Comparison of global sensitivity analysis methods for a fire spread model with a segmented characteristic

Abstract

Global sensitivity analysis (GSA) can provide rich information for controlling output uncertainty. In practical applications, segmented models are commonly used to describe an abrupt model change. For segmented models, the complicated uncertainty propagation during the transition region may lead to different importance rankings of different GSA methods. If an unsuitable GSA method is applied, misleading results will be obtained, resulting in suboptimal or even wrong decisions. In this paper, four GSA indices, i.e., Sobol index, mutual information, delta index and PAWN index, are applied for a segmented fire spread model (Dry Eucalypt). The results show that four GSA indices give different importance rankings during the transition region since segmented characteristics affect different GSA indices in different ways. We suggest that analysts should rely on the results of different GSA indices according to their practical purpose, especially when making decisions for segmented models during the transition region. All of our source codes are publicly available at https://github.com/dirge1/GSA_segmented.

References in corpus (4)

• Estimating Mutual Information
• Principal component analysis and sparse polynomial chaos expansions for global sensitivity analysis and model calibration: application to urban drainage simulation
• Variance-based sensitivity analysis: The quest for better estimators and designs between explorativity and economy
• Estimation of mutual information for real-valued data with error bars and controlled bias

Cited by in corpus (1)

• A new moment-independent uncertainty importance measure based on cumulative residual entropy for developing uncertainty reduction strategies