С»ÆÊé

Gridded snow water equivalent (SWE) products fill spatial and temporal gaps between direct measurements of SWE. These datasets are commonly produced using physically based modeling, interpolation of in-situ observations, or data fusion approaches that combine models and measurements. As remote sensing capabilities and computational methods have advanced, machine learning has emerged as an additional framework to estimate SWE in space and time. We present SnoLimits, a machine learning derived SWE (and snow depth) product at 500 m resolution in the Colorado Rockies and California Sierra Nevada during the MODIS satellite era (2001–2025). SnoLimits uses a random forest model to estimate SWE anomalies between Snotel stations and surrounding pixels, trained using ASO airborne lidar observations, fractional snow-covered area, physiographic variables, and other time-dependent predictors. A unique aspect of this dataset is that it does not rely on gridded meteorological data for inputs, unlike most existing SWE products. We compare SnoLimits to three existing SWE products (UA SWE, UCLA SWE, and ParBal SWE) to determine how a lidar informed ML product compares to established SWE products.