Developing a Regional Model of the Coastal Lowlands Aquifer System—Using Uncertainty Quantification as a Guide
Monday, December 4, 2017: 11:10 a.m.
Traditional model development proceeds from model dataset construction to the process of deterministic history-matching (calibration), where the model inputs are adjusted to adequately reproduce past observations of system state, such as water levels, fluxes, and base flow. In contrast to traditional model development, uncertainty quantification (UQ) will guide the modeling process for the U.S. Geological Survey’s Coastal Lowlands Aquifer System (CLAS) study. The CLAS study area extends geographically along the Gulf of Mexico from the Texas/Mexico border through the Florida panhandle. Because of the large extent of the aquifer system and the importance of the groundwater resource to municipal, industrial, and agricultural supplies in the study area, many local-scale groundwater-flow models have been developed in recent decades. Information from these past models provides a jumping off point for model development through use of model-provided time series of pumping, aquifer properties, recharge, and other model parameters. Thus, the intent of the CLAS study is not to create a model from scratch, but to rely on local knowledge and data to efficiently create a new regional model based on the previous models. Early in the model development, initial estimates of parameter and boundary-condition uncertainty will be used to form the prior information of the model inputs, simply referred to as the prior statistical distribution, and will allow early estimation of model forecast uncertainty as well. As the model is improved, the knowledge of forecast uncertainty will help guide additional model refinement within an iterative process. For example, if an updated hydrogeologic framework dataset is implemented, the UQ may indicate how much that dataset improved the model’s reliability. The resultant model (or model ensemble) will help quantify groundwater resources in the CLAS and provide forecast uncertainty ranges to better evaluate the reliability of the model.