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

Were currently looking for a masters student who would like to carry out their thesis project with us. This is an opportunity to work on a real-world research challenge gain hands-on experience in applied environmental research and collaborate with experts in the field. By doing your masters thesis at IVL youll gain unique insights into our research and contribute to solutions that support a more sustainable society.

Cities and property owners need fast reliable facts about their sites to plan for climate adaptation biodiversity and water management. Krontech is a research project where a practical tool is being built that turns aerial images and GIS data into clear property-level numberssuch as how much area is green or paved tree-canopy cover and where improvements would have the greatest effect.

We already combine orthophotos laser-scanned height models (LiDAR) NDVI and terrain data. We are also training AI segmentation models to produce detailed land-cover maps. This thesis will build on the existing models data and mapping workflow.

The thesis could explore one of the following directions or a related topic developed together with the student.

Investigate how seasons shadows and tree canopies affect land-cover segmentation and develop methods that make the existing approach more robust. Possible approaches include multi-season imagery shadow-aware training additional GIS or LiDAR data and uncertainty estimation for ground cover beneath trees.

Possible research questions

• How do season shadow and tree cover affect segmentation accuracy

• Can multi-season or complementary data improve model robustness

• How can uncertainty be estimated for partly or completely hidden surfaces

Develop a more detailed classification of hard surfaces such as asphalt paving stones gravel and compacted soil. Evaluate the impact of this additional detail and its uncertainty on one or more environmental indicators such as indicators related to urban heat permeability or water runoff.

Possible research questions

• Which hard-surface types can be reliably distinguished

• What data and annotation strategy are needed

• How does a more detailed classification affect selected environmental indicators

• How do classification accuracy and uncertainty affect the reliability of these indicators

The available data includes:

• High-resolution orthophotos from multiple years

• Existing land-cover annotations and model predictions

• LiDAR-derived elevation and canopy information

• NDVI and terrain data

• Building road property and other GIS layers

Access to individual datasets is subject to applicable licences and project agreements.

The ideal candidate has experience in:

• programming in Python

• training and using neural networks

Experience with computer vision semantic segmentation GIS remote sensing PyTorch QGIS GeoPandas GDAL or rasterio is beneficial.

The thesis will be carried out in collaboration with IVL Swedish Environmental Research Institute and the Krontech project partners. The final direction and scope will be agreed upon with the student based on their interests available data and academic requirements.

Host Organization

IVL Swedish Environmental Research Institute

Location: Stockholm/Gothenburg/Malmö

Credits: 30 ECTS

Group size: 12 students

Start date: Flexible

Academic supervisor/examiner: arranged with your university

Wed love to hear from you. Submit your application by clicking Register your interest.