Mapping the Future City: How AI and Geospatial Data Are Transforming Urban Planning

Urbanisation is one of the defining trends of this century, and much of the growth is happening in Africa and Asia. UN-Habitat has long highlighted the scale of informal settlement and the infrastructure gap in rapidly growing cities. Lagos is a vivid example: neighbourhoods appear, densify and change character within a few years, often far faster than official maps, cadastral records or master plans are updated.

Planning without accurate data is planning blind. You cannot site a school, size a drainage channel or design a bus route well if you do not know where people actually live and move. My studies in Geomatics Engineering at the University of Lagos focused on exactly these tools: surveying, GIS, spatial analysis and digital mapping. What has changed in recent years is that AI makes it possible to extract that information at city and national scale, quickly and repeatedly.

Building and land-use extraction from imagery

Semantic segmentation models, many descended from the U-Net architecture, label each pixel of satellite or aerial imagery as building, road, vegetation, water and so on. Instance segmentation separates individual building footprints. Large open datasets produced this way, such as Google's Open Buildings for Africa and other regions, give planners a starting picture of where structures exist, even where no official map does.

Change detection

Comparing imagery over time reveals urban expansion, new construction, encroachment on wetlands or floodplains and demolition. Siamese networks and temporal models detect changes while ignoring seasonal or lighting differences.

Population and poverty mapping

Combining building footprints, night-time lights, road density and survey data, spatial models estimate population distribution and socio-economic indicators at fine resolution. This is invaluable for allocating services where census data is outdated.

Mobility and network analysis

Graph models over road and transit networks, fed by GPS traces, mobile data and transit feeds, estimate travel demand and accessibility: how long it takes residents to reach a clinic, a market or a job centre. Accessibility analysis is one of the most powerful tools for equitable planning.

Digital twins and simulation

City digital twins integrate 3D models, sensor feeds and simulations to test scenarios such as new roads, zoning changes or flood defences before committing resources. AI accelerates these simulations and helps explore many more scenarios than traditional modelling allows.

Participatory mapping

AI is most effective when combined with human mapping. Communities and volunteers using OpenStreetMap, often coordinated by the Humanitarian OpenStreetMap Team, validate AI-suggested features and add the local knowledge that imagery cannot capture: street names, landmarks and the function of buildings.

Humanitarian and development organisations have used AI-assisted mapping to support vaccination campaigns, census preparation and disaster response, producing building and road maps for areas that had none. City governments in several countries use computer vision on street-level imagery to assess road surface quality, map informal street vending or inventory trees. The World Bank's urban development work increasingly incorporates geospatial analytics in project design, from flood-risk assessments to accessibility studies.

The common thread in successful projects is that AI outputs feed into decisions that people are already empowered to make. A beautiful map that nobody in the planning department can act on changes nothing.

• Surveillance creep. "Smart city" projects that start with traffic management can drift into pervasive monitoring of residents, particularly with facial recognition. Purpose limitation should be explicit and enforced.
• Mapping as a tool of eviction. Detailed maps of informal settlements can support upgrading and service delivery, or they can be used to target demolition. The ethics depend on who controls the data and how residents are involved.
• Model bias in unfamiliar landscapes. Models trained mainly on cities in Europe or North America often perform worse on dense informal settlements with irregular roofs and unpaved roads. Local validation is not optional.
• Data gaps and digital divides. Mobility data from smartphones under-represents poorer and older residents. Plans based only on such data can serve those already well served.
• Vendor-driven agendas. Smart-city platforms sold as turnkey solutions can lock cities into expensive systems that do not match local priorities.

1. Start with a decision, not a dataset. Identify a concrete planning question, such as where to build new schools, and work backwards to the data and models you need.
2. Use open data first. Open building footprints, OpenStreetMap, free satellite imagery from programmes like Sentinel and Landsat and open population grids form a strong, low-cost baseline.
3. Validate locally. Ground-truth AI outputs with field surveys or high-resolution imagery in representative neighbourhoods, including informal ones, and report accuracy honestly.
4. Build GIS capacity inside government. Planning departments need people who can maintain and question the data, not just consume dashboards.
5. Involve communities. Participatory mapping improves accuracy and legitimacy, and it gives residents a voice in how their neighbourhoods are represented.
6. Govern the data. Decide who can access detailed settlement and mobility data, for what purposes, and how it is protected.

When I talk with young developers about career paths, I often point to geospatial AI as an area where African engineers have a genuine edge: we understand the landscapes, languages and lived realities that global models get wrong.

The smart city conversation has too often been dominated by sensors, dashboards and control rooms. The more transformative change is quieter: AI and geospatial data giving planners, for the first time, a reasonably current picture of where people live, how they move and what they lack. For fast-growing African cities, that picture is the foundation for everything else, from drainage design to transit planning to the placement of clinics and schools.

Used responsibly, these tools can make planning more evidence-based and more inclusive. Used carelessly, they can make surveillance cheaper and displacement easier. The difference lies in governance, community participation and the presence of skilled local professionals who understand both the algorithms and the city.

• Rapid urbanisation, especially in Africa, has outpaced traditional mapping and planning processes.
• Segmentation of satellite imagery, change detection, population mapping and network accessibility analysis are core techniques.
• Open datasets and participatory mapping dramatically lower the cost of good urban data.
• Surveillance creep, eviction risks and model bias in informal settlements require explicit governance.
• Successful projects start with a planning decision, validate locally and build in-house GIS capacity.

1. UN-Habitat — United Nations Human Settlements Programme
2. U-Net: Convolutional Networks for Biomedical Image Segmentation — arXiv (Ronneberger, Fischer and Brox)
3. Open Buildings — Google Research
4. Humanitarian OpenStreetMap Team — HOT
5. Urban Development — World Bank
6. OpenStreetMap — OpenStreetMap Foundation

Cover image: original illustration created for this article. Cited works remain the property of their respective authors and publishers.