Evaluation of European regional reanalyses and downscalings for precipitation in the Alpine region

Datasets of the observed weather and climate on a regular grid are an important requisite for environmental and climate change research. In this paper we evaluate several European-scale regional reanalyses, developed in the EU project EURO4M, and compare them to existing datasets popularly used in applications today. The evaluation focuses on precipitation in the region of the European Alps, which, due to the marked topographic imprints, poses a challenging test bed. Utilizing a set of statistical indicators, we examine how the datasets represent the spatial pattern, annual cycle, frequency distribution and interannual variations. The evaluated datasets encompass new model-based regional reanalyses (UKMO and HIRLAM) and downscaling datasets (MESAN and MESCAN), one global reanalysis (ERA-Interim), and three station-based interpolation datasets (E-Obs, CRU, GPCC). The reference (APGD) is a gridded interpolation derived from very dense rain-gauge observations. The evaluation provides insight into the relative strengths and limitations of the various datasets and construction concepts. The new model-based regional reanalyses show richer spatial variations in the climatology of daily precipitation compared to the driving global reanalysis (ERA-Interim), and they correct for several unrealistic spatial features seen in global and continental interpolation datasets. But they also show biases, shifts in regional anomalies and inaccuracies in mountain-valley contrasts. Common to both regional reanalyses are overestimates of mean precipitation and wet day frequency, and an underestimate of the frequency of heavy precipitation. The 4DVar UKMO reanalysis shows a better space-time coherency with APGD compared to the 3DVar HIRLAM reanalysis. The accuracy of datasets that explicitly use rain-gauge observations (interpolation and downscaling) is found to be very heterogeneous, depending on the density of available station time series. In regions of high station density, downscaling procedures effectively correct for biases in the underlying reanalysis and reduce RMSE and SEEPS errors by a factor of 2 to 5. In areas where E-Obs and the downscaling datasets have similar input of station data the skill of the two methodologies is comparable. Interannual variations of monthly mean precipitation are highly correlated (> 0.9) with those of the reference for all European-scale datasets and even for meso-beta scale subregions. Our evaluations illustrate that the new regional reanalyses provide a valuable data resource also in a region with complex topography, but many applications will have to involve bias correction and downscaling procedures based on direct observations, ideally at high spatial resolution.