Benchmarking DASSP, a Cloud-Based Downscaling System for 30-Meter Global Soil Moisture Estimates

Knowledge of soil moisture content and soil strength at high resolution can provide critical information to support applications in agriculture, forestry, construction, recreation, and defense. This presentation describes the benchmarking and validation of DASSP (https://mobility.crearecomputing.com/), a hydrometeorological modeling approach and cloud-based computing architecture for generating high-resolution estimates of soil moisture content and soil strength. These ground state conditions are produced with global coverage at spatial scales of tens of meters, and the hydrometeorological modeling approach is applicable to either climatological, current, or forecast weather conditions.  DASSP’s ground state predictions are generated via a physics-based downscaling approach that fuses weather-scale (1/4 degree spatial scale) land surface model estimates of soil moisture and land surface water and energy fluxes, with terabytes of high resolution (1 to 3 arc-second spatial scale) geospatial data including topography, land cover, soil classification, and vegetation information.

This paper presents our approach to benchmarking the accuracy of DASSP's downscaled soil moisture estimates.  Benchmarking requires the selection of appropriate reference soil moisture data sets and the development of relevant metrics for evaluating the similarity between the reference data set and DASSP’s modeled outputs.  For our reference data, we considered a variety of sources, including global collections of in situ point measurements such as the International Soil Moisture Network (ISMN), infrequent but spatially-dense remotely-sensed measurements such as NASA’s Airborne Microwave Observatory of Subcanopy and Subsurface (AirMOSS) campaigns, and high-resolution single-catchment in situ data sets such as the Western and Grayson 1998 Tarrawarra collection.  We discuss the strengths and weaknesses of the different types of data sets for use in benchmarking DASSP’s downscaled soil moisture predictions.  We then quantify the agreement of DASSP’s predictions with the reference data sets.  We found that the single-catchment data sets were most useful for benchmarking, despite their small geographic size, as their reported data were more detailed and consistent and their biases were easier to address than the larger-scale heterogeneous databases that we considered.