What floods are extreme? Insights from a stochastic description of river flow dynamics

An increasing occurrence of floods with large magnitude has been reported in recent years by both scientific and generalist literature for river catchments around the world. Although reasons exist that could explain these phenomenon (e.g., decadal climatic cycles, growth of sealed surfaces and urbanization, modifications of the climatic forcing), the stochastic character of river flows questions the distinctiveness of these events and the reliability of the available records to estimate magnitude and frequency of floods, that may have actually been more variable than what suggested by observations. A mechanistic-stochastic model of flood frequency curves, which links a lumped description of streamflow generation and dynamics in river catchments to the occurrence of certain flood magnitudes, is used to investigate the possible appearance of discontinuities in the slope of the flood frequency curve (named step changes), typically arising from observing events with very large magnitudes. Findings indicate that step changes are more common in river basins characterized by erratic as opposed to persistent regimes, thus suggesting that hydroclimatic and landscape attributes might control the appearance of this characteristic, and that the interplay between stochasticity of the climate and catchment hydrologic response results in an intrinsically higher flood variability (and linked hazard) for erratic river catchments. The capability of the model to correctly identify the position (i.e., return period) of the step change is rather robust to the use of increasingly shorter data series randomly sampled from the whole of the available observations, indicating that an assessment of the possible existence and likelihood of step changes is feasible if a suitable physically-based description of catchment dynamics is adopted. The proposed approach can be used to objectively distinguish between normal and extreme floods (i.e., those with an exceedance probability lower than the step change), to evaluate the propensity of river basins to generate extreme floods, and to shed light on the actual flooding hazard of rivers and regions where long discharge observations are unavailable.