A research team led by NYU Associate Professor Rumi Chunara has developed an advanced artificial intelligence (AI) system that accurately tracks urban green spaces using satellite imagery offering a crucial tool for healthier city planning. The system significantly outperforms traditional methods, which have historically failed to capture up to 37% of urban vegetation.
Breaking New Ground in Urban Mapping
To validate their approach, researchers tested the system in Karachi, Pakistan’s largest city, known for its mix of dense urban sprawl and highly varied vegetation. The AI, built on enhanced DeepLabV3+ segmentation models, was trained using a technique called ‘green augmentation,’ which adjusts training data to account for different lighting and seasonal conditions. This innovation improved vegetation detection accuracy by 13.4% compared to existing AI methods.
The results are striking. The AI achieved 89.4% accuracy with 90.6% reliability, far surpassing traditional vegetation mapping techniques, which only reach 63.3% accuracy and 64.0% reliability.
A City of Unequal Greenery
Karachi’s green space per capita averages just 4.17 square meters less than half of the World Health Organization’s recommended minimum of 9 square meters. However, the disparity between neighborhoods is severe: some areas enjoy over 80 square meters per person, while five union councils have less than 0.1 square meters per capita.
The study found a strong link between economic development and greenery. Areas with more paved roads often an indicator of higher income tend to have more trees and grass. Crucially, satellite data confirmed that neighborhoods with more vegetation experience significantly lower surface temperatures, highlighting the role of urban greenery in cooling cities.
Singapore serves as a model for what is possible with intentional planning. Despite a similar population density to Karachi, it provides 9.9 square meters of green space per person, surpassing WHO guidelines.
Professor Chunara, Director of the NYU Center for Health Data Science, emphasized the system’s potential impact:
“Our system learns to recognize subtle patterns that distinguish trees from grass, even in complex urban environments. Without accurate mapping, cities cannot effectively address green space disparities.”
The team has made its methodology public, allowing other cities to adapt the model using local satellite imagery. However, retraining will be necessary to account for regional differences.
This study builds on Chunara’s extensive work using computational methods to analyze social and health disparities, including mapping systemic racism through social media and assessing telemedicine access during COVID-19.
With cities worldwide facing climate challenges and rapid urbanization, AI-driven tools like this could be essential in designing greener, more equitable environments