UNCERTAINTY IN TMDL MODELS

Abstract

Although the U.S. Congress established the Total Maximum Daily Load (TMDL) program in the original CleanWater Act of 1972, Section 303(d), it did not receive attention until the 1990s. Currently, two methods are available fortracking pollution in the environment and assessing the effectiveness of the TMDL process on improving the quality ofimpaired water bodies: field monitoring and mathematical/computer modeling. Field monitoring may be the most appropriatemethod, but its use is limited due to high costs and extreme spatial and temporal ecosystem variability. Mathematical modelsprovide an alternative to field monitoring that can potentially save time, reduce cost, and minimize the need for testingmanagement alternatives. However, the uncertainty of the model results is a major concern. Uncertainty is defined as theestimated amount by which an observed or calculated value may depart from the true value, and it has important policy,regulatory, and management implications. The source and magnitude of uncertainty and its impact on TMDL assessment hasnot been studied in depth. This article describes the collective experience of scientists and engineers in the assessment ofuncertainty associated with TMDL models. It reviews sources of uncertainty (e.g., input variability, model algorithms, modelcalibration data, and scale), methods of uncertainty evaluation (e.g., first-order approximation, mean value first-orderreliability method, Monte Carlo, Latin hypercube sampling with constrained Monte Carlo, and generalized likelihooduncertainty estimation), and strategies for communicating uncertainty in TMDL models to users. Four case studies arepresented to highlight uncertainty quantification in TMDL models. Results indicate that uncertainty in TMDL models is a realissue and should be taken into consideration not only during the TMDL assessment phase, but also in the design of BMPsduring the TMDL implementation phase. First-order error (FOE) analysis and Monte Carlo simulation (MCS) or anymodified versions of these two basic methods may be used to assess uncertainty. This collective study concludes that a morescientific method to account for uncertainty would be to develop uncertainty probability distribution functions and transfersuch uncertainties to TMDL load allocation through the margin of safety component, which is selected arbitrarily at thepresent time. It is proposed that explicit quantification of uncertainty be made an integral part of the TMDL process. Thiswill benefit private industry, the scientific community, regulatory agencies, and action agencies involved with TMDLdevelopment and implementation.