Sensitivity of water and sediment yield to parameter values and their spatial aggregation using SWAT watershed simulation model

The USDA-ARS and SARH-INIFAP conduct a binational project entitled "Developing New Conservation Management Systems" which objective is to provide agricultural production strategies that preserve natural resources and the environment. Watershed simulation models play an important role in this project because watershed response can be predicted as a function of management decisions in different scenarios. Current modeling technology requires development of large databases to parameterize watershed simulation models. Simplification of this process will contribute significantly in accomplishing the general objective on both sides of the border.; The SWAT model--Soil and Water Assessment Tool--was selected for this project since it allows for long term spatially distributed watershed response simulation. Available records on daily rainfall, surface runoff, and soil erosion from subwatersheds at the Walnut Gulch Experimental Watershed were used to study the effect of parameter value variation and its spatial aggregation on model output accuracy. The methodology included the integration of input databases with information from previous studies and field surveys. Statistical analysis of measured data included the double mass technique, model calibration, sensitivity analysis both the univariate and the multivariate approaches. The effect of aggregating spatially input data on model output accuracy was evaluated to determine the size of subwatershed for which databases must be developed in semiarid environments using the SWAT model.; Results of this research showed that Curve Number is the most influencing parameter for both water and sediment yield. Other parameters that were important include hillslope steepness and those related to channel dimensions and hydraulic properties of channel alluvium. Regarding their spatial aggregation, it was observed that model accuracy is proportional to the number of subwatersheds in a nonlinear manner. Non significant increase in model accuracy was observed when watershed discretization yields mean subwatershed size lower than 1.2 square kilometers. This value is consistent with other studies and it represents the level of spatial aggregation of the input data for which model accuracy reaches its maximum.