A novel spatio-temporal statistical downscaling model for hourly rainfall

Finer spatiotemporal resolution rainfall generation is essential for assessing hydrological impacts of climate change on medium and small basins. However, existing models have less attention on the inter-daily connection and the diurnal cycle which can strongly influence the hydrological cycle. To address this problem, a spatiotemporal downscaling model is presented which is capable of reproducing the inter-daily connection, the diurnal cycle, and the statistics on daily and hourly scales. The large-scale datasets, which are obtained from the NCEP/NCAR reanalysis data and the GCMs outputs, and the local rainfall data are analyzed to assess the impacts of climate change on rainfall. The proposed model consists of two steps, the spatial downscaling and temporal downscaling. The spatial downscaling is applied first to obtain the relationship between large-scale weather factors and daily rainfall at station scale using the k-nearest neighbor method. Then, the hourly downscaling of daily rainfall is conducted in the second step using the k-nearest neighbor method with the genetic algorithm and consideration of the inter-daily connection and the diurnal cycle. After the downscaling processes, the changes of rainfall statistics are analyzed for the periods 2046-2065 and 2081-2100 under the A2, A1B and B1 scenarios of CGCM3.1 and BCM2.0. An application to the Shihmen reservoir basin (Taiwan) has shown that the proposed model can accurately reproduce the local rainfall and its statistics on daily and hourly scales. Overall, the results demonstrated that the proposed spatiotemporal downscaling model is a powerful tool for generating hourly rainfall data from large-scale weather factors. The understanding of future changes of rainfall characteristics through this study are also expected to assist the planning and management of water resources systems.