A soil moisture downscaling playground of multiple resolution physics-based simulations

Knowing soil moisture conditions accurately is extremely important for natural hazards prediction, agriculture, and other water resources management practices. Remote sensing products have been used more and more in these contexts. Their main advantage is the spatial coverage, which allows one to obtain continental or even global products. Nevertheless, there are limitations associated with them, such as reduced penetrating depth, impact of cloudiness and snow/ice, and low spatial and temporal resolutions. To compensate for the low spatial resolution, downscaling techniques have been developed that combine different remote sensing products and/or other data considered to affect soil moisture redistribution. The main limitation in their development, is the lack of data to validate the techniques and the final product. Oftentimes in situ measurements are used for the calibration/training and for the testing/verification. These are very sparse, i.e., only available at few locations, and hard to compare directly, as both the satellite products and the downscaled estimates are volumetric and not point estimates.

Here, we create a soil moisture downscaling playground by generating soil moisture estimates with a physics-based hydrological model (ParFlow-CLM) at different resolutions, from a few kilometers to 100 meters. Having continuous gridded estimates of high- and low- resolution soil moisture with a reliable physics-based model, allows us to test and compare different downscaling techniques as well as the impact on the scaling of individual inputs/parameters. As an initial experiment, we model the East Taylor catchment (Colorado, USA) at 100m and 1000m resolution, by only changing the topography (i.e., all other inputs are resolved at 1000m), which is not only the best-known input even at high resolutions, but also the most impactful in soil moisture redistribution. The best performing downscaling technique will allow us, in an operational setup, to run the physics-based model at a coarser resolution but still have a high-resolution product in a computationally inexpensive manner. Beyond our application, the high- and low- resolution simulations generated in this work can be used for the validation of any downscaling technique also applicable with remote sensing products.