Biodiversity in forests gets a clearer, more accurate picture thanks to aerial imagery and AI
Aspen trees (Populus tremula) play a crucial keystone role in forests. In Finnish woods, aspens are relatively rare, making up only a small fraction of trees and occurring in even fewer stands. Yet these trees support a rich community: more than a thousand species of mammals, birds, insects, fungi, and mosses rely on aspen ecosystems. The rapid decay of aspen litter speeds nutrient cycling, and many organisms depend on the distinctive, gnarled, base-rich bark that aspens offer.
A research team led by Associate Professor Samuli Junttila at the University of Helsinki has developed a reliable method to distinguish aspen trees from other species using open aerial imagery data. This marks the first time researchers can identify aspens at scale with accuracy, enabling efficient mapping of their distribution without the high costs and labor of traditional field surveys.
“A scalable, affordable, and up-to-date approach for monitoring forest biodiversity is now within reach,” says Junttila.
The Global Ecosystem Health Observatory (GEHO), under Junttila’s leadership, blends remote sensing with AI to extract precise information about forest health and ecology. Their techniques have previously supported efforts to track damage from the European spruce bark beetle and to assess tree mortality.
The study’s lead author is Doctoral Researcher Anwarul Chowdhury from the University of Eastern Finland. He expresses satisfaction with the neural network’s accuracy and its potential impact.
“This work demonstrates that our methods can deliver dependable data tailored to the practical needs of forest management and conservation across all of Finland’s forests,” Chowdhury notes.
The model proved effective across seasons, reliably spotting aspens whether or not they were bearing leaves. Interestingly, the system was more often correct in identifying taller trees than shorter ones—about a 71% likelihood for fully grown, taller aspens, with even higher accuracy when the trees were leafless. This emphasis matters because tall, mature aspens are particularly valuable for biodiversity.
Looking ahead, the researchers aim to improve the model’s ability to detect young aspens. “In the future, integrating open lidar data with aerial imagery could help us recognize juvenile aspens more accurately,” Chowdhury envisions.
Dead trees also play a vital role in biodiversity
In November, a separate study from Junttila’s group, led by researchers Anis Rahman, Einari Heinaro, and Mete Ahishali, advanced techniques for identifying standing dead trees from aerial imagery with greater precision. Dead trees support many specialized and even threatened species, yet detecting them under dense canopies is challenging. By combining machine-learning methods with adaptive filtering, the team achieved better results than general-purpose forest remote-sensing models.
“Aspens and standing dead trees are important biodiversity indicators, and automated mapping from open datasets represents a major step forward for monitoring forest ecosystems,” Junttila remarks.
References:
Rahman AU, Heinaro E, Ahishali M, Junttila S. Dual-task learning for dead tree detection and segmentation with hybrid self-attention U-Nets in aerial imagery. Int J Appl Earth Obs Geoinf. 2025;144:104851. doi:10.1016/j.jag.2025.104851
Chowdhury AI, Heinaro E, Tanhuanpää T, Rahman AU, Junttila S. Mapping large European aspens (Populus tremula L.) using national aerial imagery and a U-Net convolutional neural network. Remote Sens Appl Soc Environ. 2025;40:101755. doi:10.1016/j.rsase.2025.101755
This article is republished from Helsinki University sources. Note: content may have been shortened or edited for length. For more information, please contact the cited source. Our republishing policy is available here.
