A testbed approach for benchmarking multiple gridded snow datasets and their relative value for seasonal runoff prediction

This presentation details the development of an experimental protocol for benchmarking multiple existing snow water equivalent (SWE) datasets. Given the lack of standardized SWE product and model intercomparison, a systematic community evaluation protocol is needed to provide coherent comparisons of existing products. Utilizing SWE estimates from lidar-based ASO (Airborne Snow Observatory) as ‘ground-truth’ (uncertainties notwithstanding), we evaluate the performance of more than a dozen publicly available SWE estimation approaches in the US (including SNODAS, UA SWE, US National Water Model, UCLA SWE, SWEML, NLDAS2 (VIC, Noah, and Mosaic), ERA5-Land, CU SWE, and CONUS404). Over 400 scenes of spatially continuous ASO SWE at the catchment scale are used for benchmarking the aforementioned SWE estimation methods and establishing a protocol for evaluating future products. The approach involved processing SWE products into catchment spatial resolutions, based on a common hydrofabric, to enable standardized cross-product evaluation. Multiple performance metrics are evaluated to quantify performance related to the ASO observations, including the dependence of SWE performance against elevation, aspect and land cover factors. We also assess whether, given their differences in accuracy, different products lead to different predictability for seasonal, basin-scale runoff. The catchment SWE protocol contributes to the NOAA CIROH (Cooperative Institute for Research to Operations in Hydrology) Hydrologic Prediction Testbed. As the collection of standardized results from multiple products and development groups grows, it will enable the tracking the performance and advancement of SWE estimation products, enabling evidence-based review and adoption of new SWE estimation techniques into applications, including operational prediction.