Search

We present two statistical models for downscaling flood hazard indicators derived from upscaled shallow water simulations. These downscaling models are based on the decomposition of hazard indicators into linear combinations of spatial patterns obtained from a Principal Component Analysis (PCA). Artificial Neural Networks (ANNs) are used to model the relationship between low resolution (LR) and high resolution (HR) information drawn from hazard indicators. In both statistical models, the PCA features, i.e. the linear weights of the spatial patterns, of the LR hazard indicator are taken as inputs to the ANNs. In the first model, there is one ANN per HR cell where the hazard indicator is to be estimated and the output of the ANN is the hazard indicator value at that cell. In the second model, there is a single ANN for the whole HR mesh whose outputs are the PCA features of the HR hazard indicator. An estimate of the hazard indicator is obtained by combining the ANN’s outputs with the HR spatial patterns. The two statistical downscaling models are evaluated and compared at estimating the water depth and the norm of the unit discharge, two common hazard indicators, on simulations from five synthetic urban configurations and one field-test case. Analyses are carried out in terms of relative absolute errors of the statistical downscaling model with respect to the LR hazard indicator. They show that (i) both statistical downscaling models provide estimates that are more accurate than the LR hazard indicator in most cases and (ii) the second downscaling model yields consistently lower errors for both hazard indicators for all flow scenarios on all configurations considered. The statistical models are three orders of magnitude faster than HR flow simulations. Used in conjunction with upscaled flood models such as porosity models, they appear as a promising operational alternative to direct flood hazard assessment from HR flow simulations.

With CPU times reduced by two to three orders of magnitude compared to shallow water models, porosity models are considered as efficient tools for the modelling of urban floods on the scale of a conurbation. However, they provide only upscaled hydraulic fields that yield unreliable estimates of the flood risk in terms of financial losses and hazard to human lives. Downscaling of the porosity model simulation outputs is thus necessary. The present work puts forward a downscaling approach based on the decomposition of microscopic hydraulic fields into linear combinations of spatial patterns. The coefficients of the linear combinations are predicted with an Artificial Neural Network (ANN) whose input is derived from macroscopic hydraulic fields. Principal Component Analysis is used both to decompose the microscopic fields into linear combinations of spatial patterns and to project the macroscopic fields into lower dimensional features that are fed to the ANN. This global downscaling approach, which reconstruct the whole microscopic field at once, is compared with a local downscaling approach that relies on a similar setup except that each cell of the microscopic field is estimated separately by a dedicated ANN and that there are as many ANNs as cells. The two downscaling approaches are evaluated and compared at estimating the water depth and the norm of the unit discharge on five synthetic urban configurations and one field-test case. The analyses in terms of absolute errors show that the global approach not only provides a valid downscaling scheme but outperforms, in almost all instances, the local approach.