Spatial Heterogeneity of Snow Affects Remote Sensing, Modeling, and Data Assimilation Interpretation

High resolution remote sensing of snow depth and snow covered area reveal that spatial heterogeneity of snow is common. However, our models and theory often assume spatial homogeneity of snow. Here we investigate the impact of that spatial heterogeneity on melt rates, streamflow, and data assimilation for forecasting purposes. We use high-resolution lidar from the Airborne Snow Observatory (ASO) to quantify spatial heterogeneity in alpine environments, and high-resolution visible imagery to illustrate variability in arctic and prairie settings. ASO data is then used in streamflow simulations to illustrate the impact the spatial heterogeneity has on data assimilation strategies. We assess streamflow forecasts that assimilate high-resolution snowpack or average snowpack, in models that explicitly or implicitly represent heterogeneity, and in models that are and are not calibrated to observed streamflow. We show that because even calibrated models are often inconsistent with the real spatial heterogeneity and melt rates, directly assimilating high-resolution ASO data does not achieve the best forecast performance possible. When the model is first calibrated using a combination of ASO and streamflow data, then assimilation of high-resolution snow data has a greater benefit. We finally discuss the implications that this heterogeneity has for assimilation of lower-resolution remotely sensed snow products such as passive microwave or optical snow covered area products, and that explicitly modeling that heterogeneity offers the possibility to improve the assimilation of such data for improved snow water equivalent volume and melt rate estimation.