Heavy-tailed flood peak distributions: What is the effect of the spatial variability of rainfall and runoff generation?

The distributions of many observed time series of daily precipitation and streamflow show heavy tail behaviour. This means that the occurrence of extreme events has a higher probability than would be the case if the tail was receding exponentially. To avoid underestimating extreme flood events in their occurrence probability or their magnitude, a robust estimation of the tail behaviour is required. However, this is often hindered due to the limited length of time series. One way of overcoming this is to enhance the understanding of the processes that govern the tail behaviour of flood peak distributions. Here, we analyse how the spatial variability of rainfall and runoff generation along with the tail behaviour of rainfall affect the flood peak tail behaviour in catchments of various size. To do so, a modelling chain consisting of a stochastic weather generator and a conceptual rainfall-runoff model is used. For a large synthetic catchment (>100,000 km²), long time series of daily rainfall with varying tail behaviour and varying degree of spatial variability are generated and used as input for the rainfall-runoff model. In the rainfall-runoff model, spatially variable runoff is generated by setting respective model parameters accordingly. The tail behaviour of the simulated precipitation and streamflow time series is characterized with the shape parameter of the Generalized Extreme Value (GEV) distribution.

Our analysis shows that heavy-tailed rainfall tends to result in heavy-tailed flood peak distributions, independent of the catchment size. In contrast, first results regarding the effect of the spatial variability of rainfall on flood peak tail behaviour indicate that this relation varies with the size of the catchment. In large catchments, attenuating effects, for example through river routing, might have a stronger impact than in small basins. Regarding the runoff generation, the tail of flood peak distributions tends to be heavier when a fast runoff component is triggered simultaneously in a larger share of the catchment rather than when this is the case only very localized. This in turn is linked to more homogeneous catchment characteristics and rainfall patterns. The results of this study can help with improving the estimation of occurrence probabilities of extreme flood events.