Assessing the Uncertainty in Watershed Nonpoint Source Pollution Simulations with Probabilistic Collocation Method (pcm)

Complex watershed models are widely used in nonpoint source pollution simulations. However, the simulations often involve significant uncertainty, especially in addressing non-conventional pollutants on which both knowledge and monitoring data are very limited. In this study, a stochastic response surface approach, Probabilistic Collocation Method (PCM), was considered as an efficient alternative to conventional Monte Carlo Simulation (MCS) for uncertainty analysis. A new sensitivity analysis approach PCM-AD was also developed based on PCM, which is a global technique to quantify the variance contribution of each uncertain parameter. The diazinon pollution in the Newport Bay watershed (California) was used as the case study, and a WARMF model was built. The results show that PCM can be adequately applied to the WARMF model, and PCM-AD can quantitatively identify the major sources of parameter uncertainty. Both PCM and PCM-AD are much more efficient, with respect to computational time, than traditional MCS-based approaches, if the number of uncertain parameters is around a couple of decades. It was concluded that response surface approaches like PCM is very useful in assessing the uncertainty of complex water quality modeling, and deserve further research.