NOAA Storm Events Peril Climatology — County Risk Scores

Turn 25+ years of the NOAA / NCEI Storm Events Database into a clean, per-county severe-weather climatology. For any US state (and optional county filter) over a chosen year range, this actor aggregates every recorded storm event into one row per (county × peril type) with event frequency, casualties, property/crop damage, and a composite peril risk score.

Keyless. Public-domain NOAA data. No scraping of a UI — it reads NOAA's official bulk csv.gz files directly.

Who it's for

• P&C insurance underwriters — pull hail / wind / flood / tornado frequency and loss history for a book of counties before quoting.
• Reinsurers & cat modelers — build a fast county-level peril baseline to sanity-check vendor cat models.
• Roofing, restoration & storm-chasing lead-gen — rank counties by hail and wind-storm frequency and recent worst years.
• CRE & site-selection / catastrophe risk teams — screen candidate sites for the perils that actually hit that county.

What it does (the transform)

NOAA publishes ~60,000 raw storm events per year nationally. This actor:

1. Resolves and downloads NOAA's official yearly StormEvents_details bulk files (gunzipped in-memory).
2. Filters to your states/counties/perils and year range.
3. Rolls events up into a stable (county × peril) climatology row and sorts by peril score.

Example input

{
  "states": ["Colorado"],
  "startYear": 2023,
  "endYear": 2025,
  "eventTypes": ["Hail", "Tornado", "Flash Flood"],
  "includeZones": false,
  "maxResults": 2000
}

Leave eventTypes empty for all perils, add counties: ["08031", "Denver"] to narrow to specific counties (5-digit FIPS or a name substring), and set includeZones: true to also aggregate NWS forecast-zone events (many winter-weather and marine perils are logged by zone rather than county).

Output fields

Each row is one county × peril type:

Field	Meaning
county_fips	5-digit state+county FIPS (pad2(STATE_FIPS)+pad3(CZ_FIPS))
county_name, state, state_abbr	County/zone name and state
event_type	The peril (Hail, Tornado, Flash Flood, Thunderstorm Wind, ...)
years_covered, n_years	Year range label and count of years that had data
event_count, events_per_year	Total events and annualized frequency
deaths, injuries, casualties	Direct+indirect human impact
property_damage_usd, crop_damage_usd, total_damage_usd	Reported damage in USD (K/M/B/T expanded)
damage_per_year_usd	Annualized total damage
worst_year	Year with the highest total damage for this county × peril
peril_score	Composite 0–100 peril risk score (see below)
source_files, source_url	Exact NOAA files used

About peril_score (editorial model)

peril_score is a transparent editorial composite (0–100), not an actuarial loss number. It blends three saturating, log-scaled components:

• Annualized frequency — weight 40%, saturates at 50 events/year
• Casualties — weight 25%, saturates at 1,000
• Total damage — weight 35%, log-scaled, saturates at $1B

Use it to triage and rank counties/perils; use the underlying raw fields (events_per_year, total_damage_usd, casualties) for anything quantitative.

Use as an MCP tool

This actor is a clean, chainable tool for AI agents via mcp.apify.com. Every input and output field is self-describing, so an agent (Claude, Cursor, etc.) can call it to answer questions like "how often does Denver County get damaging hail, and what was the worst year?" and feed the structured result into an underwriting or siting workflow.

FAQ

How do I get hail frequency by county? Set states to your state(s), eventTypes to ["Hail"], pick a year range, and read events_per_year and total_damage_usd per county.

What's the data source? The NOAA / NCEI Storm Events Database (public domain), read from the official bulk csvfiles directory at ncei.noaa.gov. No API key required.

How far back does it go? NOAA Storm Events data is available from 1996 onward. Set startYear accordingly.

County vs. forecast zone? By default only county events (CZ_TYPE="C") are aggregated. Enable includeZones to also include NWS forecast-zone events (CZ_TYPE="Z").

Is the peril score an insurance loss estimate? No — it's an editorial relative-risk heuristic for ranking. Use the raw frequency/damage/casualty fields for quantitative work.

How is it billed? Pay-per-result: one event per output row (one county × peril climatology record).