Purpose

The purpose of this dataset is to quantify the extent, type, and economic context of forest loss at the county scale in the eastern United States. By distinguishing between temporary and permanent deforestation, the dataset provides a foundation for evaluating the permanence of land-use change, the relative vulnerability of forests to development, and the potential costs of conservation. These data are intended to support forest landowner engagement, land-use planning, conservation prioritization, and regional reporting on forest change.

Data

A Geopackage of the data is contained in the following Zipfile:

eastern-us-counties-d4st.zip

Once unzipped, the eastern-us-counties-d4st.gpkg file can be opened in the open source QGIS software.

Access Constraints

None

Use Constraints (Data Quality and Limitations)

Urban underestimation: LANDFIRE provides more accurate forest cover data than NLCD, but does not extend back to 2010. Using NLCD forest cover in 2010 may underestimate forest canopy in urban areas, leading to systematic underestimation of forest churn.

Impervious classification: All NLCD impervious classes are treated as developed/impervious. This simplifies classification but may overestimate forest loss in cases where forest converts to low-intensity development.

Temporary vs. permanent deforestation: Temporary forest loss may later become permanent through conversion to impervious or barren land uses, or may be restored to forest. This dataset does not model future probabilities of restoration or development.

Agricultural conversions: Some conversions derived from forest → barren reflect transitions to higher intensity farming (e.g., cranberry, blueberry cultivation).

Relative values: The ratio of forest loss to fair market value (“d4st_perm_to_fmv”) is intended only for visualization and relative comparison; raw values should not be directly interpreted as absolute economic impact.

Suggested Citation

Sanda, M and Lindeman, S. Upstream Carbon. County-Level Eastern US Forest Loss since 2010. Version 1.0. Geopackage. September 2025.

Definitions of shorthand used below:

• impervious: (1 if a pixel is developed type in NLCD; fractional impervious values round up to 1)
• barren: (1 if a pixel is barren type in NLCD)
• forest_10: (1 if pixel is a forest type in NLCD 2010)
• forest_now: (1 if pixel is a forest type in LANDFIRE 22)
• disturbed: (1 if pixel overlaps a fast loss pixel from 2010-present in LCMS)
• d4st: Deforestation. Disturbed and not forest_now (forest -> disturbed -> nonforest)
• d4st_perm: Permanent Deforestation. d4st and either barren or impervious (forest -> disturbed -> barren | impervious)
• d4st_temp: Temporary Deforestation. d4st and not barren or impervious (forest -> disturbed -> not (barren | impervious); usually observed as forest -> shrubland | cropland
• new_forest: Afforestation. forest_now and not forest_10 (nonforest -> forest)

FMV (fair market value):
There are 2 FMV columns: "_vacant" and "_all".
These correspond to the respective raster sources from Nolte 2020, found here: https://doi.org/10.1073/pnas.2012865117

Source data values are in ln USD per hectare.
To convert the source raster data into interpretable values, we
1) first invert the log transform: df[col_ha] = np.exp(df[original]
2) Adjust for CPI growth (1.28 from 2017 to 2024): df[col_ha] = df[col_ha] * 1.28
3) Apply the conversion rate of acres to hectare: df[col_ac] = df[col_ha] / 2.471052

Details of Known Issues

1. LANDFIRE provides more accurate forest cover data than NLCD, but it does not support data from 2010. By using NLCD10-forest -> NLCD22-nonforest to classify deforestation as temporary or permanent, this approach consistently underestimates forest canopy in urban areas. In doing so, forest loss in urban areas may be systematically uncaptured leading to underestimation of forest churn rates.
2. NLCD provides multiple impervious classes, and all are treated as forest -> impervious. Using this approach is driven by the limits of historical LANDFIRE data; LANDFIRE data dating to 2010 would enable using LANDFIRE's various developed classes in 2022, which can enable more precise inclusion or exclusion of tree canopy. By treating all developed classes from NLCD as impervious in both 2010 and 2022, the results omit any progression in development intensity over the period and resulting forest loss, and may also overestimate the degree of forest loss when a pixel converts from forest -> developed (low intensity). The combined effect of these issues may induce either underestimation or overestimation of forest loss.
3. Temporary forest loss may become permanent. Two scenarios temporary forest loss to consider:
   a) Future Restoration: Forest -> Shrubland | Cropland -> Forest
   b) Future Development: Forest -> Shrubland | Cropland -> Impervious | Barren
   The probability of future restoration vs. development is not measured in this analysis.
4. Forest -> Barren sometimes captures forest converted to higher intensity farming (usually cranberries, blueberries).

Columns

• 'STATECO': FIPS code of the state and county (e.g. AL is 01 and Macon County Alabama is 087 -> STATECO = 01087)
• 'COUNTY_NAME': name of the county
• 'state': Abbreviation of the state
• 'pct_d4st_temp': percent of county area that experienced temporary deforestation from 2010 - 2022
• 'pct_d4st_perm': percent of county area that experienced permanent deforestation from 2010 - 2022
• 'pct_forest': percent of county area that was forest land cover (NLCD) in 2010
• 'pct_forest_10_d4st_perm': percent of forest area in 2010 that was deforested permanently (d4st_perm / forest_10)
• 'fmv_ac_vac_CN': average (area-weighted) per-acre fmv_vac value of pixels in the parcel
• 'fmv_ac_all_CN': average (area-weighted) per-acre fmv_all value of pixels in the parcel
• 'd4st_perm_to_fmv': ratio of pct_d4st_perm to fmv_ac_all_CN. This is a relative impact value that is sorted by decile for visualization; the actual values are not intended for interpretation.
• 'geometry': the county polygon

Sources

Nolte, C. “High-Resolution Land Value Maps Reveal Underestimation of Conservation Costs in the United States.” Proceedings of the National Academy of Sciences of the United States of America 117, no. 47 (2020): 29577–83. https://doi.org/10.1073/pnas.2012865117

U.S. Geological Survey. Annual NLCD (National Land Cover Database)—The Next Generation of Land Cover Mapping. U.S. Geological Survey Fact Sheet 2025–3001, 2024. https://doi.org/10.3133/fs20253001

FIA. Land Resources Explorer. Last modified August 22, 2023. https://experience.arcgis.com/experience/ddb54b68e915431182d406f9778694cb/

U.S. Department of the Interior, Geological Survey, and U.S. Department of Agriculture. LANDFIRE Existing Vegetation Type Layer. 2022. Accessed May 13, 2025. https://landfire.gov/viewer/

U.S. Census Bureau. TIGER/Line Shapefile, 2016, Nation, U.S., Counties and Equivalent Entities. TIGER/Line Shapefiles, 2024. Accessed July 15, 2025. https://www.census.gov/geographies/mapping-files/time-series/geo/tiger-line-file.html

Acknowledgements

We thank the following people for their questions and contributions to background research that informed production of this dataset: Will Martin at American Forest Foundation; Tim Stout at Northam Forest Carbon; Meghan Blumstein, PhD at University of Virginia; Daniel Smith at Wherobots; Nan Pond at Rubicon Carbon; Christopher Williams, PhD at Clark University; Michael Flaxman at FireScore AI; Maria Huyer at Mast Reforestation; and Mackenzie Heiser at Arcadia.
This publication and work described are made possible by a cooperative agreement with the Forest Service, U.S. Department of Agriculture (USDA), under the Landscape Scale Restoration authority. The contents are those of the author(s) and do not necessarily represent the official views of, or an endorsement by, the Forest Service, USDA, or the U.S. Government. USDA is an equal opportunity provider, employer, and lender.