Thesis subject

MSc thesis topic: Classifying deforestation drivers in Europe using Sentinel satellites and AI

Recent findings reveal a shift in carbon sink capacity, with boreal and temperate forests as the main contributors to global carbon uptake (Figure 1), while tropical forests have become small carbon sources due to deforestation and increased tree mortality [1,2]. However, European forests are increasingly under threat from a range of natural and anthropogenic pressures. Key drivers of deforestation include logging for timber and fuelwood, urban expansion and infrastructure development, agricultural conversion (e.g., vineyard expansion in southern France), and climate-related disturbances such as wildfire, prolonged drought, and insect outbreaks (e.g., bark beetles). These disturbances not only lead to forest loss but can also compromise the long-term carbon sink capacity of these ecosystems.

Despite rising awareness, there is still a critical knowledge gap in systematically identifying, quantifying, and mapping the specific drivers of deforestation across temperate regions, particularly in high-value forested areas. To address this, Artificial Intelligence (AI) combined with freely available Earth Observation (EO) data such as GLAD-L [4] forest alert products, Sentinel-1 SAR, and Sentinel-2 imagery offers a cost-effective solution. AI-driven models can process vast time series datasets, detect subtle patterns of forest change, and classify deforestation drivers with high accuracy [3]. By harnessing these tools, we can generate spatially explicit, up-to-date maps of deforestation and its drivers, providing actionable insights to support conservation planning.

Background

  • The student undertaking this project will be expected to access and preprocess Sentinel-1 and Sentinel-2 data over selected forested regions in France and Germany for 1-year historical period.
  • Perform visual interpretation of EO imagery to delineate and annotate polygons representing different deforestation drivers (e.g., logging, agriculture, urban development).
  • Train and validate AI models, specifically Convolutional Neural Networks (CNNs), to classify and map the drivers of deforestation using the annotated data and time-series EO products.
  • A basic knowledge of AI and machine learning workflows, and experience with CNNs is recommended for this project.

Software: Google Earth Engine, python

Objectives and Research questions

  • To develop a methodological framework utilizing AI for mapping deforestation and its primary drivers over the past three years.
  • To apply this framework in France and Germany using freely available satellite data and forest alert products.
  • To identify spatial patterns and trends of deforestation drivers to inform forest management and policy.

Requirements

  • Geo-scripting course
  • (Advanced Earth Observation)
  • Machine learning
  • Deep learning

Literature and information

  1. Yang, Hui, et al. "Global increase in biomass carbon stock dominated by growth of northern young forests over past decade." Nature Geoscience 16.10 (2023): 886-892.
  2. Pugh, Thomas AM, et al. "Role of forest regrowth in global carbon sink dynamics." Proceedings of the National Academy of Sciences 116.10 (2019): 4382-4387.
  3. Slagter, Bart, et al. "Monitoring direct drivers of small-scale tropical forest disturbance in near real-time with Sentinel-1 and-2 data." Remote Sensing of Environment 295 (2023): 113655.
  4. Hansen, M. C., et al. (2016). Humid tropical forest disturbance alerts using Landsat data. Environmental Research Letters, 11(3), 034008.

Theme(s): Sensing & measuring; Modelling & visualisation