The Pacific Northwest Research Station in Portland, Oregon, was founded in 1925. It employed 246 people, most of them scientists. It is closing.
The Forest Service reorganization will shutter 57 of the agency's 77 research facilities across 31 states. What remains will be centralized in Fort Collins, Colorado. The agency's headquarters moves to Salt Lake City. No research facility will remain in Alaska, which contains nine million hectares of national forest.
This is an inventory. A record of specific things that have already left, and what they carried with them.
The researchers
Of the 633 STEM Ph.D.s employed by the Forest Service in December 2024, 137 had departed by December 2025. That's according to an analysis by Science. Third-largest percentage loss among federal agencies with research staff.
No public breakdown exists of what fields they worked in. Fire ecology, hydrology, forest pathology, timber science, soil chemistry, wildlife biology. The number is an aggregate. The specifics left with the people.
Geoffrey Donovan, an economist at the Pacific Northwest station for more than two decades, took a buyout. "The loss of institutional capacity," he said, "is just heartbreaking."
A fire ecologist at the same station, still employed as of April, told OPB: "I have witnessed just a huge loss of research staff throughout the country through those early retirements."
The administration's April budget proposal asked to terminate forest and rangeland research entirely. The plan is to offload the work onto universities and the private sector. Among the facilities that will remain open: two labs focused on wildfire and developing forest products such as fuel and composite wood.
The experimental forests
The restructuring plans do not specify what will happen to the 82 experimental forests the Forest Service operates around the country.
Marcell Experimental Forest, northern Minnesota. Established around 1960. Sixty-five years of continuous climatic and hydrologic data collected across six peatland and upland forest watersheds. Randy Kolka, a research soil scientist who co-manages the site, has described it plainly:
"We're getting up to 65 years of data collection now, which is unheard of on the planet."
The site hosts the Department of Energy's SPRUCE experiment, the most expansive climate-change manipulation study on Earth. A phased decommissioning of SPRUCE begins in 2026 per the original research agreement, independent of the restructuring. The closure of the Northern Research Station's support infrastructure raises its own question: who maintains the underlying 65-year peatland dataset once the experiment ends and the people who built it are gone.
Bartlett Experimental Forest, White Mountains, New Hampshire. Ninety years of vegetation records. Mariko Yamasaki, a research wildlife biologist who retired from Bartlett in 2023, said: "We have a 90-year record of how vegetation has changed at Bartlett. We can speak to things in ways that nobody else can do because nobody else does long-term ecological research and management research the way Forest Service research does." Bartlett is not currently on the closure list. All research and development sites are listed as "under evaluation."
Vinton Furnace Experimental Forest, Ohio. The nearby McArthur research station, one of the 57 under evaluation, has staffed the studies at Vinton Furnace for decades. Glenn Matlack, a forest ecologist at Ohio University, noted that the experimental forest's purpose since the 1930s has been to learn how to conserve soil and stop erosion. "That activity depends on what's been learned through work in this experimental forest."
Pacific Wildland Fire Sciences Laboratory, Seattle. Part of the Pacific Northwest Research Station. Will close. Researchers were told the space would no longer be rented. They are trying to figure out what that means for them.
Between the data and the knowledge
Data is what you collect. Knowledge is what you can do with it. Between the two is a person. Usually a specific person who has spent enough years in a particular place studying a particular system that they understand what the numbers mean. Which ones to trust. Which ones need context. What was happening in the field the week the measurements went sideways. That understanding lives in the person. It doesn't transfer to the database when the person walks out.
When the person leaves, the data remains. Technically. It becomes harder to read, then hard to read, then functionally unreadable.
A 2025 paper in PLOS Biology on long-term ecological datasets put it directly: "Ecological field data requires an understanding of the system to sensibly analyze a dataset." Most ecological field studies last two to three years, the time it takes a graduate student to collect data for a thesis. Studies that exceed a decade are rare and disproportionately valuable. The same paper documented that nearly half of long-term mammal studies, 191 of 411, have already been terminated.
The research stations were where the data got read. People who understood it worked alongside each other, trained new people, maintained the chain of interpretation that made decades-old measurements usable. Close the room. The data stays where it is.
What has already become unreachable
This was happening before the restructuring. It has been happening for years. The restructuring is doing in months what attrition and retirement were doing over decades, but the underlying process is the same: the chain of interpretation breaks, and the data goes quiet.
In 2014, researchers published a study in Current Biology examining data availability from 516 published ecology studies between 2 and 22 years old. They contacted authors, asked for the underlying datasets, tested whether email addresses still worked.
The odds of a dataset being extant fell by 17 percent per year. After twenty years, roughly 80 percent of data was unavailable.
The data hadn't burned. It hadn't been deleted. Authors reported it lost, or stored on inaccessible media, or possibly in a box somewhere. The responses ranged from certainty the data was gone (stolen computer) to vague recollection that it might be on a Zip disk in a parent's attic. The decay was custodial. The person who collected the data retired, moved, changed institutions, died. The email stopped working. The floppy disk sat in a drawer without a drive that could read it.
That study examined papers up to 22 years old. Papers from the 1990s are now 27 to 36 years old.
A 2025 paper on British ecological surveys documented the same pattern: "Many important ecological data sets are at risk of loss. When they do still exist, they are often impossible to access, either due to being stored on old formats or because few people know they exist." The authors cited an earlier researcher's summary:
"Treasure troves of data, and the knowledge they could offer, are left mouldering on shelves."
