Counting Trees From Space with Artificial Intelligence
Counting Trees From Space with Artificial Intelligence
https://www.newswise.com/articles/counting-trees-from-space-with-artificial-intelligence
Publish Date: 2026-06-24 08:45:00
Source Domain: www.newswise.com
-
Transformation in Research Tracking: The article highlights how satellite monitoring has revolutionized the way researchers track deforestation, land-cover change, and ecosystem dynamics.
-
Methodology Limitations: It addresses the shortcomings of widely used global products based on medium-resolution imagery, which often fail to distinguish single trees in open landscapes, farms, savannas, and fragmented forests.
-
Advances in High-Resolution Imagery: Despite the benefits of very-high-resolution imagery, the manual annotation and cost associated with training such systems are major barriers.
-
Need for Scalable Tools: There is a critical need for scalable, accurate, transferable, and cost-effective methods to count, locate, and monitor individual trees over time amid climate change and land-use pressures.
-
New Deep Learning Approach: The study introduces a novel anchor-free deep learning method for detecting large individual trees using 3-meter PlanetScope imagery by creating Gaussian heatmaps to represent tree crowns.
-
Key Innovation – Heatmap Representation: Each tree center is modeled as a peak on a heatmap, allowing flexible modeling of various crown sizes and enhancing the system’s ability to handle noisy data.
-
Significant Dataset and Model Performance: The training dataset includes about 14 billion tree points, and the method achieves state-of-the-art performance in cover-mapping with good generalization across biomes.
-
Future Prospects and Challenges: The approach offers potential for climate science, forest management, biodiversity assessment, and carbon accounting, but challenges remain such as limited LiDAR coverage and accurate threshold selection.
