Application of ensemble-based systems for snow-mapping using NOAA-AVHRR data over Eastern Canada

Abstract

Common operational snow cover products based on optical or passive microwave sensors (IMS, MODIS SNOWMAP, NOAA GOES+SSM/I, etc.) provide maps of the snow cover extent or fractional snow cover maps. These snow cover products do not provide the probability of observing snow and its uncertainty. This information is crucial in the context of forecasting water supplies to support efficient electricity. This study's objective is to develop probability maps with ensemble-based systems, where probabilities could be used to flag the onset of spring melt. To achieve this, bagging and majority voting were implemented in the snow-mapping procedure using AVHRR-KLM data of Eastern Canada. This consists in generating 100 versions based on a random variation of the six empirical threshold parameters included in the procedure. The probability of a pixel corresponds to the number of times it was identified as snow, no-snow or cloud.