UPDATE: Project details updated during the EPSA 2025-26 edition.
Context
Recklinghausen, located in North Rhine-Westphalia, together with the Westphalian University of Applied Sciences, has developed software based on artificial intelligence (AI), which automatically derives a map of impervious surfaces from aerial images and aggregates it to any area divisions. According to statistics from the German Federal Environmental Agency, currently about 45 per cent of settlements and traffic areas in Germany are ‘sealed’ i.e. air- and watertight, and thus not available for the infiltration of rainwater. In addition, sealed surfaces are usually potential heat stores, which contribute greatly to the warming of urban areas. In view of the increasing threat of climate change, precise and continuous monitoring of sealed surfaces is an important urban planning tool for effective drainage control and avoidance of heat islands.
In the past, explicit mappings of impervious surfaces were generally recorded manually on the basis of image evaluations, but such manual mapping and its maintenance is time-consuming and costly. Given the precise and continuous acquisition of the properties via image, laser and other sensors, it is much more effective to automatically detect sealing surfaces using AI-based systems. The software generates object-sharp maps based on overflight image data.
Objectives
The aim of the automatic detection of impervious surfaces software is to enable the assessment of areas more precisely and in more detail with regard to their climate-ecological nature. In turn, these findings can serve as the basis for action concepts and adaptation measures to implement climate-effective measures for the district cities in the next step. The software provides important support in the targeted planning and evaluation of measures for more climate and environmental protection in cities, and in the implementation of sustainable, municipal land management.
Implementation
The software for automated detection and classification of impervious surfaces was developed by the district of Recklinghausen and carried out jointly with the Westphalian University of Applied Sciences. The project was carried out over 12 months within the framework of funding for a research associate at the University of Westphalia in the Department of Electrical Engineering and Applied Natural Sciences of just under €100 000. The University of Westphalia was a key part of the project, due to its expertise in the field of deep learning and computer vision in relation to geodata. In regular project meetings between the University of Westphalia and the cities involved, the objectives were defined and the progress of the project was continuously monitored. In the case of the district of Recklinghausen, employees from the areas of land registry and geo-information, as well as environment and climate protection, were involved.
Developments
Since the submission of the initial project application, the ADOIS project has progressed from a conceptual and pilot-oriented initiative to an easy to build, run, and integrate open-source solution for the automated detection of impervious surfaces.
During implementation, methodological and organisational lessons were systematically incorporated. In particular, the use of high-resolution visible colour and near infra red orthophotos combined with semantic segmentation using deep neural networks proved to be robust and scalable. Within the project, responsibilities were clearly allocated between the different corresponding tasks, such as technical development (model training, software engineering) or administrative application (data provision, validation, use in planning processes). Kreis Recklinghausen as well as other partners such as Emschergenossenschaft und Lippeverband (EGLV), and the Regionalverband Ruhr (RVR) provided substantial support to the project. Their contributions included the provision of relevant training data, the labelling and refinement of datasets, the definition of impervious surface classes, and technical consultation on system requirements. The partners were also closely involved in defining aggregation rules for the post-processing workflow and in establishing quality assurance metrics. These contributions enable regular, consistent, and comparable software updates and ensure the automated generation of high-quality impervious surface maps.
The needs of the primary target groups (district and municipal planning, cadastral, environmental, and climate protection departments) were reassessed during the project. This confirmed a strong and growing demand for frequently updated, spatially detailed, and legally robust imperviousness data to support climate adaptation, land management, and monitoring obligations. This reassessment directly informed the decision to institutionalise the solution as a reusable and openly available software tool. The results are used in the municipality Recklinghausen to build a digital twin for climate in combination with other sensor data such as temperature and humidity.
The project was aligned with emerging policy priorities related to climate adaptation, urban heat mitigation, flood prevention, and data-driven public administration. The reliance on open geospatial data and open-source software ensured coherence with digitalisation strategies and open data policies at regional and national level. The effect: today, about 20 municipalities use the software as a tool to identify potentials to eliminate urban heat islands or hotspots of potential flood hazards as well as a monitoring tool. Just recently the project received information that the City of Munich decided to integrate ADOIS as its monitoring software to track its progress in an effort to de-seal the municipal properties.
Activities and milestones were refined during implementation, shifting the focus from pure model performance to operational usability, including vectorisation, aggregation to administrative units, and automated workflows. KPIs were defined around segmentation accuracy (IoU/mIoU), spatial resolution (20 cm), processing efficiency, and successful aggregation to cadastral units.
The administrative and institutional capacity for implementation was confirmed through successful operation on standard server infrastructure without the need for GPUs during inference. Cost-effectiveness was ensured through containerisation (Docker), low hardware requirements, reuse of existing open data sources, and transparent configuration-based deployment, enabling long-term operation with limited additional resources.
Long-term impact
The ADOIS project has demonstrably achieved and in several aspects exceeded its initial objectives, generating sustainable long-term impacts for public administration and environmental policy.
All core objectives were met. These include the automated detection of impervious surfaces with a spatial accuracy of 20 cm and the reliable differentiation between building-related (high-rise construction) and other impervious surfaces (civil engineering). The result is the generation of georeferenced, vectorised impervious surface datasets as well as the aggregation of results to arbitrary spatial units (e.g. cadastral parcels) with quantitative imperviousness indicators.
High segmentation accuracy and class-wise IoU values were achieved and validated both quantitatively and through expert-based visual inspection, confirming the operational suitability of the results.
Continuous exchange with stakeholders from district and municipal administrations ensured that outputs directly supported real-world policy needs. The resulting impervious surface maps are actively used for monitoring land take, supporting climate adaptation strategies, identifying urban heat islands, and informing flood risk management.
Around 20 municipalities – including Munich, Dortmund, and Recklinghausen – as well as urban associations such as the RVR and companies such as Stadtentwässerungsbetriebe Köln already use ADOIS to analyse and monitor their properties. As a result, more than 7 million citizens indirectly benefit from this service.
Project results were evaluated using a combination of quantitative metrics (accuracy, IoU/mIoU), qualitative visual assessment, and practical applicability in administrative workflows. The ability to perform multi-temporal comparisons enables change detection and supports evidence-based planning and decision-making, demonstrating a clear and lasting policy impact.
Adjustment in the objectives
The project’s core objectives remained stable but were refined during implementation. In particular, the identification of systematic misclassifications, such as large flat roofs or parking areas being confused with buildings, led to the planned integration of height information.
Organisationally, the project significantly strengthened internal capacities. The team developed extensive expertise in AI-based geospatial analysis, especially in deep learning architectures and robust validation methods. Advanced skills were also gained in computer vision and data preprocessing. At the same time, the project deepened the team’s understanding of municipal processes, particularly in urban planning contexts.
The methodological approach and software architecture show high transfer potential for other regions and administrative bodies, especially for automated land cover mapping, environmental monitoring, and change detection. The open-source nature of the software facilitates replication and adaptation.
Within the German URBAN.KI initiative, ADOIS received funding for a follow-up phase running until June 2026, involving multiple municipalities and expanding classification capabilities. In addition, ADOIS contributes data to the BioAdapt project, funded by the Federal Ministry for the Environment, Climate Action, Nature Conservation and Nuclear Safety (Bundesministerium für Umwelt, Klimaschutz, Naturschutz und nukleare Sicherheit – BMUKN), supporting climate adaptation and biodiversity planning.
Demonstrated benefits, cost-effectiveness, and policy alignment secure long-term administrative and political support.