When working with extreme hydro-meteorological events, flood in particular, it is common procedure to classify each event by its return period/recurrence interval. This probabilistic measure has the quality to provide a standard and homogeneous measure of the different events that eventually happened in diverse geographical areas, given their relative frequency. So, a 100-year flood is a potentially devastating event both for a flood prone area, and for an area where flood events are extremely rare.
My problem is now how to set a reasonable threshold to group events characterized by a "similar" return period. For instance, let's assume I want to set three classes: mild, middle, and severe flood events, what would be a reasonable return period cut? Is there any similar procedure for structural engineering/water infrastructure design? It would be lovely if you could also provide some references justifying your choice.