A USDA Forest Service study has found that using each sawmill’s own buying history to estimate how much timber it draws from surrounding counties cuts measurement error by more than half — and eliminates the kind of wild swings that can leave planners thinking a county produced no wood at all in a given year.
That is according to new research conducted by Rapeepan Kantavichai, Consuelo Brandeis, and James Gray from the Forest Inventory and Analysis unit and Matthew Winn of the Forest Inventory and Analysis unit in Blacksburg, Virginia.
And the problem is structural.
Wood Central understands that the Timber Products Output survey does not sample every sawmill every year. When a mill goes unsampled, its county removals go unrecorded — leaving zeros in the data and sharp year-to-year swings wherever sampled mills happen to shift. Those gaps feed directly into the resource assessments that state agencies, timber investors, and federal planners use to model harvest schedules and wood supply.
The researchers tested two alternative methods, measuring accuracy in MCF, which measures thousands of cubic feet of sawlog volume. Under the Historical Procurement approach, a non-sampled mill receives a volume estimate based on sampled mills in the same stratum, then distributes that total across counties using its own most recent procurement proportions. The Radius Procurement method applies the same stratum average but spreads volume equally across every county centroid within 25 miles.
In the 2013 validation, the mean county-level error was 261 MCF under Historical Procurement, 414 MCF under Radius Procurement, and 619 MCF under the current method. The median fell from 374 MCF to 146 MCF. In the worst cases, the current method produced county-level errors exceeding 15,000 MCF; the historical method held that ceiling to 2,986 MCF.
The dataset covered 771 sawmills active across all canvass years — Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Mississippi, North Carolina, Oklahoma, South Carolina, and Virginia — with 5,000 Monte Carlo replications. Texas did not participate; Tennessee due to missing data from 2013.
Under a stress test with weaker mill-size correlation, the mean error under Historical Procurement was 324 MCF, compared with 819 MCF for the current method. The ranking is performed at every tested threshold.
Projecting 2011 procurement records forward into 2015 and 2017, errors rose across all methods as buying patterns drifted. Historical Procurement reached 405 MCF in 2015 and 530 MCF in 2017. The current method returned 786 MCF and 807 MCF for the same years.
County removal estimates serve as the basis for state resource assessments, timber supply models, and federal harvest scheduling. Errors at the scale produced by the current method feed directly into those models, skewing supply forecasts used by planners and timber investors across some of the most productive forest land in North America.
It comes as Wood Central reported that Southern Yellow Pine sawmills had fallen below cash costs, triggering curtailments across the US South not seen since the housing market collapse — conditions that leave county-level removals data increasingly difficult to reconcile with on-the-ground supply. Wood Central has also reported on why the US South — and not the West Coast — holds the long-run key to American timber supply, and on the contradiction between rising US sawmill capacity and persistently flat production across the supply chain.
For further information: Kantavichai, R., Brandeis, C., Winn, M.F. et al. Improving County Sawlog Removal Population Estimates using Historical and Distance-Based Mill Procurement Data in the Southern US. For. Sci. (2026). https://doi.org/10.1007/s44391-026-00061-z
