What Is Natural Disaster Clustering—and Why Does It Matter for the Economy

Jacob Kim-Sherman and Lee Seltzer

Understanding the economic and financial consequences of natural disasters is a major concern for researchers and policymakers. The way in which overlapping natural disaster systems interact, as exemplified by the recent fires in Los Angeles being exacerbated by strong winds, is a major area of study in environmental science but has received comparatively little attention in the economics literature. Examining these potential interactions would likely be important for financial institutions, since such assessments would, in many instances, increase the estimated financial impact of a given natural disaster. In our recent Staff Report, we develop a method of identifying disaster systems in natural disaster data, such as the Spatial Hazard Events and Loss Database (SHELDUS), and use it to argue that the economics and finance literatures may have overlooked some sources of systemic risk.

What Is Clustering?

We define clustering in natural disaster occurrences as the tendency for natural disasters to be concentrated in certain geographic regions and/or short periods of time. In this post, we focus on spatial clustering (that is, clustering in geographic regions), but the Staff Report also presents results on temporal clustering (that is, clustering across time). Clustering could have important implications for how we understand natural disasters if they have spatial spillover effects. For example, if neighboring counties are dependent on common emergency resources for aid in recovery, these resources may be strained if all of these counties are simultaneously affected by disasters. For these reasons, it is important to identify clusters of natural disasters across space in economic data on natural disasters.

A Novel Approach for Identifying Natural Disaster "Clusters"

For a given pair of neighboring counties, we ask whether they both experience damages from the same hazard type in the same month, as reported by SHELDUS. If so, they are treated as being part of the same cluster. This approach is repeated until no additional county pairs can be linked. In the left panel of the figure below, we see the cluster of counties that can be linked to Harris County, Texas (Houston) when Hurricane Harvey struck in August 2017. In comparison, the right-hand panel shows the footprint of counties identified in the official Hurricane Harvey Disaster Declaration by FEMA. The similarity of the identified cluster with FEMA's declared disaster area shows that our algorithm can identify major disasters in the data with reasonable accuracy.

Why Does "Clustering" Matter?

Natural disaster damages data that are aggregated at the cluster level may have different distributional properties compared to standard panel data, which are measured at the county level. The chart below shows a comparison of the damages in the standard SHELDUS disasters data (that is, panel data at the county level), relative to a dataset of damages aggregated at the cluster level using our method. Interestingly, we tend to observe more extreme damages when analyzing damages at the cluster level. The difference between the distributions of damages using county- and cluster-level data highlights how it may be easier to capture the effects of extreme disasters when incorporating clustering into analyses of natural disaster outcomes.

To further explore whether larger clusters tend to be more damaging, we look at how disaster damages vary according to the size of a cluster, measured by the number of affected counties in the cluster. The chart below suggests that as relative size increases, average damages grow quickly. This chart, along with additional analysis in our Staff Report, suggests that counties tend to experience disproportionately more disaster damage when they are part of clusters that experience large amounts of damage. This is suggestive of the existence of the types of disaster damage-related spillover effects that were discussed above.

Lastly, we ask whether certain hazard types appear to cause different levels of damage depending on whether county- or cluster-level data are used. The chart below displays the relationship between average damages for various hazard types based on county-level data and those based on cluster-level aggregates. The scatter points lie above the 45-degree line, implying that all hazard types appear more destructive when using cluster-level aggregates rather than county-level data. This effect is especially pronounced for certain hazard types: Droughts are the ninth-most severe hazard type when using county-level data but are the second-most severe hazard type when aggregating damages to the cluster level, possibly because the average drought occurs in a cluster of about thirty counties, relative to an average cluster size of four counties across all hazard types. Damages from droughts therefore tend to be spread out across more counties. As a result, analyses of disaster damages at the county level may lead researchers to underestimate the severity of certain hazard types when those hazard types tend to occur in large clusters.

Final Words

Inspired by an important idea from the environmental science literature, we develop a method for identifying clusters of disasters. We show that this approach is economically meaningful as it illustrates the heterogeneities in damages by natural disaster type. Failing to account for clustering may have implications for both policymakers and practitioners. For instance, if clustering is ignored, policymakers may insufficiently prepare for certain hazard types that tend to occur in large spatial clusters, such as droughts. Moreover, financial institutions may not correctly quantify natural disaster risk in their portfolios with respect to regions that are potentially exposed to low-probability, high-impact disasters. Finally, if disaster damages are correlated across different regions due to the phenomenon of spatial clustering, it may be difficult to obtain insurance for assets located in such areas. This could increase the likelihood of credit rationing in regions exposed to natural disasters, especially in markets where insurance is important, such as the real estate market. Therefore, our project sheds light on a potential source of systemic risk that banks, insurers, and policymakers may want to take into account.

How to cite this post:
Jacob Kim-Sherman and Lee Seltzer, "What Is Natural Disaster Clustering-and Why Does It Matter for the Economy?," Federal Reserve Bank of New York Liberty Street Economics, September 2, 2025, https://libertystreeteconomics.newyorkfed.org/2025/09/what-is-natural-disaster-clustering-and-why-does-it-matter-for-the-economy/.