47 Département de Géomatique appliquée Université … · 47 Département de Géomatique...

135
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

Transcript of 47 Département de Géomatique appliquée Université … · 47 Département de Géomatique...

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)

9owda wzd .?aftam (oit

tftci’t pwte &we anti tIei’t patience

n

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 SVΠproduced 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

Alti

tude

(m)

Alti

tude

(m)

Alti

tude

(m)

-

î——

2‘4

tH

t;

ç

Z

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

CL