Identification of flood events in large discharge datasets - reinsurance industry perspective

The insurance sector plays a critical role in promoting disaster resilience and recovery by providing financial protection, speeding up rebuilding and recovery, and managing the financial impact of natural disasters. To fulfill this role, insurance companies must meet the capital requirements imposed by regulators. For example, the European Solvency II regulatory framework requires insurers to hold enough capital to withstand a natural catastrophe loss with a return period of 1 in 200 years. As the historical loss data are scarce and incomplete, the insurance sector uses stochastic catastrophe models (cat models) to assess the potential cost of rare but devastating events like floods.

A stochastic event set is a crucial element of cat models. It is a collection of possible disasters with their likelihood and severity. One method to generate stochastic flood events is to use numerical models of the atmosphere to generate realistic precipitation fields, and then apply rainfall-runoff models to estimate how much water will flow into rivers and streams from precipitation and snowmelt. By running many simulations with different inputs and parameters, stochastic flood models can provide a range of possible outcomes, including floods with spatial patterns and magnitude missing in historical data.

Output of such simulations are spatio-temporal hazard grids: precipitation grids for pluvial risk and river discharge grids for fluvial risk. These grids are large as the models typically run over large geographies (countries or continents) and simulate 10,000 years or more. This contribution will (i) provide overview of existing methods how to identify flood events in such huge discharge and precipitation datasets (i.e. peak-over threshold method), (ii) show their limitations for identifying flood events, and finally (iii) propose a new methodology designed to address specific needs of reinsurance industry such as the hours-clause condition, which specifies the time period within which losses from a single event must occur in order to be covered.

As many severe floods are composed from several sub-waves (for example 2002 floods in Czech Republic), proper event identification and separation is highly relevant topic as it influences the amount of reinsurance payouts after some types of flood events and thus capital available for rebuilding and recovery.