<p>Assessment of climate change impacts on hydrological processes is often based on simulations driven by precipitation and temperature series derived from bias-adjusted output from Regional Climate Models (RCMs) using boundary conditions from Global Climate Models (GCMs).&#160; This procedure gives, in principle, locally &#8216;correct&#8217; results, but is also very demanding of time and resources. In some cases, the dynamical downscaling (i.e. RCM) followed by bias adjustment procedures fails to preserve the climate change signal found in the underlying GCM simulations, thus undermining the reliability of the resulting hydrological simulations. As an alternative, we have used the stochastic weather generator D2Gen (Mezghani and Hingray, 2009, J. Hydrol., 377(3&#8211;4): 245&#8211;60) to create multiple realisations of catchment-scale precipitation and temperature data series directly from two GCMs (MPI-ESM-LR and NorESM-M1) for the period 1951-2100. D2Gen builds on a suite of Generalised Linear Models (GLMs) to generate precipitation and temperature (i.e. predictands) as a function of explanatory climate variables (or predictors) derived from the GCM such as surface temperature, sea level pressure, westerly and zonal wind components, relative humidity and total precipitation. In this study, we have applied D2Gen on area-averaged precipitation and temperature data for 18 hydrological catchments distributed across Norway. Weather generation is then undertaken based on the expected mean modelled by the GLM plus a noise component to account for local features and random effects introduced by local physical processes that are otherwise not accounted for.&#160; The weather generator was trained for each catchment based on observed precipitation and temperature series for the period 1985-2014, and stochastic weather generation was then performed to construct catchment-scale precipitation and temperature series for the period 1951-2100 that were further used in hydrological simulations based on the HBV hydrological model for the 18 catchments.&#160;</p><p>Validation of the D2Gen results was based on comparisons with observed annual, seasonal and maximum temperature and precipitation, as well as with observed average annual and maximum annual discharge using 30-year time slices.&#160; Comparisons were also made with projected changes generated from hydrological simulations based on a) EURO-CORDEX RCM simulations (MPI-ESM-LR_SMHI-RCA4 and MPI_CCLM-CM5) for the MPI GCM; and b) high resolution (4 km) simulations with the WRF model driven by a bias-corrected NorESM GCM.&#160; Results suggest that in most catchments the D2gen approach performs equally well or sometimes even better than the traditional &#8216;bias-corrected RCM approach&#8217; in reproducing the 30-year average annual flood during the historical period. We also found that for the projection period, the simulations based directly on the GCM output (via d2gen) tend to give slightly larger projected increases in the average annual flood in rainfall-dominated catchments than does the use of bias-corrected RCM simulations. Overall, the results indicate that the D2Gen weather generator offers a feasible alternative approach for projecting catchment-scale impacts on changes in flood regimes under a changing climate.&#160; It also offers the significant advantage that it can be used directly with the CMIP-6 ensemble of GCMs without the time delay associated with the production of the next round of EURO-CORDEX based simulations.</p>
Climate change scenarios and its impact on water resources of Langtang Khola Basin, Nepal