pass InSAR 21
Figure 8: Modeled interferogram with black unes denoting known fault unes and coloured
fringes related to changes in surface elevation 23
figure 9: Standard deviation ofthe phase estirnate as a function ofcoherence and number of
independent samples L (after Barnler and Hart 1998) 25
Figure 10: Refraction change due to difference in dielectric properties between air and snow28
Figure 11: Variation of interferornetric phase difference versus sensor incidence angle. The
color curves represent the snow pack depth variation (from 5 to 100 cm) having a densïty of
0.3gc&33
Figure 12: Variation ofinterferometric phase difference versus sensor incidence angle. The
color curves represent the snow pack depth variation (from 5 to 100 cm) having a density of
0.4gcrn333
Figure 13: Variation of interferometric phase difference versus sensor incidence angle. The
color curves represent the snow pack depth variation (from 5 to 100 cm) having a density of
0.5gcm334
iv
Figuie 14: InSAR phase difference variation versus the changes in Snow Water Equivalent
(SWE) having a density of 0.3 g cm3. The plotted unes represent the sensor incidence angle
with highlight on ERS satellites incidence angle of 23° 35
figure 15: InSAR phase difference variation versus the changes in Snow Water Equivaient
(SWE) having a density of 0.4 g cm3. The plotted lines represent the sensor incidence angle
with highuight on ERS satellites incidence angle of 23° 35
Figure 16: InSAR phase difference variation versus the changes in Snow Water Equivalent
(SWE) having a density of 0.5 g cm3. The plotted unes represent the sensor incidence angle
with highlight on ERS satellites incidence angle of 23° 36
Figure 17: Schefferville study area 40
figure 18: Schematic diagram of the nested study areas for the Cold Land Processes Field
Experiment42
Figure 19: The three 25 km x 25 km study areas, and the nested Intensive Study Areas are
shown with red boundaries 43
Figure 20: The 25 km x 25 km North Park Study Area boundaries are shown in red. Nested
within the area are three 1 km x 1 km Intensive Study Areas: Potter Creek ISA, flhinois River
ISA, and Michigan River ISA (modified after NASA web site) 45
Figure 21: The 25 km x 25 km Rabbit Ears Study Area boundaries are shown in red. Nested
within the area are three 1 km x 1 km Intensive Study Areas: Buffalo Pass ISA, Spring Creek
ISA, and Walton Creek ISA (modified after NASA web site) 47
Figure 22: The 25 km x 25 km Fraser Study Area boundaries are shown in red. Nested within
the area are three 1 km x 1 km Intensive Study Areas: St. Louis Creek ISA, Fool Creek ISA,
and Alpine ISA (modified after NASA web site) 48
figure 23: Backscatter image of Rabbit Ears MSA acquired by AIRSAR instrument on
Febniary 21st 2002 (tsl6O6c.vvi2) 51
Figure 24: DEM image ofRabbit Ears MSA acquired by AIRSAR instrument on February
21st, 2002 (tsl6O6c.demi2) 52
Figure 25: Map showing snow sample locations for each ISA. Red diamonds are first day
sampling (snow depth, snow wetness, and snow surface rouglrness). The blue squares are
sampling on day 2 (snow pits, sec next figure) and green diarnonds are some special sampling
for period of 200354
Figure 26: Locations of CLPX Snow Pits within each Intensive Study Area 55
Figure 27: Methodology Flow Chart 57
V
Figtire 28: Topographic map ofWalton Creek area in 1/13 000 scale dated 1956 (modified
after www.terraserver.com) 59
Figure 29: Aerial photography of Walton Creek area dated 1999 (modified after
www.tenaserver.corn) 60
Figure 30: Illustration ofthe effect of snow cover on perceived ground topography using
differential interferometry using SRTM and Canadian DEM for Schefferville region (Québec)66
Figure 31: Difference map and corresponding topographic rnap of the Lac Vacher area 67
figure 32: Difference map and coiiesponding topographie map of mine pits 67
Figure 33: Difference rnap and topographie map ofthe east-center parts ofthe study area 68
Figure 34: DEMs and Backscatter extractions from Walton Creek ISA. Extracted from: A,
Febntary; B, March and C, September acquisitions 69
Figure 35: Sampling profile in the Walton Creek ISA September DEM’s 70
Figure 36: Altitude verification of 1294 pixels ofthe two different AIRSAR images on the
sameday70
Figure 37: Coffelation of the altitude profiles for two $eptember erossed images ofWalton
Creek ISA71
Figure 38: Validation of altitude correlation for 200 samples of Walton Creek ISA 72
figure 39: The variation of altitude ofAIRSAR data over a road for (IOPY and BG2) 73
Figure 40: The variation of altitude ofAIRSAR data over a road for (IOP2 and BG2) 73
Figure 41: Altitude eolTelation for two crossed images (Feb & Sep) ofWalton Creek ISA 74
Figure 42: Altitude correlation for two crossed images (Mar & Sep) ofWalton Creek ISA 74
Figure 43: Difference Map of September and February 75
Figure 44: Difference Map of March and September 75
Figure 45: Allocating the density by distance 76
Figure 46: Joining the snow density and snow depth data 76
Figure 47: Joining SWE ofAIRSAR data to shape file of ground data 77
figure 48: Distribution rnap of SWE differences based on comparison of remotely sensed
derived SWE and ground truth SWE 78
vi
Figure 49: Distribution map of Snow Depth error .72
Figure 50: The location of transverse profiles in Rabbit Ears MSA 79
Figure 51: The location of transversal profiles in Fraser M$A 79
Figure 52: Profile 1 near Rabbit Ears 20
Figure 53: Distribution of ground tntth SWE $2
Figure 54: Distribution ofmeasured $WE based on InSAR $2
Figure 55: Distribution of$WE differences (ground trnth and InSAR) $3
Figure 56: Evaluating the measured and estimated SWE correlation 23
Figure 57: Applied mask on the Walton ISA to separate higher altitude (outside of green &
inside of yellow zones) from lower altitudes in four sections A, 3, C and D $4
Figure 52: SWE correlations of low altitude points of 4 sections of Walton ISA $5
Figure 59: Statistical resuits ofground tnith snow deptli 86
Figure 60: Statistical resuits ofrneasured snow depth by InSAR 86
Figure 61: Statistical resuits of snow depth differences (ground truth and InSAR) 86
Figure 62: Measured vs. estimated snow depth 87
Figure 63: Snow deptÏi correlations oflow altitude points of 4 sections ofWalton ISA $8
vii
List of Tables
Table 1: Coordinates for corners of each MSA, given first in Iatitude/longitude, then in UTM.
The WGS$4 datum is used foi- ail coordinates 44
Table 2: Characteristics ofthe Meso-ceil study areas, the general physiographic regimes they
are considered to represent, and the approximate percentage of global, seasonaÏly snow
covered areas having these characteristics 44
Table 3: Location and characteristics of ISA. UTM coordinates are given for upper left (UL)
and lower right (LR) corners 49
Table 4: Different kind of imagery produced by ALRSAR data 51
Table 5: AIRSAR ftight unes and length 52
Table 6: Ground data collection date for each MSA during IOP1 and 10P2 55
viii
Acronyms
AIRSAR Airbome Synthetic Aperture Radar
BG Bare Ground
CARTEL Centre d’Applications et de Recherches en Télédétection
CDEM Canadian Digital Elevation Model
CLPX Cold Land Processes Experiment
DEM Digital Elevation Model
Diff-Map Difference Map
DInSAR DifferentiaÏ Interferornetry Synthetic Aperture Radar
DN Digital Number
ESA European Space Agency
GCP Ground Control Points
GIS Geographic Information System
InSAR Interferometry Synthetic Aperture Radar
10CC Iron Ore Company of Canada
IOP Intensive Observation Period
ISA Intensive Study Areas
JPL Jet Propulsion Laboratory
MSA Meso-celi Study Areas
NSIDC National Snow and 1cc Data Center
SDT Schefferville Digital Transect
SRTM Shuttie Radar Topographic Mission
SWE Snow Water Equivalent
TOPSAR Topographic Synthetic Aperture Radar
UTM Universal Transverse Mercator
XTI Cross-Track Interferometry mode
ix
Acknowledgments
I would like to express my gratitude to my supervisor, Hardy B. Granberg, who suggested this
topic for investigation, and whose interesting discussion, insightful critique, and final editing
ofthe thesis were instrumental to the successfuÏ compÏetion of this project.
I would especially like to express my deep gratitude to rny co-supervisor Q. Hugh I. Gwyn for
his valuable scientific advice, his guidance, and his constant availability by Internet, even
when he was on the other side of the world.
I greatly appreciate the helpful discussion and suggestion of Marcel Laperle, as well as the aid
of various professors, colleagues and personnel at the Centre d’Applications et de Recherches
en Télédétection (CARTEL) at the Université de Sherbrooke.
I want to thank Professor Ferdinand Bonn, who is no longer among us. whom I had the great
occasion to benefit of his presence and his reach of scientific knowledge (May God bless him).
My particular thanks go to my family; my wife Maryam, my sons Pouria and Parham and rny
beloved parents whose prayers for me are my main fortune.
My final acknowledgments are dedicated to Ministry of Research and Technology of the
Islamic Republic of Iran for granting me a research scholarship.
AIR$AR images and Field data were provided by NASA, Jet Propulsion Laboratory (JPL) and
National Snow and Ice Data Center (NSIDC) of United States of America, I thank them
respectively.
&4Øw
1 INTRODUCTION
1.1 Background to the study
Snow cover plays a significant foie in the energy and water budgets of the Earth on a variety
of spatial and temporal scales. The storage of large arnounts of fresh water in seasonal snow
covers is a critical element of the Earth’s hydrologie cycle. The Snow Water Equivalent
(SWE) determines how rnuch liquid water per unit area will result in case where the snow
cover is cornpletely melted and it is, therefore, an important quantity in snow-meh ninoff
prediction for hydro-power and agricultural purposes.
Seasonai snow covers through their influence on land surface albedo, radiation balance and
boundary layer stability, have profound effects on weather patterns over large areas.
On average, over 60% of the northem hemisphere land surface and over 30% of the Earth’s
land surface has seasonal snow (Robinson, et al., 1993). In many mountainous regions,
snowfall is a substantial part ofthe overail precipitation (Serreze et al., 2000), and snowrnelt is
a major source of the total amiuaÏ stream ftow. While water from snow meit can be a vitally
important natural resource, it can also become an environrnental hazard when rapid melting of
a snow pack produces ftooding.
2
The properties ofthe slow cover are ofinterest on a local scale. At Schefferville (55°N, 67°W)
the close corTelation between local snow depth variations and the distribution of permafrost
(Annersten, 1966) sparked intensive research on the spatial variations in snow accumulation
(Thom, 1969; Thom and Granberg, 1970) which led to the development of methods to rnap
snow depth using air photo sequences ftown during spring meÏt (Granberg, 1972; 1973;
NichoÏson, 1975). It also lcd to the development of an early Geographic Information System
(GIS) technique to reconstruct past snow covers for mines already under excavation
(Granberg, 1972; 1973). The vast spatial database on snow cover characteristics thus
generated together with Ïong-term snow cover monitoring by students and staff of the McGiIl
Sub-Arctic Research Station for hydrologic and other purposes (Adams et cil., 1996; Nicholson
and Thom, 1973; Nicholson and Granberg, 1973) made Schefferville the logical choice as
principal field site for snow cover modeling experirnents sponsored by the Department of
National Defence. As part ofthese studies the Schefferville Digital Transect was established in
an area NW of Schefferville (Granberg and Irwin, 1991). The modeling research further lcd to
exploration of radar as tool for terrain mapping (Granberg, 1994; Granberg et al., 1994) and
snow model validation which brought exploratory experiments by Vachon et aï. (1995) to the
SDT. h these experiments ERS-1 in 3-day orbit over the SDT provided parent images for
repeat-pass interferornetric processing over varying time intervals extensib]e from December
25, 1993 to March 28, 1994. figure 1 shows examples of coherence images from this data set.
The InSAR data created from images with three-day intervaïs showed that for the exposed
rock and alpine tundra, the scene coherence is always high but for other surface cover types,
the coherence is more variable. In the two first images (left) snow fail between the two
satellite overpasses created a loss of coherence while in the right one there was no snow fali
between the overpasses.
Subsequent analyses (Fisette, 1999) demonstrated the existence of a strong, snow-related
topographic signal, but it also demonstrated that repeat-pass InSAR is not suited for SWE
mapping. For such mapping, cross-track interferometry is required. Other research (Côté,
1998) demonstrated that interferometric coherence is a useful and unique tool to study lake ice
forming processes, in particular internai ftooding ofthe snow cover on lakes.
j
Using a different data source (RADARSAT-1), Granberg and Vachon (1998) dernonstrated
that die spatial distribution of discoutinuous permafrost can be rnapped with good accuracy
using InSAR coherence.
(c)
figure 1: Three examples ofcoherence images ofrepeat-pass interferometry with 3 days
interval, caÏcuÏated from ERS-1 parent images, (a) January 9 and 12, (b) January 12 and 15,
(c) February 14 and 17 (Images provided by Vachon et ctl., 1995)
Over the 24-day orbit interval of RADARSAT-1 snow accumulation may cause loss of
coherence, except in areas where wind continually sweeps away the snow that fails. As
previous researci has shown, such parts ofthe terrain lose their protective snow cover, causing
the average annual ground temperature to approach that of the air, which at Schefferville is
about -5 oc (Nicholson and Granberg, 1973; Granberg et al., 1994). The interferometrically
produced maps of the zones of shallow sriow (Figures. 2a, b) closeÏy match a map of
permafrost (Figure 2c) produced by the Iron Ore company of Canada (10CC) using air photo
interpretation, ground probing and geological trenching data (Granberg and Vachon, 1998).
The research at Schefferville forms the background to the current study. However, although
the research at Schefferville took the Iead in this field, research on snow cover applications of
radar interferornetry has progressed elsewhere also. Shi et aÏ., (1997) had also attempted
mapping the snow cover using repeat-pass InSAR and Strozzi et al. (1999) had found repeat
pass InSAR coherence useful in the mapping ofwet snow.
ta) (b)
4
Figure 2: (a) RADARSAT coherence image from pair image ofDecember 1 and 25, 1996, (b)
RADARSAT coherence image from pair image of February 28 and March 24, 1997, (e) Part
of permafrost map produced by 10CC in 1972-74 (modified after Granberg and Vachon,
199$), Areas swep the snow bu the wind retain coherence (white) while other parts lose
coherence due to snow accumulation
Granberg and Vachon (199$) suggested that the refraction of microwaves by dry snow should
produce a phase-shifi which is linearly related to the SWE. Guneriussen et aÏ. (2001)
elaborated this relationship further, and calculated that at the nominal ERS satellite incidence
angle (&j=23°):
= 2ki(0.$7SWE) (1)
Where As represents the changes in interferometric phase due to change in SWE and k1 is the
wave number. They ernployed data from the ERS-1 and 2 Tandem Mission to show that
snowfall and snow redistribution by wind pose problems to routine monitoring of the SWE
using repeat-pass InSAR. They also showed that the SWE does produce a measurable signal.
Later Rott et aÏ. (2003), using ERS-1 3-day repeat pass data again confirmed that the temporal
decorrelation due to differential phase delays at sub-pixel scale caused by snow fall or wind
re-distribution of snow is a main lirniting factor for use of such data.
Granberg and Vachon (1998) proposed that at vertical incidence a SWE of 32.6 mm causes a
full phase-shifi at C-band (56 mm). This value was supported by the experiments of
Guneriussen et al. (2001). In C-band this rapid phase shift poses a problem to differential
interferometry because coherence is rapidly lost. Strategies have been developed to overcome
(a) (b) (c)
5
this problem in the phase-unwrapping (Engen et aL. 2004; Larsen et aï., 2005) but they
engender substantial losses in spatial resolution. Use of a longer wavelength, sucli as L-band
would reduce this problem. Not only does a full phase phase-shifi require a distance which is
more thari four times that for C-barid, but it also increases the penetration depth (Rignot et aï.,
2001). However, it does not entirely solve the problem of coherence loss. This renders
differential InSAR flot very useful to SWE mapping.
A better approach which is explored here is to employ repeat-pass cross-ti-ack InSAR. Such
image acquisition is simultaneous, enabling use of C-hand and even shorter radar wavelengths.
This is necessary, because cross-track interferometry limits the possible length of the baseline
between image pairs. Two such datasets were made available, one fortuitousÏy in C-baud
through the Shuttie Radar Topographic Mission (SRTM) which was flown in the period
February 11-22, 2000 and therefore provides a DEM acquired under cold snow conditions at
Schefferville. Differences between this DEM and a DEM produced by Canada’s Topographic
Survey using photogrammetric techniques should in part resuit from the influence of the SWE
on the interferometric phase.
The second set of DEM’s using C and L-hand were flown for the purpose of this study during
the Cold Lands Processes field eXperiment (http://www.nohrsc.nws.gov[-cline/). The data set
was acquired using NASA’s AIRSAR in TOPSAR mode, i.e. equipped with dual receiving
antennas for cross-track InSAR. A reference data set was acquired over bare ground in late
summer for comparison with similar data sets acquired in coimection with field campaigns at
different times in the winters 2002 and 2003.
1.2 Objectives
1.2.1 General objective
The general objective of the present research is to explore, in theory, and by experiment, the
information content of repeat-pass cross-track In$AR, in particular that reÏated to SWE and
snow depth.
6
1.2.2 Specific objectives
The specific objectives ofthe research are to:
> Elaborate in theory the relationship between InSAR phase differences. SWE, and snow
depth;
Explore experirnentally the use of repeat-pass cross-track InSAR to rnap snow depth
and SWE.
1.3 flypotheses
- The phase delay of InSAR, in dry snow, is a linear function of the arnount of ice in the
path of the radar wave (Granberg and Vachon, 1998);
> InSAR backscatter, originates at the snow/ground interface in dry-snow and snow-free
conditions. In wet snow it originates at the snow surface (Ulaby et aÏ., 1982).
In dry snow conditions the SWE is expressed as a depression of the surface with
respect to the snow-free surface.
In wet snow conditions the snow depth is expressed by an elevation of the surface with
respect to the snow-free surface.
1.4 Thesis outiine
Chapter 2 presents the object of the study of this thesis. This includes the general
characteristics of ice and various facto;s affecting the snow spatial variability. A sumrnary of
our current understanding of rnicrowave-snow interaction is also given.
The theory of InSAR is given in Chapter 3. Based on this theory the expected relationships
between interferometric phase and snow water equivalent and depth are eiaborated.
