Multi-site weather generators for daily precipitation

To assess the performances of multi-site rainfall generation, three rainfall generation models (i.e., a Richardson-type weather generator and two k-NN models, one with a fixed window and the other with a moving window) were developed and their results were compared. Daily precipitation from nine and six rainfall gauges, respectively, in the Shihmen Reservoir and Tsengwen Reservoir catchments were collected. This study generated 30 sets of daily rainfall series with the same record length as the observed data. Four kinds of analyses regarding rainfall amount, dry and wet days, heavy rainfall events, and spatial rainfall correlation, were conducted to assess the generation performance of the three rainfall generation models. Analytical results reveal that the k-NN models are better than the Richardson-type weather generator, and can preserve spatial rainfall correlation among various sites. Moreover, the k-NN model with moving window is the most proper model for the study area. Thus, the k-NN model with moving window was further applied to generate future daily rainfall under climate change scenarios, with the use of the monthly rainfall change information.