Thesis project: Interpretable AI methods for naturalness evaluation of forests

MSc. Thesis project at the Dept. of Mechanics and Maritime Sciences, Chalmers University of Technology

Status: šŸŸ¢

This thesis project is assigned to Oscar StƄlnacke.
Last update: 2023-12-18

Background and scope

The preservation and management of natural ecosystems, such as forests, are of paramount importance to address environmental challenges and maintain biodiversity. As Artificial Intelligence (AI) technologies play an increasingly significant role in ecological research, there is a growing interest in utilizing AI methods for evaluating the naturalness of forests. National authorities such as the Swedish Forest Agency (Skogsstyrelsen) and Environmental Protection Agency (NaturvƄrdsverket) have several field survey methods to assess the degree of naturalness, which are based on the presence of indicator species, characteristics of the forests and trees (for example, age, species composition, stand structure), abiotic factors (for example, terrain and soil properties) and past management regimes.

The aim of this study is to develop interpretable AI methods to automatically evaluate the degree of naturalness of forests from existing national datasets (aerial photos, lidar data, etc.).

Heights of the tops of the trees from the Canopy Height Model

(Visualization of the tops of the trees, derived from the Canopy Height Model. The heights of the trees are used as input in a subsequent component of the system.)

Purpose of the work

The primary purpose of this thesis is to enhance existing AI methods for evaluating forest naturalness by incorporating interpretable components into the codebase developed by the AAI research group at Chalmers. Examples of components that can be added to the current model are:

  • Given the Canopy Height Model as input, modeling the ā€œfire resistance scoreā€, which is a measure of the effect of a fire on a given forest. Given the Canopy Height Model as input.
  • Given the Digital Terrain Model as input, modeling the ā€œterrain variety scoreā€, which is a measure of the variety of the terrain where the forest has grown.

Activities included in the thesis work

  1. Studying the problem and the available datasets
  2. Familiarizing yourselves with the existing codebase (Python)
  3. Designing new or enhance the existing methods for evaluating forestsā€™ naturalness
  4. Implementing the new methods
  5. Comparing with the state-of-the-art
  6. Integrating the new methods into the codebase
  7. Testing and validation
  8. Writing the thesis report

Profile of the student

The suitable student for this thesis work should have a solid foundation in computer science, programming, and machine learning concepts. Proficiency in programming languages like Python is crucial, as is familiarity with version control systems (e.g., Git) for code integration. The student should demonstrate the ability to work with existing code, adapt and extend it, and possess strong problem-solving skills to overcome technical challenges. Additionally, good written and verbal English skills are necessary.

Partner organizations and related projects

This thesis relates to the TolkAI research project in partnership with: