SENSITIVITY AND UNCERTAINTY ANALYSIS OF A DISTRIBUTED WATERSHED MODEL FOR THE TMDL PROCESS

The most commonly used environmental models sometimes require hundreds of parameters. This is especially true of distributed watershed models, which have parameters that spatially vary. It is not reasonable to assume that modelers will know each and every parameter value with certainty. In addition, in many cases, it is possible to obtain “equivalent” calibration results, given different combinations of parameter values.The objective of this paper is to describe a methodology to quantify the effect of multiple parameter uncertainty on the calibration output of a distributed watershed model and to present specific results for the Cannonsville Watershed. In addition, the impact of this uncertainty on the TMDL process will be investigated.A distributed watershed model, the Soil and Water Assessment Tool (SWAT Version 2000) was calibrated and validated for the Cannonsville Reservoir watershed. The Cannonsville Reservoir, which serves as a source of drinking water to New York City, has been under a ‘phosphorous-loading restriction’ due to the high phosphorus concentrations within the water body. These high concentrations are attributed to the non-point source loads from the agricultural land in the basin. The SWAT model was developed to evaluate Best Management Practices (BMPs) proposed to abate this pollution problem.After calibration, a sensitivity analysis was performed to investigate the impact of parameter changes on model output. Particular attention was given to the spatially distributed nature of the parameters, especially those related to land use and soil type. A computationally feasible algorithm and methodology for analyzing the influence of over 150 parameters on various output variables is introduced. Conclusions concerning the “most sensitive” parameters for SWAT in this study area are presented. The most sensitive parameters were then used as decision variables in a uncertainty analysis to determine the parameter combinations needed for an acceptable calibration. Results of both the sensitivity analysis and uncertainty analysis will be presented for the Cannonsville watershed.