Chapter 4 gives a description of the study areas and data sets which include both remote
sensing data, and field data. The rnethod of analysis also described.
Chapter 5 presents and discusses resuits ofthe field experiments and their analysis.
7
Chapter 6 gives a summary of the conclusions and makes recommendations for future
researcli.
1.5 Claim to originality
This study is, to the best knowledge of the author, the first attempt to employ repeat-pass
cross-track InSAR to map snow depth and $WE. Although there were problems introduced by
DEM colTection routines, and, in the Colorado case, an unstable remote sensing platfonn the
existence of a signal consistent with that expected was dernonstrated for both wet and dry
snow, indicating that satellite-borne cross-track InSAR is capable of overcoming coherence
problems experienced in atternpts at snow cover mapping using Differential InSAR.
8
2 SNOW PROPERTIE$ AND CHARACTERESTICS
2.1 Snow cover
As atmospheric snow accumulates on the ground it bonds together, fonriing an ice skeleton. It
is the properties of this skeleton which are of interest in this study which focuses on its total
height (slow depth) and its snow water equivalent (SWE), i.e. the depth of water it would
represent if liquefied. Reviews on the properties of atrnospheric snow and its subsequent
redistribution, metamorphisrn and meit are provided by, for example, the many contributing
authors in Gray and Male (1981). A summary of the factors influencing the developrnent of
the snow cover in a sea ice environment is given by Granberg (1998). Here the focus wiÏl be
on those properties which influence its interactions with electrornagnetic radiation in the
microwave range.
A snow cover is a mixture of ice, air, and sornetimes, liquid water. Many of the characteristics
and properties of snow depend upon whether it is dry or wet. When wet, it is at its melting
temperature of O °C. In dry snow, a low thermal conductivity leads to steep temperature
gradients and large diurnal variations in its surface temperature. Snow can undergo
volumetric defomiations that are basically irreversible.
Once the snow is deposited on the ground, the shape of the ice crystals begins to change, on a
process known as metamo;phism. The type of metamorphism depends on the snow
temperature and water vapour fluxes and on whether the snow is wet or dry. The metamorphic
process controls the change of shape and size ofice particles (Colbeck, 1982). Since radiation
is absorbed near the snow surface, melting proceeds from the surface downward, with meit
9
water percolating into the snow pack. When the percolating water arrives at a depth where the
snow is below the melting temperature, it freezes and releases latent heat which further warrns
the underlying snow. In this way the snow pack is gradually brought to near isothermal
conditions, at the melting temperature throughout. However, the process is oflen inteinipted
by noctumal fteezing, especially on clear, cold nights, due mainly to emission of radiation
and evaporation from the snow surface. freezing also proceeds from the surface downward,
producing a hard crust. A surface crust can also be formed by strong, cold winds. If
percolating water reaches a relatively irnpemieable boundary in the snow, it may forrn an
interior cntst or ice Yens when it freezes. Once the snow becornes wet, the grains quickly
assume the rounded, meit or melt-freeze metamorphic shape, and after the snow pack has
been wet throughout it is much more homogeneous than dry snow, although ice lenses and
layers may persist for a time.
Newly fallen snow has densities in the range 100 kg m3 to 200 kg m3 and consists of loosely
packed dendritic and plate-like ice crystals. If deposition occurs during strong winds, the
crystals are broken into srnall pieces and higher snow densities (in excess of 400 kg m3) can
resuit. After deposition a fresh snow layer tends to settie and its density increases. The
underlying layers also continue to settie and increase in density. Near the base of the snow
pack, it is common to find a layer of lower density because of temperature-gradient
metamorphism while the pack is thin. Water vapour transport due to strong temperature
gradients resuits in large faceted crystals, which can be several millimetres across with IittÏe
cohesion between them (depth hoar). In natural snow cover intermediate forrns sucli as
crystals with mixed faceted and rounded parts are more often found than fully faceted
crystals (Armstrong, 1980). In late winter and early spring, the density typicaÏly reaches a
constant value in the lower two thirds of the pack: except for lower values near the snow
ground surface. In the spring, the snow has usually attained a more uniform density
throughout the whole profile. If snow is wetted by ram and/or meÏt, grain morphoÏogy
changes rapidly and a general coarsening occurs. The typical size of the rounded particles in
wet snow is 0.5 to 2.0 mm (Dozier et al., 1987).
The natural growth of a snow pack during a winter season usually leads to stratification and
layers of different properties. In temperate climates, the ground under a snow cover usually
remains relatively warm due to the insulation provided by the snow pack. When the
10
underlying ground is unfrozen, the basal layer may be wet while the snow above is dry. In
places where the snow pack persists through the winter, or at Ïeast through multiple
snowfalls, it is buiÏt up of a sequence of layers, each having experienced different
depositional and metarnorphic environments. The different layers therefore usually have
quite distinctive characteristics, as long as they remain dry. The boundaries between layers
can 5e very sharp, especially if dust has settled on the surface, or there has been partial
melting, between snowfalls (Colbeck, 19x2).
2.2 Snow spatial variability and distribution
2.2.1 Snow accumulation
Snow cover comprises the accumulation of snow on the ground follow on from precipitation
deposited as snowfall, ice pellets etc.
Its structure and dimensions are complex and highly variable both in space and time. The
area variability of snow cover is studied at three spatial scales (Pomeroy et ut., 1995):
• RegionctÏ seule: large aieas with linear distances up to 1000 km in which dynarnic
meteorological effects such as wind flow around baniers and lake effects are
important;
• Local seule: areas with linear distances of 100 m to 1000 m in which accumulation
may 5e related to the elevation, aspect and siope of the terrain and to the canopy and
crop density, tree species, height and extent ofthe vegetative cover;
• Micro scale: distances of 10 m to 100 m over which accumulation pattems resuit
primarily due to surface rouglmess.
2.2.2 Effect of topography on snow cover distribution
Snow cover distribution varies depending on the terrain topography features. Topography
elevation, topography slope and topography aspect are the three important features which can
influence the snow cover distribution:
11
• Elevation: Where vegetation, micro-relief and other factors do not vary with elevation,
the depth of seasonal snow cover usually increases with increasing elevation because
of sirnultaneous increase in the number of snowfalÏ events and decrease in evaporation
and meit. At a specific location in a mountainous region, therefore, a strong linear
association is often found between seasonal snow water equivalent and elevation
within a selected elevation band (Porneroy et aÏ., 1995).
• Siope: Orographic precipitation rate are more depended on terrain siope and windflaw
rather than elevation. The siope angle and direction ofmountain in compare to air mass
or front are the important factttres which detemine the forcing ascent, for this reason,
sorne hill with littie elevation may receive more precipitation than a mountain with
high elevation.
• Aspect: The importance of aspect on accumulation is shown by the large differences
between snow cover amounts found on windward and leeward siopes of coastal
mountain ranges. In these regions the major influences of aspect contributing to these
differences are assumed to be related to: the directional flow of snowfall-producing air
masses; the frequency of snowfall; and the energy exchange processes influencing
snowrnelt and ablation (Gray and MaYe, 1981). Aspect also affects snow distribution
pattems due to its influence on the surface energy exchange processes. The effects are
most evident during snowmelt. Snow disappears first from those slopes that receive the
highest radiation and from those that are exposed to the movement of wami winds.
(Toews and Gluns, 1986).
2.2.3 Effect of vegetation
Tali vegetation intercepts snowfall and shorter vegetation becornes incorporated into the snow
cover. In a forest, intercepted snow may sublimate or fali to the ground. Both processes affect
the distribution of snow cover under a canopy. Sublimation reduces the amount of snow
available for accumulation whereas snow that falis from the canopy affects the spatial
distribution in depth and density of a snow cover. The latter is strongly influenced by
atmosplieric turbulence within the canopy during release. In particular regions, the
combination of these processes may produce distribution pattems that are reasonably similar
from year-to-year. The geornetries of these pattems are related to vegetation type, vegetation
12
density (projected area of canopy, trunk density or stalk density) and the presence of nearby
open areas (Pomeroy et aï., 1995).
Within forest canopies, snow depth and water equivalent vary with distance to trees, generally
decreasing witli decreasing distance to a coniferous tree tnmk and increase slightÏy with
decreasing distance to a deciduous trunk (Woo and Steer, 19$6; Sturm, 1992). Greater
accumulations tend to occur under deciduous trees and less under conifers (Barrie, 1991).
2.2.4 Effect of wind
Wind transport processes have been reviewed by Mellor (1965). In open terrain, wind
redistributes the snow cover creating sometimes large spatial variations in snow depth and
water equivalent (Granberg, 1972). During transport the snow is also fragmented which leads
to an initial higher density and stronger initial bonding (Granberg, 1972; 199$).
2.3 Microwave interactions and dielectric properties of snow
Our interest in the interaction of microwaves with snow arises from the special dielectric
properties of ice crystals and water molecules, and from the problem of propagation of
electrornagnetic waves in a heterogeneous media whose grains are srnailer than the wave
length.
An important requirement for the remote sensing of snow is the existence of signatures:
a) to identify snow covered and snow-free areas,
b) to cÏassify snow types, and
c) to qualïfy and quantify snow properties.
The search for such signatures can proceed in different ways. in a first approach, purely
empirical relationships between microwave and snow parameters are found. On the other
hand, the analytical approach tries to model the snow conditions with realistic pararneters,
and then to predict and test relationships between observed inicrowave and snow properties.
Both approaches are used in iterative steps. Only in empirical data can realistic parameters
be found and used in an analytic approach and only the analytic approach leads us to a
deeper physical understanding of the interaction of electromagnetic waves with snow cover.
13
The areal extent, snow depth and density are the prirnary pararneters of a snow cover. In
addition, the liquid water content, the rnorphology, in particular the size and shape of the
grains, surface roughness and layering are of importance.
The electromagrietic properties are the magnetic and electric polarizabilities of the media
(Hippel, 1954). Together with the geometrical arrangement of the snow components, these
parameters determine the electrornagnetic boundary values of characteristic parameters, and
therefore the propagation, scattering and absorption of electromagnetic waves.
The magnetic susceptibility ofthe snow and water components is not significant; water, for
instance, has a relative magnetic permeability of 0.999988. This smalï diamagnetic
interaction is negÏigible in comparison with the electric interaction. Similai-ly, the
electromagnetic interaction with air and water vapour is small. The relative dielectric
constant ofthe atmosphere is aiways very close to unity 1.0 <E < 1.001 (Bean and Dutton,
1966). In comparison to the uncertainties of the dielectric properties of ice and water, the
atmospheric contribution to the interaction of microwaves with the snow cover can 5e
neglected. This statement, of course, holds tnie only for the air within the snow cover. When
a rnuch longer propagation path from a satellite platforrn to the earth’s surface is considered,
atmospheric effects may have to be included in the overalÏ radiative transfer chain. Clouds,
precipitation and water vapour can produce noticeable absorption and emission especially at
frequencies above 20 GHz; and in the 60 GHz range a complex of oxygen absorption unes
greatly reduces the transparency ofthe atmosphere (Waters, 1976).
When reducing the problem to the interaction of microwaves with snow cover, the basic
electromagnetic properties are the relative dielectric constants of ice and liquid water and
their geometrical distribution in the snow cover. The relative dielectric constant of a medium
E 1S a complex number and consists ofa real (s’) and an irnaginary part (s”)
5=5-jE”(2)
where j -1 . The term s’ is usually referred to as the perrnittivity ofthe material and s” the
dielectric loss factor (Ulaby et aÏ., 1986). The reaÏ part s’ gives the contrast with respect to
free space (S’ait = 1), whereas the irnaginary part s” gives the electromagnetic loss of the
14
material. Based on Ulaby et al., (1986) treatment of the dielectric properties of snow
usually resuit in division of the snow into two groups:
• dry snow, which is a mixture of ice and air and contains no free (liquid) water
• wet snow, which contains free water.
Electromagnetically, dry snow is a dielectric mixture of ice and air and, therefore, its
cornpÏex permittivity is governed by the dielectric properties of ice, snow density and ice
particle shape (Hallikainen and Winebrenner, 1992). Since the real part of the perrnittivity
of ice jS Lice = 3.17 at frequencies between 10 MHz and 1000 GHz and is practically
independent of temperature. the perrnittivity of dry snow (Eu) is only a function of the
snow density ‘p), which, under natural snow pack conditions typically ranges between 0.2
and 0.5 g cnf3 (Ulaby et al., 1982).
At high frequencies the imaginary part (c’dS) is much smaller than the real part (Ld); then
the two quantities can easily be decoupled and the value of L’ds can be found by putting L”ds
O (MitzÏer, 1987). In fact, in the case where snow changes to wet conditions (snow
contains free water), the imaginary part shows very high sensitivity to the presence of liquid
water content (L; percent by volume). Cumming (1952) found that r.” of dry snow with a
density of 0.38 g cm3, is about 1.2 x l0 at T -1°C. With increasing liquid water content
(Lu,) from O to 0.5 %, the r.” increased by more than one order of magnitude, which
corresponds to the same order of decrease in the penetration depth (Figure 3).
Most measurements of dry snow permittivity (r.’d) refer to the measurernents of Cumming
(1952), Bobren and Battan (1982) and Tiuri et aÏ., (1984). The resuits of r.’ for dry snow
reported by Tiuri et aÏ., (1984) are shown in Figure 4 including linear and quadratic fits and
based on their work, Matzler (1987) obtained the following relationship for dry snow
permittivity:
r.’= 1 + 1.7p+0.54p2 O.lpO.6 g.cm3 (3)
15
3020
10
Ï
-t,
C
I10rn’
Volumetric liquid water content (%)
figure 3: The penetration depth of snow decreases rapidÏy with increasing liquid water
content, especialÏy in the immediate vicinity ofLw = O (modified after Ulaby et al., 1986)
1020 1 2 3 4 5
ii
E’1 1I0p
u
Figure 4: Real part ofthe dieÏectric constant of dry snow as a fttnction of the density of dry
snow relative to water (modified after Tiuri et aÏ., 1984)
16
The dielectric properties are almost independent of temperature as long as the snow rernains
dry, i.e. as long as the temperature is below -0.5°C (Ulaby et aÏ., 1986). As the temperature
approaches 0°C liquid water forms leading to wet snow. Wet snow is a three-component
dielectric mixture of ice particles, air and liquid water, and its permittivity is a function of
frequency, temperature, volumetric water content, snow density, and the shapes of ice
particles and water inclusions (Hallikainen and Winebre;mer, 1992). Since the perrnittivity
of water (Ewater = 80) is substantially higher than that of ice and air, the dielectric behaviour
of wet snow is govemed by the volurnetric fraction of water (Hallikainen and Winebrenner,
1992).
The liquid component has cornpleteÏy different dielectric properties. It is nestiing in small
fluets and veins between the contact points of neighbouring ice grains in an attempt to
reduce the surface energy. The situation for a cluster of three grains is shown in Figure 5.
Whereas the small ice grains disappear the larger ones are growing to well rounded crystals
with typical diameters of 1 mm (Colbeck et aÏ., 1990).
Snow wetness or free-water content is typically obtained using calorimetry, dilution or
dielectric measurements. Colbeck et al. (1990) gave a general classification of hquid-water
Figure 5: The unit ceil of a grain cluster showing a liquid vein at the three-grain junction and
three liquid fluets at grain boundaries (after Colbeck, 1980)
17
content: snow is said to be dry when the percentage of liquid water is 0% by volume, rnoist
at 3%, wet at 3-8%, very wet at 8-15% and saturated slush at> 15%. Liquid water is mobile
only if the so-called irreducible water content is exceeded and surface forces cannot hold
the water against gravity. The irreducible water content is about 3% and is dependent
prirnarily on snow texture; grain size and grain shape (Colbeck et aÏ., 1990).
Measurement of the spectral behaviour of the dielectric constant of wet snow has been
studied by several researchers (Linlor, 1980; Hallikainen et aï., 1982, 1983 1984, 1986;
Mtz1er et aï., 19X4). The perrnittivity and dielectric loss factors of wet snow
were measured for 62 snow samples at the frequencies between 3 GHz and 37 GHz by
Hallikainen et aÏ. (1986) using a ftee-space transmission technique. The range of liquid
water content (Lw) in the samples was between 1 and 12 percent. They applied an empirical
modified Debye-like rnodeï to caïculate the permittivity and dielectric loss factors of wet
snow. Figure 6 shows the effect of liquid water content on the dielectric behaviour of wet
snow between 3 GHz and 37 GHz for a dry snow with an average density of P 0.24 g
crn3.
t,fDpt
0 10 20 30 40
Frcuncy f ({iHz)
Figure 6: a- The spectral variation of the permittivity ofwet snow with snow wetness as a
parameter; b - The spectral variation ofthe loss factor ofwet snow with snow wetness as a
parameter. The curves are based on the modified Debye-Ïike model (after Hallikainen et al.,
1986)
As shown in Figure 6b, the maximum value of E,s was obtained at 9.0 GHz which agrees
with the relaxation frequency of water at 0°C. Because of the large increase in the
Q
Q‘no
L]
L-..
CiL
‘u
111.00.9
0.8
.. 0.7
0.6
0.5V,o-J
C
Q,
G)
cD
A b
20Frequency t t 0Hz)
18
magnitude of dielectric loss factor for water (&‘) between 3 GHz and 6 GHz, there is a
sharp increase in dielectric loss factor of wet snow (&‘) over this frequenoy raige. The
attenuation constant of wet snow increases practically linearly with frequency up to 15
GHz. From 1$ GHz to 37 GHz, the average siope is slightly srnaller.
19
3 INTERFEROMETRIC SAR
3.1 Tlieory
Since the first airborne radar interferometry application (Graham, 1974), interferometric SAR
(InSAR) has been applied increasingly in the last decade for the measurement of Earth’s
topography (technique description and review, Zebker et al., 1994a; Bamier and Hart, 1998,
surface changes - Askne and Hagberg, 1993; WegmtilÏer et ctl., 1995, and surface feature
movernents - Gabriel et aÏ., 1989; Goldstein et aÏ., 1993, Massonnet et aÏ., 1995).
InSAR exploits the phase differences of two single-look complex SAR images acquired fi-om
diffei-ent orbit positions sirnultaneously or at different times. The parent images contain
information related to phase, coherence, backscatter amplitude, texture and amplitude
differences. This information can be used to estirnate several geophysical quantities, such as
topography, snow cover, vegetation, ground dispiacement (vertical and horizontal),
hydrography, growth and dynamics of lake ice and soil characteristics.
