| This dissertation develops a quantitative approach to incorporate the uncertainties in the modeling of Total Maximum Daily Load (TMDL) into TMDL allocations. The developed methodology was demonstrated primarily on a fecal coliform TMDL study at Moore's Creek, Albemarle County, Virginia through accomplishing three research tasks. In the first task, a synthetic flow generation approach for TMDL hydrologic calibration with limited flow data is represented. The performances of three synthetic flow generators, including bootstrapped artificial neural network (BANN), maintenance of variance extension (MOVE), and drainage area ratio (DAR), were evaluated given various small data sets of flow observations and flow predictors spanning different time periods. The results show that BANN outperforms MOVE and DAR in the low and medium flow predictions and that the best performance can be achieved by replacing the highest 10% of BANN synthetic flows with corresponding DAR predictions.; The second task investigates parameter uncertainties and prediction uncertainties in the Moore's Creek HSPF model using the generalized likelihood uncertainty estimation (GLUE) approach. GLUE identified 381 acceptable parameter sets in the Moore's Creek HSPF model, which could lead to various acceptable model predictions.; Finally, a robust optimization approach for TMDL allocations is developed. A robust genetic algorithm (GA) optimization model linked with a response matrix approach was used to incorporate the multiple acceptable parameter sets in the Moore's Creek HSPF model into TMDL allocations and to minimize pollutant load reductions given various levels of reliability with respect to the water quality standards. The trade-offs between reliability levels and total load reductions of allocation scenarios were evaluated, and the optimized load reduction scenarios were compared with the scenario generated by a trial-and-error approach and approved by the USEPA. The results demonstrated the advantage of the robust optimization model, which can more efficiently search for effective and reliable TMDL allocation scenarios than a trial-and-error process. The optimized load reduction scenario requires 30% less load reductions than the scenario approved by the USEPA at the same reliability level. Moreover, this work shows that current TMDL practice of using an arbitrary selected MOS could result in significant under-design or over-design. |
