Random forest regression used to downscale MODIS snow cover data

Researchers have developed a method to produce fine-scale, 20-meter fractional snow cover (FSC) by downscaling MODIS Normalized Difference Snow Index (NDSI) data to Sentinel-2 resolution using random forest regression.

While MODIS provides daily global archives at 500m resolution, Sentinel-2 offers higher spatial resolution at 20m but only every five days. The new approach uses dynamic features, such as MODIS NDSI and day-of-year, alongside static topographic features to predict Sentinel-2 NDSI.

Tested at an alpine study site, the model achieved an R2 of 0.795 and an RMSE of 0.155, outperforming common resampling methods. The results indicate that combining topographical data with low-resolution NDSI can improve the characterization of snow cover dynamics in mountain landscapes.