The InSAR geometry in the plane of observation is shown in Figure 7. SAR data are acquired
at the locations designated 1 and 2 and are processed to complex images S (x, y) A1 exp {çl }
and S, (x, y) = A, exp {02 } where (x,y represents the azimuth and range image coordinates.
Subject to constraints on surface change and proximity of the orbits to one another, the two
images may be registered and combined to produce an interferogram (Equation 4):
20
I(x,y) (4)
The phase of the interferogram is the phase difference between the two images. The
unwrapped phase ((L)) may be related to the path difference (3)between two imaging locations.
If the baseline, defined in ternis of orbit separation (B,) and tilt angle (), is known, then the
measured difference in the interferogram phase (cÏ) between two points may be related to the
change in range dR and the change in surface height dz (Ulander and Hagberg, 1993):
2kB,,[dz+cos9.dRJdØ+dØR, (5)
RsinO
where R is the range to the first point,
e is the local incidence angle,
k = 2rt/2 is the radar wave number,
2L is the radar wave length.
And thus
B,, Bcos(&—) (6)
Where B,, is the perpendicular (normal) component ofthe baseline and () is the tilt angle.
The tenn dR may be systernatically removed from the interferogram to retain only the tenam
dependent phase fluctuation d. In this case, each phase fringe (change in phase from O to 2n)
represents a relative change in elevation of
d _%RsinO/ 7/23,, t)
For successful interferometry, the phase in the two images must be well correlated. The
coherence, a measure ofthe phase correÏation, is defined as:
conjugate ofcomplex number has the sarne real part but opposite imaginary part. The conjugate ofre”’ is re”.
21
(s1 (x, y)5 (x, y)7=
Where (s (x, v)S (x, y) is the spatial average over a processing window.
Lineof
sight h
n
(8)
y
Figure 7: Geometry ofrepeat-pass InSAR
The coherence lies in the range Oyl. If the coherence is large (close to unity), the phase in
two images is highly correlated between the two passes. If the coherence is srnall (close to
zero), the phase is flot well conelated. Surface changes during the time between passes, such
as temporal changes in the geometric and electric properties of the surface, may reduce the
desired coherence. This contribution to the measured coherence, usually refeiied to as the
scene coherence, can provide information on geophysical surface properties, however, other
causes such as phase aberrations in the processor, differing imaging geornetries, and
heterogeneous atmospheric conditions, may also lead to phase decorrelation.
X
22
3.1.1 Differeutial Interferometry
A development of the interferornetry scheme can be used to detect changes in topography.
Suppose that two interferograms are fonned of a given target scene, taken at different times,
and one subtracted from the other. If nothing has changed during the intervening period, there
will be a constant phase difference everywhere. If, on the other hand, elevation changes have
occurred, then these will show up as fringes in the differential interferogram.
This gives an extraordinarily sensitive means of measuring elevation changes. In fact,
Differential Interferometry (DInSAR) is a measure of the small dispiacernents which have
occurred in the scene between two image acquisitions (Gabriel et al., 1989). If there are
changes during the image acquisition times, the interferometric phase will record any
dispiacement of the ground along the radar une of sight that has occurred during the time
interval between the two images.
The sensitivity of the phase to the topography increases with an increase in the baseline
separating the two antenna locations at times of the data acquisitions. To generate a surface
dispiacernent rnap it is necessary to remove the topographie signal from the interferogram.
This can be achieved by either:
• simulating the topographie phase using a digital elevation model and with
knowledge of the geometry of the interferometric system (two pass rnethod, e.g.,
Massonnet et al., 1993; Murakarni et aï., 1996);Or
• by generating two interferograms out of three or four SAR images of the same
area, and computing the phase difference of the two or to eliminate the
topographic phase signal common to both interferograms (three or four pass
rnethod, Gabriel et aÏ., 1989; Peltzer and Rosen, 1995).
The sequence ofDInSAR images ofthe Landers earthquake (June 18, 1993) provided an ideal
test for differential interferometry. Because the shallow depth of the epicentre, the earthquake
produced spectacular surface rupture in an arid area less than three months afler the ERS-1
satellite began acquiring rada;- images in its 35-day orbital cycle (Massoirnet et aï., 1994a).
23
With 20 fringes in the shape of a crushed butterfly, the first earthquake interferogram (Figure
8) illustrated the coseismic deformation field (Massonnet et al., 1993). The original study
launched many others (Peltzer et aï., 1994; Zebker et al., 1994b; Massonnet et al., 1994a;
feigi et aï., 1995).
Each fringe denotes 2$ mm of change in range. The number of fringes increases from zero at
the northern edge of the image, whcre no coseismic dispiacement is assumed, to at least 20,
representing 560 mm in range difference, in the cores of the lobes adjacent to the fault. The
asymmetry between the two sides of the fault is due to the curvature of the fault and the
geometry of the radar. Black fines denote the surface rupture mapped in the field (Massonnet
et aï., 1993).
Figure 8: Modeled interferogram with black unes denoting known fault unes and coloured
fringes related to changes in surface elevat ion
(Modified after Massonnet et al., 1994a)
24
3.1.2 Errors and limitations ofInSAR
The major limiting factors for topographic mapping and deformation monitoring using InSAR
are:
• The limited measurement accuracy ofthe interferometric phase (phase noise;
Barnier and Hart, 199$; Rodriguez and Martin, 1992);
• The variation in the wave propagation conditions over the data acquisition period
(atmospheric signal; Zebker and Rosen, 1997; Hanssen and Feijt, 1996; Goldstein,
1995; Mouginis-Mark, 1995);
• Repeat-pass temporal change ofthe backscatter property ofthe surface (temporal
deconelation; Zebker and Villasenor 1992).
3.1.2.1 Phase noise
The phase differences which constitute the interferogram can be corrupted by phase noise,
firstÏy due to the finite signal-to-noise ratio in each of the two images, and secondly due to the
fact that, for a distributed target, the two signais corresponding to a given pixel are partially
decorrelated (Zebker and Villasenor, 1992). This occurs because the contributions from the
scatterers within the pixel add up with a slightÏy different phase reÏationship, since they are
viewed from a slightly different angle. A larger baseline can resuit in a greater decorrelation
effect. The phase noise can, of course, be reduced by averaging multiple looks, but at the
expense of increasing the pixel size.
The effect of pixel deconelation on phase noise has been evaluated by Barnier and Hart
(1998). Figure 9 shows the phase noise standard deviation as a function of coherence and
number of looks. It is interesting to note that (for high coherence) averaging of L independent
complex interferogram samples reduces phase noise significantly.
3.1.2.2 Atmospheric perturbations
The determination of surface-changes using JiiSAR is based on the assumption that the
radar signal propagation is unaffected by the atmosphere. This, however, is not the case.
Path delays occur in both the ionosphere and the troposphere. Ionospheric path delay is
caused by variations in the total electron content along the path and by moving ionospheric
disturbarices. The former depends on the time of day and influences the whole scene
25
homogeneously. The latter may cause localized artefacts. The rnost severe atrnospheric
effect, however, occurs in the troposphere. Its path delay consists of two components, the
dominant dry part (about 2.3 ni in vertical direction) and the small but highly variable wet
part which is caused by the strong temporal and spatial variability of the water vapour
concentration. It can take on magnitudes up to 30 cm. Artefacts in interferograms have been
reported quite often and some of them have been assigned to atmospheric perturbations as
in Massonnet et aï., (1994b), Massonnet and Feigi (1995) and Hanssen and Feijt (1996).
1 ai
80
60
4a
Figure 9: Standard deviation ofthe phase estimate as a function ofcoherence and number of
independent samples L (after Bamier and Hart 1998)
These effects can cause misinterpretation in the order of centimetres of the defonnation
fields at the surface derived from InSAR. For instance, a differential interferornetry chai-t of
the KiÏauea volcano system, generated by data of the April and October mission of SIR-C in
1994 was first interpreted as a demonstration of a volcanic lift of several centimetres. Later
this tumed out to 5e just the atmospheric effects. These effects have been reported by
Hanssen and Feijt (1996). Although recognized as one of the most important and
challenging research topics for InSAR, only a few quantitative results have been published
so far (e.g. Goldstein, 1995; Massonnet and Feigl, 1995; Tarayre and Massonnet, 1996).
Especially in wet regions like the Netherlands, SAR images exhibit artefacts due to the
temporal and spatial variations of atmospheric water vapour. Other tropospheric variations,
20
coherence
26
such as pressure and temperature, as well as ionospheric perturbations also induce
distortions, but the effects are smaller in magnitude and more evenly distributed throughout
the interferogram than is the wet tropospheric term.
3.1.2.3 Temporal decorrelation
The main lirniting factor for repeat-pass InSAR is the temporal change of the backscatter
property of the surface between two satellite passes, the so-called temporat decorretation.
Reasons for this may be due to volume scaflering in vegetated areas, especialÏy forest,
changes in vegetation phenology, variations in sou moisture, lava flow, fieezing and
thawing, and man-made changes. Temporal decorrelation is most prevalent for water
surfaces and the lowest for desert or other arid areas with sparse vegetation. A loss of
correlation, i.e. of scene coherence, makes the gathering of phase information more difficuit
or even impossible. For instance, snow covered, agricultural and other vegetated areas
suffer from severe temporal decorrelation. For snow decorrelation it couÏd be even a few
hours depending on the change in temperature.
In many SAR images, decolTelation effects are visible even after just one day (ERS tandem
mission), after a few weeks the complete scene can be almost deconelated. Qualitatively, it
is lumwn that:
1) desert is better than dense forest;
2) dry conditions are better than wet;
3) long radar wave length is better than short;
4) urban areas and solid scatters such as houses, rocks, and corner reflectors hardly
decorrelate;
5) water decorrelates within 0.1 s which means that repeat-pass InSAR cannot be
applied over ocean surfaces;
6) agricultural activities lead to a loss of coherence for individual agricultural
fields.
An example of scene coherence is demonstrated in Figure 1.
27
3.2 The influence ofwet and dry suow on InSAR DEM’s
3.2.1 Dry condition
Interaction of radar with snow is a function of the snow properties. When snow is in a dry
condition, its backscattering is minimal at C-band (Bemier and Fortin, 1991). Hence, the
snow/ground interface is the main reflector. However, its higli refractive index creates a phase
delay which varies linearly with the snow pack water equivalent (Granberg and Vachon,
1998). In this section the theory ofthese interactions will be elaborated.
Earlier elaborations ofthe distortion ofDEM’s by dry snow (Shi et al., 1997; Strozzi et aï,
1999; Guneriussen et aï., 2001) sec it as a resuit of refraction, analogous to that observed in
the optical domain. AccordingÏy, assuming that the images are obtained when the gi-ound is
covered with homogenous dry snow, the phase term in equation 5 is given by:
tota1 = øtopo+ Ç5d
(9)
42z-=—AR+ç (10)
Where
2. is the radar wave length;
AI? the two-way difference in slant-range vector from sensor to the ground pixel, and
ç5 is the two-way propagation difference in free air and snow.
The radar beam vector will change because of the change in refraction due to difference in
dielectric properties between air and snow (Figure 10).
Now ql, the dry snow phase term in equation 10 represents the difference in the two-way
propagation with and without snow:
ds%ds(11)
2$
Line of sight
t1s= _42T[/\R —(, +AR)] (12)
Where
2: radar wave length;
O: incidence angle;
z: snow cover depth;
n: snow refraction index.
yX
Figure 10: Refraction change due to difference in dielectric properties between air and snow
Using the geornetry illustrated in figure 10, we obtain:
—4r sin2 O. —n= zcosU+
(5 I t? 2—sin e,(13)
29
In dry snow conditions the complex pennittivity is governed by the dielectric properties of ice,
snow density, and ice particle shape (Haïlikainen and Winebrenner, 1992). This pennittivity is
independent of the temperature because the permittivity of ice is & 3.17 in a range of
ftequencies ftorn 10 MHz to 1000 GHz. Thus the permittivity of dry snow (e’d$) is only a
function of density (p), which, under natural snow pack conditions usually varies over the
range 0.2 to 0.5 g crn3 (Ulaby et aÏ., 1982).
Contrary to permittivity of dry snow (u), the imaginary part (ds) is much smaller. in
which case the value ofd can be found by puffing E’ — O (Mitz1er, 1987).
Based on Mitzler (1987) the dielectric constant of dry snow is:
n2 E1+1.7p+0.7p2 (14)
Where (p,) is the snow density and () is dielectric constant (real part). For snow density in the
0-0.6 g/crn3 range, a squar foot ofequation 14 given as:
n=Jr=1+0.85p (15)
Inserting equation 15 into equation 13, gives the reÏationship between interferornetric phase
and changes in snow density and snow depth:
— 4r sin2 û — (i + 0.85p)z cos&.+
(16)z
,JZ.g5p)2_sjn2o
Since snow water equivalent given as:
$WEzrz.p (mm)
(17)
Equation 16 becomes:
30
— 4 SWE.p1 (Sj112— ï)— 0.85SWE
$WE.p1.cos1 + t (18)
J1+0.85p) —sin2O1
A simpïified forrn ofequation 18 is:
—4yr cos2&.+0.85pds = cosO1— .$WE (19)
cos2 O + ï.7p + 0.7 p2
This equation reveals that there exists a direct relation between the InSAR phase difference of
dry snow and changes in snow water equivalent SWE. In other words, the interferornetric
phase difference of dry snow is a function of the snow depth. Presence of the deeper dry snow
on the ground produces more delay of the phase during a round-trip.
The sensitivity of the algorithm was tested to sec how different parameters would control the
phase difference obtained from radar round-trip (siant range) in the dry snow. Ibis is feasible
because, the combination of the incidence angle (O) and snow density (p) may have different
influence on the InSAR phase differences.
In a first sensitivity analysis we verified several combinations of these parameters, and indeed
the incidence angle of sensors plays a very important role in phase differences of InSAR data
due to hydrographical features of the medium. As we see in equation 19, this parameter has a
crucial influence on InSAR phase.
Each sensor platform has a different incidence angle (O). To see how the actual and future
operational radar satellites wouÏd react in regard ofthe developed algorithm (equation 19); we
considered a wide range of incidence angle in the assessment. One of the most frequently used
in interferometry SAR is the ERS Tandem Mission using ERS-1 and 2 was designed to
provide the interferometric data with temporal delay of one day. The incidence angle of the
two sensors was fixed at 23° upon which many theoretical studies in the InSAR domain are
based. ENVISAT also has the sarne incidence angle.
31
Another satellite data source used frequently in interferometry SAR is the Canadian
RADARSAT- 1. This SAR instrument can shape and steer its radar beam at C-band. A wide
variety of beam widths are available to capture swaths of 45 to 500 km, with a range of 8 to
100 meters in pixel size and incidence angles of 20 to 49 degrees.
Using the rang of incidence angles available from these satellites we calculated the variation
of hSAR phase differences in relation to incidence angles changes from 20 to 50 degrees. To
facilitate the understanding of the SIIOW characteristics role on interferometric phase
difference, we also varied the dry snow density with three typical ones (0.3, 0.4 and 0.5 g cm3)
as well as the snow depth from 5 cm to 100 cm with an interval of 5 cm.
In the second assessment of the sensitivity of equation 19, we traced the variation of hiSAR
phase differences versus the Snow Water Equivalent (SWE). We postulated a snow cover with
100 cm depth and with three density conditions (0.3, 0.4 and 0.5). AIl these variations in SWE
and density were also anaÏysed at various incidence angles. We particularly emphasized the
23° incidence angle of ERS I and 2.
3.2.2 Wet condition
When the snow changes to a wet condition (presence of liquid water), die irnaginary part
shows very high sensitivity to the amount of liquid water content (L; percent by volume).
Cumming (1952) found that with increasing liquid water content of a dry snow from zero to
0.5 percent, the e” increased by more than one order of magnitude. This means that the
depth penetration of microwaves decreases significantly and the scattering cornes primariÏy
from the air-snow interface. Thus, in ideal conditions, the depth of wet snow may be
rneasured simply by subtracting the DEM produced for snow-free terrain from that
generated for terrain covered by wet snow.
The roughness of the wet snow surface lias a strong influence on the backscatter,
consequently in interferometric processing, wet snow cover is considered as a distortion
superimposed on the topography. Therefore the “snow phase” in figure 10, (y) is a
reduction in the two-way path given as:
32
4yrG÷ ——-—(ARG —AR) (20)
‘t
Where:
G+Ws is the phase difference related to the altitude ofthe ground covered by wet snow;
ARG is two-way path of snow ftee ground;
And is two-waypath ofwet snow cover.
Thus, in summary, the DEM produced when the snow cover is wet represents the ground
surface elevation, to which the depth of snow bas been added. Thus, differential InSAR can
measure both snow depth and density, the former on occasions when the snow surface is wet,
and the latter when the snow cover is dry.
3.3 Theoi-etical analysis
In this section we analyse the theoretical researcli phase where we explored the relationship of
interferornetry SAR phase difference with Snow Water Equivalent. The phase difference
sensitlvity of developed algorithm (equation 19) was verified in comparison with the variation
of existing parameters.
In the first stage the crucial influence of variation of incidence angle of the sensor on radar
siant range was recogiised.
The curves in Figure 11 show the variation of interferornetric phase difference for sensor
incidence angle (from 20 to 50 degrees). The coloured curves represent the snow cover depth
from 5 to 100 cm having the density of 0.3 g cm3.
33
-eeoC
-ae(D(D
Q.
-euoC
Dl(D(D
o-
400
350
p= 0.3 g cm3
00
EQ
ow
EEE30 35 40 45 50
Incidence angle 0°
Figure 11: Variation of interferometric phase difference versus sensor incidence angle. The
color curves represent the snow pack depth variation (from 5 to 100 cm) having a density of
0.3 gcm3
Figures 12 and 13 are representations of interferometric phase difference sensitivity for the
same range of sensor incidence angle (from 20 to 50 degrees) but for a snow cover having the
density of 0.4 and 0.5 g cm3 respectively.
p= 0.4 g cm3
100
EQ
ow-o
oCu,
25 40 45 5030 35Incidence angle 00
Figure 12: Variation of interferometric phase difference versus sensor incidence angle. The
coÏor curves represent the snow pack depth variation (from 5 to 100 cm) having a density of
0.4 g cm3
34
800
700
600
500eo
•0www-Co-
400 F
Eo-Co.w
Figure 13: Variation of interferometric phase difference versus sensor incidence angle. The
color curves represent the snow pack depth variation (from 5 to 100 cm) having a density of
0.5 gcm3
In the second test ofthe sensitivity ofthe algorithm (equation 19), the InSAR phase difference
was verified versus the changes in the quantity of snow water content.
Figures 14, 15 and 16 were generated for different snow densities of 0.3, 0.4 and 0.5 g cnï3.
