Regression-based downscaling of spatial variability for hydrologic applications

There is an obvious imbalance between, on the one hand, the importance of spatio-temporal variability of precipitation for river flows and, on the other, their representation in current empirical downscaling models that are applied for climate scenarios. The imperfect variability results from incomplete forcing of the large scales. The last IPCC report mentioned three regression-based methods that try to overcome the imperfection of point-wise variability: randomization, inflation, and expanded downscaling, Here, we analyze and compare these methods with respect to their spatial variability and how that relates to river runoff. Using the downscaled temperature and precipitation for observed and simulated large-scale forcings (climate scenarios), we applied the hydrologic model HBV for two river basins in Germany. We discuss the obvious and hidden model imperfections regarding present and future precipitation climate, along with their relevance for runoff. The overall picture is quite diverse, and it appears that temporal characteristics, i.e. time-lagged effects, are at least as important as spatial characteristics. We conclude that, although the models agree in a number of essential projections for river flow, a more consistent picture requires the full spatio-temporal variability as it depends on the large scale atmosphere.