Space‐Time Simulation of Hourly Rainfall From Areal Statistics Accounting for Spatial Heterogeneity and Anisotropy

Abstract A stochastic space‐time model for hourly rainfall is proposed with the aim of generating inputs for distributed hydrological models used to estimate extreme floods. The model is based on the multi‐stage procedure adopted in the STREAP model of Paschalis et al. (2013, https://doi.org/10.1002/2013wr014437 ): the first stage generates continuous time series of areal rainfall statistics (intensity, wet area), while the second stage disaggregates these areal statistics into high‐resolution rainfall fields using a single‐step procedure. This paper focuses on the spatial disaggregation process and enhances the STREAP model by introducing two components that cannot be ignored in many regions: spatial heterogeneity and anisotropy. An auto‐regressive spatial Gaussian process is used to account for space‐time dependences. Then, a meta‐model combines the simulated Gaussian fields and the areal rainfall statistics, leading to the generation of anisotropic rainfall fields. Local parameters are also introduced to account for spatial heterogeneity, leading to the definition of new calibration and simulation procedures. A rainfall data set with hourly and kilometric resolution is used to calibrate and evaluate the model in six contrasted regions. The evaluation framework aims at assessing the capacity of the model to reproduce the statistical properties of rainfall at different temporal and spatial scales. In most regions, the introduction of spatial heterogeneity was found to be key to reproduce rainfall properties that are essential for hydrological applications, such as at‐site distributions (including extremes) and the co‐occurrence of extremes within a region at different aggregation time scales.