Global hydrological models and the Great Alpine Region(GAR): review and performance evaluation of the ISIMIP model ensemble

One of the direct consequences of climate change lies in its impacts on the hydrological cycle: the recurrence of extreme positive and negative anomalies of precipitation is set to become more and more frequent, while the rising of global temperatures is already impacting on freshwater availability,or excess, in many regions of the world. The Great Alpine Region (GAR), i.e. the geographical area comprising the European Alps and their close proximity, is in this context a particularly impacted spot as it is showing high levels of both exposure and vulnerability to climate change. This is especially relevant because of the importance of the Alps for the freshwater resources across the whole continent. In this thesis we perform a thorough data analysis on hydrological data at 0.5°x 0.5° resolution for the ISI-MIP-2a protocol's model ensemble over the period 1971-2000 in the GAR. The purpose of this work is to provide insights into the current capability of Global Hydrological Models (GHMs) and their reliability on the selected scale and area. To achieve that, we perform an exploratory analysis of the data providing graphs and maps of different hydrological signatures for the region. Secondly, we compare the models' output data with a large observational dataset provided by the Global Runoff Data Center (GRDC). Lastly, the models' performance is ranked over two different experiments and we set a link between the performance of the models and the morphological properties of the catchments. To this end, daily runoff generated at each pixel of the model's grid is compared to daily runoff data from 141 catchments of the GAR. Three hydrological signatures are selected to perform the comparison: a quantile for the high flows ($ Q_{95}$ ), one for the low flows ($ Q_5$) and lastly the Mean Monthly Runoff curve. For each signature the modified Kling Gupta Efficiency (KGE') is computed to account for the misfit between the observed and modeled data. The performance on the three signatures allows for a relative ranking of the ensemble participating the project. Results highlight that the ensemble mean tends to outperforms each single model in general but has specific downsides related especially to the representation of the monthly regime curve. All models struggle to represent correctly the low flows quantiles while the high flows quantiles are more consistently reproduced by the models. Mean monthly regime curves rank in the middle of the three indicators and show two main important results: one is the consistent negative bias in mean values across all models, which overall considerably underestimate the total volume of runoff produced, secondly the difficulty of the models to correctly reproduce the timing of both peak and low flows.