We present two statistical models for downscaling flood hazard indicators derived from upscaledshallow water simulations. These downscaling models are based on the decomposition of hazardindicators into linear combinations of spatial patterns obtained from a Principal ComponentAnalysis (PCA). Artificial Neural Networks (ANNs) are used to model the relationship betweenlow resolution (LR) and high resolution (HR) information drawn from hazard indicators. Inboth statistical models, the PCA features, i.e. the linear weights of the spatial patterns, of theLR hazard indicator are taken as inputs to the ANNs. In the first model, there is one ANNper HR cell where the hazard indicator is to be estimated and the output of the ANN is thehazard indicator value at that cell. In the second model, there is a single ANN for the wholeHR mesh whose outputs are the PCA features of the HR hazard indicator. An estimate of thehazard 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 waterdepth and the norm of the unit discharge, two common hazard indicators, on simulations fromfive synthetic urban configurations and one field-test case. Analyses are carried out in termsof relative absolute errors of the statistical downscaling model with respect to the LR hazardindicator. They show that (i) both statistical downscaling models provide estimates that aremore accurate than the LR hazard indicator in most cases and (ii) the second downscalingmodel yields consistently lower errors for both hazard indicators for all flow scenarios on allconfigurations considered. The statistical models are three orders of magnitude faster than HRflow simulations. Used in conjunction with upscaled flood models such as porosity models, theyappear as a promising operational alternative to direct flood hazard assessment from HR flowsimulations.