Algorithm theoretical basis documentation for an operational snow cover product from Sentinel-2 and Landsat-8 data (Let-it-snow)

This document describes the algorithm of the Let-it-snow (LIS) processing chain to generate the snow cover extent product for the Theia land data center. The algorithm takes as input a Sentinel-2 or Landsat-8 image of surface reflectance corrected from atmospheric and slope effects, the associated cloud mask (level 2A product provided by Theia) and a digital elevation model. The output is a single band raster at the same resolution of the input image giving the snow presence or absence and a cloud mask. The output cloud mask is different from the input cloud mask because some pixels can be reclassified as snow or no-snow by the algorithm.

The snow detection algorithm works in two passes: first the most evident snow cover is detected using a set of conservative thresholds, then these snow pixels are used to determine the lowest elevation of the snow cover. A second pass is performed for the pixels above this elevation with a new set of less conservative thresholds.

The processing chain also generates a vectorized version of the snow mask after pass 1 and 2 and a color composite that is overlaid by these polygons. These secondary products are intended for expert validation purpose.