Cities like Dhaka, Bangladesh, are getting hotter as rapid urban growth replaces natural surfaces with concrete and buildings. This Urban Heat Island (UHI) effect does not impact all neighborhoods equally. Some areas, especially low-income and densely built zones, experience significantly higher temperatures, increasing risks to health and well-being.

To understand the causes of urban heat, researchers often use artificial intelligence (AI) tools like SHAP, which identify general factors such as vegetation loss or building density. However, these tools do not show how the influence of these factors varies across different parts of a city.

Our study applies a newer method called GeoShapley, which introduces spatial awareness into AI analysis. It reveals not only what contributes to heat but also where these effects are strongest. For instance, vegetation was more effective in cooling certain neighborhoods than others. GeoShapley also identified localized thresholds and spatial clusters that SHAP could not detect.

By generating detailed, location-specific heat maps, this approach provides urban planners with actionable insights. These findings can guide targeted cooling strategies, such as where to plant trees or reduce built-up areas. Our method can serve as a scalable and transferable framework for other climate-vulnerable cities seeking equitable heat mitigation solutions.