Metamodel-Based Sensitivity Analysis: Polynomial Chaos Expansions and Gaussian Processes

Abstract

Global sensitivity analysis is now established as a powerful approach for determining the key random input parameters that drive the uncertainty of model output predictions. Yet the classical computation of the so-called Sobol' indices is based on Monte Carlo simulation, which is not affordable when computationally expensive models are used, as it is the case in most applications in engineering and applied sciences. In this respect metamodels such as polynomial chaos expansions (PCE) and Gaussian processes (GP) have received tremendous attention in the last few years, as they allow one to replace the original, taxing model by a surrogate which is built from an experimental design of limited size. Then the surrogate can be used to compute the sensitivity indices in negligible time. In this chapter an introduction to each technique is given, with an emphasis on their strengths and limitations in the context of global sensitivity analysis. In particular, Sobol' (resp. total Sobol') indices can be computed analytically from the PCE coefficients. In contrast, confidence intervals on sensitivity indices can be derived straightforwardly from the properties of GPs. The performance of the two techniques is finally compared on three well-known analytical benchmarks (Ishigami, G-Sobol and Morris functions) as well as on a realistic engineering application (deflection of a truss structure).

Trade-Offs between Geographic Scale, Cost, and Infrastructure Requirements for Fully Renewable Electricity in Europe

“…To remove the biases these low fidelity model runs introduce, we perform 10 additional runs on the original, high-fidelity resolution and use a multi-fidelity approach 71 to retrieve a single surrogate model for continental and national scales. The estimated cross-validation error of the surrogate model is below 5%, and thus, we deem the surrogate sufficiently accurate 69 to derive total- and first-order Sobol’ indices.…”

“…Because of the high computational requirements, in particular the time our model takes to run, this approach would be prohibitive for our study. Thus, we employ a method described in Sudret 68 and Le Gratiet 69 to perform a polynomial chaos expansion of our original model to derive a surrogate model. We use the MATLAB package UQLab.…”

Tropospheric ozone in CCMI models and Gaussian process emulation to understand biases in the SOCOLv3 chemistry–climate model

“…Greenslade et al (2017) calculate the mean fraction of total tropospheric ozone attributable to STE at three sites between 38 and 69 • S as 1 %-3 %, and show that during individual STE events, over 10 % of tropospheric ozone may be directly transported from the stratosphere. Due to its global tropospheric lifetime of ∼ 22 days, ozone is subject to intercontinental transport (Auvray and Bey, 2005), and this is modulated by decadal climate variability (Lin et al, 2014). Ozone is lost from the troposphere either by dry deposition or photochemical destruction.…”

Resampling strategies to improve surrogate model‐based uncertainty quantification: Application to LES of LS89

“…In order to test the new resampling methods, 3 analytical functions, see Table , with increasing numbers of input dimensions are presented, namely: (i) Rosenbrock ; (ii) Ishigami ; and (iii) g‐function . They are all widely used because they are nonlinear and nonmonotonic.…”