The range of SWE is related to the snow cover depth which is considered to be from 5 to 100
cm. In the same figures we plotted different sensor incidence angle, with emphasis on the 23°
of ERS satellites (solid black une). Generally, most satellite’s incidence angles are located
between 20 and 50 degrees.
p= 0.5 g cm3
300 F
200 F
100F
Incidence angle O
35
400 I5o
350
p=O3gcm V’300
250 .
],_,VZz 4O
C0 5 10 15 20 25 30
Snow Water Equivalent (cm)
Figure 14: InSAR phase difference variation versus the changes in Snow Water Equivalent
(SWE) having a density of 0.3 g cm3. The plotted unes represent the sensor hicidence angle
with highuight on ERS satellites incidence angle of 230
-eeu
e
C C
wt
e
ww
e
Q.
20
Snow Water Equivalent (cm)
Figure 15: InSAR phase difference variation versus the changes in Snow Water Equivalent
(SWE) having a density of 0.4 g cm3. The plotted unes represent the sensor incidence angle
with highlight on ERS satellites incidence angle of 23°
36
800
700
p=0.5gcm3
600
500
4O0 .
Afl
200
1I
Snow Water Equivalent (cm)
Figure 16: InSAR phase difference variation versus the changes in Snow Water Equivalent
(SWE) having a density of 0.5 g cnï3. The plotted unes represent the sensor incidence angle
with highlight on ERS satellites incidence angle of 23°
3.4 Interpretation and discussion of theoretical research
In Figure 11 we demonstrate the fluctuation of phase delay based on the sensor incidence
angle within the range of 20 to 50 degrees. The different curves in Figure 11 also show the
variation of interferometric phase difference versus snow pack depth variation (5 and 100 cm).
for the first three snow thicknesses (5, 10 and 15 cm), the relationship is linear and stable,
implying that the incidence angle does flot have a significant role in phase delay in this range
of snow thicknesses. However, with increasing snow depth the changes become more
significant.
With increasing incidence angle greater than 30°, the phase delay increases exponentially. We
observe that the phase wrapping at 49 degrees incidence angle occurs at approximately 100 cm
snow which corresponds to a change in snow water equivalent of 300 mm (SWE=pz).
Previous research concluded that C-hand phase-wrap occurs every 32.7 mm of water
equivalent at vertical incidence angle (Granberg and Vachon, 199$) and 49.2 mm at 23°
incidence angle (Guneriussen et al., 2001).
37
The question now arises as to the role of snow density in this algorithm. To answer this we
exarnined the situation ofa snow pack having an average density of 0.4 and 0.5 g crn3. Figure
12 and Figure 13 show the variation of snow density variation on interferometric phase delay.
The influence of a snow cover with 0.4 g cm3 density is apparent in that the phase wrapping
starts at approximately 43 degrees incidence angle for 100 cm of snow and continues until 50
degrees foi- snow depth of 65 cm. These are equal to 40 to 26 cm of SWE respectiveÏy.
Fïgure 13 represents the snow density of 0.5 g crn3, in it we observe more sensitivity between
phase delay and snow density. At 3$ degrees of incidence angle, the phase wrapping occurs
for 100 cm of snow which equals a SWE of 50 cm. Ail incidence angles over 38 degrees
creates a phase wrapping at lesser snow depths. We observe that the phase wrapping at 50
degrees incidence angle appears for approximately 50 cm of snow which represents 25 cm of
SWE.
In both cases an incidence angle less than 30 degree does flot show a significant effect on
phase delay but at larger incidence angle it plays an exponential role on InSAR phase
difference.
In order to highlight the sensitivity of the algorithm to the snow water equivalent parameter,
we traced the impact of SWE on changes to the interferornetric phase delay. Figure 14 is a
presentation of InSAR phase difference due to changes in $WE for different sensor incidence
angles. This figure reveals a linear reiationship between them. A higher SWE generates more
phase delay and this will be more important with increasing the incidence angle. The ERS
incidence angle of 23 degree which is plotted as a solid black une expresses the weakness of
this sensor foi- estimating the dry snow water equivalent.
The maximum of 30 cm SWE is set up based on the average snow density of 0.3 g cm3. If we
want to sec how the InSAR phase delay changes versus the SIIOW water content with a higher
snow density, Figure 15 and Figure 16 are graphies for a snow pack with an average density
of 0.4 and 0.5 g cm3 respectively.
3$
The phase delay is more pronounced by a snow pack with higher density Figure 15. The
increase in the incidence angle is stili an important help for SWE to measure the phase
difference. With a snow density 0.5 g cm3 the phase difference is even more sensitive to SWE
(Figure 16). Even at the 23 degree incidence angle of ERS images it becomes quite
remarkable. For example: 50 cm SWE having an average density of 0.5 g cm3 creates
approximately a 100 degree phase difference.
3.5 Conclusion of theoretïcal researcli
The algorithrn that we have developed shows a direct relationship between interferornetric
phase difference and snow water equivalent. This relationship becomes more significant when
the sensor incidence ange! increases. The differences between the theoretical resuits (equation
19) and the pubÏished experirnenta! work clearly indicate the need for further detailed field
work in conjunction with stable satellite data.
Snow density plays a primary roTe on InSAR phase delay. An increase in snow density
produces significant increases in phase delay.
There is a linear relation between SWE and InSAR phase delay; their correlation is especially
evident at higher incidence angles. This also implies that the ERS data may not be able to
distinguish a phase difference for a snow pack of 100 cm but RADARSAT-1 with a higher
incidence angle would be more reliable to estimate the dry snow water equivalent.
The conclusion of this theoretical research reveals that the first hypothesis of the research is
valid as has been also confirmed by Granberg and Vachon, 199$ and Guneriussen et aT, 2001.
The RADARSAT-1 twined with RADARSAT-2 will provide very sophisticated
interferornetric data to estirnate the dry snow water equivalent because of the large choice in
incidence angles, offering high resolution imagery and interferometric coherence due to their
short temporal delay between image acquisitions.
39
4 METHODOLOGY
4.1 Introduction
This chapter describes the study areas and datasets as well as the n;ethodology for
experimental research. Two study areas were chosen, one at $chefferville, Canada, and the
second in Colorado, USA. The general physiographies of the study areas are presented and the
rernote sensing and ground truth data are described.
4.2 Schefferville study area and data
As noted in the introduction, the region near Schefferville (55°N, 66°W) bas seen much snow
and SAR research. It was therefore a natural choice for one of the two field experiments. The
scene selected (Figure 17) includes lakes, exposed rocks, tundra, and boreal forests and is in
the zone of discontinuous permafrost (Granberg et al., 1994; Granberg and Vachon, 199$).
A band of alpine tundra extends diagonaÏly from upper lefi to the lower right of the scene. The
altitude varies from 475 m to $40 m above the sea level. The maximum altitude associated
with Mont frony is in the west while the North-East is characterised by a moderate topography
with several vast lakes. The Howeils River flows through the $outh-West section at an altitude
of 500 m. In the center of the area, the original topography is partly remodelled by mine pits
and mine waste dumps.
40
Schefferville Topographic Map66 fl. 66 5Q7’
z
1• • —j — -.
‘I
.-... - . . .•
S-.
‘e-..
-- t- - — -
C ‘- .. I .‘V-
—-•‘-•‘- :- k-.:,;--•:-..,
2 3 4
4 6Km
liTu’‘nz
C)’oz
. -
- r •.—•.
\ ‘-‘i \Y
665OV67 5V1 67W 66
2
figure 17: Schefferville study area: alpine tundra extends diagonally from upper left to the
lower rïght ofthe scene (red circle), the Howeils River flows through the South-West section
and North-East ofthe site contains several lakes in a moderate topography
The site is located in a transition zone between tundra and boreal forest. The forest is mostly
black spruce on a background of lichen where the ground is dry, or moss where it is humid.
Ridge crests are mostly treeless and its vegetation is strongly influenced by snow abrasion
(Granberg et aï., 1994). In the forest-tundra environment, snow accumulation patterns are
controlled by strong winds that accompany the snow storms. Zones of locally greater wind
specd, notably ridge crests, experience loss of snow due to wind erosion and remain nearly
devoid of snow throughout the winter (Thom and Granberg, 1970; Granberg, 1973). In
contrast, the snow is transported along the surface, often as migrating snow dunes, and
41
accurnulates in areas downwind of the ridge crests which experience snow accumulations
sornetirnes exceeding 4 m of dense snow grow in the downwind direction (Granberg, 1973;
Granberg and Vachon, 1998).
4.3 Data
For the Schefferville study area two DEM’s were used. One is the Canadian Digital Elevation
Model (CDEM), produced by the Canadian Topographie Survey using photogrammetric
methods with 20 m resolution and a vertical accuracy of 20 m.
The second was produced by the Shuttie Radar Topography Mission (SRTM) using C-band
radar in a cross-track configuration with the second receiver anteima rnounted at the end of a
60 m retractable boom (Hensley et al., 2000). The cartographie products derived from the
SRTM data were sampled over a grid of 1 arc-second by 1 arc-second (approximately 30 m by
30 m), with linear vertical absolute height enor of less than 16 m and linear vertical relative
height enor of less than 10 m (Rodriguez et aÏ., 2006). The DEM used here was re-sampled to
match the higher resolution of the CDEM from SRTM DEM data available free from the
NASA SRTM web site at a resolution of 90 m.
4.4 Colorado study areas and data
A full description ofthe CLPX experiment is given at the website:
http://www.nohrsc.nws. gov[cline/
Ail data sets from the CLPX are found at:
http://nsidc.org/datalclpx/#data
Here only a cursory description will be given.
The Colorado study area is Iocated in the complex terrain of northem Colorado and southem
Wyoming (104°- 108.5° W, 38.5°- 42° N) where elevation gradients and aspect variations
provide a wide range of conditions over relatively short distances. Study sites were selected
that range from Ïow-relief (flat topography), unforested areas with shallow snow covers, to
high-relief (complex topography) and densely forested areas with deep snow covers.
42
The site contains 5 sets ofnested study areas in a large 3.5° x 4.5° region including nine 1x1
km intensive Study Areas (ISA) within three 25x25 km Meso-celi Study Areas (MSA), in an
approximately 215x170 km region (Figure 18). The ISA represent different physiographic
regions. Activities at these sites included intensive spatial sampling of snow properties
coincident with airborne observations. Airbome observations provided both passive ami active
microwave coverage of each of the three MSA.
Figure 1$: $chematic diagram ofthe nested study areas for the Cold Land Processes field
Experiment (modified after NASA web site)
I Locaf-Scale
Observation Site (CSOS)(Iha)
I
9 Intensive StudyAreas (ISA)
f1.krn x i-km)
3 Meso-celI Study
Ateas (MSA)
(25-km x 25-km)
I h—; ‘I
I ‘bi \ I SmaIf-Regional
I StudyArea
/ \ (t5°x2.5°)
I Large-RgionaIStudy Area(3,50 x 4.5°)
43
4.4.1 Characteristics of Meso-ceil Study Areas (MSA)
Nested within the small-regional study area are three 25 km x 25 km Meso-ceil Study Areas
(MSA) named: 1-North Park, 2-Rabbit Ears and 3-fraser (Figure 19). The coordinates of the
three MSA are listed in Table 1.
Figure 19: The three 25 km x 25 km study areas, and the nested Intensive $tudy Areas are
shown with red boundaries (modified after NASA web site)
The primary objective in selecting these areas was that each individual area should represent a
distinct cold-region physïographic regime, while together the three areas should represent as
broad a range of regimes as possible. Topography, forest cover, and snow characteristics were
the three major criteria for selecting the areas (Table 2). The MSA’s include four of the major
global snow cover classes of Sturrn et al. (1995), which together comprise 88% of the
•
___
i• —
—1 — ..
r- •
44
seasonally snow covered areas of the Earth. The remaining 12% is the maritime snow class,
which is flot found in the experiment study region.
Table 1 t Coordinates for corners of each MSA, given first in latitude/longitude, then in UTM.
The WGS84 datum is used for ail coordinates
MSA f UpperLeft Lower Left Upper Right Lower Rïght
400 48’ 00” N 40° 34’ 30” N
KDD MM SS) -106° 26’ 00” W -106° 26’ 00” W h106° 08’ 30’ W h106° 08’ 30” W
North Park 378,978 E 378,574 E 68403,3$7 E
(UTM) 4,517,548N 4,492,493N 4,517,170N 4,492,088N
40° 34’ 00” N 40° 20’ 30” N 40° 34’ 00” N 40° 20’ 30” N
(DD MM SS) -106° 50’ 00” W -106° 50’ 00” W H106° 32’ 30” W H106° 32’ 30” W
Rabbit Ears 344,736 E 344,170 E 369,46$ E 369,064 E
(UTM) 4,492,250 N 4,467,141 N 4,491,2 19 N 4,466,737 N
Fraser 40° 00’ 30” N 39° 47’ 00” N 40° 00’ 30” N 7’00”N
(DD MM SS) -106° 01’ 30” W -106° 01’ 30” W -105° 44’ 00” W i05° 44’ 00” W
Fraser 412,437 E 412,215 E 437,461 E 437,184 E
(UTM) 4,429,124 N 4,404,072 N 4,428,902 N 4,403,878 N
Table 2t Characteristics ofthe Meso-ceil study areas, the general physiographic regimes they
are considered to represent, and the approximate percentage of global, seasonally snow
covered areas having these characteristics
CharacteristicsRegime Coverage
North Low-relief grassiand, with shrubs in wetter areas Prairie 53
Park and few trees. Shallow, windswept snow covers Tundra
and extensive frozen sous.
Rabbit Moderate-relief area with mixed Woodlands, 10
Ears coniferous/deciduous forests and large forest gaps. Low
Moderate to deep snow covers. Mountains/Foothills,Boreal Forest
rHigh-relief area with predominantly coniferous, Boreal Forest, 25
subalpine forests at middle to upper elevations and Intermountain
alpine tundra at highest elevations. Lower Regions,
elevations (broad valley bottom) are unforested, High Mountains, I
irrigated grazing land Moderate to deep snow Alpine
covers, increasing with elevation.
45
4.4.1.1 North Park MSA
North Park is a broad, high-elevation, parkland approximately 40 km in diameter. The MSA is
centrally located in North Park with a mean elevation of 2499 m (Figure 20). Most of the
MSA has low local relief (50 m) with a maximum of 312 m, due largely to the presence of Jow
foothiils in the south-eastern part ofthe MSA.
Figure 20: The 25 km x 25 km North Park Study Area boundaries are shown in red. Nested
within the area are three 1 km x 1 km Intensive $tudy Areas: Potier Creek ISA, Illinois River
ISA, and Michigan River ISA (modified afier NASA web site)
1(4,2” W lô4Ic. 104’IZW W 13 S4 I AGW W
?% ‘.
l ‘ b
ç
______________
-
/ j j _‘__‘1
.0 V,s_ ‘‘
r- . \ i-
z -.. / - ,,
n3
I . —‘ 1
t - b
IlhnojsRÏverlSA I,
si r/ I’.“Ç7 PollerCreek ISA / 4 -/
___
-— T
-
“.‘
-
°‘
1 ‘2ItW I * -
I_I_I fr— -O” .4i
46
North Park has a very low snow accumulation due to the precipitation “shadow” caused by the
sulTounding mountains.
The area forms the headwaters of the North Platte River. Several srnall meandering rivers
drain the sunounding mountains and flow through the flat topography of North Park, where
they join the North Platte River just north of Walden. Consequently, parts of the MSA are
reÏatively wet.
The MSA has littie forest cover; most of the vegetation is sage-grassiand, with willow along
riparian areas. Snow packs in this area tend to be shallow and windblown, and are typical of
prairie and arctic- and alpine-tundra snow covers (53% of the global seasonal snow cover
(Sturm et aï., 1995). The MSA is crossed by two main highways and several secondary roads
that rernain open during winter, providing access throughout the area. The town of Walden,
located near the centre of the MSA, is the principal source for services in the area.
4.4.1.2 Rabbit Ears MSA
The mean elevation ofthe Rabbit Ears MSA is 2725 m. The topography consists prirnarily of
low to moderately rolling huis extending north-south throughout the middle of the MSA.
These are flanked by much flatter valleys at lower elevations along the eastern and western
edges ofthe MSA (figure 21).
The land cover of the Rabbit Ears MSA is the most diverse of the three. The forest cover
consists of rnixed stands of spruce-fir, lodgepole pine, aspen, and scrub oak forest interspersed
with broad glades and rneadows.
Tue area receives heavy snowfall during the winter and spring, and often develops the deepest
snow packs in Colorado at the higher elevations near Buffalo Pass. Snow packs in this area
have depths and other characteristics representative of regions where orographically induced
precipitation plays a dominant role, i.e. about 10% ofthe global seasonal snow cover (Sturrn et
aÏ., 1995).
47
‘:
\-.
I—-. ‘ Buffalo Pass ISA
Spnng Creek ISAk
S
.Çï. ‘I
-7ç’’.’
5._J .1
2.-
S .5S S
(jçWaInISA
Figure 21: The 25 km x 25 km Rabbit Ears Study Area boundaries are shown in red. Nested
within the area are three I km x 1 km Intensive Study Areas: Buffalo Pass ISA, Spring Crcek
ISA, and WaÏton Creek ISA (modïfied afler NASA web site)
The MSA is crossed east-west by Highway 40 over Rabbit Ears Pass. Other primary and
secondary roadways provide access throughout the southem haïf of the MSA. The northem
haif is less accessible, and requires snow mobiles for efficient travel. The town of Steamboat
Springs, along the western edge ofthe MSA, is the principal source for services in the area.
ç‘ j
‘
S--.
j : Rabbit Ears MSAS.- -
S
- — .- ‘ -‘ç 1
48
4.4.1.3 Fraser MSA
The fraser MSA is a topographically complex area, with a large north-south topographie
gradient (Figure 22). Its mean elevation is 3066 m, with a maximum relief of nearly 1400 m
and a maximum elevation of 3962 m.
Figure 22: The 25 km x 25 km fraser Study Area boundaries are shown in red. Nested within
the area are three 1 km x 1 km Intensive Study Areas: St. Louis Creek ISA, Fool Creek ISA,
and Alpine ISA (modified afler NASA web site)
-
St Louis Cœek ISA
•••/
cœeISA.
