Modeling watershed non-point source pollution: complexity, uncertainty and future directions

Non-point source pollution (NPSP) is a major cause of surface water quality degradation. Watershed models (e.g. the Soil and Water Assessment Tool, SWAT) have been increasingly used to simulate NPSP and support the pollution prevention. These models originated from hydrologic models but add significant complexity. Their simulations usually involve substantial uncertainty especially when observational data are scarce, which largely limits the models application. Based on our past and ongoing studies, this presentation discusses the following issues: 1) effective and efficient methods to quantify the uncertainty associated with the model simulations; 2) cost-effective strategies to reduce the uncertainty through data acquisition and assimilation; 3) directions to improve the current NPSP models. While discussing the first issue, Probabilistic Collocation Method (PCM) based approaches of uncertainty analysis (UA) and data assimilation will be presented, and the important role of management concerns in the UA will be discussed. Regarding the second issue, approaches to optimize data acquisition and assimilation, based on the concept of value of information (VOI), will be introduced, and the tradeoff between uncertainty and cost will be discussed. While addressing the last issue, two key points will be made. First, the complexity of the NPSP models does not necessarily lead to good simulation results, but is very likely to introduce significant uncertainty and the parameter identifiability issue. Thus, the model complexity has to be tailored to the data condition. Second, some core modeling assumptions should be re-examined through further studies on the physical process of NPSP. For example, our recent experimental studies showed that the enrichment theory widely adopted in NPSP models has significant limitations. This presentation calls for more efforts on developing a new generation of watershed NPSP models.