Scientists at Altai State University have devised an AI-powered method that uses satellite imagery to accurately identify outbreaks of forest diseases and pest infestations. The regional Forest Protection Center has already adopted this technique, as it enables the assessment of tree damage in remote areas that forest pathologists cannot reach by off‑road vehicles or helicopters. For more information, read the Rossiyskaya Gazeta (RG) report.

The total forest area in Altai Krai spans nearly 4.5 million hectares, with over 29,000 hectares currently suffering from various forms of damage. At first glance, the proportion of "unstable" forest stands—a term used by forest pathologists—appears small, representing just 0.64 percent of the total area. However, experts are alarmed by the accelerating spread of diseases and pests in recent years. For instance, in the Novichikha forestry district, tree dieback has expanded by 5,000 hectares over the past five years, while in the Gorno-Kolyvansk forestry district, bark beetles damage has doubled in just eight years.

Biologists at Altai State University (ASU) identify several drivers behind the growth in ​​damaged coniferous forests. The first is a massive rise in groundwater levels that occurred in 2016, which flooded roughly 24,000 hectares of pine forest and whose effect are still evident today. Another factor is the spread of the Ussuri bark beetle (Polygraphus proximus), an invasive dangerous pest believed to have been introduced to Altai from the Russian Far East back in 2012. This pest can drive a healthy tree to a critical condition within just two to three years.

The complete cycle of total forest dieback takes spans ten years, and researchers have learned to track changes in forest health across this specific timeframe using satellite imagery. With the advent of artificial intelligence, they have developed a universal algorithm that can remotely identify stages of tree decline with high precision. The developer, Lyudmila Dolgacheva—a senior lecturer in the Department of Physical Geography and Geoinformation Systems at ASU—previously worked in forest protection and has firsthand knowledge of the industry's challenges. For over a decade, she has been seeking ways to safeguard the planet's "green lungs" from destruction.

We took a classic method of satellite image interpretation—using spectral analysis to locate and measure damaged forest areas—and combined it with AI capabilities, Lyudmila Dolgacheva told an RG correspondent. The AI ​​can assess the degree of damage based on geometric features, such as the shape and size of tree crowns. This created a tandem linking the Russia’s 'Kanopus-V' system with Europe’s Sentinel-2, powered by the robust U-Net neural network. Thanks to intelligent processing of satellite imagery, data accuracy has reached 95 percent, enabling informed and effective decisions on forest protection against diseases and pests. For instance, if only an early stage of tree damage is detected, a treatment application suffices; however, if AI identifies a critical stage, sanitary felling is required.

Before this algorithm was developed, forest pathologists would travel to affected sites after receiving satellite imagery to conduct detailed on-the-ground assessments—counting damaged trees and evaluating their condition. However, accessing every part of the forest is not always possible; ground-based monitoring is simply unfeasible in hard-to-reach locations, such as flooded, swampy, or high-altitude areas. This is where artificial intelligence comes in, providing an accurate picture of the situation—down to the condition of individual trees—without the need for field visits. The algorithm’s output is overlaid onto a map using the open-source software QGIS, giving foresters a ready-made "disease map" to support urgent decision-making.

To start with, we focused on just two types of damage, noted Lyudmila Dolgacheva. These are waterlogging in flooded areas and infestation by one of the most dangerous pests—the Ussuri bark beetle (Polygraphus proximus). This choice was deliberate: these specific types of damage most frequently occur in hard-to-reach forest areas. However, we plan to eventually train the neural network to accurately identify all possible forms of damage to forest stands, although we may need to develop new research methods to achieve this.

According to Irina Rotanova, an associate professor in the Department of Physical Geography and Geoinformation Systems at Altai State University (AltSU), the use of modern geoinformation technologies and artificial intelligence to address practical issues of natural resource management is currently one of the most relevant areas in geography and geoecology. Such developments will enable foresters to protect forest resources from pests and the vagaries of nature with scientific precision rather than by guesswork.

The AI-based methods developed by biologists to detect areas of forest damage have already been incorporated into AltSU’s training curricula for forestry specialists.