Challenges
Urban planning increasingly requires high-resolution, semantically rich data to support climate adaptation, infrastructure development, and emerging airspace management for drones.
Norrköping Municipality faced the challenge of creating a digital twin that could serve both as a planning tool and as a foundation for future drone traffic management systems. Existing datasets were insufficient for training AI models capable of detailed semantic segmentation of urban surfaces, materials, and vegetation.
The project aimed to address this gap by collecting multimodal airborne data and developing AI-driven methods for sensor fusion, enabling accurate classification and visualization of complex urban environments.
Solutions
The solution involved a large-scale airborne data collection campaign using a fixed-wing aircraft equipped with RGB, thermal, and hyperspectral sensors. RGB imagery was captured at 4 cm GSD (Ground Sampling Distance), thermal at 20 cm, and hyperspectral at 90 cm, providing complementary layers of information. These datasets were harmonized and fused to generate training data for a semantic segmentation AI model. Compared to RGB-only baselines, the fused approach significantly improved classification accuracy for surface types such as asphalt, concrete, and gravel, as well as vegetation categories and building elements.
The classified outputs were converted into GIS layers and integrated into municipal planning workflows, supporting analysis of green infrastructure and urban heat islands. In parallel, the data were used to create a visually and materially credible digital twin in Unreal Engine, enabling realistic simulations for stakeholder engagement and scenario planning.
The project also connected with research platforms such as Visual Sweden City Platform and UTM-City, ensuring alignment with future airspace management requirements.
Results and Benefits
The project delivered transformative benefits for Norrköping Municipality. By providing semantically rich GIS layers, the initiative strengthened the decision-making support in climate adaptation, enabling precise mapping of green structures and heat islands. This improved prioritization of interventions and enhanced communication with policymakers and citizens. The digital twin, enriched with mask-driven material assignments and vegetation placement, offered unprecedented realism in simulations, increasing confidence in planning outputs even among non-experts.
Beyond urban planning, the semantic layers laid the foundation for future drone autonomy by enabling GPS-independent navigation and semantics-guided route planning, critical for safe BVLOS (Beyond Visual Line of Sight) operations in GNSS-degraded environments. The integration with UTM-City positions Norrköping as a frontrunner in managing low-altitude airspace, a responsibility that municipalities will increasingly assume.
The project also catalyzed research momentum, with datasets now supporting ongoing collaborations with FOI and forming the basis for new R&D proposals in autonomous navigation and semantic mapping. In summary, the initiative established a complete pipeline from data collection to actionable insights, creating immediate municipal value while paving the way for robust, interpretable autonomous systems and sustainable urban development.
Perceived Social/Economic Impact
The project enhances urban resilience by enabling data-driven climate adaptation and infrastructure planning. It supports safer drone operations, fostering innovation in logistics and emergency response. By integrating advanced AI and sensor fusion into municipal workflows, the initiative strengthens Sweden’s position in smart city development and creates opportunities for local businesses engaged in geospatial technologies and drone services.
Lessons Learned
One of the most important lessons was the value of investing in multimodal data collection, combining RGB, thermal, and hyperspectral imagery to create richer semantic layers and improve classification accuracy.
Early engagement with municipal departments proved essential for integrating these data products into planning workflows and climate adaptation strategies. The project also highlighted the benefits of sensor fusion for training AI models, as this approach significantly enhanced segmentation of surfaces, vegetation, and building elements compared to relying on a single data source. Finally, aligning technical development with ongoing research platforms such as Visual Sweden and UTM-City ensured that the work was scalable and positioned for future impact.
The experience showed that relying solely on RGB imagery for semantic segmentation is insufficient, as it limits accuracy for complex urban features. It also became clear that the complexity of data harmonization and security considerations should never be underestimated, since geospatial data can be sensitive and requires careful handling. Another insight was the importance of planning for public data release and API standardization early in the process, as these steps are critical for long-term adoption and interoperability.
“The dataset is considered unique globally and is expected to place Norrköping at the forefront of creating digital twins. The flight over Norrköping will play a crucial role in the collection of valuable urban and environmental data” says Erik Telldén, Researcher, Linköping University and Visual Sweden