:
-
L,
-‘i .j:L; :
J .‘)‘(z’,- -
49
The Continental Divide winds through the high-elevation alpine areas of the Front Range,
along the southem margin of the MSA. The Fraser River and its tributaries (Vasquez Creek,
St. Louis Creek, and Crooked Creek) flow northward from the Divide through a finely
dissected series of ridges and glacial valleys throughout the southem haif of the MSA. Here,
the Fraser River valley broadens into a wide glacial outwash plain. Vegetation at lower
elevations to the north consists of inigated grasslands, sparse and mixed forests.
The highest elevation areas in the south are predominantly alpine tundra or bare rock. The
mountain areas in between are dominated by dense coniferous forest, generally with spruce-fir
forests on the wetter north-facing siopes and lodgepole pine on the drier south-facing siopes.
The Fraser MSA is typically cooler than both the Rabbit Ears and North Park MSAs, due
rnainly to its higher elevation. Snow packs in this area typicaÏly display only minimal
modification by wind and considerable depth hoar developrnent (25% of the global seasonal
snow cover (Sturrn et aï., 1995). The eastem two4hirds ofthe MSA is accessible via Highway
40, which crosses Berthoud Pass over the Continental Divide in the southeast corner of the
MSA, and transects the MSA northward, following the Fraser River. Several secondary roads
in this part of the MSA also remain open during winter. The western third of the MSA is less
accessible. The towns of Fraser and Winter Park are the principal sources for services in the
area.
4.4.2 Characteristics of Intensive Study Areas (ISA)
The ISA (Table 3) were selected so as to represent different physiographic characteristics.
Factors potentially affectïng microwave response were also considered, such as the relative
wetness or dryness of the site, and the type, density and distribution of forest cover.
Table 3: Location and characteristics ofISA’s. UTM coordinates are given for upper left (UL)
and lower right (LR) corners
Site ULUIM LRUTM Characteristics
NP Creek [,653 E 3 88,653 E Dry grassiand with isolated shallow ponds, on a flat aspect
(North Park) 4,503,210 N 4,502,2 10 N with low relief; shallow windswept snow cover.
Tiffinois River [5394,504 E Wet grassiand with widespread meandering riparian areas, on
(North Park) 4,506,210 N 4,505,210 N a flat aspect with low reÏief shallow windswept snow cover.
50
[NM higa 3 99,494 E 400,494 E Dry sage-grassiand (larger vegetation) with moderate relief:
River 4,500,540 N 4,499,540 N shallow windswept snow cover.
(North Park)
RB Buffalo Pass [57,076 E 358,076 E Dense coniferous forest interspersed with open meadows; low
(Rabbit Ears) 4,488,891 N 4,487,891 N rolling topography with deep snow packs.
[jSpring Creek 350,558 E 351,558 E Moderate density deciduous foi-est (aspen); moderate
(Rabbit Ears) 4,488,451 N 4,487,45 1 N topography on west-facing siope, with moderate snow packs.
Walton Creek 359,645 E 360,645 E Broad rneadow interspersed with srnall, dense stands of
(Rabbit Ears) 4,474,079 N 4,473,079 N cornferous forest; low rolling topography with deep snow
packs.
IFS F— 425,253 E 426,253 E Moderate-density coniferotis (lodgepole pine) forest, on a flat
Creek 4,420,309 N 4,4 19,309 N aspect with low relief.
I (Fraser)
fF Fool Creek 425,488 E 426,488 E Moderately high-density coniferous (spruce-fir) forest, on wet
(fraser) 4,415,259 N 4,414,259 N north-facïng siope.
FA Alpine 425,903 E 426,903 E Alpine tundra, with some subalpine coniferous forest;
(Fraser) 4,411,844 N 4,410,844 N generally north-facing with moderate relief.
4.4.3 AIRSAR data
The Airborne Syrithetic Aperture Radar (AIRSAR) system was operated in cross-track
interferornetry mode (XII, or TOPSAR) in both C and L band sirnultaneously. It was ftown
on a DC-$ aircraft at 8 km altitude over the three MSA. Details of the operationai
chai-acteristics and processing rnethods are given by:
(http://nsidc.org/dataldocs/daac/nsidc0l53clpxairsar.gd.htm1).
4.4.3.1 Data and imagery file
Basically, the integrated TOPSAR data product consists of two different data types as shown
in Table 4. The ts####_c.derni2 file represents the DEM calculated from interferometry
process using C-band and the ts#### c.VVi2 file represents the backscatter image ofthe scene
in C-band as well. Ail TOPSAR products begin with the letters ‘ts’ followed by the scene
identifier (ts####). This identifier changes when the site, flight direction or data acquisition
date change.
The AIRSAR ftight direction is identified by the angle between ftight direction and north
direction (second coiumn, e.g. North Park 270-1). If ail three ISA’s of a MSA were not
covered by first AIRSAR light une, a second light une was added to the data set. In order b
51
recognize the different light unes of a data sceie acquisition the flight direction angle is
followed by a number 1 or 2 and a new identifier is associated with it.
Nine data sets were used in this research (Table 3).
Table 4: Different kind of imagery produced by AIRSAR data
Data Flight direction Description
i7c.demi2 JNorth Park 270-1 - C-hand DEM
ts1627c.vvi2 jNorth Park 270-l - C-band VV, backscatter image
629 c.demi2 North Park 270-2 - C-hand DEM
9c.vvi2 Park 270-2 - C-hand VV, backscatter image
The DEM data (ts#### c.derni2) are in INTEGER*2 format, and represent the elevation ofthe
terrain above a spherical approximation to the WGS-$4 ellipsoid.
Figure 23 is a sample of backscatter image and Figure 24 is a sample of a DEM image of
ATRSAR data set. Both of them are related to Rabbit Ears MSA with 41 km x 12 km
dimensions.
Figure 23: Backscatter image ofRabbit Ears MSA acquired by AIRSAR instrument on
February 21st, 2002 (tsl6O6c.vvi2)
52
During each IOP the three MSA’s were captured by the AIRSAR instrument separately. The
data set included nine images which were used in this researci (Table 5). One of the
disadvantages of the data base is the variation in flight length because in many cases they do
flot cover the saine length over the sites.
TabJe 5: ARSAR flight unes and length
MSATOPSAR Date
I Flight Length
ID Number (km)
I ts1629 02/21/2002 40
North Park tsl 591 03/25/2002 40
tsl 544 09/17/2002 30
[ tsl 570 02/23/2002 50
• Rabbit Ears t Tsl 589 03/25/2002 t_____ 30
ts1543 09/17/2002 [ 30
ts1451 02/19/2002 50
Fraser Ts1586 03/25/2002 j 30
Ts1532 09/17/2002 40
4.4.4 Ground data
The collection of in situ measurements of snow characteristics is a major and critical
component of the field experiment. The objective of the ground data was to provide high
quality; spatially intensive data sets for calibration and validation of the remote sensing
Figure 24: DEM image ofRabbit Ears MSA acquired by AIRSAR instrument on February21st, 2002 (tsl6O6c.demi2)
53
images and extracted resuits. The ground data collection component was primarily focused on
quantifying the spatial mean, variance, and distribution of snow at scaies up to 1 km2 (i.e. the
Intensive Study Areas).
A major goal of the ground data collection was to ensure that ground measurements are
representative of the snow conditions at the time of the microwave rernote sensing overfliglits.
Risks of significant changes in ground conditions increases with each passing day, therefore
the ground data collection strategy was to complete ail measurernents in a given MSA within a
two-day period. Within a given two-day period, risks of significant change are flot equal for
different variables. First, snow surface wetness and roughness at the tirne of the over light can
strongly influence the remote sensing signal, and can also be very dynamic. Second, many
internai snow pack characteristics are much less dynamic than surface characteristics
(depending on conditions). Third, recent changes in snow pack characteristics (e.g. new snow
accumulation) can ofien 5e identified easily from snow pit information, but are more difficult
to identify from surface observations. Therefore, the ground data collection strategy also
prioritized snow measurernents to 5e made on day 1 (the target day for airborne data collection
in the study area) and the snow pits on day 2. Ail three ISAs within a given MSA were
sampled simultaneously within a two-day period.
4.4.4.1 Snow depth and surface wetness
Sampiing on day 1 (same day as InSAR data collection) consisted of intensive spatial
measurement of snow depth and surface wetness. The ISAs were stratified using a 10 x 10
(100 m interval) grid. X and Y coordinates of a starting point within each 100 m grid ceil were
selected at random. A transect interval of 5, 10, 15, 20, or 25 m was selected at random for
each grid ceil. Day 1 yielded 500 randorniy seiected sample points distributed throughout the
ISA.
Two orthogonal transects of two points each were defined using the starting point and transect
interval. The direction of each transect was determined by the location of the starting point
within the grid ceIl. Transects were oriented to ensure that the transect remained within the
grid ccli. A stratified random sampling framework was estabiished for the collection of ail
snow samples in the ISA (Figure 25).
54
4.4.4.2 Snow pits
Sampling on day 2 consisted 0f SflOW density, snow depth and snow wetness. The ISAs were
stratified using a 4 x 4 (250 m interval) grid. X and Y coordinates of single sample points
within each 250 m grid were selected at random. Day 2 yielded 16 sample points in each ISA
(figure 26), providing a reasonably large sample of the spatial variations in snow density.
This information was used with the snow depth measurements to determine the distribution of
Snow Water Equivalent (SWE) throughout each ISA.
Details ofthe ground data collection for each ISA are given in Table 6.
E
o
D
0 100 200 300 400 500 600 700 800 900 1000
Distance (m)
Figure 25: Map showing snow sample locations for each ISA. Red dïamonds are first day
sampling (snow depth, snow wetness, and snow surface roughness). The blue squares are
sampling on day 2 (snow pits, see next figure) and green diamonds are some special sampling
for period of 2003
1000
900
ISA Sample Locations
Dut, ———-—---———
—-—————-———-———---——-—- * 9 — 0*0
D
700--------————-— O
Oc., *0
600
500: °
<
-. D
400D
k
g k r300 r -
200----—--—-——-
---—--———--—— -—-—------——- ——
55
Typica ISA
..
tc ChaIe
j /
sO3•
O7
O3•
‘9”
Figure 26: Locations of CLPX $now Pits within each Intensive Study Area
Table 6: Ground data collection date for each MSA during IOP1 and 10P2
Intensive MSASnow depth Snow pit
Observation Period Collection date Collection date
North Park 02/21/2002 02/21/2002
IOP1 Rabbit Ears 02/23/2002 02/24/2002
fraser 02/1 9&2012002 02/20/2002
North Park 03/27/2002 03/27&28/2002
10P2 Rabbit Ears 03/29/2002 03/30/2002
Fraser 03/25&2612002 03/26/2002
56
4.5 Method of analysis
4.5.1 SRTM InSAR data
During the period (11-22 february, 2000) the snow at Schefferville was dry, enabling
assessment of the effect of dry snow. Differences in altitude between the SRTM DEM and
CDEM are visually inspected for sigris of a snow-related signal. The difference was draped
onto the CDEM to facilitate the analysis oftopographic influences.
4.5.2 CLPX InSAR data
The InSAR processing ernployed pairs of ATRSAR images to generate the DEM. This process
was appÏied on three image series: firstly on the images captured in the first Intensive
Observation Period (Top 1, February 2002) when the snow was in a dry condition; secondly in
10P2 (March 2002) where the snow was starting to meit and third was in the Bare-Ground
period (BG2, fail 2002) when the ground was snow free.
The DEM from B G2 served as reference. Subtracting it from the DEM’ s with snow cover
would Ïeave the SWE and the snow depth as residual in respectiveiy dry and wet snow. The
analysis, as shown in the second column of Figure 27 and invoïves the following five steps.
1. DEM altitude processing, collection, and georeferencing;
2. Misfit assessment and reduction;
3. Analysis of snow spatial variability in the ISA’;
4. Extraction of ISA sub-scenes, $WE, and snow depth for cadi MSA;
5. Tmporting ail data into a Geographical Information System (GIS) for
analysis.
NA
IRSA
RD
ATA
Col
orad
o
InSA
R
Corr
ecte
dD
EM
FEB
—M
AR
-SE
P
Fig
ure
27:
Met
hodo
logy
fIo
wC
hart
/SR
TMda
taSc
heff
ervi
l le
InSA
RD
EM
s)
Dat
a DIn
SAR
InSA
RM
ufti
-Pat
h
Pro
cess
ing
Expen
men
t
Cro
ssIm
ages
Roa
dP
rofi
le
Com
para
ison
Ext
ract
ion
Alt
itud
eM
odel
Cor
rect
ion
Gen
erat
ion
Geo
rnet
ric
Mod
el
Cor
rect
ion
Val
idat
ion
__
__N
DIf
lSA
R fMsw
j
HF
F]
L
Ext
ract
ion
of
road
tran
sver
sal
prof
iles
InSA
RC
orre
cted
DE
M
j
.1In
SA
Ran
dG
round
Dat
a
inte
gra
tion
inG
IS
fSta
tist
iad
5$
4.5.3 Elevation correction
The AIRSAR DEM’s are in an NTEGER*2 format, and represent the elevation of the terrain
above a spherical approximation to the WGS-$4 ellipsoid. To translate the 1NTEGER*2
values supplied in the file to elevations in meters, one has to apply the following calculation:
Hs = (elevation increment) * DN + (elevation offset)
Where DN is the INTEGER*2 number from the file, elevation increment is the elevation
found in field 7 ofthe DEM Ï;eader, and elevation offset is the elevation supplied in field 8 of
the DEM header.
ENVI software was used to process and analyse the AIRSAR images. In the first step the VVI
(Amplitude) and DEM data were imported to the ENVI environment. The new header did flot
contain the information conceming the translation of the TNTEGER*2 values to elevations in
meters, which are supplied in the ts####c.demi2 file. This infonnation was extracted
separately.
4.5.4 Interactive image georefereucing
Another important process is georeferencing the DEM’s. Four sets of data have to be
compared including IOP1, IOP2 and BG2 interferometric data as weIl as the ground data.
From each MSA, we needed to extract the three ISA’ s using only their upper left and lower
right coordinates. This meant that the interferometric data could be precisely georeferenced for
ISA extraction which was compared to the ground data rnaps.
To georeference the nine images (MSAs) we required another georeferenced image or map in
UTM coordinates as a base. The base map and air photos are available on the internet in UTM
system (www.tenaserver.com). The interactive rnap was available from 1/1000000 scaÏe to
1/13 000 scale (from 1952 to 1987). The web site also offered the possibility to change
between the map and the air photo of the scene. The air photos are in mosaic form with 1m
pixels.
59
The air photos are more recent (1999) than the topographic maps (1956) and are similar in
appearance to radar amplitude images. However due to the differences in backscafter of
objects on the amplitude images of different seasons, some features could flot be identified in
the air photos. An example of the map and its related air photo for south of Walton Creek ISA
and its nearby road is given in figures 28 and 29.
Figure 28: Topographic map ofWalton Creek area in 1/13 000 scale dated 1956 (modified
after www.terraserver.com)
After plotting a series of points on the amplitude image to corresponding points on the map or
air photo, the image was re-sampled using the nearest neighbourhood approach.
The amplitude image was used for geo-referencing because this was the only image presenting
backscatter of site features which let us identifSi the geometry of terrain and assign the
coordinates to the Ground Control Points (GCP’s). During interferometric processing ail
extracted images (Amplitude, Coherence and Interferogram) are co-referenced. This means
60
any pixel on the DEM represent the altitude of the corresponding pixel on the amplitude
image.
Figure 29: Aerial photography ofWalton Creek area dated 1999 (modified after
www.terraserver.com)
4.5.5 Misfit between TOPSAR DEM’s
At the analysis stage it became apparent that the fit between the different TOPSAR DEM’s
was less than excellent. This, of course, greatly complicates the extraction of snow-related
signais. Some ofthe error is attributable to the stability ofthe radar pÏatforrn, but sorne ofit, as
indicated by Imel (2002) is inherent in the TOPSAR system.
4.5.5.1 Theory of multi-path contribution
When TOPSAR data are acquired, the phase difference between the top and bottom antennas
is used to calculate the elevation angle of the scattering centre in each pixel. The baseline at
each point along the platform’s path is reconstnicted, given the record of the piatform attitude
and knowÏedge of the antenna phase centre positions in the platfonri-fixed coordinate system.
•%•4j.’
•‘...
. .“ - -
-
61
Irnel (2002) showed that the multi-path eiror varies from close to zero to as rnuch as +3 ni.
This, of course, creates problems in snow depth and water equivalent detennination.
The TOPSAR interferometric phase is due to the difference in the paths between the pixel
scattering centre and each antenna phase centre:
(D=4w.AR/2 (21)
Where:
A]? is the path difference,
il is the radar wave length, and
‘D is the interferometric phase.
In the absence of contamination by rnulti-path transmission, this path difference is:
A]? = n.B (22)
Where (n) is a unit vector in the look direction from the radar to the pixel; (B) is the vector
from one antenna phase centre to the other (the “baseline vector”).
Now, if a multi-path signal is present, say from signals retuming from the scattering scene
which bounce off of a wing or an engine cowiing before aniving at the receiver antenna, then
some fraction ofthe signal will be present with an interferometric phase conesponding to:
AR’=nB’+L (23)
Where A]?’ is the path length travelled by the multi-path signal, (B’) is the baseline between the
transmit anteirna and the source of the multi-path (wing or engine) and (L) is the distance
between the muÏti-path source and the receiver antenna.
To compute the phase actually measured by the system, one would have to sum over ah such
possible paths, scaling the contributions by the relative scattering cross sections of each sottrce
of multi-path. This is a difficuit problem, and was not atternpted during the lights.
Nevertheless, as a resuit of having a signal contribution from a baseline other than the
62
expected one, there will be height variations with range that do not con-espond to scattering
related to topography. Imel (2002) showed that the multi-path enor varies from close to zero
to as rnuch as +3 rn. However, it seemed important to see how this eiTor contributes in our
study area which had totally different conditions.
4.5.5.2 Misfit analysis and common referencïng of DEM data
With this objective, two September DEM’s of Rabbit Ears MSA (Walton ISA) which were
acquired on the same day but in perpendicular light directions (from the 3G2 period) were
compared. The altitudes of 1294 pixels were extracted across the images in a continuous
profile from the two crossed images. By tracing two profiles extracted from two DEM’s the
variation in the height differences could be assessed.
If the differences in height were constant, it was considered an offset and removed from one of
the images to remove the rnulti-path error. However, the multi-path contribution varied along
the track. In order to define the magnitude of the multi-path contribution a two-step analysis
was performed. In the first step, the feasibility of the method was assessed using the
Septeinber data for the Walton Creek ISA.
Due to variations in height differences, a specific model was needed for each data set to ensure
that ail pixels of a given ISA resulted in the nearest values to its twin DEM to aÏÏow
comparison. To this end a pair of September images were used, in which the altitudes would
be identical unless the multi-path error is present.
