Downscaling climate projections over large and data sparse regions: Methodological application in the Zambezi River Basin

Climate impact studies often require climate data at a higher space-time resolution than is available from global and regional climate models. Weather generator (WG) models, generally designed for mesoscale applications (e.g., 10(1)-10(5) km(2)), are popular and widely used tools to downscale climate data to finer resolution. One advantage of using WGs is their ability to generate the necessary climate variables for impact studies in data sparse regions. In this study, we evaluate the ability of a previously established state of the art WG (the AWE-GEN-2d model) to perform in data sparse regions that are beyond the mesoscale, using the Zambezi River basin (10(6) km(2)) in southeast Africa as a case study. The AWE-GEN-2d model was calibrated using data from satellite retrievals and climate re-analysis products in place of the absent observational data. An 8-km climate ensemble at hourly resolution, covering the period of 1976-2099 (present climate and RCP4.5 emission scenario from 2020), was then simulated. Using the simulated 30-member ensemble, climate indices for both present and future climates were computed. The high-resolution climate indices allow detailed analysis of the effects of climate change on different areas within the basin. For example, the southwestern area of the basin is predicted to experience the greatest change due to increased temperature, while the southeastern area was found to be already so hot that is less affected (e.g., the number of 'very hot days' per year increase by 18 and 9 days, respectively). Rainfall intensities are found to increase most in the eastern areas of the basin (1 mm.d(-1)) in comparison to the western region (0.3 mm.d(-1)). As demonstrated in this study, AWE-GEN-2d can be calibrated successfully using data from climate reanalysis products in the absence of ground station data and can be applied at larger scales than the mesoscale.