Spatiotemporal clustering of streamflow extremes and relevance to flood insurance claims: a stochastic investigation for the contiguous USA

Recent research highlights the importance of Hurst-Kolmogorov dynamics (else known as long-range dependence), characterized by strong correlation and high uncertainty in large scales, in flood risk assessment, particularly in the dynamics of flood occurrence and duration. While several catastrophe modeling professionals nowadays incorporate scenarios that account for previous historical extreme events, traditional flood risk estimation assumes temporal independence of such events, overlooking the role of long-range dependence that has been observed in hydrometeorological processes. This study delves into the validity implications of these assumptions, investigating both the empirical properties of streamflow extremes from the US-CAMELS dataset and the ones of flood insurance claims from the recently published FEMA National Flood Insurance Program database. Analyzing the US-CAMELS dataset, we explore the impact of streamflow’s clustering dynamics on return periods, event duration, and severity of the over-threshold events and corroborate empirical findings with stochastic simulations reproducing the observed dynamics. Results show that for all clustering indices, the divergence between the properties of the observed and the shuffled (randomized, considered as independent) time series is pronounced in many gauges. The latter suggests a deviation from the independence assumption and a clear indication for the existence of clustering in streamflow extremes which is further quantified through a stochastic investigation based on the HK dynamics, indicating a persistent behavior. Furthermore, the apparent existence of clustering mechanisms in streamflow extremes is shown to be associated with spatiotemporal clustering in related insurance claims in the USA, yet with spatially variable patterns reflecting different flood generating mechanisms.