Mapping snow cover in forests using optical remote sensing, machine learning and time-lapse photography

The accurate spatial information of snow cover is useful for understanding the impact global warming, and it high significance hydrological disaster prediction, water resources management, climate change research. Normalized Difference Snow Index (NDSI) based approach has been used extensively around world mapping snow, they displayed accuracy in open areas. However, capturing forests remains problematic due to obstruction effects forest canopy, which causes area be seriously underestimated. In this paper, we present a new algorithm on machine learning (ML) technology improve binary (BSC) forests, using remotely sensed surface reflectance ground truth data. A time-lapse photography network with two-hour resolution was established eastern Qilian Mountains northwestern China obtain data both We trained Random Forests (RF) 500-m Moderate Resolution Imaging Spectroradiometer (MODIS) from bands 1–7 generate BSC results (RF-BSC). Then evaluated RF-BSC NDSI-derived maps three different NDSI thresholds (i.e., 0.10, 0.29, 0.40) against ground-truth indicate that proposed performance area, compared NDSI-threshold approach. can retrieve 67% all real pixels, while NDSI-based only detect 8–14%. also find seems sensitive changes solar illumination conditions coverage. This study suggests fusion optical remote sensing ground-based observations an effective improving at regional scales.