|
Abstract:
|
We proposed a new probabilistic sensitivity analysis for large multivariate output data with a focus on a computationally demanding model. The use of multivariate Gaussian process as a surrogate model is employed to replace the expensive computer model. To improve the computational efficiency and performance of the surrogate model, compactly supported correlation functions were used, which introduced sparsity into the correlation matrix. The technique adopted a more computational efficient sampling approach by using a robust adaptive Metropolis algorithm to speed up the convergence of the target distribution within a Bayesian framework. The method was used for the computation of first-order and the total contribution of each input parameter to the model output variance. To validate the performance of the proposed method, the method was applied to a multivariate dataset from the IMPRESSIONS Integrated Assessment Platform (IAP2) model which is an extension of the CLIMSAVE IAP. Results show that the proposed methods are efficient and accurate for the computation of global sensitivity indices of complex models with correlated input variables.
|
