Improving Mountain Snow and Land Cover Mapping Using Very-High-Resolution (VHR) Optical Satellite Images and Random Forest Machine Learning Models

Very-high-resolution (VHR) optical imaging satellites can offer precise, accurate, and direct measurements of snow-covered areas (SCA) with sub-meter to meter-scale resolution in regions complex land cover terrain. We explore the potential Maxar WorldView-2 WorldView-3 in-track stereo images (WV) for snow mapping at two sites Western U.S. different regimes, topographies, vegetation, underlying geology. trained random forest models using combinations multispectral bands normalized difference indices (i.e., NDVI) produce maps priority feature classes (snow, shaded snow, water, exposed ground). then created area products from these compared them coarser satellite fractional (fSCA) Landsat (~30 m) MODIS (~500 m). Our generated accurate classifications, even limited available bands. Models on a single image demonstrated model transfer, best results found in-region transfers. Coarser-resolution MODSCAG fSCA identified many more pixels as completely (100% fSCA) than WV fSCA. However, while also snow-free (0% fSCA, only slightly underestimated number pixels. Overall, our demonstrate that strategic observations VHR such complement existing operational data map evolution seasonal cover.