Latin hypercube sampling with the SESOIL model

The seasonal soil compartment model SESOIL, a one-dimensional vertical transport code for chemicals in the unsaturated soil zone, has been coupled with the Monte Carlo computer code PRISM, which utilizes a Latin hypercube sampling method. Frequency distributions are assigned to each of 64 soil, chemical, and climate input variables for the SESOIL model, and these distributions are randomly sampled to generate N (200, for example) input data sets. The SESOIL model is run by PRISM for each set of input values, and the combined set of model variables and predictions are evaluated statistically by PRISM to summarize the relative influence of input variables on model results. Output frequency distributions for selected SESOIL components are produced. As an initial analysis and to illustrate the PRISM/SESOIL approach, input data were compiled for the model for three sites at different regions of the country (Oak Ridge, Tenn.; Fresno, Calif.; Fargo, N.D.). The chemical chosen for the analysis was trichloroethylene (TCE), which was initially loaded in the soil column at a 60- to 90-cm depth. The soil type at each site was assumed to be identical to the cherty silt loam at Oak Ridge; the only difference in the three data sets was themore » climatic data. Output distributions for TCE mass flux volatilized, TCE mass flux to groundwater, and residual TCE concentration in the lowest soil layer are vastly different for the three sites.« less