Modeling climate change impacts on hydrological variability using an efficient multi-site GCM downscaling method

Abstract:

The coarse resolution of GCM outputs cannot match the high resolution input requirement of hydrological models and thus are inappropriate for impact assessment of climate change. Though numerous downscaling techniques have been used to gap the mismatch, the methods based on single site cannot be used by the distributed hydrological models for hydrological extreme simulation since the flood in one subbasin can be offset by the adjacent ones due to the ignorance of multi-site spatiotemporal correlation of meteorological variables. This study developed a multi-site downscaling method based on a two-stage weather generator (TSWG) through three steps: (i) spatially downscaling GCMs with a transfer function method; (ii) temporally downscaling GCMs with a single-site weather generator; (iii) reconstructing the spatiotemporal correlations with a post-processing and nonparametric shuffle procedure. Five GCMs (CanESM2, CSIRO_3.6.0, GFDL_CM3, HadGEM2-AO and MPI-ESM-LR) under four RCPs (RCP2.6, RCP4.5, RCP6.0 and RCP8.5) were used to generate climate scenarios for the period of 2011-2040. The hydrological simulation was carried out by SWAT in the Jing River catchment on the Loess Plateau. Future annual mean precipitation would change by -7.7% to 9.2%, annual mean maximum and minimum temperature would increase by 1.4-1.8 ℃ and 1.1-1.4 ℃, respectively. Overall, future climate tended to be warmer and drier under most GCMs and RCPs, and this trend would be more significant for flood season; however, the variations of monthly precipitation would be greater than present. The annual mean streamflow would change by -18% to 38% and be more variable. The monthly streamflow would be more variable for most months due to the increase of monthly maximum streamflow and decrease of monthly minimum streamflow. Therefore, the stremflow in the Jing River should be paid more attention for its possible disasters. The multi-site downscaling method proposed in this study is efficient and performs well for its spatiotemporal correlation reconstruction and hydrological variability simulation, which provides a powerful tool for impact assessment of climate changes.