Long-term Variability in the Arrival Rate of Flood Events as Evidenced by Flood Clustering

Hydrologists have observed that large floods seem to occur in clusters in many records suggesting that flood risk varies from decade-to-decade. This research seeks to demonstrate the existence of flood clustering in the United States. The variance of the inter-arrival time of floods larger than 5-year flood threshold was used as a test statistic for clustering with a regional bootstrap analysis that preserved the cross-correlations among flood flows at all sites in the same year. In the two regions considered in detail, clustering is statistically significant at the 7% level in the Mid-Atlantic region and at the 1.5% level in the geographically larger Southeast Gulf-Coast region, where a stronger regional signal was observed. The level of clustering observed is consistent with a slow periodic variation in flood risk with a 1:2 or 1:3 change in the flood arrival rate, which is very plausible. On the other hand, regional tests of trend and of year-to-year persistence using the lag-one autocorrelation coefficients did not produce a statistical significant result in either region, even at the 10% level.