One (nonstatistical) way to handle uncertainty: end-member versions of a groundwater/surface water model for decision support
Abstract: Hydrological models are inherently characterised by uncertainty related to input and calibration data, model structure, and parameters. This uncertainty propagates to model results and, hence, it must be assessed. In this presentation, after an overview of the sources of uncertainty for groundwater models and a brief discussion of advanced statistical techniques for evaluating uncertainty, a simpler method is presented based on end-member models. The core idea is to create multiple model versions by varying the distribution of inputs whose values are both key to the desired predictions and estimable based on available data. In the case study presented, the model objective is to simulate the interactions between a river and a line of shallow wells proposed for its riparian area. The MODFLOW model predicts how much water the wells would draw from the river. The key input is the connectivity of permeable deposits along the river - one end-member model minimizes this connectivity whereas the second maximizes it. This procedure is then evaluated in terms of the sources of uncertainty first described. The talk concludes with a surprise use to which the models were put by a regulatory agency for actual decision support.
Speaker's bio: Daniel T. Feinstein has worked in the field of quantitative hydro-geology for 30 years. He studied groundwater modeling at the New Mexico Institute of Mining and Technology and at the University of Wisconsin-Madison before working as a consultant on remediation projects for Papadopulos & Associates and Geraghty & Miller. In 1997 Dr. Feinstein joined the USGS where he specialized in interpretive studies involving regional groundwater modeling, simulation of groundwater/surface-water interactions, and contaminant transport. He recently headed up a pilot modeling study of the Lake Michigan Basin as part of a USGS program to determine the status, trends, and prospects of water availability in the United States. His current research interests include statistical models which emulate process-driven simulations of groundwater age and stream depletion by wells, and transport models which forecast heat flow under climate change. Dr. Feinstein is an adjunct professor at the Geosciences Department of the University of Wisconsin-Milwaukee and teaches modeling courses in Italy.
(2 MB)