Safety Index methodology

How the score is built, where the numbers come from, and the limits of the data.

The composite score

Each park gets a single 0–100 Safety Index. Higher is safer. The score is a weighted average of two subscores:

ComponentWeightSource
Hazard Score60%NOAA Storm Events Database (2024)
Crime Score40%FBI Crime Data Explorer (2023)

We weight hazards higher than crime because campers die from floods, wildfires, and severe storms much more often than from violent crime — and because rural-county FBI numbers don't tell you much about a remote campground.

Hazard Score

We count NOAA Storm Events recorded in each state during 2024 and weight them by likely impact on a camper:

  • Wildfire ×3.5
  • Flash Flood ×3.0 · Flood ×2.5 · Storm Surge ×2.5
  • Hurricane / Typhoon ×4.0 · Tropical Storm ×2.0
  • Tornado ×2.0 · Avalanche / Landslide / Debris Flow ×2.0
  • Lightning ×1.5 · Excessive Heat ×1.5
  • Hail ×1.0 · High Wind ×0.6 · Thunderstorm Wind ×0.8
  • Winter Storm / Blizzard / Ice Storm ×0.5–0.8

The state's total weighted hazard count maps to a subscore from 0 to 100 on a fixed bucket scale (states under 100 weighted events score 95+; states above 3,000 score below 25). Texas (≈6,000 weighted events) lands at the bottom; Rhode Island (≈75) lands at the top.

Crime Score

We pull monthly violent-crime rates per 100,000 population from the FBI's Crime Data Explorer for 2023, then sum across the 12 months for an annual rate. Brackets:

  • Under 150 / 100k → score 100 (very low)
  • 150–250 → 88 (low)
  • 250–400 → 70 (moderate)
  • 400–600 → 50 (elevated)
  • 600–850 → 30 (high)
  • 850+ → 12 (very high)

Granularity

Each park's coordinates are reverse-geocoded to a county via the US Census Geocoder. Hazard incidents are then filtered to that county's NOAA records — so two parks in the same state but different regions don't share the same hazard number. Crime data currently stays at state level (county-level violent-crime rates aren't cleanly exposed by the FBI CDE API; on the roadmap to add via agency-level aggregation).

Limitations

  • NOAA reporting density. Counties with denser weather-spotter networks report more events, which can bias rankings slightly. We mitigate this with the event-type weights — a tornado outweighs ten reported high winds.
  • One year of weather data. The current build uses 2024. Climatologies should ideally be 10-year averages; we'll roll forward as data permits.
  • No park-specific incidents. NPS doesn't publish park-by-park incident data. If you have a source, we want it.
  • Multi-state parks. Yellowstone spans WY/MT/ID; we use the geocoded centroid (typically the most-developed county) as the primary location.

Data sources

Index built Sun, 26 Apr 2026 07:51:41 GMT. Source data refreshes weekly.