Snow Cover Mapping From Chinese Geostationary Meteorological Satellite FY-2F VISSR-II Data Using Random Forest Classifier

Accurate monitoring of snow cover is essential for understanding climate dynamics and supporting hydrological and environmental management. The Fengyun-2F (FY-2F) satellite is one of China's first geostationary meteorological satellites, equipped with the Stretched Visible and Infrared Spin Scan Radiometer-II (VISSR-II), which provides moderate-resolution imagery at high temporal resolution but has only five spectral channels. To enhance the snow cover mapping capability of the FY-2F VISSR-II sensor, this study develops a novel algorithm in which multi-temporal observations within a single day are first fused to generate a daily synthesis image, thereby reducing cloud contamination. Based on this synthesized image, snow cover detection is then carried out using a random forest model driven by multi-source spatiotemporal features. The algorithm was rigorously evaluated against in-situ snow depth observations, Landsat 8 snow cover maps, MODIS snow cover products, and FY-4 A snow cover product. The results indicate strong performance of the FY-2F snow cover product, with an overall accuracy (OA) of 90.8% and an F1-score (FS) of 88.9% against in-situ measurements, and an OA of 94.6% with an FS of 91.9% compared to Landsat 8 snow cover maps. When validated against the four MODIS-based reference products and FY-4 A snow cover product, the FY-2F snow cover product achieved OA values ranging from 92% to 99%. The novelty of this work lies in the effective integration of auxiliary data to compensate for the limited bands. It provides a valuable reference for snow cover mapping from early geostationary meteorological satellites with a limited number of spectral bands.