Multi-sensor/multi-temporal Approaches for Snow Cover Area Monitoring

The overall idea behind the work presented is to combine the use of optical and SAR sensors and utilise the best features of each sensor when possible in order to map snow cover area (SCA) more frequently and with better spatial coverage than would otherwise be possible. Optical remote sensing sensors are able to map snow cover quite accurately, but are limited by clouds. SAR sensors penetrate the clouds, but current satellite-borne sensors are only able to map wet snow accurately. In this paper we describe the methodology developed and the results of applying this for SCA mapping through the snowmelt season 2004 in South Norway. The results include the use of ENVISAT ASAR and Terra MODIS. Common for all the experiments is that the sensor fusion has taken place at the level of geophysical parameters. A few algorithms for multi-sensor time-series processing have been developed. One approach is to analyse each image individually and combine them into a day product. How each image contributes to the day products is controlled by a pixel-by-pixel confidence value that is computed for each image analysed. The confidence algorithm is able to take into account, e.g., information about observation geometry, probability of clouds, prior information about snow state and reliability of the classification. The time series of day products are then combined into a multi-sensor multi-temporal product. The combination of products is done on a pixel-by-pixel basis and controlled by each individual pixel’s confidence and a decay function of time for the product. The “multi-product” should then represent the most likely status of the monitored variable. INTRODUCTION The seasonal snow cover is practically limited to the northern hemisphere. Here, the average snow extent during the winter months ranges from 30 to 40 million km. The water equivalent volume of this snow mass ranges from 2000 to 3000 km. In the mountainous areas and in the whole north of Europe, snowfall is a substantial part of the overall precipitation, e.g., in Finland 27% of the annual average total precipitation is snow. In Norway, about 50% of the precipitation in mountainous areas is snow. Monitoring of the seasonal snow is important for several purposes. In northern regions, the snow may represent more than half the annual runoff, putting specific demands on the models and other tools employed in managing this water resource. Risk of flooding enhances this demand, both in areas with stable winter coverage, and in areas only occasionally covered with snow. Snow covered ground affects the energy exchange processes developing weather and climate, both locally and in large regions, and is an important element in meteorological and climatological modelling tools. The snow pack itself causes avalanches every year in alpine regions, enforces a high priority road clearing service both in cities and in rural areas, and affects many other aspects of human life. Optical remote sensing sensors are able to map snow cover quite accurately, but are limited by clouds. Synthetic Aperture Radar (SAR) sensors penetrate the clouds, but current satellite-borne sensors are only able to map wet snow accurately. The research institutes Norwegian Computing Center (NR) and NORUT IT have together developed algorithms for snow variable mapping applying a combined multi-sensor multi-temporal approach. The overall idea is to utilise the best feaEARSeL eProceedings x, issue/year 2 tures of each sensor when possible in order to map snow variables more frequently and with better spatial coverage than would otherwise be possible. In this paper we briefly describe the variable retrieval algorithms we use for SAR and optical data, and then describe the multi-sensor multi-temporal approach we have developed followed by a presentation of experiments performed and finally discussion and conclusions.