Original analysis · Barcelo & Associates

Updated September 8, 2026

What 170,843 flood claims say about Houston — including the parts that surprised me.

Almost everything written about flood risk in Houston cites the same three sources. This does not. Every figure below was computed from the raw claim records FEMA publishes for Harris County — all 170,843 of them — using fields that, as far as I can find, nobody else has analysed at this level. Three findings survived scrutiny. Two did not, and I have published those too, because which analyses you throw away is the honest measure of the ones you keep.

170,843 FEMA claim records analysed128,809 residential claims that paid3 findings that survived scrutiny2 that did not, published anyway
Scope

What is being counted, stated once.

Every figure describes residential claims that paid more than $0 — 128,809 of them — drawn from FEMA's NFIP Claims dataset filtered to Harris County. Residential only, because a warehouse loss and a bungalow loss answer different questions. Paying claims only, because 30,753 of the county's claims paid nothing, and averaging over filed claims understates what a paying claim actually cost. Medians are used in preference to averages throughout, because a handful of catastrophic losses drag the mean well above the typical claim.

Finding one

More than a third of paid claims did not come from a named storm.

37.5%of paid claims, no named storm
$1.50Bpaid on unnamed events
$13,211median, unnamed event
$47,207median, named storm

Houston plans for hurricanes. Of 128,809 paid residential claims, 48,254 — 37.5% — were recorded against something that was not a hurricane or a tropical storm at all: ordinary rainfall, a seasonal storm, a downpour that never earned a name. Those claims account for $1.50 billion in payments.

The median tells you why this matters more than the total does. A named-storm claim paid a median of $47,207; a claim from an unnamed event paid $13,211. These are not the catastrophic losses that make the news — they are the ordinary ones, and they are frequent. A household that has decided flood coverage is a hurricane-season question is protecting itself against 62.5% of the record and leaving the rest.

The share varies enormously by neighbourhood, and that variation is real: it runs from 14.3% in Bellaire to 58.7% in Memorial. Each neighbourhood page on this site now carries its own figure, and the ZIP lookup carries all 136.

By neighbourhood

The same figure, neighbourhood by neighbourhood.

A county average hides the one thing that matters for a decision: the share runs from 14.3% to 58.7% across these 15 neighbourhoods. In 38 of the county's 136 ZIP codes, more than half of paid claims did not come from a named storm.

Neighbourhood ZIP % not a named storm % pre-flood-map Median paid
Bellaire 77401 14.3% 10.4% $55,843
Garden Oaks 77018 19.3% 11.7% $26,111
Houston Heights 77008 20.8% 16% $37,402
Braeswood 77025 21% 15.6% $72,703
Oak Forest 77092 24.1% 15.2% $34,968
Sagemont 77089 29.8% 10.9% $23,094
Spring 77379 35.6% 1.6% $74,925
Bear Creek 77084 41.5% 3.7% $80,342
Energy Corridor 77079 42.8% 16.4% $82,476
Cypress 77429 45.7% 6.3% $71,993
Meyerland 77096 45.9% 19.4% $100,500
Kingwood 77339 48.7% 8.7% $60,000
The Woodlands 77381 57.5% 3.9% $10,930
Katy 77450 58.1% 1.2% $13,176
Memorial 77024 58.7% 10.4% $30,352

Each row is its own ZIP code, computed the same way as the county figure. A ZIP spans thousands of properties — this informs whether to carry coverage, not any claim about a single address.

Finding two

Homes built before the first flood maps show a median payment 2.4 times higher.

$54,280median, pre-FIRM homes
$22,502median, post-FIRM homes
2.4×the gap
13,793paid pre-FIRM claims

A pre-FIRM building is one that predates its community's first Flood Insurance Rate Map — it was never built to a mapped elevation standard, because there was no map to build to. Across Harris County, 13,793 paid residential claims were on pre-FIRM buildings, with a median payment of $54,280. The 115,016 post-FIRM claims had a median of $22,502.

That is the kind of gap that should immediately make you suspicious, because pre-FIRM homes are also older, smaller and cheaper — and cheaper homes should produce smaller claims, not larger ones. So I checked the confound directly, and it runs against the finding rather than producing it: sorted by construction decade, the county's 1950s-70s homes show lower median payments than its 1990s-2010s homes. Pre-FIRM buildings pay out more despite being in the decades that otherwise pay out less. The elevation standard is doing the work.