A profile of altitudes was extracted across each of the two crossed images and compared in
order to calculate their conelation. A conelation model was created and then the model was
inverted and applied to the first image in order to decrease the enor due to the DEM misfit. As
validation, another 200 altitude points were extracted and their correlation was assessed. The
second step consisted of evaluating the rnethodology on the pair of images from two different
IOP’s, ofwhich one represents the bare ground.
b provide common reference for detailed assessrnent of snow cover effects ploughed roads
were ernployed, assuming that the pixels would have the same altitude regardless of season.
63
Thus, the altitudes of 150 pixels along a road near the Walton ISA were extracted from
DEM’s for February and September to reference both to the road elevations. A similar strategy
was used to analyse snow cover effects along profiles crossing ploughed roads.
The rnisfit between the TOP$AR DEM’s limits the choice of experimental ISA’s because the
given ISA has to be near to the road so that a common reference can be obtained. The best
suited ISA was Walton Creek located in Rabbit Ears M$A situated approxirnately one
kilometre from the road.
4.5.6 Snow spatial variability
To assess the capacity of JiiSAR data to detect snow spatial variability, profiles crossing a
ploughed road were extracted from each DEM for the three seasons (IOP 1, 10P2 and BG2)
using the road as common reference, or zero point. The point elevations for some distance on
both sides of the zero point could then reasonably be assumed to be related to snow cover
effects. Accordingly, the wet snow profile (March) should appear above the Fall profile, and
the dry snow profile should be below the Fall profile. The transverse profiles included 30
pixels of 5m each, (75 m each side of the road) with pixels number 15 and 16 at the zero point.
4.5.7 SWE and snow depth estimation
After georeferencing, sub-images for the ISA were extracted in order to explore the snow
signals using maps of differences between the different DEM’s. To facilitate analysis both
SAR and ground data were integrated into a Geographical Information System.
4.5.8 Data integration
The snow measurement data for all ISA were stored in the same file ready for mathernatical
processing. These analyses consisted of calculating the SWE of each field point where snow
depth had been measured. Based on equation 17 we need to have the snow density of each
point to calculate its $WE.
In order to assign a snow density to each of the 500 snow depth measurements, “allocation by
distance” was used. To this end 16 polygons were calculated, centered on the snow pit
64
locations, identifying depth sites in closest proxirnity to sites of density measurement. In the
next step the generated polygons were converted to raster formats having the same resolution
of radar images (5 rn) and a density value assigned each pixel, enabling calculation of the
SWE for each pixel.
The rernote sensing and ground data were then integrated into the Geographic Information
System. This was a direct process because ail the data sets were built with the sanie coordinate
system (UTM), all remote sensing data could be overlayed on the colTesponding ground data
map, enabling comparison and statistical analysis.
65
:
5 RESULTS AND DISCUSSION
5.1 Resuits at Schefferville
The difference between the SRTM and the CDEM is shown in (Figure 30). The grey level is
proportional to the difference. The medium grey toile on the larger lakes represents zero
difference. Darker tones represent parts where the SRTM DEM is below the CDEM and
brighter toiles parts where the reverse is true. In theory, the cold SflOW covered terrain should
be rather dark at this time of the year. However, the effect of the sno cover was cornpensated
for in the DEM processing. Known elevations of large lakes were arnong the topographic data
used for the correction (which is why they appear uniformly medium grey). As the lakes are
normally frozen at this time of the year, and the main radar reflection is from the ice/water
interface (Côté, 199$), the SRTM DEM correction routine, assuming the radar-deterrnined and
the known lake levels to be equal, over-corrected the topography, which is the likely reason
for the pale ridge crests. This is because the current logic applies also to lake ice which, like
snow, is transparent when cold. Nevertheless, dark areas occur, in particular near lakes and
downwind of tundra areas, suggesting effects of snow accumulation. In part the dark tone may
be the resuit of standing biomass not included in the CDEM, should that biomass be frozen. At
Schefferville in Febniary, it can grade from frozen to unfrozen, with its effects on the radar
signal varying accordingly.
66
A detail of the difference map is shown in Figure 31 a together with its conespondingtopographie map in Figure 31b. Comparing these two figures, we can recognize three lakes(Vacher, Sauvaget and Matemace) and other smaller ones. The darker areas (red zones)suggest the presence of dry snow. Lakes flot used in the SRTM DEM are black as a resuit oflake ice growth. Some wetlands are similarly affected.
Examinmg more closely the central part of Figure 30 we sec several white and black spots(Figure 32, see inside the red circle). The black spots located mostly in the center of the rnapare iron ore mine pits excavated by the Iron Ore Company. Because these holes were notpresent in the CDEM, they conectly appear as depressions in the SRTM DEM with respect tothe CDEM. The white spots are mine dumps created during the mine excavation.
zU)U)
U)
oU)
U)
4
Figure 30: Illustration of the effect of snow cover on perceived ground topography usingdifferential interferometry using SRTM and Canadian DEM for Schefferville region (Québec)
nzmm
n
Schefferville Diff-map
67°5W 67’W 6655W 665OW
0 1 2 3 1EiH.:L
V..2 6
67
Zones with generally higher wind speeds, especiaÏly ridge crests, continually lose any
accumulated snow due to wind erosion and remain very nearly devoid of snow throughout the
winter. The snow accumulates downwind, often creating very deep accumulations in valleys
on the alpine tundra (Granberg, 1973). This suggests the cause of the pronounced banding
seen in the alpine tundra areas (Figure 33, see red circles).
V.. -
-‘
-L. 2-
, _..--.4s - •.‘ V
figure 31: Difference map and corresponding topographic map ofthe Lac Vacher area. Thedarker areas (red zones) suggest the presence of dry snow.
I
II
L..
:•--
VI
kfigure 32: Difference map and corresponding topographic map of mine pits (inside the red
circle). The black spots are pits and white spots are mine dumps created during the mineexcavation
6$
Figure 33: Difference map and topographie map ofthe east-center parts ofthe study area. Thered circles are the cause ofmiscorrection of SRTM data
In summary, differences between the SRTM DEM and the CDEM are consistent with well
known snow and lake ice conditions at Schefferville. Although rendered less directly visible
by the correction of the SRTM DEM, there is a snow cover signal consistent with what was
expected for a cold snow cover.
5.2 Resuits from Colorado
Once ah images were georeferenced, we extracted the ISA’s from each MSA for every IOP.
The results of this process are three DEM’s (February, March and September) of the Walton
Creek ISA which was used for this part of the research. The three DEM’s and their
conesponding backscafter images are illustrated in Figure 34.
k
69
A: DEM image (Left) and backscatter image (Right) from ts 1570
B: DEM image (Lefi) and backscatter image (Right) from ts 1589
C: DEM image (Left) and backscatter image (Right) from tsl 543
Figure 34: DEM’s and Backscatter extractions from Walton Creek ISA. Extracted from: A,february; B, March and C, September acquisitions
As a first step in the analysis, the degree of misfit between snow-free DEM’s was assessed.
Using two September DEM’s of Walton Creek ISA flown cross-wise, co-registered profiles
70
were sampled in each of them in such way as to cover most of the topographic variations of
the site (Figure 35).
Co-plotting as two continuous profiles shows that the two DEM’s are very similar but that
they are vertically offset (Figure 36).
3020
3000
2980
2960
2940
2920
2900
2880
2860
2840
Figure 36: Altitude verification of 1294 pixels ofthe two different AIRSAR images on thesame day
Figure 35: Sampling profite in the Walton Creek ISA September DEM’s
Cross Image AltitudeREWalton
1 74 147 220 293 366 439 512 585 658 731 804 877 950 1023 1096 1169 1242
Number ot sample— Sep (1538) — Sep (1543) I
71
The two profiles were then compared to each other (Figure 37). Regressing one ofthe profiles
against the other yielded the following expression which was applied to one of the images to
decrease the misfit.
Y = 0.914X + 280.18 (24)
To validate this resuit and to see how effective the applied model was, another 200 altitude
points were extracted randomly and their correlation verified (Figure 38).
3020
3010
— 3000E
2990
2980
2970
2960
2950
— 2940
2930
2920
Multi-Path Assessing
2880 2900 2920 2940 2960 2980 3000 3020
ts1543 altitude profile (m)
Figure 37: Correlation ofthe altitude profiles for two September crossed images of WaltonCreek ISA
The second step consisted of evaluating the methodology on the image pair from two different
seasons (TOPs). In one image the ISA is snow covered (dry or wet) and in the second one from
September the ground was bare. The altitudes of 150 pixels along a road near the Walton ISA
were extracted from InSAR DEMs for the two periods in February and September (Figure
39). Here again the profiles are very similar but there are differences ofbetween 3.7 and 9 m.
The same verification method was tested in the same area for the same sampte points but for
two other IOPs, from March and September. The results show general similarities between the
72
profiles but with absolute differences ofbetween of 3 m and 7.5 m (Figure 40) which could be
caused by multi-path contribution. The differences iltustrate the variations between TOPSAR
DEM’s that can be present over the ISA. To remove the variation, a simple regression model
was developed, using the road near the Walton Creek ISA, along which 375 points were
extracted. A scatter plot is shown in Figure 41. The operation was repeated again to correct
the March DEM (Figure 42). The models were then appÏied to the February and March
DEM’s and difference maps were calculated, subtracting the September DEM ftom those of
february and March (Figs 43, 44).
Validation of Cross Image Correlationf Modeled)
2990 +
29es
2990
2975
w2970
•
2965
‘tC2960In
•2955
2950
2945
2945 2950 2955 2960 2965 2970 2975 2980 2985 2990 2995
ts1543 altitude profile f m)
Figure 38: Validation of altitude correlation for 200 samples of Walton Creek ISA
73
Road altitude profilesRabbit Ears fWalton)
2880
FebSep
2655- -
t, t, t-- . t, n t- — n t- — n t— t t-- u, œ t, t-- u, n I— —
Number of samples
Figure 39: The variation of altitude of AIRSAR data over a road for (IOPÏ and BG2)
2876
2885
2866
2864
Mat_Sep
C t, t- — u, œ n t- — 0 o n t— — C O t, t- .- C o n t- — C O t, t-.- C O t, t- .- u, O)n n n n n -u- ‘u- ‘u- u, t, t, u, u, t- t- t, n u, o o o — — n n n n u, -u- ‘u- -u
Number of samples
Figure 40: The variation of altitude ofAIRSAR data over a road for (10P2 and 3G2)
2875
2860
Road altitude profilesRabbite Ears (Walton)
2880
2878 - --
2874
Ee2872 - --
2870
74
2870 2880 2890 2900 2910 2920
February road altitude f m)
•FEB-SEPI
Figure 41: Altitude correlation for two crossed images (Feb & Sep) of Walton Creek ISA
28602860 2870 2680 2890 2900 2910
Match road altitude f m)
‘Mat SEP
Creating mode to decrease Multi-Patherror over Walton ISA
2905
— 2900
E2895
p2890
2885
22880
.0 2875E
2870
0 2865
2860
4
f=O.883X+327.1
Creating model to decrease Multi-Path error overWalton ISA
2905
•2900
G 2895
z2890
2885
22880
.0 2875E
28700.
2865
2920
Figure 42: Altitude correlation for two crossed images (Mar & $ep) of Walton Creek ISA
75
Figure 43: Difference Map of September and February. Gray level represents the relativeSWE variation
Figure 44: Difference Map of March and September. Gray level represents the relative snowdepth variation
76
5.3 Integration of remote sensing and field data
Before starting to compare the Diff-Maps with ground data (SWE and snow depth), we
prepared the attributes of the shape files in which the SWE of ground data were calculated for
500 points. The graphical illustration of the process of allocating the density of 16 points to
their neighbourhood pixels is demonstrated in Figure 45. The association ofthe snow density
measurements to 500 snow depth measurements is shown in Figure 46.
Figure 45: Allocating the density by distance
-t --
, :_e
J
•
r
rA
J
rJ
• r
L . •.
Figure 46: Joining the snow density and snow depth data
77
5.4 SWE and depth estimation
In order to detect the signal related to SWE and snow depth, the difference maps for February
and March respectively were exported to a GIS environment and joined with ground data
shape files. The resuit is two new shape files, each containing the 500 punctual data
representing the measured snow water equivalents and depths and those estimated for the
same points (Figure 47) using InSAR. From these two sets of data, the differences between
InSAR estimates and measurements were calculated (Figures 48 and 49).
Both figures exhibit a pronounced North-South gradient showing that the vertical co
registration, and therefore the accuracy of the InSAR SWE and snow depth estimates
deteriorates with distance from the road (Figures 48 and 49). However, it is encouraging that
near the road the differences between the estimated and field measured quantities are usually
small. To further assess the snow signal detectable by InSAR a different method was therefore
employed. Profiles transverse to ploughed roads were extracted and co-registered at their
center points, which were located on the road (Figure 50, 51). Each profile is 150 m long,
consisting of 30 elevations 5 m apart.
• : 44.4*r);s?
Figure 47: Joining SWE of AIRSAR data to shape file of ground data
7$
SWE Error - Rabbit Ears (Walton ISA)
Â
Legend
— 25m
SWE_ErrorX
- Projection:UTM zone 13
Depth Error - Rabbit Ears (Walton ISA)
gg—
g
2
p.;:. •
•1 .‘._ )gP •
ii:’ -
:: ‘
339460
Source: GLPX DataAuthor: AH E. G. 0 62.5 125 250 375 550
Figure 42: Distribution map of SWE differences based on comparison ofremotely sensedderived SWE and ground truth SWE
359400 350100 350*09
!_
N360000 310200 368409 350600 310000
ALegend
Dep_Error
Projection:UTM zone 1]
2
5j
I
-
—.—
—at I!I
.—
: ‘0
: •
‘.i_.•0
—
Source: CLPX DataAuthor: Ah E. G.
31t_ 31*409 310199—r-—- -
36011• S
0 62.5 125 250 375 50]
Figure 49: Distribution map of Snow Depth error
79
Figure 51: The location of transversal profiles in Fraser M$A
Nine sets of transverse profiles were extracted in the Rabbit Ears MSA. Each set includes
three profiles for respectively february (blue), September (red) and March (yellow). Similar
profiles were extracted in the fraser MSA. Their location is shown in Figure 51. The full sets
--
,2
Figure 50: The location of transverse profiles in Rabbit Ears MSA
/
$0
of graphs of these profiles are provided in Appendix 1 and 2. Only one example is give here
(Rabbit Ears Profile 1) in Figure 52.
2874
2872
2870
2868
2866
w2864
2862
2860
2858
2856
2854
figure 52: Profile 1 near Rabbit Ears
The profiles show that, in general, the February elevations are well below those of the
September topography and this is in accordance with the second and the third hypotheses of
the research. The March elevations are seen sometimes above, sometimes below those in the
September DEM. However, tantalisingly, the February difference is smaller than expected and
the March difference ïs also flot exactly as expected
Among the 9 February profiles at Rabbit Ears (Appendix 1) four of them (1, 7, $ and 9) show
the signal expected for dry snow. In profiles 3 and 6 the signal is less obvious and profiles 2, 4
and 5 give mixed results.
—FEB SEP MAR
1 2 3 4 5 6 7 8 910111213141516171819202122232425262728293031
Number ofsamples
81
The March profiles also give mixed resuits, suggesting that in the variable Colorado winter
climate oniy parts of the 8110W cover may have been wet. In March, both wet and dry snow
was reported from snow pits of the nearby ISA, suggesting that the experimental conditions
may have been less than optimal. There is also a possibility that road sait may have been
rnixed with the snow within a throw distance for snow ploughs of about 20 ni.
The Fraser profiles show similar resuits. Among the 7 sets of profiles (Appendïx 2), numbers
1, 2 and 3 show the expected dry snow signal in February while the March profiles are only
partially above the September profile. Profiles 4, 5 and 7 show the expected resuits in only
half the profile. In February the left side of the profile is located below the September profile,
while the right side is above it. Profile 5 displays a bump which is also present in the March
profile when the profile generally conforms to the expected wet snow signal. A similar case
was observed at Rabbit Ears. In profile set number 6, the february and September profiles do
not show significant differences. The March profiles at Fraser show a pattem which is
generally similar to that observed at Rabbit Ears.
In summary, the road-crossing profiles show the expected results, but not unambiguously so.
There are unexpected variations which may have their cause in a variety of factors such as an
only partly wet snow cover in March and possible residual wetness from earlier, wanri spells
in February. To this, possible effects of road sait may be added to an unknown misfit between
the TOPSAR DEM’s. The vegetation also adds noise to the snow signal. Irnportantly,
however, there appears to be a snow signal amongst the noise in particular for profile 5 of the
Fraser M$A where we observe a clear signal for wet snow. These examples would thus
confirm hypotheses 2 and 4.
5.5 The nature of the 5110W depth and water equivalent signais
This section examines the observed and calculated snow depth and water equivalent signals
for the purpose of further elucidating their character. Although the altitudinal and other
TOPSAR DEM co-registration enors are too large to make a calibration attempt meaningful, a
further examination of the relationship between DEM differences and snow properties is of
interest. The estimated and observed data will be analysed statistically. First their correlation
82
will be examined and then, by means of stratification an attempt will be made to reduce the
misfit.
5.5.1 Snow water equivalent
The WaÏton ISA is characterized by broad meadows interspersed with small, dense stands of
coniferous trees. The SWE distribution determined from february field observations is shown
in Figure 53, that which was estimated from the February DEM in figure 54, and the
difference between the two in Figure 55.
As seen earlier, one DEM is tilted with respect to the other and this produces a large error
which is responsible in large part for the differences between the observed and calculated
snow water equivalents. Nevertheless, there is a correlation between the two, albeit noisy
(Figure 56). Not unexpectedly, it is not statistically significant.
