Puerto Rico & USVI
About this dashboard
This dashboard accompanies Modeling Population Vulnerabilities to Climate Change-Driven Hurricanes and Tropical Storms in the North Atlantic Basin (Mullins & Uelmen 2026), posted as a preprint on medRxiv. It maps modeled mortality risk from tropical cyclones at census-block and census-tract scale across 21 Atlantic and Gulf coast states and territories, including Puerto Rico and the U.S. Virgin Islands, under six storm-intensity scenarios (Tropical Storm and Saffir–Simpson Categories 1–5).
Scan to read
Preprint · medRxiv
Modeling Population Vulnerabilities to Climate Change-Driven Hurricanes and Tropical Storms in the North Atlantic Basin
Hurricanes and tropical storms are among the deadliest and costliest disasters in the United States, and climate change is expected to make the strongest storms more frequent. Combining records from 24 of the most damaging storms to strike the U.S., Puerto Rico, and the USVI since 1992 with detailed demographic information for communities within 25 miles of the coast, we measured each storm's rainfall, wind, and flooding and used boosted-tree machine-learning models to estimate mortality risk for every census block. Risk rises with storm severity and concentrates in densely populated coastal cities — notably in Puerto Rico, Louisiana, Florida, and the Mid-Atlantic — and in low-lying inlets, peninsulas, and sounds.
Two geographies, one map. The model estimates risk for every U.S. census block in the study area — 1,186,384 blocks per scenario. Because a million-polygon map is unreadable at national scale, the dashboard is multiscale: zoomed out it shows census tracts (block estimates aggregated to their tract, mapped as the tract mean), and past roughly 1:250,000 it switches automatically to the underlying census blocks. Watch the legend heading — it reads "tracts" or "blocks" so you always know which geography you are reading, and the class ranges update accordingly. Hovering a block shows its 15-digit GEOID and rate; clicking adds population.
The mapped value at both scales is mortality per 100,000 residents. Areas are grouped into quintiles computed within each scenario — fifths of the distribution, ranked 1st through 5th. We label them by quintile rather than by severity adjectives on purpose: these are modeled deaths, and calling any nonzero mortality "very low" implies a reassurance the model does not provide. The classes are also relative, not absolute — boundaries differ between scenarios and between the two geographies, so a 5th-quintile block and a 5th-quintile tract are not the same threshold, and a 1st-quintile area is the lowest-ranked fifth under that scenario rather than an area facing no risk. Click any tract for its values across all six scenarios; block popups report the active scenario. The Share button copies a link restoring your current view. The aim is practical: helping residents, planners, and emergency managers explore local risk and prepare — from evacuation planning to infrastructure hardening and public-health communication.
The 23 storms behind the model. Risk estimates are trained on the observed impacts of 23 historical storms (one boosted-tree model per storm) that struck the U.S., Puerto Rico, and the U.S. Virgin Islands between 1992 and 2024 — from Andrew (1992) through Beryl (2024). Use the Historical storm selector to read each storm's landfall, intensity, deaths, and damage, or open Compare all 23 storms for the full sortable table. A recurring pattern is visible there: for many storms — Frances, Rita, Irma, Michael, Beryl — indirect deaths outnumbered direct ones, reflecting evacuation, prolonged power loss, generator carbon monoxide, and interrupted medical care rather than storm forces themselves. Storm figures come from NOAA/NHC Tropical Cyclone Reports and NOAA NCEI records; each card links to its source report.
Data use & citation. These data are free to use. If you use them in any form — analysis, reporting, teaching, or a derived product — we ask that you cite both the manuscript and the dataset. They are companion records: the manuscript documents the model, the Dryad deposit holds the data it produced, and citing only one leaves the other untraceable.
ManuscriptMullins, S., & Uelmen, J. (2026). Modeling Population Vulnerabilities to Climate Change-Driven Hurricanes and Tropical Storms in the North Atlantic Basin. medRxiv. https://doi.org/10.64898/2026.08.16.26360517
DatasetUelmen, Johnny; Mullins, Shaelyn (2026). Data from: Modeling population vulnerabilities to climate change-driven hurricanes and tropical storms in the North Atlantic Basin [Dataset]. Dryad. https://doi.org/10.5061/dryad.qv9s4mwwq
The preprint has not yet been certified by peer review. This page will be updated with the journal citation once the manuscript clears peer review.
Contact. Questions and media inquiries: Johnny Uelmen — uelmen@wisc.edu.
Build v18-23-storms-2026-09-23. Produced by the Uelmen Disease Ecology Lab, University of Wisconsin–Madison. Data & analysis: Shaelyn Mullins. Source data: uelmen.sharedigm.com/data/hurricane.