New Programming Environments for Uncertainty Analysis

Mary C. Hill1, Eileen P. Poeter2, E.R. Banta3, John Doherty4, Steen Christensen5, Richard L. Cooley6,
D.M.Ely7, Justin Babendreier8, George Leavesley3, Matthew Tonkin9, Ray Julich7

1 US Geological Survey USGS), mchill@usgs.gov, Boulder, Colorado, USA
2 Colorado School of Mines & International Ground Water Modeling Center, Golden, Colorado, USA
3 USGS, Lakewood, Colorado, USA
4 Watermark Numerical Computing, Inc., Corinda, Queensland, Australia
5 University of Aarhus, Aarhus, Denmark
6 USGS, retired.
7 USGS, Tacoma, Washington, USA
8 U.S. Environmental Protection Agency, Ecosystems Research Division, National Exposure Research
Laboratory, Environmental Protection Agency, Athens, Georgia 30605, USA
9 SS Papdopulos and Associates, Yarmouth Port, Massachussetts

ABSTRACT

We live in a world of faster computers, better GUI's and visualization technology, increasing international cooperation made possible by new digital infrastructure, new agreements between US federal agencies (such as ISCMEM), new European Union programs (such as Harmoniqua), and greater collaboration between US university scientists as CUAHSI evolves. These changes provide new resources for tackling the difficult job of quantifying how well our models perform. This talk introduces new programming environments that take advantage of these new developments and will change how we develop methods for uncertainty evaluation. For example, the programming environments provided by EPA’s COSU API and USGS’s JUPITER API, and Sensitivity/Optimization Toolbox provide enormous opportunities for faster and more meaningful evaluation of uncertainties. Instead of waiting years for ideas and theories to be compared in the complex circumstances of interest to resource managers, these new programming environments expedite the process. In the new paradigm, unproductive ideas and theories will be revealed more quickly, productive ideas and theories will more quickly be used to address our increasingly difficult water resources problems. As examples, two ideas in JUPITER API applications UCODE_2005 and OPR-PPR are presented: uncertainty correction factors that account for system complexities not represented in models, and PPR and OPR statistics used to identify new data needed to reduce prediction uncertainty.