Machine-learning forest map suggests fewer large trees in North America than previously estimated
image:
(a) Methodological workflow for continental-scale tree density estimation. (b) Modular framework of the deep learning and remote sensing pipeline used in this study.
view moreCredit: Jingjing Liang, Forest Advanced Computing and Artificial Intelligence (FACAI), Department of Forestry and Natural Resources, Purdue University, USA
How many trees are there in North America's forests? A new study published in Forest Ecosystems brings researchers closer to answering that question by combining forest survey data, satellite observations, and machine learning to map tree density across Canada, the United States, and Mexico.
Tree density is a key indicator of forest structure and plays an important role in studies of carbon storage, biodiversity, ecosystem functioning, and forest management. While national forest inventories provide valuable information, differences in data availability and sampling methods can make it difficult to assess tree density consistently across large geographic regions.
To address this problem, the research team combined forest inventory data from more than 600,000 forest inventory plots with environmental information derived from satellites and other sources, including climate, soils, and terrain. They then compared several machine-learning approaches to determine which could best predict tree density across North America at a resolution of about 3 km.
Among the models tested, a feedforward neural network (a type of machine learning model) outperformed other approaches in predictive accuracy and was selected to generate the continental maps.
The maps reveal clear regional patterns. The highest tree densities occur in boreal and temperate conifer forests across Canada, Alaska, and the Pacific Northwest. Moderate densities are found in many eastern forests, while deserts and other dry regions support far fewer trees.
Using this approach, the researchers estimate that North America contains between 339 billion and 514 billion trees with DBH＞10 cm. This range is lower than a widely cited previous estimate of 603 billion trees for the continent.
The study examined where predictions are more reliable. Areas with abundant inventory data, particularly in the United States and southern Canada, show lower uncertainty. In contrast, regions with greater environmental complexity or less comprehensive survey coverage, such as northern boreal forests and mountainous regions, display higher uncertainty.
One of the study's key findings is that the estimates of tree abundance is sensitive to forest definitions and tree-size thresholds. Results varied according to the forest maps used and the minimum tree size included in the analysis. These technical choices are often overlooked but have major consequences for policy and carbon accounting.
The framework supports forest monitoring, biodiversity assessment, and carbon accounting applications. Its modular design allows updates as new inventory data and satellite observations become available, providing a consistent approach for tracking forest conditions over time.
By integrating harmonized data, remote sensing observations, and explicit uncertainty analysis, this study establishes a reproducible foundation for large-scale forest assessments across North America.
DOI Link:
https://doi.org/10.1016/j.fecs.2026.100466
Journal
Forest Ecosystems
Article Title
Continental-scale mapping of forest tree density in North America using remote sensing and deep learning with uncertainty quantification
How can scientists predict tree growth for species they have never seen before?
SciOpen
image:
Data used in this study were collected from the long-term permanent forest inventory and monitoring plots established and maintained by the FIA program on the larger islands of the Puerto Rico (main island of Puerto Rico, Culebra, Mona and Vieques) and US Virgin Islands (St. Croix, St. John, and St. Thomas) archipelagos.
view moreCredit: Sheng-I Yang, Warnell School of Forestry and Natural Resources, University of Georgia, Athens, USA
Tropical forests are home to an astonishing number of tree species, but many of them are rare. This diversity poses a major challenge for forest researchers who need to predict future tree growth: how do you build a growth model for a species you have barely seen before?
A new study from researchers at the University of Georgia and the U.S. Forest Service provides answers. Published in Forest Ecosystems, the study evaluates four different strategies for projecting tree diameters for "unobserved" species that were not included when a growth model was originally developed.
Using 20 years of forest inventory data from Puerto Rico and the U.S. Virgin Islands, the team tested four common approaches:
Approach I—Fixed-effects only predictions
Approach II—Post-hoc species-specific adjustments
Approach III—Separate models with/without species effects
Approach IV—Hybrid data grouping with mixed modeling
The results showed that all four strategies can reliably project future tree diameters for new species. However, approaches that incorporate species-level information tended to produce more precise predictions.
Among the four strategies, the researchers suggest that post-hoc calculation of species-specific adjustments (Approach II) is particularly useful because predictions can be calibrated. The trade-off is that this method requires at least one repeated measurement for the new species.
For situations where no repeated measurements are available, the study found that grouping rare species into an "others" category (Approach IV) also works well. The researchers determined that a threshold of 25 observations per species was optimal for defining which species should be grouped together (i.e., species with fewer than 25 observations were grouped into an “others” group).
The study also revealed broader patterns in Caribbean forest growth. Trees in Puerto Rico generally showed greater diameter growth than those in the U.S. Virgin Islands, likely reflecting differences in climate and forest conditions. Moist and wet forests tended to support faster growth than dry forests, while forest diversity and competition among trees also influenced long-term growth rates.
According to the research team, improving tree growth prediction is becoming increasingly important as Caribbean forests face intensifying environmental pressures, including stronger hurricanes, prolonged droughts, and climate change.
"This work provides a working example for selecting the proper methodology to handle rare species," said Sheng-I Yang, the lead author of the study. "That's critical for the continuous monitoring of diverse and vulnerable forest resources, not just in the Caribbean, but in tropical forests worldwide."
As climate change and more frequent hurricanes threaten tropical forests, the ability to accurately project tree growth becomes increasingly important for predicting forest recovery, carbon storage, and ecosystem resilience. The study suggested that further research is needed to evaluate the proposed methods across different forest types and environmental conditions in other tropical forests.
DOI Link:
https://doi.org/10.1016/j.fecs.2026.100458
Journal
Forest Ecosystems
Article Title
Examining strategies to project tree diameter for unobserved species in diverse tropical forests using mixed-effects models
No comments:
Post a Comment