FielU
SWE_rnele_1 1Statistics:
481Mnmum: o.o;8oMaximum: 0.918400
198,296480Mean: 0,405899Standd Deviation: 0,101808
Heid
SWErnag 1St&istics:
Coutt 491Minimum: -0.877197Maximum 5.871338Sum: 1097,252441Mean: 2,234730Standard Deviation: 1,205084
Frequency Distribution
150
‘ 100
zu.I50
0
0,0 0,1 0,3 0,4 0.5 0,6 0,8 0,9
oundSWE(m)
Figure 53: Distribution of ground truth SWE
Frequency Distribution
80
60
40
20
O
-0.9 0.1 1,0 7,9 2,9 3,8 4,8 5,7
lSWE(m)
Figure 54: Distribution of measured SWE based on InSAR
$3
-1,2 -0,3 0,6 1,6 2,5 3,4 44 5,3
SWEEnur(m)
Figure 55: Distribution of SWE differences (ground truth and InSAR)
Ewo,o
oo
1,0
0,9
08
0,0
SWE CorrelationRabbit Ears - Walton ISA
-2.0 -1.0 0,0 1,0 2,0 3,0 4,0 5,0 6,0 7,0
Image SWE(m)
Figure 56: Evaluating the measured and estimated SWE correlation
Stratifying the Walton ISA into low and high altitude areas with respect to the September
backscatter amplitude image (Figure 34c) and the topographie map and also dividing it into
four sections (figure 26 A-D) gave the mask in Figure 57. Using this mask, the correlation of
each sub-section of the site was calculated. The stratification increased the correlation to
significant levels for the individuai strata (Figure 58). Even though the increase of correlation
for each stratum is flot good enough to conclude with a high level of confidence that the
signais are reiated to the SWE the increased correlation from almost zero to r 0.40 shows
Field
SWE,Error
Stakisbcs;
Count:Minim’,un:Maamum:SunMean:Standasd Deviahon:
Frequency Distribiflion
491-1.2476975,469628897,9559611.8288311.195941
20
60
40
20
o
0,7
0,6
0,5 --
0,3
02
0,1
y=0,Ollx.0,381R=0,13
SWE image
linear(SWE Image)
84
that the DEM misfit produces a significant part ofthe errors. As we saw earlier in section 4.5.5
the multi-path error is a main source of AIRSAR DEM misfit.
Figure 57: Applied mask on the Watton ISA to separate higher altitude (outside of green &inside of yellow zones) from lower altitudes (inside the green zone) in four sections A, B, C
and D
$5
SWE Correlation SWE Correlation
Walton Section: B Walton Section: C
0,60 0,80
0.70 070y = 0,053x + 0,278 y = OCOiX + 0,404
R=0,40 R=0,000,600,60
0,500,50
0,400,40t
t
2 0,30 2 0,30ww
0,20 0,20
0,10 0,10
0,00 0,00
0,00 1,00 2,00 3,00 4,00 5,00 6,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00
Image Altitude Difference (m) Image Altitude Difference (m)
SWE CorrelationSWE Correlation Walton Section: D
Walton Section: A0,80 0,80
0,70 0,70y = DOux + 0,403 y = -0,046x + 0,474
R 0,390,60 0,60 R -036
0,50 .. 0,50w
0,40 040
g2 0,30 0,30w j,
0,20 0,20
0,10 0,10
0,00 0,000,00 1,00 2,00 3,00 4,00 5,00 6,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00
Image AltItude Difference (m) Image Altitude Difference (m)
Figure 5$: SWE correlations of low altitude points of 4 sections of Walton ISA
5.5.2 Snow depth
The snow depth distribution for the Walton ISA was determined from field observations
(Figure 59). Differences between the Mardi and $eptember DEM’s were used to generate the
ftequency distribution of SAR-estimated snow depths seen in Figure 60. The dïfference
between the estimate and observation is shown in Figure 61.
$6
Statistics:
Cotrit:Mimtin:Maxm*zn:Sum:Mean;Standard Deviation:
5050.1700003.000000768,3200001,5214260,347885
Figure 59: Statistical resuits of ground truth snow depth
Stakistics:
Minimum:Maximum;Sum:MeanStandard Deviation:
5051.3048328,8712262076.7216804.1123201,421322
80
60
40
‘20
O
Count;Minimum;Maximum:Sum:Mean:Standad Deviation:
Figure 60: Stati stical resu Its of measured snow depth by InSAR
505-0.5550687.3557811308.4016802.580834t440544
Field
6Depth .1 frequency Distribution
160
.100
z! 50
o0,2 0,6 1,0 1.3 1,7 2,1 2,5 2,9
GroundSnowflqidi (m)
Fi&d
ImQ,Depth Frequency Distribution
1,3 2,4 3,4 4,4 5,5 6,5 7,6
Izuaw SDowDqh(m)8.6
Field
Dep_Errof
Statistic:
Frequency Distribution
60
50
‘40
g-30Vsr.
10
o-0,6 0,5 1,6 2,7 3,8 4,9 6,0 7,1
SnowflcpthEuur(m)
Figure 61: Statistical resuits of snow depth differences (ground truth and InSAR)
$7
As in the case of the snow water equivalents the estimated snow depth distribution is
dominated by misfit between the two DEM’s. However, as in the case of the snow water
equivalents, there is a noisy correlation between the two (Figure 62). Again, there is no
statistically significant correlation. Nor could any significant correlation reasonably be
expected under the circumstances.
Walton_DepthCorrelation
2*_. a:Pftl J
3,0 5,0
Image Snow Depth (m)
Figure 62: Measured vs. estimated snow depth
However, stratifying the samples using the same mask applied for SWE section (Figure 57)
the correlations were calculated and a significant increase for each stratum was observed
(Figure 63).
3,5y0,016x+ 1,454
R = 0,06
I-
.
. .& .
. ..
20 • Pt •_
p •
•
—-
•.•
10 .• •• •• •
•05
..
0,0
-1,0 1,0 7,0 9,0
8$
Depth Correlation Depth Correlatïon
Walton Section: B Walton Section: C
250 250
y=O,140x + 1028 y=0,195x +0796
R=0,63 R=0,422,00 2,00
1,50
1,00 1,00
0
0,50 0,50
0,00 0,00
0,0 1,0 2,0 3,0 4,0 5,0 6,0 7,0 0,0 1,0 2,0 3,0 4,0 5,0 6,0 7,0
Image Altitude Difference Cm) Image Altitude Difference (m)
Depth Correlation Depth CorrelationWalton Section: A Walton Section: D
2,50 2,50
y =0,066x + 1,498 y =0,343x +0045R=0,39 R=0,62
2,00 2,00
1,50 ‘- ‘ 1,50
o o
• 1,00 . 1,00C Cz z2 £C, C,
0,50 0,50
0,00 0,00
0,0 1,0 2,0 3,0 4,0 5,0 6,0 7,0 0,0 1,0 2,0 3,0 4,0 5,0 6,0 7,0
Image Altitude Difference (m) Image Altitude Difference (m)
Figure 63: Snow depth correlations of low altitude points of 4 sections of Walton ISA
Once again the correlations are improved. This shows that the error is flot randomly
distributed and, again, that there is a snow depth signal among the DEM co-registration noise.
The resuit further suggests that for routine applications there will be, as suggested also by the
road cross-sections, issues to resolve with regard to the degree of 5flOW wetness.
89
As snow wetness generates a signal in the amplitude domain, the assessment ofwhether or flot
the snow is sufficiently wet or dry (to produce a reliable snow depth or snow water equivalent
signal) should be fairly straightforward. In a sirnilar vein, the terrain could be stratified with
regard to vegetation and other fiO1SC factors to improve both snow depth and snow water
equivalent estimates. However, given the problems with the present cross-track InSAR data
sets, such an analysis is better deferred to a future project. It was welI beyond the scope ofthe
present study to attempt to re-compute DEM’s from the raw SRTM data. Airbome cross-track
InSAR. it has been demonstrated, is not suited for snow cover mapping.
90
6 CONCLUSIONS
As has been shown in previous research at $chefferville, InSAR contains snow cover
information which destroys coherence even for small changes in SWE between parent image
acquisitions. As a consequence, attempts to use repeat-pass InSAR data to map the $WE at
Schefferville and elsewhere have been frustrated. Coherence loss also prevents the use of
differential InSAR techniques.
In this thesis, the use of repeat-pass cross-track InSAR has been explored. In cross-track
InSAR the parent images are acquired simultaneously and this solves the problem of
coherence loss. Comparing independently produced DEM’s, one with, and the other without
snow, wouÏd solve also the DInSAR coherence problem.
As has been shown elsewhere, cold snow is largely transparent at radar wavelengths and
therefore the ground/snow interface is the main reflector even under deep snow covers.
However, because of the high refractive index of ice (1.79), the two-way passage through the
snow cover causes a phase shift which is directly proportional to the SWE. When wet, the
snow surface becomes the main reftector. Accordingly, differences between the elevations of
two DEM’s (one summer, one winter) measure either SWE or snow depth, depending on the
state ofthe snow.
Two sets of cross-track InSAR DEM’s were employed to explore this idea further at two
different field sites. At Schefferville a photograrnmetrically produced DEM was compared to
91
the SRTM DEM. At the CLPX field site in Colorado multiple airbome cross-track DEM’s
were compared.
The resuits showed first that attempts, in the radar processing, to tie the interferometrically
produced DEM’s to the known topography partly eliminates the SIIOW cover signal. This
affected both the $RTM DEM and the TOPSAR DEM’s. The lesser stability of the aircraft
platforni contributed additional co-registration problems, showing that for snow cover
mapping by repeat-pass cross-track Jn$AR a stable satellite platform is required.
Although for these reasons the snow cover signal was rendered quite noisy it was possible to
demonstrate, at least qualitatively, the presence of a snow cover signal consistent with that
which was expected. Re-processing of the raw $RTM data specffically for the purpose of
snow cover detection was outside the scope ofthis thesis but would be a logical further step.
The analysis further suggests that additional factors, such as the state (frozen or flot) of not
only the snow cover, but also the vegetation cover and underlying sous may need to be
accounted for in order to optimise this technique which, as yet only in theory, offers snow
cover information which is unprecedented in terms of accuracy and spatial resolution.
92
Adams, R.$., Spittiehouse, D.L. and Winkler, R.D. (1996) The effect of a canopy on the
snowmelt energy balance. Froc. 64th Annual Meeting Western Snovv Conference, April
1996, Bend, Oregon, p. 171-174.
Annersten, L. (1966) Interaction between surface cover and permafrost (Biuletyn Glacjalny,
No. 15, p. 27-33.
Armstrong, R.L. (1980) An analysis of compressive suain in adjacent temperature-gradient
and equi temperature layers in a natural snow cover. Journal of Gtaciology, vol. 26, n°
94, p. 283-289.
Askne, J. and Hagberg, 1.0. (1993) Potential of interferometric SAR for classification of
Landscape surfaces. Froceeding of the International Geosciences and Rernote sensing
symposium (IGARSS ‘93), Tokyo, Japan, 18-21 august 1993 (Piscataway, NI: IEEE), p.
985-987.
Bamier, R. and Harti, Ph. (1998) Synthetic Aperture Radar Interferometry. Inverse Froblems,
vol. 14, p. R1-R54.
Barrie, L. (1991) Dry deposition to snowpacks. In: Seasonal snowpacks: Processes for
compositional change. Proceeding of the NATO advanced research workshop on
processes of chemical change in snowpacks, Maratea, Italy, July 1990, Springer-Verlag,
Berlin. Series G, Ecological Sciences, n° 28, p. 1-20.
93
Bean, B.R. and Dutton, E.J. (1966) Radio Meterology. New York, Dover Pub, 435 p.
Bernier. M. and Fortin, J.P. (1991) Evaluation of the potential of C and X band SAR data to
monitor Dry and Wet sriow cover, International Geosciences ancÏ Remote Sensing
Symposium (IGÀRSS ‘9]), ESPOO, Finland, June 3-6 1991, p. 2315-231$.
Bohren, C.f. and Battan, L.J. (1982) Radar Backscattering of Microwaves by Spongy 1cc
Spheres. Journctl ofthe Atmospheric Sciences, vol. 39: p. 2623—262$.
Colbeck, S.C (1980) Liquid distribution and the dielectric constant ofwet snow. In Proceeding
of the NASA workshop on the Microwave Remote Sensing of Snowpack Properties,
Goddard Space Flight Centre, p. 2 1-39.
Colbeck, S. C. (1982) An overview of seasonal snow metamorphism. Rev. Geophysics Space
Phvsics, vol. 20, p. 45-6 1.
Colbeck, S.C., Akitaya, E., Arrnstrong, R., Gubler, H., Lafeuille, J., Lied, K., McClung, D.
and Morris, E. (1990) The International Classification for Seasoual Snow on the Ground.
Issued by: The International Commission on Snow and 1cc of the International
Association of Scientific Hydrology and Co-Issued by International Glaciological
Society, 37 p.
Côté, S. (1998) Aiialyse du processus de croissance de la glace des lacs avec les données du
satellite ERS-1 en mode amplitude et interférométrique : région de Schefferville.
Mémoire de maitrise, Université de $herbrooke, Sherbrooke, 87 p.
Cumming, W.A. (1952) The Dielectric Properties of 1cc and Snow at 3.2 Centirneters. Journal
ofAppÏiedPhysics, vol. 23, p. 768-773.
Dozier, J., Davis, R.E. and Perla R. (1987) On the objective analysis of snow microstructure,
in Avalanche Formation, Movement and Effects. International Association of
94
Hydrological Sciences, Wallingford, UK, (H. Gubler, Ed.), IAHS Publication, n° 162, p.
49-59.
Engen, G., Guneriussen, T. and Overrein, 0. (2004) Delta-K interferornetric $AR technique
for snow water equivalent (SWE) retrieval. IEEE Geoscience ancÏ Remote Sensing
Letters, vol. 1, n° 2, p. 57—61.
Feigi, K.L., Sergent, A. and Jacq, D. (1995) Estimation of an earthquake focal mechanism
from a satellite radar interferogram: application to the December 4, 1992 Landers
aftershock. Geophys. Res. Lett., vol. 22, p. 1037-1048.
Fisette, T. (1999) Influence de la neige sur la précision des modèles numériques d’altitude
(MNA) crées par interférométrie RAS, Mémoire de maîtrise, Département de
géographie et télédétection, Université de Sherbrooke, 87 p.
Gabriel, A.K., Goldstein, R.M. and Zebker, H.A. (1989) Mapping Small Elevation Changes
Over Large Areas: Differential Radar Interferometry. Journal of Geophysical Research,
vol. 94, n° 37, p. 9183-9191.
Goldstein, R.M. (1995) Atmospheric limitations to repeat-track radar interferometry. Geoph.
Res. Lett., vol. 22, p. 2517—2520.
Goldstein, R.M., Engelhard, H., Kamb, B. and Frolich, R.M. (1993) Satellite Radar
Interferometry for Monitoring 1cc Sheet Motion: Application to an Antarctic 1cc
Stream. Science, vol. 262, p. 1525-1530.
Graham, L. (1974) Synthetic interferometer radar for topographic mapping. Proceedings ofthe
IEEE, vol. 62, n° 6, p. 763-768.
Granberg, H.B. (1972) Snow depth variations in a forest-tundra cnvironment near
Schefferville, P.Q., M.Sc. Thesis (unpub.) McGill University, 131p.
95
Granberg, H.B. (1973) Indirect mapping of the snowcover for pennafrost prediction at
Schefferville, Quebec. In: Permafrost, The North American Contrjbutjon to the Second
International Conference. National Acaderny of Sciences, Washington, p. 1 13-120.
Granberg, H.B. (1994) Mapping heat loss zones for permafrost prediction at the
northemlalpine lirnit of the Boreal forest using high resolution C-band SAR. Remote
Sensing ofEnvironment, vol. 50, n° 3, p. 280-286.
Granberg, H.B. (199$) Snow cover on sea ice. In: Physics of ice-covered seas (M. Leppiranta
ed.), vol. 2, Helsinki University Printing House, p. 605-649.
Granberg, H.B., Judge, AS., Fadaie, K. and Simard, R. (1994) C-band backscatter from
northern terrain with discontinuous permafrost: The Schefferville Digital Transect.
Canadian Journal ofRemote Sensing, vol. 20, p. 245-256.
Granberg, H.B. and Irwin, G.l. (1991) Construction of a digital transect at Scheffer-ville to
validate a geographic snow information system. In ProceecÏings, J 51’!? C’cutadian
Conference on Remote Sensing, Calgary, p. 84-87.
Granberg, H.B. and Vachon, P.W. (1998) Delineation of Discontinuous Permafrost at
Schefferville Using RADARSAT in Interferometric mode. Seventh International
Permafrost Conference, Yellowknife, NWT, 23-27 lune 199$, p. 341-345.
Gray, D.M. and Male, D.H. (1981) Handbook of Snow: Principles, Processes, Management
and Use, Pergarnon Press Canada, Willowdale, Ontario, 776 p.
Guneriussen, T., Høgda, K.A., Jolmsen, H., and Lauknes, I. (2001) InSAR for estimation of
changes in snow water equivalent of dry snow. IEEE Transactions on Geoscience ancÏ
Remote Sensing, vol. 39, n° 10, p. 2101—2108.
Hallikainen, M., Ulaby, F. and Abdel-Razik, M. (1982) Measurements of the Dielectric
Properties of Snow in the 4-1$ GHz Frequency Range. European Microwave
C’onference, l2th, 1982, p. 151-156.
96
Hallikainen, M., Ulaby, F. and Abdel-Razik, M. (1984) The Dielectric Behavior of Snow in
the 3- to 37-GHz ranges. 1984 IEEE International Geosciences Remote Sensing
Symposium (IGARS$ ‘24) Digest, Strasbourg, France, 27-30 August, p. 169-176.
Hallikainen, M.T. and Winebreimer, D.P. (1992) The physical basis for sea ice rernote
sensing: Microwave Remote Sensing of Sea 1cc. Geophyscal Monograph. Series, AGU,
Washington, D.C, vol. 68, p. 29-46.
Hallikainen, M.T., Ulaby, F.T. and Abdelrazik, M. (1986) Dielectric properties of snow in the
3 to 37 GHz range. IEEE Transactions on antennas and propagation, vol. AP-34, n° 11,
p. 1329-1339.
Hallikainen, M.T., Dobson, M.C. and Abdel-Razik, M. (1983) Modeling of Dielectric
Behavior of Wet Snow in the 4-1$ GHz Range. 1983 IEEE International Geosciences
Remote Sensing Symposium (IGARSS ‘83) Digest, San Francisco, CA, 31 Aug.- 2 Sep.
Hanssen, R. and Feijt, A. (1996) A first quantitative evaluation of atmospheric effects on
SAR interferometry. In: Proc. FRINGE 96 Workshop on ERS SAR Interferometry,
ZUrich, Switzerland, 30 Sept-2 Oct. 1996, ESA Publications, Noordwijk, SP 406, n° 1,
p. 277—282.
Hensley, S., Rosen, P. and Gurrola, E. (2000) Topographie map generation from the Shuttie
Radar Topography Mission C-band SCANSAR interferometry. Proceedings of SPIE,
vol. 4152, p. 179-189.
Hippel, A.R.V. (1954) Dielectric and waves. John Wiley & Sons, New York, 304 p.
