Improving Calibration, Sensitivity and Uncertainty Analysis of Data Based Models of the Environment

Cornell University, Ithaca NY

Investigators

Abstract

0229176 Shoemaker Site-specific pollution transport models are based on field data and the calibration of the parameter values that cannot be directly measured. This combination of field data and calibrated model is a powerful tool in environmental analysis. However, one of the difficulties in using such models is that the calibration process can be computationally very demanding if it is done thoroughly. An even more serious computational obstacle is the quantification of the uncertainty associated with model forecasts of pollution transport. This proposal develops a new computational procedure (acronym ROCUS) that can be applied to a range of pollution transport problems. In the proposed research, ROCUS will be applied to a large watershed in New York, for which the PI's research model results are currently being used in formulating regulatory policy. ROCUS will also be used for anaerobic bioremediation of chlorinated ethenes using field data collected by DOD and an 18 species reactive transport model developed by the PI and her students. The ROCUS procedure involves a group of computationally efficient algorithms for calibration, combined sensitivity and uncertainty analysis that is integrated through the use of response surface methods. One of the hypothesis of the study is that computation time to find good values for model calibration can be significantly reduced by replacing methods currently in use with one or both of the two new optimization algorithms to be investigated. These two methods both involve the use of a Radial Basis Function response surface coupled with a new optimization algorithm. Previously published papers and the PI's computational experiments support this hypothesis. The extent to which this hypothesis is valid will be assessed in the proposed research by a systematic comparison of accuracy and computational speed of the two response surface optimization algorithms with two existing heuristic methods. These tests will be performed on the two environmental applications and on standard global optimization test problems. The applications represented are very serious pollution problems, and the results will add to scientific understanding and to methods for environmental protection. The teaching and research objectives will be obtained by incorporation of research results into two courses the PI teaches. Research infrastructure contributions will follow from the interdisciplinary nature of the research. In addition, the PI has a long history of working to promote underrepresented groups, especially women engineers, and this research project will continue that effort.