AI Measures Urban Growth from Satellite Photos

A new artificial intelligence system can track urban development across entire cities using satellite imagery, providing businesses and governments with accurate tools to monitor rapid urbanization. As the global urban population is expected to rise from 55% today to nearly 70% by 2050, with much of this growth concentrated in developing countries like India, China, and Nigeria, automated monitoring has become increasingly important for site selection, market analysis, and sustainable planning.

The key finding is that researchers have developed the Atlas Urban Index (AUI), which uses vision-language models to provide reliable urban development scores for specific regions. Unlike traditional methods like the Normalized Difference Built-up Index (NDBI), which often struggle with environmental factors like cloud cover and seasonal variation, this new approach delivers consistent measurements across different locations and time periods.

The methodology involves collecting Sentinel-2 satellite imagery for regions of interest and processing them into standardized units called Geohash 5 (GH5) cells, each covering approximately 5km×5km areas. Researchers obtain imagery from the first and third quarters of each year, filtering out cloudy images to ensure quality. They then extract RGB composites from the satellite data and convert them to JPEG format for analysis.

For scoring urban development, the system uses OpenAI's GPT-4o-mini vision-language model with two calibration strategies. First, it provides reference images representing different urbanization levels to ensure spatial consistency. Second, it supplies the most recent past image alongside the current one to maintain temporal consistency and mitigate effects of noise and seasonal variation. This dual calibration allows the system to overcome challenges that plague traditional pixel-based methods.

Results from analyzing Bangalore regions demonstrate the system's effectiveness. At Kempegowda International Airport, the AUI showed steady growth from 7.2 in 2016 to 8.2 in 2025, accurately reflecting the physical development observed through construction of Terminal 2, road networks, and commercial infrastructure. In contrast, NDBI failed to reliably capture this progression, showing inconsistent values that didn't align with actual development.

Similarly, at Bannerghatta National Park, the AUI captured the region's transformation from largely natural landscapes to limited urban development, with scores rising from 3.5 to 4.1 between 2016 and 2025. This gradual increase reflected the actual physical changes in the region, while traditional methods showed erratic patterns that didn't correspond to real development trends.

The context of this breakthrough matters because accurate urban monitoring forms the foundation for data-driven decision-making in evolving environments. Real estate developers depend on such measures to guide site selection for projects, while retailers and restaurants use development patterns to identify promising markets and monitor existing store performance. Government agencies require robust tools to evaluate policy outcomes, allocate resources effectively, and plan for sustainable urban growth.

The need becomes particularly critical in regions experiencing informal construction, where government records may be absent or incomplete. In these cases, independent monitoring through satellite imagery becomes the only viable method for tracking urban change over time.

Limitations noted in the paper include the current demonstration focusing only on Bangalore regions with similar environmental characteristics. The authors acknowledge that testing across more diverse geographic contexts—such as desert cities in the Middle East or tropical regions in Southeast Asia—would be needed to ensure cross-context consistency. Additionally, the current approach relies on manual selection of reference regions, though future work could automate this process.

The researchers suggest that the pipeline could be enhanced by incorporating additional data sources, such as Overture Maps Foundation data, to help refine the urban index further. The method could also be extended to forecast future development patterns, making it valuable for real estate planning and retail strategy. Eventually, the vision-language model approach could be used to create large-scale datasets for training supervised learning models, offering a more scalable and cost-effective solution for urban monitoring.