Emily A. Fogarty, James B. Eisner, Thomas H. Jagger, and Anastasios A. Tsonis
Abstract Hurricanes affecting the United States are examined with methods of network analysis. Network analysis is used in a variety of fields to study relational data, but has yet to be employed to study hurricane climatology. The present work is expository introducing network analysis and showing one way it can be applied to understand regional hurricane activity. The network links coastal locations (termed "nodes") with particular hurricanes (termed "links"). The topology of the network is examined using local and global measures. Results show that certain regions of the coast (like the state of Louisiana) have high occurrence rates, but not necessarily high values of connectivity. Regions with the highest values of connectivity include southwest Florida, northwest Florida, and North Carolina. Virginia, which has a relatively low occurrence rate, is centrally located in the network having a relatively high value of betweenness. Six conditional networks are constructed based on years of below and above average values of important climate variables. Significant differences in the connectivity of the network are noted between phases of the El Nino-Southern Oscillation.
Hurricanes that make landfall in the United States pose a significant threat to life and property. The frequency and intensity of hurricanes at the coast has been studied extensively (Elsner and Kara 1999; Lyons 2004; Keim et al. 2007). In fact, over the long term the United States gets hit on average by one or two hurricanes per year. The strongest hurricanes (category three or higher on the Saffir-Simpson hurricane damage potentsial scale) occur less frequently, with the United States getting hit on average by three every five years. Studies have focused on how the frequency and intensity of coastal hurricanes fluctuate with climate variations (Gray et al. 1993; Lehmiller et al. 1997; Elsner and Jagger 2004; 2006).
J.B. Elsner and T.H. Jagger (eds.), Hurricanes and Climate Change, 153
doi: 10.1007/978-0-387-09410-6, © Springer Science + Business Media, LLC 2009
For instance, it is well known that pre-season values of the North Atlantic oscillation (NAO) portend the risk of hurricanes reaching the United States (Elsner and Jagger 2004). Results from these studies are important for quantifying the risk of a catastrophic hurricane.
While these studies are important for assessing the regional or local risk of a hurricane strike and how it varies with climate, they say nothing about the relationships of risk between regions or how such relationships change with climate variations. For instance, a hurricane moving out of the Caribbean Sea may affect more than one coastal region. Over the long run this introduces correlation between the frequencies of hurricanes at different locations. Knowing which regions tend to get hit in unison can help with risk assessment especially for those in the business of hurricane-related insurance.
Network analysis allows us to examine hurricane landfalls in a relational way. For instance how are Florida hurricanes related to Texas hurricanes, if at all? If every hurricane that strikes Florida goes on to strike Texas or North Carolina, then the risk of losses between Florida and elsewhere is correlated. This is important to know since insurance companies need to diversify their exposure over uncorrelated regions so as to minimize the impact of a single event on their book of business. It is our contention that interesting connections between coastal hurricane paths and climate analysis that have yet to be seen by more conventional approaches might be available through a network analysis.
Some previous studies have considered coastal hurricanes in a relational way. Elsner and Kara (1999) examined the occurrence of hurricanes that hit both Texas and Florida in a single season. They also looked at the occurrence of hurricanes hitting both Florida and North Carolina. They found that while the frequency of Florida to North Carolina hurricanes has remained rather constant, the frequency of Florida to Texas hurricanes decreased during the second half of the 20th century. However, there was no attempt to analyze the complete network of multiple landfalls. In studying typhoons affecting China, Fogarty et al. (2006) used a factor analysis model to understand the correlated risk between coastal provinces. They found that when hurricane activity is high in the southern provinces it tends to be low in the northern provinces and this seesaw in activity is related to the El Nino-Southern Oscillation (ENSO) phenomenon.
Network analysis offers a way to look at the correlated risk of hurricanes in a more direct and systematic way than these previous studies. Here we demonstrate one way network analysis can be applied to understand regional hurricane activity. This is the first such study of its kind so in section 2 we begin with an introduction to the basic ideas behind networks. Following this, in section 3, we examine the data on U.S. landfalls providing summary statistics and plots of frequency. In section 4 we show how to construct an adjacency matrix from an incidence matrix and show how the adjacency matrix leads to a network of landfalls. In section 5 we show how to compute local and global metrics associated with the topology of the network including the diameter and the prestige of individual nodes. In section 6 we examine how these metrics change with climate covariates including the NAO.
Was this article helpful?