An evaluation of uncertainty in water quality modeling for the Lower Rio Grande River using Qual2E-UNCAS and neural networks (Texas, Mexico)

Modeling of environmental systems such as the Lower Rio Grande River (LRGR) requires a great deal of understanding of the complex processes that take place in the river. While water quality models like Qual2E can simulate the important complex processes taking place in a water body, there remains many factors that affect their reliability. This project was focused on estimating the uncertainties and limitations in water quality modeling for the LRGR using Qual2E-UNCAS (Uncertainty Analysis). Uncertainties associated with modeling certain water quality parameters using Neural Networks for the Lower Rio Grande were also investigated in this project.; Uncertainty in Qual2E for the LRGR was estimated using Qual2E-UNCAS which included Sensitivity Analysis, First Order Error Analysis (FOEA) and Monte Carlo analysis. Sensitivity analysis and FOEA were used to identify the most sensitive parameters in the prediction of dissolved oxygen concentrations and biological oxygen demand within the Reaches of LRGR. Several default values had to be used in modeling the LRGR due to the unavailability of information on input parameters such as flow coefficients and reaction rate constants used in Qual2E. This induced significant uncertainties in modeling, which were evaluated using Sensitivity Analysis. Monte Carlo Simulation was also used to effectively quantify the output uncertainties of Qual2E for the LRGR and compared to other measures of uncertainty. The feasibility of using Artificial Neural Networks (ANNs) to model the water quality parameter chlorophyll-a for the LRGR was evaluated as a result of Qual2E's limitations in modeling non-point source pollution. In this study, ANNs were successfully applied to predict chlorophyll-a in the LRGR. Results of this study quantified the uncertainty in water quality modeling using Qual2E and a stochastic model (ANN) and also provided the scenarios where these two models could be used to effectively model water quality in LRGR.