The Drought Relief Gap

The most expensive peril, and the least visible one

Drought is the most expensive thing that happens to American crops. In the measured storm-damage record it accounts for $29.1 billion of $46.9 billion — 62% of all recorded crop damage, more than freeze, hail, wind, flood and tornado combined.

It is also the peril least visible in the record most people reach for first. A federal disaster declaration is the public statement that something bad happened in a place. So we put the two records side by side: every weekly drought reading for 3,211 counties from 2000 to 2024 — 4.19 million county-weeks — against all 5,252 federal declarations.

Of 80,275 county-years, the number carrying a federal drought declaration is zero. Not few. None. Over the same span, 37% of those county-years carried a declaration of some other kind.

First, we checked it wasn’t our own mistake

A zero like that is exactly what a broken data join looks like. If our method simply failed to match drought declarations to counties, we would see the same empty result and mistake it for a finding.

So we ran the same machinery over the earlier record. It matched 38 drought declarations to 1,336 counties without difficulty — every one of them before 1994. The tool works. It finds nothing after 1993 because there is nothing there.

The last drought declaration issued to any US state was Texas, September 1993, covering 149 counties. Its official title was “Extreme Fire Hazard.” The only drought declaration anywhere in the 25-year measurement window went to Chuuk, in the Federated States of Micronesia.

The relief is real. It just runs somewhere else.

This is not a story about drought going unanswered. It is a story about two systems that work in completely different ways.

A disaster declaration has to be requested by a governor and approved by a president. The USDA’s Livestock Forage Program works the other way around: eight consecutive weeks at D2 on the drought map is the trigger. Nobody has to ask.

By that rule the measured record lit up 23,560 county-years — 29% of everything measured — and 3,085 of 3,211 counties qualified at some point. The longest unbroken run of D2-or-worse drought we found in a single county lasted 296 consecutive weeks: five and a half years.

We should be precise about what that program does and does not cover. It compensates grazing loss — pasture that failed — not the row-crop and orchard damage that makes drought the costliest peril in the first place. The crop side has its own instruments: federal crop insurance, NAP, Secretarial Disaster Designations. None of those are measured here. We single out livestock forage for one reason: its trigger is the measurement itself, so it can be read straight off the same map.

Drought does reach the record — wearing a different label

Dry ground burns. Fire is the largest declaration category of the last 25 years. So we asked whether the fire declarations were quietly landing in drought.

For every declaration, we looked up how dry that county was in the week it opened — and scored it against that same county’s record for that same month of the year. That comparison removes the two obvious traps at once: some counties are simply dry, and some months are simply dry.

54% of fire declarations open in a week with D2 drought or worse somewhere in the county. For every other peril that figure runs between 4% and 15%.

We then tried to break it. Most fire declarations are fire-management grants, and one wildfire siege can generate a dozen of them in one state in one week — which are not a dozen independent facts. Collapsing each siege to a single observation pulls the fire figure from the 61st percentile down to the 57th. It does not remove it. Nor is it an artefact of the grant program: the effect is stronger among the 46 full major-disaster fire declarations than among the 1,249 grants.

One honest exception. Of the eight perils we tested, tropical storm also sits slightly on the dry side, at 0.542. With only 25 declarations behind it, that could easily be chance, and we report it rather than dropping it. Fire is the only peril that separates by a margin the data can actually support.

What this says, and what it doesn’t

It says: the federal declaration record is not a drought record, and has not been one since 1993. A county can spend five straight years in serious drought and never appear in it. Anyone using declarations as a stand-in for what a place has weathered will erase the single most expensive peril there is — and will erase it hardest exactly where drought lasts longest.

It does not say anyone failed. The Stafford Act is built for sudden events that break infrastructure; drought is neither sudden nor structural, and agricultural drought relief was deliberately routed elsewhere. What we have measured is how complete that routing turned out to be.

And it is not a claim that fire money is drought money. Fire declarations open in dry weeks because drought makes fires. Those grants reimburse firefighting costs to state agencies; they do not reach the drought-damaged farm. What we are describing is a coincidence in the bookkeeping, not a transfer, and certainly not anyone choosing a label.

Finally, the two records are different kinds of thing on purpose. A declaration is a political and statutory act with thresholds, a request and an approval. The drought map is an assessment of conditions. The gap between them is evidence about routing — not evidence that either one is wrong.

Reproducibility

Every number on this page comes from the results.json file in the drought-relief-gap workspace, and a checker script verifies that correspondence mechanically rather than by eye. Pipeline: the US Drought Monitor county grid and the FEMA declaration record are read from the TerraPulse dex layer; declarations are matched to counties by name and state at a 97.7% resolution rate, with every unmatched area enumerated; each declaration is scored against its own county’s same-calendar-month history; and every test is run at declaration level and again with wildfire sieges collapsed, because grants issued in one siege are not independent. Bootstraps are seeded, and the analysis produces byte-identical output across repeated runs. No model, forecast or reanalysis field is used anywhere.