PhD Candidate in Explainable AI for Earth Observation

With the growing integration of Artificial Intelligence (AI) in geospatial decision-making processes, ensuring transparency, accountability, and trust in model outputs has become imperative, especially in critical domains such as disaster risk management, social vulnerability mapping using satellite imagery, and ecological monitoring of invasive species.

You will explore how Explainable AI (XAI) methods can efficiently interpret geospatial models' behaviour. The central aim is to explore how such techniques can provide intuitive, human-interpretable explanations for complex model predictions derived from Earth Observation data, including high-resolution satellite and aerial imagery.

You will engage in a comparative analysis of XAI techniques in two key directions: their capacity to enhance the interpretability and usability of model outputs for non-technical stakeholders—partners, such as policy-makers, environmental managers, and disaster response coordinators—including the potential integration of Large Language Models (LLMs) to translate technical explanations into more accessible, natural language narratives — and their potential to iteratively improve model performance by integrating feedback from explanation-based diagnostics.

The research may be applied to diverse domains, ranging from damage assessment in post-disaster scenarios to detecting environmental change processes in vulnerable ecosystems.

You will present (both preliminary and mature) results at appropriate meetings and conferences, publish results in scientific workshops, conference proceedings, and journals, contribute to teaching on topics related to your work, and supervise and mentor involved MSc students as needed.

• An MSc degree in Geoscience, Computer Science, Artificial Intelligence, Mathematics, or a related field
• A curiosity in deep learning, image processing, GIS, and a willingness to delve deeper into these topics
• Passion for research and a background in geospatial modelling, machine learning, with programming experience in Python
• The ability to work on real-world problems in an interdisciplinary and internationally oriented environment
• Excellent communication skills and a strong command of the English language. Fluency in Dutch or willingness to learn Dutch is a plus

Interested and motivated candidates are encouraged to apply, even when they still need to obtain all the desired skills. You can develop relevant skills on the job through dedicated learning and doctoral training.

The Geo-Information Processing Department (GIP) works on the design and development of methods & techniques for processing (acquiring, organising, analysing) heterogeneous collections of spatiotemporal data, and in the implementation of open geo-information solutions (models, visualisations, and services) that help to understand key societal problems.