Simple approach to emulating complex computer models for global sensitivity analysis. (December 2015)

• Record Type: Journal Article
• Authors: Stanfill, Bryan; Mielenz, Henrike; Clifford, David; Thorburn, Peter
• Is Part Of: Environmental modelling & software. Volume 74(2015:Dec.)
• Page Start: 140
• Page End: 155
• Publication Date: 2015-12
• Subjects: Agricultural simulator -- APSIM -- Generalized additive model -- Meta-model -- Uncertainty quantification; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2015.09.011
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Sensitivity analysis is an important step in understanding how uncertainty is propagated through complex computer models. Unfortunately, the most reliable sensitivity analysis techniques take a significant amount of time to execute due to the large number of computer model evaluations required. Emulators can be used to speed up the process by replacing the computer model with a statistical model that mimics the computer model and is computationally efficient. In this manuscript we propose two emulator-based sensitivity index estimators that require minimal set-up and are computationally inexpensive to compute. We demonstrate their accuracy with computer models that have known sensitivity index values and illustrate their application in practice with the agriculture systems simulator APSIM. Highlights: We propose two methods for estimating sensitivity indices based on a simple emulator. Our methods can easily be implemented in R and do not require prior specification. They are more accurate than existing Monte Carlo methods with the same number of model runs. We illustrate our methods using the agriculture systems simulator APSIM.