NOAA Storm Events API — Daily NCEI Storm-Event Diff & Loss-Revision Feed

Daily diff of the NOAA NCEI Storm Events Database for re-insurance cat-bond desks, P&C loss-reserve teams, climate researchers, and AI severe-weather agents. Watches the official NCEI bulk CSV feed for damage / fatality / F-scale / narrative revisions and emits one JSONL change record per detected change.

What this does

• Tracks every storm event in the NOAA NCEI Storm Events Database with full historical diff coverage (property damage / crop damage / injuries / deaths / magnitude / tornado F-scale / narrative landing / event-type revisions).
• Cuts loss-revision detection latency from weeks to one day so cat-bond underwriters and loss-reserve teams catch preliminary→final damage figures and F-scale upgrades the same day NCEI re-publishes.
• Replaces $25-50k/yr commercial cat-modeling subscriptions (CoreLogic Hazard HQ, Verisk AIR Worldwide) with a metered pay-per-result Apify actor at $3-540/month per peril basis.

Use cases

NOAA storm events database API for re-insurance cat-bond desks

A cat-bond underwriting desk tracking a parametric-trigger basket of tornadoes EF3+ in Tornado Alley states (TX / OK / KS / NE / IA) filters this actor on event_types: ["Tornado"] +

states: ["TEXAS", "OKLAHOMA", "KANSAS", "NEBRASKA", "IOWA"]

Storm events damage revision feed for P&C loss-reserve audit

A regional P&C insurer's reserving actuary filters on event_ids: [<portfolio-affected events>] +

change_types: ["damage_property", "damage_crops", "narrative_landed"]

NCEI storm events diff for climate-research methodology audit

A university climate-research group studying NCEI's mean preliminary→final correction lag filters on

change_types: ["tor_f_scale", "narrative_landed", "event_type_revised"]

Cat bond loss revision tracker for AI severe-weather agents

A severe-weather AI agent embedded in a parametric-insurance platform consumes the diff stream as a retrieval-augmented fact source. Each diff record is a structured event the agent can reason over ("NCEI just upgraded the May-15 Norman tornado from EFU to EF3 with $42M property damage finalization") rather than an unstructured "NCEI re-published the 2025 file" HTML page. The agent uses change records to prompt the platform's underwriters proactively rather than waiting for the next quarterly cat-bond reset.

Input

See .actor/input_schema.json for the full schema. Key fields:

Realistic example input (Tornado Alley severe-season basket):

{
  "event_types": ["Tornado"],
  "states": ["TEXAS", "OKLAHOMA", "KANSAS", "NEBRASKA", "IOWA"],
  "change_types": ["damage_property", "tor_f_scale", "deaths_total", "narrative_landed"],
  "year": 2026
}

Output

JSONL dataset; one record per detected change:

{
  "actor": "actor-4-noaa-storm-events-diff",
  "schema_version": "1.0.0",
  "extracted_at": "2026-05-22T12:00:00Z",
  "source_url": "https://www.ncei.noaa.gov/stormevents/listevents.jsp?eventType=ALL&eventId=1234567",
  "event_id": "1234567",
  "episode_id": "200001",
  "state": "Texas",
  "event_type": "Tornado",
  "begin_date": "10-MAY-26 17:32:00",
  "change_type": "tor_f_scale",
  "before": {"scale": "EFU"},
  "after":  {"scale": "EF2"},
  "detected_at": "2026-05-22T12:00:00Z"
}

Apify Console preview shows a 7-column table view (event ID / state / event type / begin date / change type / detected at / source URL) so non-engineers can sanity-check runs before piping into BI.

Pricing & limits

• $0.02 per result (per detected diff record).
• Free 7-day trial so you can verify the diff fidelity against your own EVENT_ID watchlist before committing.
• Recommended schedule: daily during US severe-weather season (April–November); weekly off-season for older-year retroactive re-publishes.
• Estimated monthly cost:
  • Low-end research user (~30 EVENT_IDs, weekly cadence): $0.50–$1/mo
  • Mid-market loss-reserve audit (~200 events, daily cadence): $90–540/mo
  • Re-insurance cat-bond desk (~2,000 events, daily cadence): $360–2,700/mo
  • Replaces CoreLogic Hazard HQ ($25k/yr) and Verisk AIR Worldwide ($50k+/yr) at substantially lower cost with full per-event diff fidelity rather than monthly aggregate tables.

Data source & freshness

• Source: NOAA NCEI Storm Events Database public bulk CSV feed (StormEvents_details-ftp_v1.0_dYYYY_cYYYYMMDD.csv.gz files at https://www.ncei.noaa.gov/pub/data/swdi/stormevents/csvfiles/).
• Update cadence (source-side): NCEI re-publishes the current year's file roughly every 2–4 weeks during severe-weather season as NWS field offices submit damage surveys, narrative finalizations, and F-scale upgrades. Older years (5+ years back) re-publish 1–2× per year on methodology updates.
• Actor cadence (recommended): Daily 06:00 UTC during severe season; weekly off-season. The actor compares against the previous snapshot — running more often than NCEI re-publishes is harmless but emits zero records.
• Public-domain license: NOAA NCEI Storm Events Database is US government public-domain under 17 USC §105 (no copyright on government works). NCEI imposes no programmatic-redistribution restriction. Actor self-limits to a polite 1-rps default.

