Automatic mapping of urban climate–adaption measures from open geodata

Dutch municipalities must be climate-resilient and water-robust by 2050 under the Deltaplan Ruimtelijke Adaptatie (DPRA). Wadis play an important role: these vegetated swales collect rainwater and allow it to infiltrate into the soil. They are among the most common urban climate-adaptation measures, but there is no legal obligation to register them. Wadis are currently mapped manually in a geographic information system (GIS). This is time-consuming and produces an incomplete overview, with wadis on private land in particular remaining largely unmapped.

The Hague University of Applied Sciences (Smart Sensor Systems research group) collaborates with Hanze University of Applied Sciences to develop AI models that automatically detect and delineate wadis and their key components. The models use publicly available elevation data (AHN-4/5, collected using laser measurements) and nationwide, geometrically corrected aerial imagery (orthophotos). The detected wadis and their maintenance-relevant characteristics are then integrated into ClimateScan, the largest open database for climate-adaptation measures in the Netherlands. 

Goal

The goal is to develop and validate AI models that can map urban wadis across the Netherlands. Using only publicly available elevation data (AHN-4/5) and aerial imagery (orthophotos), these models determine where wadis are located, their boundaries, and which characteristics are relevant to maintenance.

Problem statement

How can deep learning be used efficiently to automatically detect wadis from open datasets, using a method that can also be applied at scale? And which approach yields the most accurate and practical results for public authorities and businesses?

Two sub-questions guide the work:

1. Which type of AI model best maps the boundaries of wadis: U-Net, which classifies images into categories such as ‘wadi’ and ‘not wadi’, or Mask R-CNN, which detects and delineates individual wadis?
2. What is the most effective way to combine elevation data and aerial imagery: by providing both to the model together, or by first detecting wadis in aerial images and then using elevation data to validate and refine the results?

Target group

• Municipalities steering DPRA progress monitoring (Nijmegen as pilot; nationally relevant).
• Water authorities and engineering consultancies.
• SMEs and contractors designing or maintaining climate-adaptation infrastructure.
• Researchers and students working on geospatial AI and urban climate adaptation.

Intended outcomes

• A dataset of identified and annotated wadi locations in Lent (Nijmegen), linked to data from the Actueel Hoogtebestand Nederland (AHN).
• Four trained AI models for automatically detecting wadi’s and identifying their characteristics.
• A benchmark comparing the models and selecting the best-performing model.
• Integration of wadi detections and their characteristics into ClimateScan.
• Open-source model code, enabling other researchers to use and further develop the approach.
• A submitted article for a professional journal or conference, sharing the research findings.

Duration

1 April 2026 - 31 March 2027

Team

The Hague Univarsity of Applied Science (THUAS) – AI Partner
Research group Smart Sensor Systems:

• John Bolte — lector, Smart Sensor Systems
• Amey Vasulkar — THUAS lead and senior researcher, AI model development

Hanzehogeschool Groningen — project lead
Research group Klimaatadaptatie, Kenniscentrum Noorderruimte:

• Aaron Sibma — project leader
• Eelco Verschelling — co-lead, researcher
• Floris Boogaard — lector, Klimaatadaptatie

Gemeente Nijmegen — practice partner

• Bart Tromp — researcher, data preparation and validation
• Sidney Stax — municipal contact

Partners

• Hanzehogeschool Groningen — lead applicant (penvoerder), Lectoraat Klimaatadaptatie, Kenniscentrum Noorderruimte.
• De Haagse Hogeschool — AI model development, Lectoraat Smart Sensor Systems.
• Gemeente Nijmegen — practice partner and data provider.
• RobbytuProjects — IT partner, develops and maintains the ClimateScan platform.
• Donkergroen — climate-adaptation design and construction; supplies hundreds of reference wadi projects nationwide.

Funding

Regieorgaan SIA — KIEM HighTech 2025

Participating programmes

• HBO Bètatechniek — Civiele Techniek / Built Environment
• AI and Data Science minors and graduation projects at De Haagse Hogeschool
• Innovatiewerkplaatsen (IWP) at Hanzehogeschool Groningen

Contact

Amey Vasulkar — [email protected]

Figure: Annotated cross-section of a wadi: stormwater storage above the filter layer, drainage layer, and infiltration into the groundwater.