Climate and Solar Signals in Property Damage Losses from Hurricanes Affecting the United States

Climatologists have constructed statistical models using climate variables to anticipate the level of coastal hurricane activity (Elsner and Jagger 2006) and to account for changes inhurricane intensity (Jagger and Elsner 2006; Elsner et al. 2008). Thus, we speculate that it might be possible to detect climate signals in historical damage losses caused by hurricanes. The purpose of the present paper is to show evidence that specific climate and solar signals are indeed detectable in records of damage losses from hurricanes along the US coastline (Jagger et al. 2010).

Elucidating a connection between environmental conditions and economic threats from natural hazards is an important new and interesting line of inquiry (Leckebusch et al. 2007). Although others have discovered climate signals in damage losses using bivariate relationships including El Nin˜o and wind shear (Katz 2002; Saunders and Lea 2005), this paper is the first to look at the problem from a multivariate perspective. The work is based on a recent study that uses pre-season environmental variables to anticipate insured losses before the start of the hurricane season (Jagger et al. 2008). Here, we examine the multivariate relationship between a set of pre-determined environmental variables and damage losses from hurricane winds.

Our loss modeling strategy is to combine separate models of event counts with event losses. For modeling annual losses, we use a compound Poisson distribution with the Poisson distribution on the annual number of loss events with a log-normal distribution for each loss amount. The annual loss is therefore represented as a random sum of independent losses, with variations in total loss decomposed into two sources, one attributable to variations in the frequency of events and another to variations in loss from individual events. For modeling extreme losses over longer time spans, we use an exceedance model with a Poisson distribution for the annual number of losses exceeding a given threshold and a generalized Pareto distribution (GPD) for the excess loss of each exceedance.

We begin in Sect. 2 by examining the normalized damage loss data. In Sect. 3 we consider the environmental data associated with climate patterns and solar activity. In Sect. 4, we justify our decision to model small and large losses separately. In Sect. 5, we develop a model for annual losses and in Sect. 6 we develop a model for extreme losses. Summary and conclusions are provided in Sect. 7.

Thomas H. Jagger, James B. Elsner & R. King Burch. 2011. Climate and solar signals in property damage losses from hurricanes affecting the United States. Natural Hazards. July 2011, Volume 58, Issue 1, pp 541-557