 |
2007 Ground Water Summit |
Wednesday, May 2, 2007 : 1:40 p.m.
Assessment and Management of Model Uncertainty
A regional groundwater flow model was developed along the Texas Gulf Coast where historical pumping has been estimated to exceed 1,000,000 acre-ft/year across a 9800 square mile region. Construction and calibration of the model is hampered by a lack of historical pumping records, a scarcity of hydraulic pumping tests, and a paucity of information on historical aquifer water levels. This places limitations on the ability of the calibration process to provide reliable estimates of hydraulic property detail throughout the model domain, which in turn may lead to high levels of predictive uncertainty. To overcome this difficulty, and to make use of the large amount of information available from historical drilling throughout the study area, we developed algorithms for estimating aquifer hydraulic properties using regional-scale parameters constrained by geologic/geophyscial information. Algorithms were also developed to assign model pumping rates based on qualitative information pertaining to its historical magnitude and distribution. Calibration (implemented using PEST ’s regularized inversion capabilities) was then undertaken to estimate the distribution of hydraulic properties throughout the model domain, based on all available data. PEST’s predictive uncertainty analysis capabilities were then employed to determine the uncertainty associated with critical model predictions taking account of (1) the sensitivity of these prediction to various hydraulic properties, boundary conditions and input variables employed by the model, (2) the innate variability of these quantities as assessed through geological and other studies, and (3) the extent to which knowledge gained through the calibration process was able to reduce, or not reduce, the parameters/inputs to which key predictions are most sensitive.