Imel, D.A. (2002) What’s wrong with My AIRSAR Data? ARSAR Earth Science and
applications Workshop, 4-6 March 2002, Jet Propulsion Laboratory.
97
Larsen, Y., Malnes, E. and Engen, G. (2005) Retrieval of snow water equivalent with Envisat
ASAR in a Norwegian hydropower catchment. IEEE International Geoscience and
Remote Sensing Symposium (IGAR$S 2005), Seoul, Korea, 25-29, vol. 8, p. 5444-5447.
Linlor, W.I. (1980) Permittivity and Attenuation of Wet Snow Between 4 and 12 GHz. J.
Appt. Phys., vol. 51, p. 2811-2816.
Massonnet, D. and feigl, K.L. (1995) Discriminating geophysical phenomena in satellite
radar interferograms. Geoph. Res. Lett., vol. 22, p. 1537-1540.
Massonnet, D., Adragna F. and Rossi, M. (1994b) CNES General-Purpose SAR Conelator.
IEEE Trans. Geosciences andRemote Sensing, vol. 32, n° 3, p. 636-643.
Massoimet, D., Feigl, K.L., Rossi, M. and Adragna, F. (1994a) Radar interferometric
mapping of deformation in the year after the Landers earthquake. Nctture, vol. 369, p.
227-230.
Massonnet, D., Rossi, M., Carmona, C., Adragna, F., Peltzer, G., Feigi, K. and Rabaute, T.
(1993) The Dispiacement Field of the Landers Earthquake Mapped by Radar
Interferometry. Nature, vol. 364, p. 138-142.
Massonnet, D., Briole, P. and Arnaud, A. (1995) Deflation of Mount Etna Monitored by
Spacebome Radar Interferometry. Nature, vol. 375, p. 567-570.
Matzler, C. (1987) Applications of the Interaction of Microwaves with the Natural Snow
Cover, Remote sensing reviews, vol. 2, 391 p.
Mitzler, C., Aebischer, H. and Schanda, E. (1984) Microwave Dielectric Properties of Surface
$now. IEEE1 Oceanic Engin., vol. 0E-9, p. 366-371.
Melor, M. (1965) Blowing snow. USA CRREL Cold Regions Science and Engineering Part
III, Section A3C, Monograph, 95 p.
9$
Mouginis-Mark, P. (1995) Analysis of volcanic hazards using Radar Interferometry. Earth
Observation, Quarterly, vol. 44, p. 6-10.
Murakami, M., Tobita, M., Saito, T. and Masharu, H. (1996) Coseismic Crustal Deformation
of 1994 Northridge Califomia Earthquake Detected by Interferometric 1ERS-1 SAR.
Proceedings ofthe lst SAR Interferometric Workshop, Tokyo, Japan, p. 47-66.
Nicholson, f.H. (1975) snow depth mapping from aerial photographs for ttse in permafrost
prediction. In proceedings, 32’ Eastern Snow Coiference, p. 124-126.
Nicholson, F.H. and Granberg, H.B. (1973) Pennafrost and snow cover relationships near
Schefferville. In: Permafrost, The North American Contribution to the Second
International Conference. National Academy of Sciences, Washington, p. 151-15$.
Nicholson, F.H. and Thom, B.G. (1973) Studies at the Timmins 4 permafrost experirnental
site. Tri: Permafrost, The North American Contribution to the Second liternationa1
Conference. National Acaderny of Sciences, Washington, p. 159-166.
Peltzer, G. and Rosen, P. (1995) Surface displacement of the 17 May 1993 Eureka Valley,
California, earthquake observed by SAR interferometry. Science, vol. 268, p. 1333-
1336.
Peltzer, G., Hudnut, K.W. and Feigi, K.L. (1994) Analysis of surface dispiacement gradients
using radar interferometry: New insights into the Landers earthquake. I Geophvs.
Res., vol. 99, p. 21971-21981.
Pomeroy, J.W., Marsh, P., Jones, H.G. and Davies, T.D. (1995) Spatial distribution ofsnow
chemical load at the tundra-taiga transition. b, (K.A. Tonnessen, MW. Williams and
M. Tranter, eds.) Biogeochernistry of Seasonally Snow-covered Catchments. IAHS
Publication n 228. IAHS Press: Wallingford, UK, p. 19 1-206.
99
Rignot, E., Echelmeyer, K. and Krabill, W. (2001) Penetration depth of interferometric
synthetic-aperture radar signais in snow and ice. Geophysical Research Letters, vol. 2$,
n° 1$, p. 3501-3504.
Robinson, D., Dewey, K. and Heim, R. (1993) Global snow cover monitoring: An update.
BuÏl. Arnerican Meteorologicat Society, vol. 74, p. 16$9-1696.
Rodriguez, E. and Martin, J.M. (1992) Theory and design of interferometric synthetic
aperture radars. IEEE Froc., vol.139, p.147-59.
Rodriguez, E., Morris, C.S, Beiz, J.E. (2006) A Global Assessment of the $RTM
Performance. Fhotogrctrnmetric Engineering and Rernote Sensing. vol. 72, n° 3, p.
249-260.
Rott H, Nagler, T. and Scheiber, R. (2003) Snow mass retrieval by means of SAR
interferometry. Froc. of fRJNGE 2003 Workshop, Frascati, Italy, 1—5 December,
(European Space Agency ESA).
Serreze, M., Clark, M., Armstrong, R., McGinnis, D. and R. Pulwarty (2000) Characteristics
ofthe western United States snowpack from snowpack telemetry (SNOTEL) data. Water
Resources Research, vol. 35, n° 7, p. 2145-2160.
Shi, J., Hensley, S. and Dozier, J. (1997) Mapping snow cover with repeat-pass synthetic
aperture radar. Froceedings oJIGARSS, Singapore, p. 62$-630.
Strozzi, T., Wegmuller, U. and Matzler, C. (1999) Mapping wet Snow covers with SAR
interferometry, Int. Journal ofRemote Sensing, vol. 20, n° 12, p. 2395-2403.
Sturrn, M. (1992) Snow Distribution and Heat Flow in the Taiga. Arctic and Alpine Research,
vol. 24, n° 2, p. 145-152.
Sturm, M., Holmgren, J., and Liston G. (1995) A seasonal snow cover classification system
for local to global applications. I Climate, vol. 8, n° 5, p. 126 1-1283.
100
Tarayre, H. and Massonnet, D. (1996) Atmospheric propagation heterogeneities revealed by
ERS-1 interferometry. Geoph. Res. Lett., vol. 23, p. 989-992.
Terraserver website (2006) Topographie map and aerial photography. www. terraserver. com.
Thom, B.G. (1969) New permafrost investigations near $chefferville, P.Q. Rev. Geogr.
Montreal, V. 23, p. 3 17-327.
Thom, B.G., and Granberg, H.B. (1970) Pattems of snow accumulation in a forest-tundra
environment, central Labrador-Ungava. Eastem Snow Conference. Proceeedings of the
27th1 Aniiual Meeting, p.76-86.
Tiuri, M., Sihvola, A., Nyfors, E. and Hallikaiken, M.T. (1984) The complex dielectric
constant of snow at microwave ftequencies. IEEE Journal of Oceanic Engineering, vol.
0E-9, n° 5, p. 377-382.
Toews, D.A.A. and Gluns, D.R. (1986) Snow accumulation and ablation on adjacent forested
and clearcut sites in southeastem British Colombia. Proceedings of Western Snow
Conference (USA,), n° 54, p. 1001-1011.
Ulaby, F.T., Moor, R.K. and Fung, A.K. (1982) Microwave Remote Sensing, active and
Passive, vol. 2: Radar Remote Sensing and Surface Scattering and Ernission Theory,
Artec House Inc., p. 457-1064.
Ulaby, F.T., Moor, R.K. and Fung, A.K. (1986) Microwave Remote Sensing, active and
Passive, vol. 3: From theory and Applications, Artec House Inc., p. 1065-2 162.
Ulander, L.M.H. and Hegberg, J.0. (1993) Use of ll’SA for radiometric calibration over
slopping terrain. CEO SAR calibration workshop proceeding, Sep. 20-25, ESTEC, ESA
WAP-04$, Noodwijk, The Netherlands, p. 147-159.
101
Vachon, P.W., Geudtner, D., Gray, A.L. and Touzi, R. (1995) ERS-1 Synthetic Aperture
Radar Repeat-Pass Interferometry Studies: Implication for RADARSAT. Journal
Canadien de Télédétection, vol. 21, n° 4, p. 441-454.
Waters, J.W. (1976) Absorption and emission by atmospheric gases. Astrophysics, vol. 12,
Part B: Radio telescopes, p. 142-176.
Wegmuller, U., Wemer, C.L., Nesch, D. and Borgeoud, M. (1995) Land-Surface Analysis
Using ERS-1 SAR Interferometry. ESA Bulletin, n° $1, p. 27-30.
Woo, M.K. and Steer, P. (1986) Monte Carlo simulation of snow depth in a forest. Water
Resources Research, n° 22, p. 864-868.
Zebker, H. and Rosen, P. (1997) Atmospheric artifacts in interferometric SAR surface
deformation and topographie maps. I Geophys. Res., vol. 102, n° B4, p. 7547-7563.
Zebker, H. and Villasenor, J. (1992) Decorrelation in interferometric radar echoes. IEEE
Trans. Geosci. Rem. Sens., vol. 30, p. 950-959.
Zebker, H.A., Rosen, P.A., Goldstein, R.M., Gabriel, A. and Wemer, C.L. (1994b) On the
derivation of coseismic displacement fields using differential radar interferometry: the
Landers earthquake. I Geophys. Res., vol. 99, p. 19617-19634.
Zebker, H.A.; Wemer, C.L.; Rosen, P.A. and Hensley, S. (1994a) Accuracy of topographie
maps derived from ERS-1 interferometric radar. IEEE Transaction on Geosciences and
Remote Sensing. vol. 32, p. 823-836.
102
Snow spatial variabïlity resuits
Rabbit Ears MSA
The following figures represent the extracted road transverse profiles for 3 seasons (February
with bleu color, March with yellow color and September with red color). The X axis is
labelled with the number of pixels beside the road. Pixel 15 and 16 are the road centre with
altitudes assurned equal in ail seasons. The Y axis is the altitude of each pixel in meters.
/ /
Alt
itud
eCm
)A
ltit
ude
(m)
Alt
itud
e(m
)M
MM
jM L
’, à ‘J,
00 00 ‘O o
00
M L’)
ÇID C
/1
4; I,
/ ,1
M L’,
à C,’
00 00 C M
c,
M L’)
M L’)
à u’ 00 “4 00 C o à
00
M L’)
7 (f
Ï s/
/
4/ /
K
2871
2810
2869
E
2868
2867
2866
Rabbit Ears Profile 2
2871
104
FEB SFP
2870
2869
E
2868
2867
2866
2865 —
1 2 3 4 5 6 7 8 910 11 12 13 14 15 16 17 18 192021 22232425262728293031
Number cf samples
---S MAR
2871
2870
2869
E
- 2868
2867 —
2866 -
28651 2 3 4 5 6 7 B 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
Number of samples
FE8 ----SEP MAR
28651 2 3 4 5 6 7 8 910111213141516171819202122232425262728293031
Number of samples
Alti
tude
(m)
MM
M
/t
M M 00 00 00 o
ZM
o 00 DO -Vu
‘fl
M M M
Alti
tude
fm)
MM
r-M
ovi
ev
i
7Ï ‘s
Alti
tude
fm)
Ï
00 00 m
ï
/
M M ‘Ii
00 DO e
ZM
oui
(000
W
u M M (li
M
j
M M vi 00 .0 00 10 o
00 n .
00
0 €0 M M M
100 00 n
C
t
Rabbit Ears Profile 4
2900
2895
2890
2885
p2880
- 2875
2870
2865
2860
2855
2850
106
—FEB —-SEl’
—SEl’ MAR
•‘a
------ -
‘-ç
I 2 3 4 5 6 7 8 9 10111213141516171819202122232425262728293031
Number 0f samples
—FEB —-SEP MAR
‘
-
: EZ:
I 2 345678 910II12I3I4I5I6I7I819202I22232425262728293031
Number 0f sampes
2900
2895
2890
2885
?2880
2875
2870
2865
2860
2655
2850
2900
2895
2890
2885
2880
2875
2870
2865
2860
2855
2850
I 2 3 4 5 6 1 8 9 I0I11213141516171819202122232425262728293031
Number 0f sampes
2875
2910
I 2 3 45678 91011 121314I5161718192021222324252627282930
Number 0f samples
SEP MAR
-—-
107
—FEB -—SEP
Rabbit Ears Profile 5
2910
29œ
2695
-UD
2890
2685
2860
2910
2900
2900
..2895
D28go
2885
2660
2875
1 2 345678 91011121314151617181920212223242526fl282930
Number 0f samples
—FEB —5FF’ MAR
2900
2900
2695wD
2890
2885
2860
2675
1 2 345678 9101112131415161718192021222324252627282930
Number cf samplesn
Alti
tude
(m)
Alti
tude
fm)
Alti
tude
(m)
e
)/
PV%
ÏN
)
(D -1 D)
D
o (j,
D) 3-4
-o-.
(D
N)
N) w N)
-4
/C,) C
o 00
N)
L.) w C,
-4 D)
‘D D N)
L.) w o, -4 D)
CC N)
N) w N)
M -4
3
z 3 o. (D ocn
D) 3 •0 (D (n
M L.) w o, -4 D) ‘o e N)
L.) w o, -4 C,
‘D N)
N)
C.)
//
/ )(I /
G,
E(G.0
G,
Io,F.
C,,
L,,
wEu,.0
LI,
C,,
o,L,)
L,,
o,F..
I,,
C,)
ow
r—C‘J,
-4-,
o.wo
LI,
U)G
,
Eu,.0
LI,
C,)
(.4(.4G,
C-.
LI,
C,,
G,
F.
LI)
C,)
LI)O
LI)Q
LI)
(w)apflI
(w)
apnpiv
(w)
Rabbit Ears Profile $
2892
2890
2889
E
2886
28M
2882
2880
2892
2890
2888
E
2886z
2864
2882
2888
E
2886z
2884
2682
1 2 3456789101112131415161718192021222324252627282930
Number ot samples
—SEP MAR
..-
110
BSEP
2886
2892
2890
I 2 3456789 10 11 12 131415161718192021 222324252627282930
Number of samples
—FEB —SEP MAR
___________
-.
2880
f 23456789 1011 12131415161718192021 222324252627282930
Number of samples
ç
M cc cc cc o
o Q)cc
cc
-u cc M M b M
L’, C
Alti
tude
(m)
Alti
tude
(m)
Alti
tude
(m)
MM
b-b
cccc -c
cccc
cccc
Ncc
cc
MM
MM
cccc
M 1 MM
cccc
cc—
112
Snow spatial variability resuits
Fraser MSA
The following figures represent road transverse profiles for 3 seasons (february in blue,
March in yellow, and $eptember in red). The X axis is numbered in pixels beside the road.
Pixel 15 and 16 are the road centre which their altitudes presumed equal in ah seasons. The Y
axis is the altitude of each pixel in meters.
n
MM
MM
MM
Alti
tude
(m)
Alti
tude
(m)
Alti
tude
fm)
/
MM
MM
MM
/
)Ï%
ç J.
0 ni
-c -c C t’J
M (11 O)
O) cc o
z—
D-
—(D
.
o ——
o O)
—
(D—
(ncc M M M M u, M O
)M -c M O
)
/ /
M c-z, O) -c Q) cc o
D-
—(D o -
—o
O)
O)
—
3-c
(D—
0cc M o M M M M C.
,M M u, M O
)M —
4M O
)
J
M C.)
u,
O)
--4 O)
cc o
ou’
_1
U
O (D.
(ncc M M M M C.
,
M u’ M o,
0 n t
)
Alti
tude
(m)
Alti
tude
(m)
Alti
tude
(m)
MM
MM
MM
M
Tj
-t ci,
(D -t -t C (DM w tu C
,
-1 s cc o
C—
3w
otui
-s,
— C,
W 3-4 12 M M 12 12 12 12
0 ri, t I
M w tu, C,
-4 s cc o
C—
3w
o-—
(D•
1—
o -s,
—w D
)—
3-4
0cc 12 M 12 M M C)
, 12 12 12
M w u,
ci
-J s cc o
ZM
C_
3w
o°’
-s,
—O
c,
Is
o—
03
cc 12 M M M M w M 12 12 M -4 12 12
w 0 m t
m t
M W (n C)
-4 C)
C o
ZM
C—
3W
o (nC
)w
—3-1
t—
s-C
) M N M M (n M C)
M W (n C)
-1 C).
C o
o“
3-4
t—
o
M M W M (n M -4 k)
D
Alti
tude
(m)
Alti
tude
(m)
MM
MM
Alti
tude
(m)
MM
Mk)
MM
MM
(t rt -t C (D
M W (n C)
-4 C)
C o
C—
3W
ci, n,o
t—
—o
—w
r—
53
-4z
t— C
)0-a C M N M M (n M -4 M C
)
C)
n, t
Alti
tude
fm)
Alti
tude
fm)
Alti
tude
(m)
MM
MM
(flO
r/
jr
j
M L., e -4 e e
ZM
o (ne
e 3-4
-o (nu M
ÇdD
-( -t C Ut
LJ
/
M w o e -4 e
n,u
3w
e
o ue
eu
3-4
—je
Ut
u M M M -4
4
)5
M w e e -4 e C
ZM
3-4
-o (nu M M e
J
J
n, t
ç-
)2 y
J
t
Alti
tude
fm)
j
Alti
tude
fm)
//
/
J
Alti
tude
fm)
r/
/z
M (J,
Oc cc o
ou
i oc
-D-
C,,
— cc M M C.,
M Q,
Ï
/
A
//
M C.,
‘ii oc 00 cc o
00 t
C,)
Oc
—
u,— cc M M C.
,
M M Q,
-t
V)
-t
-t C
00
I t
M Q’ oc 00 cc e
ZM
oui
3—
,i-0
0w
— cc M M t.,
M u, M 00
00 u, n t
fraser Profile 7119
—FB —SEP
2940
2930
2920
E
- 2910
2900
2890
2880
2940
2930
2920
E
- 2910
2900
2690
2880
1 2 3456789 101112131415161718192021222324252627282930
Number 0f samples
- --SEP MAR
E
-U
2940
2930
2920
2970
2900
2890
2880
1 2 3456789 101112131415161718192021222324252627282930
Number 0f samples
—FEB SEP MAR
I 23456769101112131415161718792027222324252627282930
Number 0f samples
top related