Abstracts
Résumé
L'objectif de cette étude est de vérifier le potentiel des images radar à synthèse d'ouverture (RSO) pour estimer l'équivalent en eau du couvert nival sur le bassin de la rivière La Grande (Baie de James, Québec). Il s'agit d'un milieu dominé par une forêt ouverte d'épinettes noires, des brûlis et des tourbières. Cette information intéresse grandement Hydro-Québec qui gère plusieurs installations hydro-électriques dans cette région subarctique. Durant deux ans, six campagnes de terrain ont été réalisées sur le bassin de la rivière La Grande et une dizaine d'images RSO du satellite européen ERS-1 ont été acquises, étalonnées et géoréférencées, afin de déterminer la relation entre les coefficients de rétrodiffusion des images radar (hiver et automne) et la résistance thermique du couvert nival. Cette relation constitue la première partie d'un algorithme d'estimation de l'équivalent en eau. Elle utilise plus spécifiquement le rapport de rétrodiffusion, qui est la différence entre une image avec neige et une image sans neige. La deuxième partie de cette algorithme déduit l'équivalent en eau du couvert de neige à partir de sa résistance thermique et de sa densité, en se basant sur la relation physique établie par les mesures de terrain. L'équivalent en eau du couvert nival a donc été estimé pour quatre images de février et mars 1994 et 1995. L'erreur moyenne sur l'estimation de l'équivalent en eau de la neige au sol est de 2% à 3% (-5 à 7mm) sur l'ensemble des sites d'échantillonnage avec un écart-type de 14 à 19% (-35 à 45mm). Ces résultats ont encouragé Hydro-Québec à poursuivre la recherche avec les données du satellite canadien RADARSAT (opérationnel depuis le 1er avril 1996) et à développer un prototype pour la cartographie de l'équivalent en eau du couvert nival à partir d'images radar.
Mots-clés:
- Équivalent en eau,
- Neige,
- Télédétection,
- Radar,
- RSO
Abstract
The goal of this study was to evaluate the potential of synthetic aperture radar (SAR) images for estimating the snow water equivalent (SWE) on the La Grande river watershed (James Bay area, Québec). This information is of major interest for Hydro-Québec, which exploits many hydroelectric complexes throughout this subarctic region. The La Grande watershed is composed of moderately dense to opened black spruce forests, opened areas, burned areas and peat bogs. Over two years (1994-1995), six field campaigns were carried out on a study site located between the LG4 and Laforge1 reservoir, in the center of the La Grande river watershed. The field measurements were of two types: 20 snow lines (depth, snow water equivalent (SWE), density) and 8 snow profiles (depth, density, grain size, temperature, dielectric constant). With these data, the thermal resistance of the snowpack was calculated for every test-site, using the depth, density and thermal conductivity of each layer.
Concurrently, more than 10 SAR images (European Satellite ERS-1) of the study site were acquired, calibrated and georeferenced. The backscattering coefficients of all winter images were extracted. Using a reference image (snow-free), backscattering ratios were calculated. They are the difference between a winter image and a snow-free image. This process is used to reduce the impact of vegetation and topography. Then, the relationship between the backscattering ratios and the snowpack thermal resistance of february and march 1994 are established, as the first part of an algorithm developed to estimate the snow water equivalent. The second part of the algorithm infers the snowpack water equivalent from its thermal resistance and density, based on the physical relationship established with field data. This approach is based on studies conducted by INRS-Eau in a southern Quebec agricultural area (BERNIER and FORTIN (1998)). The hypothesis are based on the following:
- The snowpack characteristics influence the underlying soil temperature;
- The dielectric constant of the soil varies with the soil temperature under 0°C;
- The radar signal is influenced by the soil dielectric constant;
- Thus, the snowpack characteristics (thermal resistance) influence the radar signal.
However, due to variations of soil humidity on the date of the reference image (september 1994), two slightly different relationships were obtained. One for open areas and open forests and one for burned areas and peat bogs. This shows the importance of using a good reference image, with homogeneous soil conditions. It could be better to obtain an image later in the fall, when the soil is frozen.
The relationships established here are preliminary, as they use a small dataset. It is estimated that a better regression should be obtained with the acquisition of more images and with a greater range of snow characteristics. However, the algorithm is applied to test the applicability of the method.
First, the algorithm was applied on the test-sites, using the images from February and March of 1994 and 1995. The mean error on the estimation of the snow water equivalent is 2% to 3% ( 5 to 7mm), with a deviation of 14% to 19% ( 35 to 45mm). The results are comparable for both years, even if the algorithm is based on 1994 data only. Secondly, the algorithm is applied on the whole images. A classification of a Landsat-TM image is used to identify the land cover of every pixel, which determines the regression and the snow density to be used in the algorithm. Four maps of the SWE are produced and resampled to a resolution of 500m. These are compared with the field measurements from the four nearest Hydro-Quebec snow survey sites. The SWE measured by Hydro-Quebec are all within the most dominant SWE class of each map. Further validation of the results will be possible when the algorithm can be applied on a sub-watershed, which is the actual scale used by Hydro-Quebec.
However, the results of this study were sufficiently promising to Hydro-Quebec to support a follow up research with data from the canadian satellite RADARSAT (operational since april 1996). Meanwhile, to improve the algorithm, it is important to obtain a good and homogeneous reference image, to better assess the impact of the land cover and to acquire a dataset with a greater range of snow characteristics.
Keywords:
- Snow Water Equivalent,
- Snow,
- Remote Sensing,
- Radar,
- SAR