Economic impact of model predictive uncertainty on tmdl assessments: A case study in the Lake Elsinore San Jacinto Watershed

The economic and environmental consequences of over or underestimating model predictive uncertainty in TMDL assessments are potentially significant. Current methods for arbitrarily assigning a margin of safety (MOS) factor to model predictions are likely underestimating uncertainty and misguiding regulators in the TMDL decision making process. In order to investigate the economic impact of varying levels of assumed uncertainty, a hypothetical TMDL for sediment load reduction in the San Jacinto Watershed was selected as a case study. Using the Soil and Water Assessment Tools (SWAT) model, three different assumed amounts of uncertainty, (1) an arbitrary 10% MOS factor, (2) model parameter uncertainty determined by ParaSol, and (3) model predictive uncertainty determined by SUNGLASSES, were assigned to model sediment load predictions. Cost estimates for extended detention basin (EDB) implementation in the watershed were determined for three different TMDL goals based on conservative, moderate, and liberal estimates of sediment loads into Canyon Lake. In general, cost projections for achieving all three TMDL goals increased as the assumed amount of model uncertainty increased. In this study, the 10% MOS factor captured all of the parameter uncertainty associated with SWAT and predicted slightly higher cost estimates for EDB implementation than parameter uncertainty estimates. The 10% MOS factor was shown to only represent about 35% of the total estimated predictive uncertainty in SWAT. This result suggests that regulators are likely underestimating the amount of uncertainty present in model predictions and thus underestimating TMDL abatement costs.