Testing Predictive Skill of Groundwater Flow and Transport Simulations

Friday, 19 December 2014: 2:10 PM
Mary C Hill, University of Kansas, Department of Geology, Lawrence, KS, United States, Ming Ye, Florida State University, Scientific Computing, Tallahassee, FL, United States, Laura Foglia, Darmstadt University of Technology, Darmstadt, Germany and Dan Lu, Oak Ridge National Laboratory, Oak Ridge, TN, United States
Many environmental systems evolve over decades, centuries and longer, in contrast, for example, to weather predictions, which can be compared to reality weekly, daily, and even hourly; hurricane predictions, which can be compared to reality over weeks and days; and El Nino predictions, which can be compared to reality over years. There are many methods that people suggest using to develop models of environmental systems with long-range consequences, yet rarely are these methods tested for their predictive skill in practical problems. Of interest is how to determine if there are model development methods that tend to produce such long-term predictions with greater skill than other model development methods. Here three tests of groundwater flow and transport are discussed and compared. One test uses spatially defined cross-validation to inspect predictive capability, while two tests use paired complex and simple models. Results are analyzed in the context of how increasing model complexity affects predictive skill for the problems considered.