Item

Abstract

The synthesis of daily flows is a challenging task if non-stationarity needs to be considered. Nonparametric k - nearest neigh-bour (k-nn) resampling techniques have been applied successfully for the generation of stationary daily streamflow before. The objective of this study to build a k-nn bootstrap approach for the simulation of daily streamflow for future climate with a focus on optimal predictor selection. Observed daily streamflow is resampled conditioned on observed and simulated climate variables from regional climate models considering past and future scenarios. The resampling is done at seasonal time scales with subsequent disaggregation to daily values. Different approaches for optimal predictor selection are compared including linear correlation, non-linear partial informational correlation with and without variable transformations. The approach is validated using general flow statistics considering current climate and a pseudo future. The testing of the method is carried out for some catchments in the Harz mountains in Germany comprising five streamflow gauges with long daily observations. Climate data from both observations and the German Weather Service core ensemble of climate models are used for conditioning. Results show that the performance of the approach can be improved with optimal predictor selection and variable transformations. Changes in the flow regime can be captured reasonably well, given that resampling is done solely from historic data.