The Thermometer
They Can't Erase.

Federal agencies are scrubbing climate records. The data still exists — in more places than they know. A divergence-first analysis of what's being deleted, what's survived, and what the market hasn't priced.

In early 2025, NOAA began removing datasets. Not archiving — removing. Pages that had existed for years returned 404s. Download links stopped working. The U.S. Global Change Research Program website went dark. USDA climate research portals followed. The EPA's climate change indicators page — a resource used by state planners, insurance actuaries, and infrastructure engineers for over a decade — disappeared without a redirect.

The official framing: budget cuts, data audits, policy review. The actual pattern: systematic removal of longitudinal temperature, precipitation, and sea-level records that formed the empirical basis for climate risk pricing across hundreds of billions of dollars in financial instruments.

You can delete a government URL. You can't delete the underlying physics.

What Was Removed

The deletions weren't random. They clustered around four data categories that share a common thread: they're all primary inputs for financial risk modeling.

These aren't obscure academic datasets. They're the primary inputs for crop insurance pricing, municipal bond risk modeling, reinsurance underwriting, and wildfire exposure assessment. The entities that price those instruments were using this data yesterday. Today they're getting 404s.

// divergence signal

Federal government removes climate risk data from public access.

Reinsurance companies continue pricing climate risk — using the data they'd already downloaded.

The market knows what the government is pretending it doesn't know.

Where the Data Still Lives

The Internet Archive's Wayback Machine had crawled most of these datasets before deletion. Within days of the first removals, a coordinated academic effort — dubbed the "Data Refuge" movement, originating from the University of Pennsylvania — began systematic preservation of at-risk federal climate data. Librarians, not hackers. Archivists, not activists. People who understand that data loss is permanent and political cycles are temporary.

NOAA mirror sites operated by university research departments preserved the raw files. The European Centre for Medium-Range Weather Forecasts (ECMWF) maintains its own independent ERA5 reanalysis dataset — completely outside U.S. jurisdictional reach. ERA5 provides hourly estimates of atmospheric, land, and oceanic climate variables from 1940 to present, at 31km global resolution. It doesn't need NOAA's permission to exist.

Berkeley Earth's independent temperature record extends to 1850 and is maintained in Berkeley, California, not Washington. The World Meteorological Organization's Global Climate Observing System maintains records submitted by 193 member nations. You cannot delete the thermometer in Helsinki. You cannot delete the buoy data from the Pacific Weather Observation Program. You cannot delete what ECMWF already ingested in 2023.

The data is distributed. It always was. What was removed was the convenient government aggregation layer — the one that made it easy for journalists, researchers, and small municipalities to access the underlying science without needing academic database subscriptions that cost $10,000–$50,000 annually.

The Drought Beneath the Drought

Focus on the temperature data and you miss the bigger signal. While NOAA datasets were being scrubbed, the western United States entered its 27th year of what climatologists call a "megadrought" — the driest period since at least 800 AD, verified through tree-ring analysis that predates any government database.

The Colorado River — which supplies water to 40 million people across seven states and Mexico — has seen its flow decline roughly 20% since 2000. Lake Mead hit its lowest recorded level in 2022 and hasn't meaningfully recovered. The 1922 Colorado River Compact, which allocated water rights based on an abnormally wet period, is now governing a system with fundamentally less water than the legal framework assumes exists.

// the water math

The Colorado River Compact allocates 16.5 million acre-feet/year across 7 states + Mexico.

Average actual flow since 2000: ~12.4 million acre-feet/year.

That's a 4.1 million acre-feet annual deficit — roughly the entire allocation of Arizona and Nevada combined.

The legal framework assumes water that doesn't exist. The market is beginning to notice.

The USDA drought monitor archives that were removed are precisely the records you'd need to track this multi-decade trend. Without public access to severity indices dating back to 1895, a county planner in rural Arizona trying to assess long-term water availability is now operating with less historical context than a Munich Re analyst in Germany who downloaded the full archive in 2023.

Where Drought Meets Markets

The financial implications cascade across domains:

What the Market Hasn't Priced

Here's the divergence: the removal of public data access doesn't change the underlying risk. It changes who can see it.

Large reinsurers — Munich Re, Swiss Re, Berkshire Hathaway Reinsurance — maintain multi-decade proprietary climate datasets. Munich Re's NatCatSERVICE has tracked natural catastrophe losses since 1980 across 30,000+ events. They had already downloaded everything NOAA published. Their pricing doesn't change because a government URL went dark.

In fact, the data removal benefits them. Their informational advantage just widened. They can see the full picture. Their counterparties increasingly cannot.

Small municipal bond issuers in flood-prone counties? They relied on public access. Agricultural lenders pricing crop insurance for the next growing season? They're now operating with less data than the counterparties on the other side of the trade. Community banks in rural counties across the Great Plains, whose loan portfolios are concentrated in agriculture and real estate? They used USDA drought data to stress-test their own exposure. That dashboard is gone.

Information asymmetry is mispricing. Mispricing is edge.

The entities that rely on public government data for risk pricing are now at an information disadvantage relative to entities that already had private data infrastructure. That disadvantage will manifest in mispricings — in municipal bond spreads, in reinsurance treaty pricing, in agricultural commodity options, and in the credit quality of water-dependent municipalities that haven't yet adjusted their planning horizons.

The Leading Indicators

The physical world doesn't wait for bureaucratic consensus to produce economic signals. If you know where to look, the repricing is already starting:

The Longer Play

Removing data doesn't remove reality. The 2025 Atlantic hurricane season happened whether or not NOAA's historical comparison datasets were publicly accessible. The Colorado River compact is under strain whether or not the USDA drought severity archives are online. The tree rings in bristlecone pines across the Great Basin still encode 4,000 years of precipitation data that no executive order can reach.

The market will eventually reprice. It always does — not because someone updated a government database, but because the physical world doesn't wait for bureaucratic consensus. A 404 page doesn't stop a river from running dry. It doesn't prevent a wildfire from jumping a containment line into a residential zone. It doesn't make a crop insurance policy solvent when the drought severity data it was priced against has been understated.

The edge is in the gap between when the data disappears from public view and when the underlying reality manifests in prices. That gap is currently open. It won't be forever.

Data over politics. The thermometer doesn't care who's in office. It just keeps reading the temperature.

This is not investment advice. This is pattern recognition. What you do with the pattern is your problem.

// sources

  1. [1] Internet Archive / Wayback Machine — Archived NOAA datasets (pre-deletion snapshots)
  2. [2] ECMWF ERA5 Reanalysis — Global climate reanalysis, 1940–present, 31km resolution
  3. [3] Berkeley Earth — Independent surface temperature record, 1850–present
  4. [4] World Meteorological Organization — Global Climate Observing System, 193 member nations
  5. [5] Munich Re NatCatSERVICE — Natural catastrophe loss database, 30,000+ events since 1980
  6. [6] NIFC National Interagency Fire Center — Historical wildfire perimeter and frequency data (partial archive)
  7. [7] Bureau of Reclamation — Lake Mead water level data, daily readings
  8. [8] Colorado River Research Group — Independent hydrological analysis and flow projections
  9. [9] Williams et al. (2022) — "Rapid intensification of the emerging southwestern North American megadrought in 2020-2021," Nature Climate Change
  10. [10] University of Pennsylvania Data Refuge — Academic preservation of at-risk federal datasets
  11. [11] @TheBondFreak (Randy Woodward) — 30+ years of bank credit analysis, RV loan delinquency tracking
← All research Subscribe via RSS