Dams and Flood Reduction. A modelling study of the effects of dams on flood reduction in Germany.

The main intent of this thesis is the evaluation, through modelling scenarios, of how large-scale water flows patterns in Germany would be modified if reservoirs were missing. This issue is not too popular among the scientific panorama, that, being quite divided on the question of dams bringing more advantages to people and economy or disadvantages to the environment and the ecosystems, often focus their studies on a single catchment or small geographical areas. To evaluate the relationship between rivers discharge and the action of the dams, machine learning tools are exploited to run SALTO. SALTO is a distributed conceptual rainfall‐runoff model on large scale that was first locally and then regionally calibrated with the PASS approach, an algorithm that works without any previous definition of the dominant catchment descriptors controlling regional patterns (Merz, Tarasova and Basso, 2020). The calibration function is defined by a weighted sum of Kling-Gupta efficiency and a metric that focuses on flood quantiles evaluated for the return period of 5 years. With this model it is possible to simulate the runoff both in normal conditions and in a scenario in which the action of the dams is ignored. Analysis of model efficiencies after the local calibration show that three quarters out of the total catchments assume values higher than 0.75, suggesting good performances, and that to larger basins generally correspond lower results. Comparisons between the model efficiency after the regional calibration in normal conditions and when neglecting the action of the dams, show that the second one is in many cases higher than the first, and this reflects the higher complexity that the algorithm face when it considers basins. By effectively plotting hydrographs for selected catchments, the difficulties of the model in reproducing the runoff arise. Nevertheless, their analysis, together with flow duration curves, highlight some peculiarities in the behaviours of the streamflow in those catchments containing dams, that distinguished themselves from the watersheds without barriers. Anyway, the key point of the study is the analysis of flood quantiles (determined for return periods of 2, 5, 10, 20, 50, 100 and 500 years), based on runoff computed in three different cases: from observations, from simulations considering the effects of the dams and from simulations ignoring them. Two kinds of tests are carried out: the computation of relative differences between the first two abovementioned datasets, and the calculation of relative differences between the two distinct simulations. To reach some plausible conclusion, all the catchments for which the second analysis resulted to be a negative number (ambiguity of dams that increase the magnitude of the runoff) or a value greater than 10 (dams that would reduce the flux of tens, hundreds or even higher orders of magnitude) are neglected. With filtered results, it can be concluded that, for the majority of catchments, the model slightly underestimates the runoff with respect to the one relative to observations for smaller return periods up to 50 years; on the other hand, on average, dams make the runoff decrease of around 20% to 30%, according to the different return period. Of course, when considering results of all catchments, many no-senses and absurdities arise, and this imprecise outcome could be imputed to impressive approximation in data or to inadequacies during the calibration procedure.