The International Biological Programme, a major coordinated research effort, failed to leave a lasting legacy because the data collected was never made available for reuse. The data existed. The interpretive community that could have used it did not persist. The data became inert.
The national forest census
The FIA is the Forest Service's congressionally mandated national forest census, in operation since 1930. It holds the largest continuous body of forest inventory data in the world. Its carbon estimates are what the United States reports to the United Nations Framework Convention on Climate Change.
The program operates through four research stations. Two of them, the Pacific Northwest and the Northern, are directly affected by the restructuring. FIA in California is managed by the Pacific Northwest Research Station, the station that is closing.
As of April 2026, the FIA's EVALIDator tool posted a notice: "EVALIDator may experience extended downtime starting in April while we move the application to a new hosting environment. Please be aware of this interruption and schedule work accordingly."
No published reporting has yet documented disruption to FIA field plot measurement cycles. Whether the operational continuity holds will answer itself over the next year or two.
What the insurance carriers took with them
The same pattern shows up outside federal science. When an interpretive community departs, it takes embodied knowledge with it. The soil scientist who knows which sensor drifts in August has a counterpart: the actuary who spent fifteen years building a parcel-level risk model from claims data in the Sierra foothills.
In California, private insurers wrote 28 percent fewer homeowners policies in 2023 than in 2020. In 46 of 58 counties, nonrenewals outnumbered new policies. State Farm, Allstate, Farmers, Travelers, Chubb, Nationwide, USAA. The FAIR Plan, California's insurer of last resort, grew from 242,000 residential policies in September 2021 to 642,000 in September 2025.
Colorado launched its own FAIR Plan on April 10, 2025. Premiums in the state had risen 58 percent in five years. Oregon, Washington, Montana, Idaho are seeing the same pressure.
What left with the carriers is proprietary. ZIP-code-level and parcel-level loss histories. Internally developed risk models built over decades of claims data. Field-inspected condition assessments. None of that transfers to the FAIR Plans that absorb the displaced policyholders. The FAIR Plans must reconstruct actuarial analysis independently, for a pool of properties that is, by definition, the highest-risk subset of the market.
An NBER working paper from 2024 documented striking variation in how insurers classify and price wildfire risk. Firms with coarser risk models charge higher prices in high-risk areas or choose not to serve them at all. The carriers with the most sophisticated understanding of the risk are the ones who left. What remains is a system that knows less about what it's insuring than the system it replaced.
No published study documents specifically what happens to a carrier's accumulated actuarial knowledge when it exits a state market. That absence is itself a kind of answer.
The precedent
During the first Trump administration, a similar relocation plan at the Bureau of Land Management moved the agency's headquarters from Washington to Grand Junction, Colorado. Eighty-seven percent of the main office's employees left. The Biden administration moved the bureau back in 2021.
No published study has documented what institutional knowledge was lost in the interim, or how long recovery took, or whether it did.
Conservation advocates have explicitly invoked this precedent. Chris Swanston, chief scientist of the Save the Redwoods League, a conservation nonprofit, said that if researchers are allowed to relocate to nearby universities or nonprofits, "there's real hope we can avoid a bunch of retirements and departures." And if not, "then we'll see losses."
Ann Bartuska, a former undersecretary for research at USDA, said some consolidation makes sense but was surprised by the extent.
What becomes unreadable
The person who spent twenty years at a research station and knows which plots flood in wet springs. The actuary who built a risk model from a decade of claims in a county that no longer has a private insurer. The grad student who would have been hired to continue the fieldwork but wasn't, because the position was eliminated before it was posted.
Resources for the Future noted in 2025 that governments provide data and identify problems "through long institutional memories and collaborations with external scientists who help develop solutions." The phrase is institutional memories. The memory lives in the people. Databases hold numbers. What the numbers mean walks out when the person does.
The Forest Service still has a chief and a budget and a name. The FAIR Plan still writes policies. The servers still hold the data. But the 17-percent-per-year decay curve was already running before any of this, and the restructuring removes the people who were still holding the chain together. The shuttered station, the eliminated position. Those losses get reported. The custodial loss has been running for decades, quiet, and it is the larger one.
The day after Memorial Day is as good a time as any to take inventory. Just to write down what was here and isn't, so that when someone needs it later, they'll at least know what to look for. Even if nobody's left who can tell them what it means.
-
The 82 experimental forests: The restructuring plans do not specify what happens to the Forest Service's experimental forests, and all research sites remain "under evaluation", meaning the full scope of closures and data-transfer plans is still emerging.
-
Fire season without capacity: The Forest Service enters the 2026 wildfire season having completed 38 percent less mitigation work than in previous seasons, while forecasters project 5.5 to 8 million acres burned and conditions that will push fires to spread faster and grow larger.
-
FIA field measurement continuity: Two of the four research stations coordinating the Forest Service's national forest census program are directly affected by the restructuring, and whether the 95-year-old inventory's field plot measurement cycles survive intact remains unreported.
-
Insurance knowledge after withdrawal: An NBER working paper documented striking variation in how insurers classify wildfire risk, and the carriers with the most sophisticated parcel-level models are disproportionately the ones that have exited high-risk state markets.