WATERSHED MODEL PARAMETER SENSITIVITY, AUTO CALIBRATION AND PREDICTION UNCERTAINTY USING INVERSE TECHNIQUES IN TURKEY CREEK WATERSHED

Watershed models need calibration before they are utilized as a decision making tool in the planning and management of water resources. Manual approaches are still used for calibration; however, they are time consuming and require experienced personnel. This paper describes application of an automated approach to estimating the parameters of a Watershed Analysis Risk Management Framework (WARMF) model of the Turkey Creek Watershed in Colorado based on daily stream flow measurements using UCODE. The WARMF model calculates daily runoff, shallow ground water flow and water quality of a river basin. UCODE performs nonlinear regression in which a weighted least-squares objective function is minimized with respect to the parameter values using a modified Gauss-Newton method or a double-dogleg technique and can be used on any application model. Using UCODE it was possible to achieve a good fit to stream flow observations for the Turkey Creek watershed. Simulated daily stream flow matched the observed stream flow fairly well with an R² value of 0.77. Additional advantages of using UCODE include provision to do parameter selection through sensitivity analysis and determination of prediction uncertainties. The parameters with the high sensitivities were selected for estimation, effectively reducing the number of parameters to be considered. The calibrated model was used to predict the effect of changes in land use on stream flow. The changes in land use involved conversion of forest in parts of the watershed to development including commercial and residential use. This scenario was used to assess uncertainty in the predictions produced by the calibrated model. The land use conversion caused an increase in peak flows and a reduction in flow during low flow (base flow) conditions. Application of the linear uncertainty estimation software included with UCODE indicates narrow predictive uncertainty because the confidence intervals were on the order of +/-10% of the predicted values except for few cases, which occurred during low flow conditions.