Global Sensitivity Analysis Methodology Research Articles

Novel global sensitivity analysis methodology accounting for the crucial role of the distribution of input parameters: application to systems biology models

SUMMARYThe reliability of model predictions is affected by multiple sources of uncertainty; therefore, most of the efforts for modeling biological systems include a sensitivity analysis step aiming to identify the key contributors to uncertainty. This generates insight about the robustness of the model to variations in environmental conditions, kinetic parameters, initial concentration of the species, or any other source of uncertainty. Local sensitivities measure the robustness of the model to small perturbations on the inputs around their nominal value. There are several numerical methods for the calculation of local sensitivities, but the calculated values should be identical within the numerical accuracy of the method used. In contrast, as will be shown in this contribution, the results of different global sensitivity analysis methods can be very different and highly dependent on the distribution considered for the inputs under evaluation. In this work, derivative‐based global sensitivities are extended to be able to consider an accurate probability density function for the parameters based on the likelihood function. This strategy enforces the areas of the parameter space most likely to reproduce the desired behavior, minimizing the importance of parameter sets with low probability of being optimal to dominate the sensitivity ranking. A model of a biochemical pathway with three enzymatic steps is used here to illustrate the performance of several relevant global sensitivity analysis methods considering different probability density functions for the parameters and revealing important hints about which method and distribution to choose for each type of model and purpose of the analysis. Copyright © 2012 John Wiley & Sons, Ltd.

Global sensitivity analysis of stochastic computer models with joint metamodels

The global sensitivity analysis method used to quantify the influence of uncertain input variables on the variability in numerical model responses has already been applied to deterministic computer codes; deterministic means here that the same set of input variables always gives the same output value. This paper proposes a global sensitivity analysis methodology for stochastic computer codes, for which the result of each code run is itself random. The framework of the joint modeling of the mean and dispersion of heteroscedastic data is used. To deal with the complexity of computer experiment outputs, nonparametric joint models are discussed and a new Gaussian process-based joint model is proposed. The relevance of these models is analyzed based upon two case studies. Results show that the joint modeling approach yields accurate sensitivity index estimators even when heteroscedasticity is strong.

Sensitivity Analysis of a Fire Spread Model in a Chaparral Landscape

Due to a unique combination of environmental conditions, the chaparral shrublands of southern California are prone to large, intense wildland fires. There is ongoing work in the fire research community to establish whether fuel accumulation or weather conditions are the determining factor in the prevalence of large chaparral fires. This study introduces a framework for contributing a modeling perspective to understanding these alternative hypotheses. As models formalize our understanding of the physical process of fire spread, the sensitivity of the models to the meteorological and fuel inputs should be indicators of their relative importance. A global sensitivity analysis (GSA) was conducted on HFire, a spatially explicit raster model developed for modeling fire spread in chaparral fuels, based on the Rothermel spread equations. The GSA provided a quantitative measure of the importance of each of the model inputs on the predicted fire size. The results indicate that, under extreme weather conditions, wind speed was over three times more influential on predicted fire sizes than any other single model input. This finding supports the idea that fires burning under Santa Ana conditions are primarily driven by high wind speeds. Future research will involve extending the GSA methodology to quantify the relative importance of these inputs in terms of the long-term fire regime in chaparral ecosystems.