Considering observed and future nonstationarities in statistical downscaling of Mediterranean precipitation

Abstract

Winter precipitation in the Mediterranean area for the twenty-first century was statistically downscaled under the explicit consideration of nonstationarities. Nonstationarities arise from substantial modifications of the atmospheric circulation, which lead to significant changes of regional precipitation characteristics. For the detection of nonstationarities in the relationships of the large-scale circulation and regional precipitation in the observational period, statistical model performance under potentially nonstationary conditions was compared to model performance under stationarity. To account for nonstationarity in the future projections, circulation characteristics in general circulation model (GCM) output used to downscale precipitation were also analysed. The correspondence of GCM and observed circulation characteristics was used to specifically select appropriate downscaling models. Statistical model performance was affected by nonstationarities, which was most pronounced not only in the north-eastern Mediterranean regions but also in western Mediterranean North Africa. Furthermore, it was found that variability in the GCM data used for the projections is at least as large as seen in the observational period. This finding underlines the need to explicitly take nonstationarities in statistical downscaling into account. As downscaling result we obtain mainly a reduction of the probability of rain and a rather indifferent pattern regarding the change of the 75 % up to the 95 % quantiles for most regions of the Mediterranean area until the end of the twenty-first century were mainly obtained. However, due to the nonstationarities, results depend strongly on the specific time periods under consideration.