OpenFEMA Data Sets Hub

It is the platform record, not one table. FEMA's OpenFEMA program publishes the agency's statistical estate as a catalog of 49 datasets, and this record catalogs the catalog: seven categories - Disaster Info, Emergency Management, Individual Assistance, Public Assistance, Hazard Mitigation, NFIP and Misc - with every entry carrying its own description, file distributions and field-level data dictionary.

Eight or so titles carry most property & casualty weight. The NFIP extracts dominate by volume: FIMA NFIP Redacted Policies v2 at roughly 73.6 million policy transactions, and the redacted claims extract at about 2.72 million claim records with loss dates from 1970. Around them sit Disaster Declarations Summaries near 70,000 records, the Individuals and Households Program registrations at roughly 25.9 million, housing assistance splits for owners and renters, Public Assistance funded project detail, Hazard Mitigation Assistance mitigated properties, and the National Risk Index.

Within Datadory's Property & Casualty Insurance slice, this record owns the breadth layer: every other FEMA record on our shelf is one drawer of this cabinet. Browse the rest of the shelf at the property casualty insurance data hub.

Two records were captured and verified during the August 2026 research pass, and together they show the two halves of the catalog: the declaration that opens an event and the claim that closes one.

The first is a Fire Management Assistance declaration: FM-5672-AK, federal disaster number 5672, the Mukluk Fire in Alaska, effective August 18, 2026. Seven columns identify the event completely - designation string, number, state, declaration type, peril, title, date - which is why declarations make such clean event calendars.

The second is an NFIP flood claim: loss date August 25, 2017, Texas, rated in flood zone AE, occupancy type 6, with a net building payment of $149,870.03. Gross damage estimates sit beside net payments in the full extract; the net figure is the receipt - what the program actually sent after deductions.

Both rows above are real, not synthetic. Your sample extends them with the full row width of whichever tables you scope, and the schema in the sample is the schema that ships.

A catalog of 49 datasets publishes no single schema - each entry carries its own field-level data dictionary, and the claims extract alone documents seventy-two fields. The table above therefore reads across the catalog rather than pretending it is one wide table.

On the declarations side, seven columns do all the identification work: the composite femaDeclarationString, the disasterNumber that every assistance record joins on, the declaration type and peril, the title, and the effective date.

On the transactional side, the NFIP extracts share a documented spine: loss dates (dateOfLoss, yearOfLoss), the rated flood zone, occupancy coding, and the money columns - gross and net payments for building, contents and Increased Cost of Compliance, deductibles, and coverage carried. Census geography rides on every row as the join spine, stable and built for aggregation.

Definitions here are marked estimated at hub level and get confirmed extract-by-extract when your sample arrives - the honest order of operations for a catalog this size.

Geography - the United States and its territories. NFIP records locate to state, reported ZIP, county FIPS and approximate coordinates; declarations resolve to designated counties and sub-county areas. The census geography is deliberately the analytical grain, which makes joins to your own portfolio references clean rather than lucky.

Temporal - the deep end of this shelf. NFIP claims reach back to 1970, five decades of American flooding; declarations and the assistance programs generally begin in the late 1990s, varying by dataset. Depth and speed live on different tables, and knowing which is which is part of scoping a request properly.

Granularity - record level throughout: one row per claim, per policy transaction, per assistance registration, per funded project. Repeat-loss properties appear once per claim, which is exactly what repeat-loss analysis wants. Against a catalog mean quality score of 7.81 across 1,744 datasets, this record scores 9/10 - carried by its span, its documentation and the two verified rows on file.

API, files, or your warehouse. Daily, weekly, or hourly.

API, files, or your warehouse. Daily, weekly, or hourly.

Forty-nine tables arrive speaking one house style: typed, documented against their per-dataset dictionaries, and keyed so the claims extract attaches to the policy book, the declaration calendar and your own territory references without name-matching. Take the full catalog once as files, or keep the tables you use current on whatever cadence your models need.

Every delivery carries the field dictionary and validation rows for the tables in scope. Name the states, perils, programs and years you care about and the sample returns shaped to them before any commitment.

Five jobs earn their keep from this breadth:

• Event calendaring - declarations turn scattered losses into a clean per-event frame, with declaration numbers grouping claims, policies and aid registrations onto one calendar.
• Severity and payout modelling - fit empirical curves from the government's own claim receipts, separating gross from net and paid from closed-without-payment.
• Portfolio screening - score mortgage and asset books by county, tract and ZIP loss history before exams, deals or renewals.
• Aid-flow auditing - trace federal assistance from declaration through registration to award, household by household.
• Mitigation research - pair Hazard Mitigation Assistance projects against subsequent losses to ask whether the buyouts and elevations paid.

Insurers, reinsurers and cat analysts get the whole federal flood story in one catalog - exposure, outcome and trigger. Investors and quants get event-attributed losses back to 1970 for stressing catastrophe-linked books. Developers and builders ship products on documented schemas instead of reverse-engineering a government portal. Market researchers and consultants ground regional flood-market work in county-scale federal numbers. Across all of them the constant holds: one catalog, one house style, every table documented.

This record owns the front door. Its neighbours each hold one room. FEMA NFIP Redacted Claims v2 is the realized-loss layer in full width; FEMA NFIP Redacted Policies v2 is the premium side - the exposed book those claims landed on. FEMA NFIP Geospatial Hub (ArcGIS) owns mapped regulatory hazard - where flooding is governed, versus where it happened - and the pairing is worked through in OpenFEMA Data Sets Hub vs FEMA NFIP Geospatial Hub.

Outside FEMA, First Street Foundation Climate Risk Data models forward-looking hazard per building, NOAA NCEI Billion-Dollar Disasters Time Series tabulates catastrophic events as adjusted totals, and NAIC Research and Insurance Data supplies regulator-grade market analysis. Only the catalog answers what the whole federal estate holds - because only FEMA publishes its own index.

Four things worth settling upfront.

First, this is a catalog record. Samples are cut per table - claims, policies, declarations, assistance - so name the tables you want and the sample returns each with its own dictionary attached, rather than one mythical mega-schema.

Second, two rows are on file. The Mukluk Fire declaration and the 2017 Texas claim are verified; the rest of the row width arrives with your sample, scoped to named states, perils and years.

Third, depth varies by table. The claims extract reaches to 1970; several assistance and mitigation tables start later. If your study needs a common start date, say so at scoping and the panel gets aligned deliberately rather than hopefully.

Fourth, terms travel with the delivery. FEMA publishes usage conditions alongside the data and provides it as-is, without warranty of accuracy; deliveries carry the conditions so redistribution decisions are made against the record, not folklore.