Remotely-sensed observational constraints on snow-covered area and snowpack albedo improve snowpack and streamflow simulations in the NOAA National Water Model

Abstract:

The NOAA National Water Model (NWM) simulates major hydrologic processes for operational forecasting and water management across the conterminous United States. With NWM version 3.0 currently in development, accurately simulating snowpack state variables (e.g. snow water equivalent (SWE), albedo) and snowmelt-driven streamflow remains challenging. Errors in snow accumulation, distribution, and melt rate propagate into simulated streamflow time series.

We reduce snow model errors by imposing observation-based constraints on NWM fractional snow-covered area (fSCA) and snowpack albedo and analyzing the impacts on SWE and streamflow. We first conducted a set of parameter sensitivity experiments to identify model parameters that strongly influence snowpack dynamics. Of these parameters, we selected the set that control the shape of the snow depletion curve and seasonal snowpack albedo evolution using 15 years of STC-MODSCAG (Spatially and Temporally Complete MODIS Snow-Covered Area and Grain Size) data over three snow-dominated domains in the western United States (Sierra Nevada, Snake River Basin, Upper Colorado River Basin). We then implemented these derived values as spatially distributed parameters in the NWM’s snow model (Noah-MP). Model simulations in small (<50 square km) test basins within each domain yielded seasonal streamflow time series with lower, earlier peaks--improving agreement with observations. We used a random forest algorithm to explore the major landscape and climatological factors influencing fSCA and albedo parameter variability. Classifiers included mean annual precipitation, air temperature, and wind speed,as well as slope, aspect, and elevation. Results indicate that the majority of variability is explained by these classifiers; however, significant unexplained variance suggests forcing and/or model structural errors remain. Results from these experiments help to qualify NWM snow model skill, improve streamflow forecasting for water management, and inform NWM data assimilation strategies to better accommodate model parameter uncertainties.