The same pattern appears, less starkly, in whether a building is elevated: 6,551 claims on elevated buildings paid a median of $16,742 against $25,714 for the 122,258 that were not. That comparison is more heavily confounded — elevated buildings are concentrated in the high-risk zones where the rules require it — so treat it as suggestive, not settled.

Finding three

Underinsurance is rarer than the sales pitch suggests — and concentrated exactly where it hurts.

5.1%damage above coverage, ordinary year
11.1%the same, during Harvey
31.4%of policies already at the NFIP cap
$235Mof damage above the coverage carried

Comparing recorded building damage against the building coverage actually carried, on 127,932 claims: damage exceeded coverage in 6,466 of them, or 5.1%. That is a smaller number than most agents imply, and I would rather report it than inflate it. The median household carried $150,000 of building coverage against a median damage figure of $26,514.

The exposure is not in the ordinary year. During Harvey the underinsured share more than doubled, to 11.1% of claims, with a median shortfall of $15,533. And the structural constraint sits above all of it: an NFIP policy caps building coverage at $250,000 and contents at $100,000. 31.4% of the policies in this analysis were already at that building cap — they could not have bought more NFIP coverage if they wanted it — and 7.9% of those still had damage exceeding it.

Across the whole set, $235 million in recorded damage sat above the coverage carried. That is the honest shape of the problem: most households are adequately covered for an ordinary year, and the gap opens in the catastrophic one, at the top of the market, where the federal cap stops.

What I threw away

Two analyses that did not survive, and why.

Both of these would have made better headlines than anything above. Neither is in this study.

A water-depth-to-payout curve

The obvious next question is what a foot of water costs, and the dataset has a depth field. It is not usable: it carries negative values, the units are inconsistent between records, and the resulting curve is not monotonic — payments at 7-12 inches come out below payments at 4-6. A clean “every inch costs X” number would have been the most quotable line on this page. It would also have been wrong.

A construction-decade risk ranking

Sorting claims by the decade the house was built produces a tidy chart in which newer homes look worse: 1990s-2010s homes show median payments around $60,000 against roughly $25,000-30,000 for 1950s-70s homes. Read as risk, that is backwards. It measures house size and replacement value — a bigger, costlier home costs more to repair from the same flood. Payment size is not a risk rate, and presenting it as one would mislead every reader who is deciding what to buy.

Limits

What this study structurally cannot tell you.

So what

The three questions this leaves on the table.

Do you carry flood coverage at all?

It is a separate policy from homeowners in every case — no Texas homeowners policy pays for rising water — and across Harris County the majority of paid claims were rated outside the high-risk zone, where nobody required it.

Do you know when your house was built, relative to the maps?

It is the single strongest signal in this dataset, and it is a fact you can look up once and never revisit.

If you are near the NFIP cap, have you priced what sits above it?

31.4% of the policies here were already at the $250,000 building limit. Excess flood coverage above the federal cap is a different product and a different conversation — one worth having before the season, not during it.

Questions

Questions about this study

Who did this analysis?

I did — Ricardo Barcelo, at Barcelo & Associates Insurance in Houston. It is computed from FEMA's public OpenFEMA API rather than licensed from a vendor, which is why the methodology is published alongside it and why the discarded analyses are published too. Anyone can reproduce it from the endpoint and filters listed on the methodology page.

Is this endorsed by FEMA?

No. This product uses the Federal Emergency Management Agency's OpenFEMA API but is not endorsed by FEMA. FEMA publishes the raw records; the aggregation, the choices behind it, and any errors in it are mine.

Does it include individual property records?

No, and it cannot. OpenFEMA's terms prohibit republishing anything that could identify an individual property, so only aggregates appear here. The smallest group published anywhere on this site is a ZIP code with at least 25 paying claims.

Why residential only?

Because mixing a warehouse loss into a median that a homeowner is going to read makes the median mean nothing. Commercial and residential claims behave differently in both frequency and size, and this study is written for households.

How often will this be updated?

FEMA updates the dataset continuously as claims close. This analysis was retrieved on September 8, 2026, and the retrieval date is stamped on every figure that depends on it. Between retrievals a number can move by a claim or two; nothing here is stated with more precision than that supports.

Next step

Want this read against your actual policy?

Bring the declarations page and the year the house was built. Ricardo Barcelo, Insurance Consultant, works out of 1235 North Loop W in Houston and will read it against what has actually been paid where you live, in English or Spanish.