First Results in Calculating Urban Green Spaces with Machine Learning and Geographic Information Systems

Green spaces in urban areas are essential for maintaining healthy and sustainable living environments, providing a multitude of environmental, social, and economic benefits. Urban green spaces contribute to air purification, temperature regulation, biodiversity preservation, and improved mental and physical well-being while mitigating urban challenges like the "heat island" effect.
This study focuses on assessing urban green spaces within the construction boundaries of the city of Sofia, Bulgaria, using machine learning (ML), satellite imagery, and geographic information systems (GISs).
This research utilizes satellite data from Sentinel-2 imagery, GIS tools, particularly QGIS, data preprocessing, and semi-automatic classification using the Spectral Angle Mapper (SAM) algorithm. The results were cross-referenced with data from CORINE Land Cover (CLC), a standardized European land classification system.
This study demonstrates how integrating multiple data sources and machine learning (ML) technologies improves the accuracy and efficiency of green space analysis. Semi-automatic classification methods trained with user-defined samples successfully distinguished land cover types, allowing for detailed mapping of vegetation, urban areas, and water bodies. This approach provides valuable insights for sustainable urban planning and natural resource management.
By applying these methods, we estimated the distribution and characteristics of green areas in Sofia, Bulgaria, highlighting the potential for GIS and remote sensing technologies to support evidence-based decision-making. The findings underscore the importance of integrating modern tools and data systems in urban development plans to address environmental and social challenges effectively.

This study demonstrates how integrating multiple data sources and machine learning (ML) technologies improves the accuracy and efficiency of green space analysis. Semi-automatic classification methods trained with user-defined samples successfully distinguished land cover types, allowing for detailed mapping of vegetation, urban areas, and water bodies. This approach provides valuable insights for sustainable urban planning and natural resource management.