Abstract
Cloud cover in remotely sensed snow cover is a persistent issue limiting accurate observation of snow cover. Cloud gap filling (CGF) seeks to infill cloud covered pixels, yet methods have historically focused on binary snow detection. Continuous normalized differenced snow index (NDSI) snow cover provides more information about snow cover than binary detection and is a valuable variable for modeling snowmelt runoff. Our goal was to use the dynamic seasonally recurrent snow depletion pattern to produce cloud free continuous NDSI snow cover for the period of 2000-2020 in the Upper Snake River Basin, Wyoming. We test two approaches to CGF for continuous NDSI snow cover, a one-to-one approach where one model is derived for each NDSI snow cover value, and a one-to-many approach where one model is derived to CGF all NDSI snow cover values. We then compared CGF results with the Moderate Resolution Imaging Spectroradiometer (MODIS) CGF product. We demonstrate the dynamic seasonally recurrent snow depletion pattern produces continuous daily NDSI snow cover with 96.23% accuracy over the NDSI snow cover range of 10-50 as calculated against independent cloud free (<10% cloud covered area) images from 2018 to 2020. Our model estimates of snow coverage were up to 10.8% lower than the MODIS CGF product. Accurate daily imagery of snow extent is essential for snowmelt modelers to forecast runoff and manage water resources because up to 1/6th of the world's population relies on water from snowmelt.