Global sensitivity analysis of a 3D street canyon model—Part I: The development of high dimensional model representations

Abstract

Traditional methods for global uncertainty and sensitivity analysis (SA) such as Monte Carlo methods based on full model simulations are often not suitable for application in environmental modelling due to the nonlinearity and computational complexity of the models. The high dimensional model representation (HDMR) method was developed to express the input–output relationships of a complex model with a high dimensional input space. HDMR provides a model replacement that can be easily employed within global SA. In this work an optimisation method is developed as an extension to the existing set of HDMR tools to improve the accuracy of the mapping process. An application from the field of urban flow and dispersion is chosen to demonstrate the effectiveness of the approach. Model replacements of a k– ε model simulating the flow field in an urban street canyon are constructed. The necessary computational effort for their construction is considerably lower than required by traditional global SA methods. First- and second-order sensitivity indices can then be calculated using the replacements without the need for additional full model runs. Comparison with a large sample of full model runs demonstrates that the output statistics of the full model are well represented by the model replacements. The proposed HDMR method therefore provides a powerful tool for general application to global SA of environmental models.