ABSTRACT:

Mountain snowpack in the western United States represents a critical store of terrestrial water, but current observational methods limit our ability to quantify snow water equivalent (SWE). Snow stations provide real-time measurements at sparse points in space, while airborne lidar surveys capture spatially continuous snapshots at infrequent points in time. Together, these datasets enable machine learning to ascertain the spatial patterns of snow from lidar while anchoring estimates to real-time conditions via snow stations. We present SnoLimits, a daily 500 m SWE and snow depth dataset spanning the MODIS era (2001–2026), created using a random forest model trained on physiographic and dynamic predictors. Spatial validation demonstrates SnoLimits outperforms existing products (UASWE, ParBal, UCLA SWE) when compared to withheld lidar surveys, with lower RMSE and higher correlation. Temporal validation at snow stations indicates performance better than or comparable to UCLA SWE. SnoLimits is intended for hydrological modeling, water resource applications, and snow model intercomparisons in Colorado and California. Operational SWE and depth data are available with a two-day latency period.