A comparative study of multiple approaches to soil hydraulic parameter scaling applied at the hillslope scale

Soil hydraulic parameters were upscaled from a 30 m resolution to a 1 km resolution using four different aggregation schemes across the Little Washita watershed in Oklahoma. A topography-based aggregation scheme, a simple homogenization method, a Markov chain Monte Carlo (MCMC)-based stochastic technique, and a Bayesian neural network (BNN) approach to the upscaling problem were analyzed in this study. The equivalence of the upscaled parameters was tested by simulating water flow for the watershed pixels in HYDRUS-3-D, and comparing the resultant soil moisture states with data from the electronically scanned thin array radiometer (ESTAR) airborne sensor during the SGP97 hydrology experiment. The watershed was divided into pixels of 1 km resolution and the effective soil hydraulic parameters obtained for each pixel. The domains were then simulated using the physics-based HYDRUS-3-D platform. Simulated soil moisture states were compared across scales, and the coarse scale values compared against the ESTAR soil moisture data products during the SGP97 hydrology experiment period. Results show considerable correlations between simulated and observed soil moisture states across time, topographic variations, location, elevation, and land cover for techniques that incorporate topographic information in their routines. Results show that the inclusion of topography in the hydraulic parameter scaling algorithm accounts for much of the variability. The topography-based scaling algorithm, followed by the BNN technique, were able to capture much of the variation in soil hydraulic parameters required to generate equivalent soil moisture states in a coarsened domain. The homogenization and MCMC methods, which did not account for topographic variations, performed poorly in providing effective soil hydraulic parameters at the coarse scale.