This paper introduces a new method to up-scale dependent loss distributions from natural hazards to higher spatial levels, explicitly incorporating their dependency structure over the aggregation process. The method is applied for flood risk in Europe. Based on this "hybrid convolution" approach, flood loss distributions for nearly all European countries are calculated and presented. Such risk-based estimates of extreme event losses are useful for determining suitable risk management strategies on various spatial levels for different risk bearers. The method is not only applicable for natural disaster risk but can be extended for other cases as well, i.e., where comonotonic risks have to be "summed up" without loss of risk information.