AI, Remote Sensing & Old-Growth Forests

Map of mountain village nested in a valley

Figure 1: The village of Rucar nested in Romania’s Carpathian Mountains.

Machine Learning and Remote Sensing for Old-Growth Forest Detection: A Carpathian Mountains Case Study

Thomas Ratsakatika¹, Srinivasan Keshav², Mihai Zotta³, Emily Lines¹
¹ Department of Geography, University of Cambridge, ² Department of Computer Science and Technology, University of Cambridge, ³ Fundația Conservation Carpathia

Abstract

Old-growth forests are irreplaceable ecosystems typically characterised by diverse tree structures, large old trees, coarse woody debris, and the absence of significant anthropogenic disturbances. Their complexity and stability provide critical ecosystem services and unique habitats for biodiversity. Despite their importance, 47 million hectares of old-growth forests were lost globally between 2000 and 2020, intensifying the need to map and monitor those that remain. While previous studies have used machine learning to detect old-growth forests with varying accuracy, there is a lack of systematic, quantitative comparisons of machine learning and remote sensing techniques for this purpose.

This thesis will use the Făgăraș Mountains in the Romanian Carpathians as a case study to systematically analyse machine learning and remote sensing techniques for advancing the detection of temperate and montane old-growth forests. First, it will investigate how different classification algorithms, including random forests, convolutional neural networks and vision transformers, impact the accuracy of old-growth forest detection from Sentinel-2 optical data. Second, it will examine how different earth observation data types and resolutions, including Sentinel-1 synthetic aperture radar and ultra-high resolution images, contribute to detection accuracy. Third, using at least 300 mobile LiDAR scans as a ground truth, it will assess the capabilities and limitations of Earth observation data for measuring specific indicators of old-growth forests, such as structural complexity and the presence of large old trees. This research will be conducted in collaboration with Fundația Conservation Carpathia, a Romanian conservation organisation, and the findings aim to inform conservation strategies within the Carpathian Mountains and beyond.