FAQ

What's the difference between the NOAA Storm Events Database and the NWS Storm Prediction Center API?

The Storm Events Database is NCEI's authoritative post-event archive — every confirmed severe-weather event with finalized damage / fatality / F-scale data. The SPC API emits forecast/preliminary products (convective outlooks, watches, tornado-warning polygons) in near-real-time. This actor consumes the post-event archive, not the forecast feed. For real-time warning ingestion, use a separate watch-and-warning actor or the official NWS API.

How does this compare to the NCEI Search API?

The NCEI Search API (ncei.noaa.gov/cdo-web/api/v2/) returns metadata records, not bulk event detail rows. The bulk CSV feed this actor consumes carries the full ~50-column event record including DAMAGE_PROPERTY / DAMAGE_CROPS / NARRATIVE_* fields the Search API does not expose. Use the Search API to discover event IDs by station / location / date range; use this actor to track diffs over time.

What's an NCEI publish-stamp and why does it matter?

NCEI's filename embeds a cYYYYMMDD stamp marking the publish date. When NCEI re-publishes a year's file, the stamp changes and the filename is the only signal that the underlying data shifted. The actor records the chosen stamp in its run summary so buyers can correlate diff records with publish events end-to-end.

How is DAMAGE_PROPERTY formatted?

NCEI uses a 1.50K / 200M / 2.5B shorthand encoding (K=thousand, M=million, B=billion). The actor's parse_damage() normalizes to integer dollars but preserves the raw cell so buyers see what NCEI actually wrote. A diff record's before / after blocks carry both the parsed integer and the raw shorthand.

What's the F-scale finalization workflow?

NWS uses EFU (unrated) for tornado preliminary values that will later be upgraded to EF0–EF5 after a damage survey team visits the path. The actor's tor_f_scale change-type catches this finalization event. Mean preliminary→final lag is about 14 days; P90 is roughly 47 days.

Does this cover storm-event sub-tables (locations, fatalities)?

No — v1 covers only the main details table (one row per event). The companion StormEvents_locations-*.csv.gz (per- location lat/lon detail) and StormEvents_fatalities-*.csv.gz (per-fatality age/sex breakdowns) are NOT joined here. Buyers who need those should request a v2 follow-up actor — open a GitHub issue or email and we'll prioritize.

What about events NCEI drops on re-publish?

NCEI does occasionally drop events when re-publishing (most commonly duplicate CWA reports). v1 does NOT emit those as diff records; a retracted change_type is on the v2 roadmap if buyers ask. Run-summary line counts surface the implicit retraction count if you need to flag it manually.

Can I run multiple independent watchlists from one Apify account?

Yes. Each watchlist gets its own snapshot_key value. The actor reads/writes the previous snapshot from snapshot_key in your Apify Key-Value Store, so a "Tornado Alley" run and a "Hurricane corridor" run from the same account don't trample each other.

Is older-year coverage sparse?

Pre-1996 coverage has fewer columns populated. The parser tolerates missing columns and emits empty dicts where the NCEI source itself has nothing to diff. Buyers tracking methodology-update re-publishes on older years should expect fewer change records but cleaner signal.

Is there a rate-limit risk?

The actor self-limits to roughly 1 req/s and respects standard NCEI politeness conventions. We've never hit a rate limit at typical filter sizes. The bulk CSV is a single GET; the bulk of runtime is local CSV parsing, not network.

Companion actors in this portfolio

changewire maintains a portfolio of federal-data diff actors. All share the same daily-snapshot + JSONL change-record pattern, watchlist filtering, and per-record change-type semantics. Pick the data domain you need; mix and match across a single Apify account:

• changewire/clinicaltrials-protocol-diff — NIH-registered trial protocol diffs.
• changewire/grantsgov-opportunity-alert — Daily diff of federal grant opportunities (status, deadline, award amount, eligibility).
• changewire/fda-orange-book-extraction — FDA Orange Book pharmaceutical applications + patent listings, daily diff. (currently hard-blocked at the Akamai WAF; awaiting routing fix)
• changewire/uspto-ttab-docket-extractor — USPTO TTAB trademark docket records + filings + parties + deadlines.
• changewire/epa-echo-compliance — EPA ECHO facility compliance status + violations + enforcement cases.
• changewire/samgov-contracts-diff — SAM.gov federal contract opportunities, daily diff. (currently hard-blocked on identity-tied SAM.gov API key; awaiting user-machine key generation)

Running locally (for contributors)

cd actors/actor-4-noaa-storm-events-diff
apify run --purge  # uses .actor/input_schema.json defaults

State-scoped run:

$apify run --input '{"states": ["OKLAHOMA", "KANSAS"], "year": 2025}'

Event-type run:

$apify run --input '{"event_types": ["Tornado", "Hurricane (Typhoon)"]}'

Legal

NOAA NCEI Storm Events Database is US government public-domain (17 USC §105 — no copyright on government works) with no NCEI ToS restriction on programmatic redistribution. Actor obeys a polite 1-rps default. No PII in output (NCEI itself redacts narrative fields below privacy thresholds before publication).