The provision of highly resolved and accurate climate information for climate change impact assessments represents one of the major research and application fields in the climate research and downscaling community. Although regional climate models (RCMs) have shown their capability to reproduce mesoscale and even finer climate variability satisfactorily, they still feature considerable differences between model results and observational data. Concerning threshold related extreme indices, which are often required for climate change impact research projects the respective errors can be non-linearly increased. Empirical-statistical post-processing approaches (model output statistics) offer an immediate pathway to mitigate these model deficiencies. However, by definition statistical approaches are limited to their calibration range and thus cannot adequately account for new extremes in a changing future. This study presents one selected empirical-statistical downscaling and error correction method (DECM), namely quantile mapping (QM). QM is evaluated for its performance regarding one to three day maximum precipitation amounts in European sub-regions. These precipitation indices are essential e.g. for the assessments of floods or landslides. For this purpose, QM is applied to daily precipitation data from various state-of-the-art RCM simulations from the ENSEMBLES project (http://ensembles-eu.org). The evaluation indices are derived from corrected daily data and compared to observations regarding the magnitude of the events, their frequency as well as their temporal sequence. Furthermore, a simple extension of QM is analyzed regarding QM's potential to generate values outside its calibration range. The results strongly recommend the combination of dynamical and statistical approaches for any climate change impact assessments. We show that uncorrected RCMs partly feature drastic errors in magnitude as well as in the number of events. With QM these errors are strongly reduced independent of the RCM considered as QM adapts the entire probability distributions. Errors in the temporal sequence of events are not reduced by QM as it is inherit from the RCM. In addition, we demonstrate that using a simple empirically based constant extrapolation of QM's error correction function, new extremes outside the calibration range can be generated. In our application, these new extremes are even less biased than the uncorrected RCM simulation results. By this means a major drawback of QM is removed and the confidence in QM also for correcting future climate scenarios is increased.

The ENSEMBLES data used in this work was funded by the EU FP6 Integrated Project ENSEMBLES (Contract number 505539) whose support is gratefully acknowledged. In addition we would like to acknowledge the EU FP6 project CLAVIER, the EU FP7 project ACQWA, and the Austrian Climate Research Program (ACRP) project Deucalion for funding this research activity.