Simultaneous Heat and Water (SHAW) Model: Model Use, Calibration, and Validation

Abstract

A discussion of calibration and validation procedures used for the Simultaneous Heat and Water (SHAW) model is presented. Three calibration approaches are presented and compared for simulating soil water content. Approaches included a stepwise local search methodology, trial-and-error calibration, and an automated multi-objective parameter optimization algorithm; the automated algorithm was used to optimize four separate parameter sets with increasing vertical heterogeneity in the soil profile, i.e., considering from one to four soil layers, respectively, within the soil profile. In the stepwise approach, parameters for each soil horizon were individually varied to determine which parameter could minimize the root mean square deviation (RMSD) between measured and simulated soil water content of the top 20 cm. Subsequently, all other parameters were varied while holding constant the parameter that minimized the RMSD in the previous iteration. Iterations continued until the RMSD was minimized. For the trial-and-error calibration, plots of simulated and measured soil water content were examined, and soil parameters of each soil horizon or individual soil layers were varied to obtain a better fit and to minimize RMSD of the top 20 cm as well as the top 60 cm. The automated multi-objective parameter optimization algorithm searched throughout a feasible parameter space for parameter combinations that minimized each of several RMSD objective functions, and then effectively minimized the tradeoffs between the objective functions. Variation in simulated daily soil water content between the simulations ranged from 0.018 to 0.026 m3 m-3 at the different depths, with more variability between simulations being observed within the top 10 cm. Much of the variability between the calibrated simulations was attributed to the calibrations that assumed uniform properties in the 0-75 cm soil horizon, i.e., the stepwise calibration and the single-layer automated optimization; variation between these and the other simulations ranged as high as 0.030 to 0.043 m3 m-3 near the surface. Advantages and disadvantages of the three calibration approaches are discussed.