Abstract
Foehn has an impact on various climatological variables like temperature and humidity in the highly populated valleys of the Eastern Alps. With increasing global warming, the question arises as to how well climate projections are able to produce conditions for foehn and how their occurrence changes with climate change. This study uses four XGBoost models to classify foehn in EURO-CORDEX climate models of the CMIP5 generation for localised and widespread foehn events in Rhine and Inn valley located in the Eastern Alps, whereas ERA5 reanalysis data in combination with training data derived from semi-automated weather stations are used in the training process. Weights for individual models are derived by analysing the performance of EURO-CORDEX models in their ability to produce foehn-enabling conditions and considering their independence from each other in the historical period. Bias analysis shows a systematic negative bias in annual foehn occurrence for the GCMs ICHEC-EC-EARTH and MOHC-HadGEM2-ES, whereas others show no systematic bias. The training data confirm other studies by showing that the two selected valleys show differences in foehn occurrence and in the portion of widespread events. A weighted trend analysis for future behaviour of foehn shows a statistically significant increase in widespread events in Rhine valley and a decrease in total events in Inn valley under increasing warming conditions. Further, a shift in foehn seasonality in both valleys can be observed with a higher frequency in the spring months and a lower frequency from July to October, also depending on the climate change signal.