Georgia study models historic American chestnut distribution from 1804–1832 land surveys

An observational species-distribution study using Georgia land-lot surveys from 1804–1832 found that American chestnut (Castanea dentata) occurred across all five of the state’s physiographic provinces, with higher sampled abundance in the Piedmont than in three mountainous provinces. The researchers used the records to model associations between historical chestnut occurrences, soil properties and landform characteristics, identifying elevation as the most influential predictor in nearly every model.

The study covered the area represented by the Georgia land lottery maps, approximately 70% of the state. Surveyors recorded “witness trees” around land-lot boundaries and intersections, identifying species on hand-drawn maps. The researchers systematically searched all 392 digitized maps and distinguished American chestnut from chestnut oak. Among 717,901 sampled trees, 15,710, or 2%, were identified as chestnut. This extensive record provided pre-clearing observations that may predate widespread chestnut blight and, possibly, widespread effects of Phytophthora root rot.

The survey data documented a markedly uneven distribution. Chestnut represented 3.1% of sampled trees in the Piedmont, 2.8% in the Blue Ridge, 1.0% in the Ridge and Valley, 1.0% in the Appalachian Plateau and 0.25% in the Coastal Plain. The Coastal Plain calculation included only the portion within the documented occurrence area. The Piedmont proportion was 11% higher than the Blue Ridge proportion and more than 200% higher than the proportions in the Ridge and Valley and Appalachian Plateau. These findings contrast with anecdotal accounts and range maps that have commonly excluded much of the Georgia Piedmont and Coastal Plain.

For habitat modelling, the researchers paired chestnut locations with 18 edaphic and landform predictors derived from the Gridded Soil Survey Geographic database and 10-metre digital elevation data. Soil variables included particle size, drainage, clay content, erosion class, pH, frost-free days, parent material and soil taxonomy. Landform variables included elevation, slope, aspect expressed as eastness and northness, and curvature. None of the predictors showed collinearity problems. Because the surveys did not document confirmed absences, the analysis used 47,130 geographically random background points, three times the number of chestnut locations.

Species-distribution models were constructed in the sdm package for R. The researchers compared boosted regression trees, generalised additive models, generalised linear models, maximum entropy and random forest, with each run replicated three times and 30% of the data reserved for testing under subsampling and bootstrapping procedures. Random forest with bootstrapping consistently produced higher area-under-the-curve scores and better separation between chestnut occurrences and background points, so only random-forest results were reported. Performance was evaluated with AUC, correlation score and true skill statistic.

The full statewide model achieved an AUC of 0.96, a correlation score of 0.74 and a true skill statistic of 0.77. Elevation had a normalised importance score of 0.66, compared with 0.04–0.10 for other variables, and was followed by number of frost-free days, slope, percentage clay, curvature, pH, northness and eastness. Response curves associated chestnut occurrence with increasing elevation, slope, eastness, pH, clay percentage and either convex or concave curvature, while the full model associated it with fewer frost-free days. These are modelled associations, not evidence that the individual predictors caused historical occupancy.

Models were also fitted separately for the Appalachian Plateau, Blue Ridge, Ridge and Valley, Piedmont and Coastal Plain. Their AUC values ranged from 0.86 to 0.95, with the Coastal Plain models producing the highest scores. Elevation was the strongest influence in every model except the full Coastal Plain model, in which slope ranked first. Other relationships varied among provinces and between full and reduced models, including responses to aspect, frost-free days, pH and clay content. Northness was positively associated with occurrence in the statewide and Ridge and Valley models but negatively associated in the Coastal Plain and Appalachian Plateau, indicating context-dependent relationships among environmental variables.

A reduced statewide model was used to create the habitat-suitability map because missing data in four soil variables prevented the full model from producing a complete predictive map. It placed the greatest concentration of highly suitable habitat in the upper Piedmont and Blue Ridge, reflecting elevation’s dominant statewide influence. The province-level maps are more geographically nuanced. These habitat estimates remain models rather than direct measurements of historical tree abundance, and their interpretation is bounded by uncertainty in occurrence locations, random background sampling, surveyor preferences, incomplete soil records and the absence of contemporary climate data.

The authors concluded that American chestnut was more broadly distributed and adaptable across Georgia community types than records based on surviving trees, sprouts or isolated historical accounts suggest. They recommend that restoration practitioners consult the individual physiographic-province models and maps rather than relying only on the statewide map. The study does not demonstrate that disease-resistant planting will succeed or that modelled suitability eliminates pathogen pressure; improved chestnut stock remains vulnerable to both chestnut blight and Phytophthora root rot. Its restoration value is instead in identifying where historically suitable conditions occurred and in documenting the variety of Georgia habitats in which the species was once present.

Subject of Research: Historical modelling of American chestnut distribution and habitat suitability in Georgia using soil and landform predictors

Article Title: Modelling the historic distribution and habitat of American chestnut ( Castanea dentata ) in Georgia, USA using edaphic and landform predictors

Article References: Klaus, J. M., & Klaus, N. A. (2025). Modelling the historic distribution and habitat of American chestnut (Castanea dentata) in Georgia, USA using edaphic and landform predictors. Frontiers of Biogeography, 18, Article e161937. https://doi.org/10.21425/fob.18.161937

Image Credits: AI Generated

DOI: 10.21425/fob.18.161937

Keywords: American chestnut, Castanea dentata, species distribution modelling, Georgia, historic land surveys, habitat suitability, soil predictors, landform predictors, forest restoration, Modelling, historic, distribution