Master's thesis: Robust and uncertainty-aware urban land-cover mapping at property scale

Urban climate adaptation, biodiversity planning and water management depend on reliable information about individual properties. Within the Krontech research project, IVL and its partners are developing a tool that combines aerial imagery and geospatial data to quantify land cover at property level. This master's thesis will extend existing AI segmentation models and mapping workflows, focusing on robustness or more detailed hard-surface classification and their associated uncertainties.

The assignment

• Agree on a thesis direction and scope with IVL, taking student interests, available data and academic requirements into account.
• For the robustness direction, assess how seasonal variation, shadows and tree cover influence segmentation accuracy, and investigate improvements using multi-season imagery, complementary GIS or LiDAR data, shadow-aware training or uncertainty estimation for obscured ground.
• For the hard-surface direction, investigate how reliably asphalt, paving stones, gravel and compacted soil can be distinguished, including suitable data and annotation strategies.
• If pursuing detailed hard-surface mapping, evaluate how classification detail, accuracy and uncertainty affect selected indicators of urban heat, permeability or runoff.
• Build on existing annotations and model predictions using available orthophotos, LiDAR-derived information, NDVI, terrain data and GIS layers.

Who we are looking for

• Master's student or pair of students undertaking a 30 ECTS thesis.
• Experience programming in Python and training and applying neural networks.
• Experience with computer vision, semantic segmentation, GIS or remote sensing is an advantage.
• Familiarity with PyTorch, QGIS, GeoPandas, GDAL or rasterio is beneficial.

What we offer

• Practical experience addressing an applied environmental research problem.
• Collaboration with IVL researchers and Krontech project partners.
• An existing modelling workflow, trained-model outputs and geospatial datasets to build on; access to particular datasets depends on licences and project agreements.
• Opportunity to shape the final research scope together with IVL.

Practical details

Locations: Stockholm, Göteborg or Malmö. Scope: 30 ECTS for 1–2 students, with a flexible start date. The academic supervisor and examiner are arranged through your university. Apply using the “Register your interest” button.