Beyond Snow Depth: Assimilating Measured Albedo and Vegetation Height In Operational iSnobal Modeling

This work presents a unique integration of two remote sensing products into an operational snowpack modeling pipeline, with tangible and clear improvements to the energy balance snowpack model iSnobal. The core operational product is the combination of aerial lidar measured snow depths from the Airborne Snow Observatories (ASO) and density fields from the physically-based energy-balance snowpack model iSnobal. This fusion results in a highly accurate basin-scale snow water equivalent (SWE) product, which is being used by water resource managers in operational decision making in the Western United States. The first new application presented here is real-time albedo measured by ASO coincident with snow depth surveys, which replaces iSnobal albedo fields when available. The second application is the year-to-year static incorporation of ASO measured vegetation heights, which replace and improve upon the existing LandFire dataset. The original application and results are from WY 2023.

M3 Works runs the physically-based snowpack model iSnobal operationally in collaboration with ASO in 29 basins in California, Colorado, and Oregon. Model inputs are from NOAA’s High Resolution Rapid Refresh, as well as GFS 14-day forecasts in a snowpack forecast application. Model outputs are daily, and include all snowpack state variables, as well as products such as Surface Water Input, which is the combination of snowmelt and rain on bare ground (all water inputs to the ground surface). The central mission of the project is to improve information available to water resource managers by creating more accurate information about basin-scale snow water equivalent (SWE) than has previously been available. The fast operational turnaround time of ASO measured snow depths (~72 hours) combined with daily modeled snowpack results from iSnobal gives real-time, high resolution (50m), and highly accurate basin-scale SWE. In the California ASO program, surveys occur up to six times per basin throughout the snow season. Accuracy of snow volume on the order of 10 thousand acre-feet (TAF) can have ramifications for water management in basins such as the Kings and San Joaquin, in which a typical snowpack can peak at 2,000 TAF or more. Snow depth is well constrained by ASO lidar measurements, so any improvements in modeled snow density in iSnobal are important and have immediate impact in reducing SWE uncertainty.

The assimilation of ASO-measured albedo into the real-time iSnobal modeling pipeline was first accomplished in WY2023 in multiple basins in California and Colorado. The default albedo parameterization in iSnobal has long been recognized as not accurately characterizing the energy inputs into the snowpack, and default model behavior has consistently undermelted snowpacks in the ablation season. To cope with this, a date-based albedo decay parameterization was built into iSnobal, but to use it requires knowledge of peak SWE in the basin and the timing of melt. This fact renders it a bulky tool to use operationally in real time. When ASO surveys capture albedo, the model visible and infrared albedos are replaced with measured values in a similar manner as ASO snow depth ingestion into the model, which is a direct replacement of model values. For WY2023 the measured infrared and visible albedos were consistently lower than model albedo values, and the measured albedos were instrumental in constraining snowmelt processes in a water year that saw peak snowpacks that were up to 200% of normal. Compared against side-by-side model runs with the same model configuration and inputs, the inclusion of measured albedo improved model performance by up to ~8% compared to basin total volumes after each flight, and also improved model snow depth R² and RMSE. Measured albedo can also be used to constrain and refine the existing date-based albedo decay methods for times when measured albedos may not be available.

In addition to measured albedo, the iSnobal modeling framework was adapted to include ASO lidar measured vegetation height, which also resulted in meaningful improvements to the snowpack energy balance. Vegetation height is an important control on both radiative and thermal energy fluxes, and is a year-to-year static input to the model. Legacy iSnobal is set up to incorporate vegetation products such as the LandFire dataset. In the course of operational modeling, certain vegetation classes and locations consistently under-melted in the model than was measured by ASO surveys, even with the incorporation of ASO albedo products. Measured vegetation heights are frequently lower than LandFire products, and incorporation of these products improved model performance by up to ~5% compared to basin total volumes after each flight, and also improved model snow depth R² and RMSE.

This work describes the ingestion and application of two remote sensing products into operational snowpack modeling that have previously been more frequently applied in research applications. The ingestion of both products demonstrated clear improvements in model performance, particularly in the melt season, and will be an integral part of the ASO-M3 Works measurements and physically-based modeling collaboration in the future.