This study developed a machine learning approach to improve satellite estimates of snowpack properties like snow grain size and snow cover in forests. Using high-resolution Landsat data and lower-resolution VIIRS and MODIS images, the authors trained models to separate snow and vegetation signals in coarse satellite pixels. Their method accurately predicted subgrid snow reflectance and improved estimates of snow indices and vegetation measures. This approach helps overcome bias caused by mixed landcover, especially in forests, and supports more reliable daily snow monitoring over large areas using operational satellite data. It is especially useful where high-resolution data are limited.