The custom AI model was fed a dataset of images from Google Earth and was able to detect forest cover changes with an accuracy of 94.37 per cent.

Charles Darwin University (CDU)

Artificial Intelligence (AI) is the newest tool in the arsenal to prevent the degradation and depletion of forests, with new research revealing how the technology can help protect the ecosystem.

Charles Darwin University (CDU) researchers collaborated on an international study, led by the University of Sri Lanka, to develop an AI model which detects changes in forest cover, or the amount of land surface covered by trees.

According to the United Nations, between 2000 and 2022 there was a net forest area loss of 100 million hectares.

Researchers took U-Net architecture – which is used for image segmentation often in biomedical image analysis – and adapted it to compare past and present pictures of the ecosystem and detect where forest loss has occurred.

This custom model was fed a dataset of images from Google Earth and was able to detect forest cover changes with an accuracy of 94.37 per cent.

Researchers also tested the model on other datasets, which had an accuracy rate of 97.82 per cent and 98.44 per cent respectively.

Co-author and CDU Associate Professor in Information Technology Bharanidharan Shanmugam said the model was ideal for real-world applications because it produced high accuracy rates despite needing fewer training samples.

“Traditional methods for forest cover monitoring often struggle with accuracy and efficiency. Many rely on manual interpretation, which is time-consuming and prone to errors,” Associate Professor Shanmugam said.

“Our research provides a powerful tool for governments, environmental agencies and conservationists to detect and monitor deforestation more effectively.

“By leveraging deep-learning techniques, our model enables rapid analysis of satellite images, allowing authorities to identify high-risk areas and respond to deforestation before irreversible damage occurs.

“Unlike traditional approaches that require extensive manual effort, our method automates the process, making large-scale monitoring more feasible and cost effective.”

The study was a collaboration between the University of Sri Lanka, CDU, Friedrich-Alexander University in Germany, University of Peradeniya in Sri Lanka, and the University of Otago in New Zealand.

Co-author and CDU Lecturer in Information Technology Dr Thuseethan Selvarajah said another advantage was the model can function with limited labelled data.

“This makes it highly adaptable for use in regions where high-quality training datasets may not be available,” Dr Selvarajah said.

“Whether deployed in tropical rainforests, boreal forests, or temperate woodlands, the model can provide valuable insights for conservation efforts.

“By integrating this technology into existing environmental monitoring frameworks, governments and conservation organisations can enhance their ability to protect forests, enforce regulations, and mitigate the long-term impacts of deforestation.

“In the broader context, this research contributes to global efforts in combating climate change and preserving biodiversity.”

Change Detection for Forest Ecosystems Using Remote Sensing Images with Siamese Attention U-Net was published in the international journal Technologies.

INDEPENDENT NEWS IS IMPORTANT.

Enjoy reading Eco Voice? Please help us by purchasing a GIFT Voucher or send one to a friend and encourage people to purchase trees or seeds via The Native Shop – www.nativeshop.com.au