Spatial downscaling of rainfall fields using a multiple-point geostatistics-based approach

High-resolution gridded rainfall product at sub-daily and kilometer scales is required for many hydrological applications. In ungauged catchments, gridded rainfall data are often obtained through remote sensing, primarily satellites, whose spatial resolution is too coarse and requires to be downscaled to a finer resolution. The challenge is not only to downscale the rainfall intensity but also to downscale the spatial structure of rainfall fields, as both elements are essential for assessing the surface hydrological response. For this purpose, we further developed the stochastic multiple-point geostatistics (MPS) method, which enables the downscaling of long-term coarse-gridded rainfall using only a few years of high-resolution rainfall observations. We describe the methodology and demonstrate an application whereby long time series (1998-2019) of hourly CMORPH rainfall dataset are downscaled from 7 km to 1 km resolution based on training images from the 1-km CMPAS dataset available for a much shorter period (2015-2020), taking the area of Beijing as a case study. We show that the downscaled rainfall fields are following the expected spatial structure. Moreover, the downscaled rainfall intensities are consistent with station-based rainfall observations. And the heavy rainfall intensities at the 99th quantile match those expected due to the change in spatial scale and the application of an areal reduction factor. The results indicate that MPS preserves the spatial structure and downscales rainfall intensities well, especially for heavy rainfall, even if limited high-resolution training data is available. The proposed downscale approach can be applied to other rainfall datasets and in other regions.