Probability-based approaches for incorporating uncertainty into water resource models

Uncertainty in information used to make decisions is unavoidable; however it can be reduced by integrating information from multiple sources, and model techniques incorporating uncertainty and variability can produce more useful probabilistic outcome estimates. This work demonstrates the use of methods for decreasing uncertainty and for using probabilistic outcome data effectively in understanding the water quality and quantity in the Catawba River system in western North Carolina.;Sparse monitoring data and error inherent in water quality models makes the identification of waters not meeting regulatory standards difficult. This work uses the Bayesian Maximum Entropy (BME) method of modern geostatistics to integrate water quality monitoring data together with model predictions to determine the likely status of a water (i.e. impaired or not impaired) and to estimate the level of monitoring needed to characterize the water for regulatory purposes. Although the model predictions used to augment the measured data has a high degree of uncertainty, their inclusion reduces the uncertainty in chlorophyll a estimates enough that the likely impairment status of all sections in all but one reservoir can be determined. For the remaining reservoir, probabilistic predictions of future chlorophyll levels are used to illustrate how monitoring costs can be reduced using a BME framework.;Rainfall-inflow models used for analyzing water availability often have complex forms that can inhibit a thorough analysis of uncertainty in model results because of long model run times and the large number of parameters that are not known with precision. This work demonstrates a rainfall-inflow model that uses reduced spatial and temporal resolution to facilitate model construction and to allow for a robust assessment of model uncertainty. Uncertainty is captured in 2000 116-year inflow scenarios generated using Markov Chain Monte Carlo methods and scenario-specific estimates of model residual error. These scenarios were incorporated into a multi-reservoir management model. Although the median system behavior agrees with prior work that did not include uncertainty, including a distribution of possible outcomes results in a doubling of the estimate of the number of times reservoirs fall below target minimum levels and an increase in the likelihood of reaching critical levels.