Statistical models for predicting future water quality and flow with uncertain inputs and measurements

As required by section 303(d) of the Clean Water Act, each state must calculate for all impaired waterbodies a Total Maximum Daily Load (TMDL) which specifies the total amount of contaminant load that the waterbody can receive and still meet water quality standards. Three critical deficiencies related to the TMDL process limit water quality improvement, including limited progress developing TMDLs, the lack of a defined Margin of Safety (MOS), and the lack of uncertainty analysis. This thesis develops modeling techniques to address these three TMDL deficiencies in an effort to realize improved national water quality. The work is centered on two problems: estimating a regional minimum MOS for point source TMDLs that accounts for changes in low flow as a result of climate change; and second, developing a water quality model for long-term, low-frequency monitoring records that incorporates uncertainty analysis.;Second, a Bayesian water quality model is developed to create TMDLs from long-term, low-frequency monitoring records. This new approach is designed to increase progress in TMDL development by accommodating previously ignored long-term monitoring records and has the added benefit of propagating uncertainty from model parameters and natural variability through to the predicted results. The Bayesian model is applied to a TMDL calculation in a local, urban watershed with pathogen impairment. The new Bayesian approach is used to specify pathogen load reductions under different flow conditions, with the option of pooling data between flow conditions using a hierarchical structure for watersheds missing data under one flow condition. In addition to demonstrating the new Bayesian model's ability to specify the necessary load reduction under uncertainty, a new technique for predicting future water quality given a control strategy intervention with uncertain performance is introduced. The new Bayesian approach is less data intensive than the existing models and allows for the comparison of intervention strategy in the TMDL process. The policy recommendations and new modeling techniques are presented here for consideration by the US EPA in the creation of technical guidance for TMDL implementation to increase TMDL progress and assure that water quality standards are achieved.;Using a regional regression approach, the change in load of a conservative contaminant resulting from changes in low flow due to climate change is predicted for the Mid-Atlantic region of the United States. The predicted regional change in low flow and contaminant load in the near future due to changes in temperature and precipitation is a predicted decrease of 8%, greater than the recommended 5% TMDL MOS. This predicted change in contaminant load sets a lower bound for the MOS needed to protect existing water quality from anticipated changes in temperature and precipitation. The regional regression approach results in small model error for the near-future climate predictions and is useful for comparing multiple climate scenarios for any region with relatively small changes in the climate in the near future.