The EU’s new nature restoration law and Sweden’s biodiversity targets put strong emphasis on recovering semi-natural grasslands and open habitats – systems that store carbon and host pollinators and other species. However, they depend on well-timed grazing to maintain their potential to host these species. Many of those areas are in need for restoration and would be considered to be included in new restoration projects. Yet verifying where and when nature restoration occurs is difficult; field inspections are expensive and sparse, and self-reports are noisy. Earth observation (EO) can potentially streamline this significantly – satellites like Sentinel-2 provide free, multi-spectral imagery every few days over all of Sweden. The challenge, and opportunity, is to turn these raw time series into operational, trustworthy signals about restoration on the ground.
The thesis is part of the 2-year ongoing project “AI-based remote sensing for monitoring nature restoration and landscape elements at farm level” (https://www.rymdstyrelsen.se/innovation/beviljade-bidrag/rymdtillampningsprogrammet-2024-3/ai-basera…), where Arla is a key actor and stakeholder, and with stakeholders also including the Swedish Board of Agriculture and the Swedish Environmental Protection Agency.
Related reading:
- EU Nature Restoration Regulations: https://environment.ec.europa.eu/topics/nature-and-biodiversity/nature-restoration-regulation_en
- Swedish Board of Agriculture about nature restoration (Swedish): https://jordbruksverket.se/vaxter/odling/biologisk-mangfald/naturrestaurerings-forordningen
- The Swedish Environmental Protection Agency about nature restoration (Swedish): https://www.naturvardsverket.se/amnesomraden/mark-och-vattenanvandning/eu-forordning-for-att-restaur…
Description
In this master thesis, you will develop and train machine learning (ML) models for monitoring nature restoration in Swedish pastures, based on multi-year time series of satellite data. The work will include e.g. (i) preprocess existing nature restoration data to make it ML-ready; (ii) develop and implement ML method(s) in modern AI frameworks such as PyTorch; (iii) train and evaluate the ML methods and compare results with simpler / non-ML-based approaches.
The work requires students with skills in machine learning, image processing, and preferably also remote sensing, GIS and/or ecology. You will be expected to start out with a literature study, then begin with simpler models and eventually extend or develop more advanced solutions. As this is a master thesis project with a research organization, we will help you reach a high level of research excellence, and a successful project will ideally result in writing a joint research paper in addition to the master thesis.
Read more about the project at Master’s thesis; Machine learning and Earth observation data for monitoring nature restoration | RISE