A transferable machine learning framework integrating multi-dimensional vegetation thresholds to predict green...

Urban overheating, which affects more than two billion people, is a growing problem for cities in arid and semi-arid regions. The majority of current research relies on two-dimensional satellite indices like Normalized Difference Vegetation Index (NDVI), which do not capture the vertical vegetation structures that shape thermal comfort at street level, despite the fact that urban greening has demonstrated promise as a cooling strategy. Quantitative cooling thresholds specific to hyper-arid climates are also lacking; these are the benchmarks that urban planners genuinely require in order to make well-informed design choices. To close these gaps, this study creates a predictive framework. XGBoost machine learning (R² = 0.84) and SHAP-based threshold analysis were used in conjunction with multi-dimensional vegetation metrics, including NDVI, Green View Index (GVI), and a recently developed Combined Greenery Index (CGI). The main case study is the 16 km² King Salman Garden that is being planned for Riyadh, Saudi Arabia. The analysis provides design benchmarks for cities with Köppen BWh climates by identifying critical non-linear thresholds, CGI ≥ 0.25 and GVI ≥ 25%, beyond which cooling effects increase sharply. The role of wind direction in green space planning is highlighted by model projections that show a mean daytime air temperature reduction of 3.6 °C ± 0.9 °C, extending 1.6 km downwind and 0.8 km upwind. Additionally, the findings identify an optimal range for irrigation efficiency between 65 and 75% of reference evapotranspiration (ET₀), which allows for 83% of maximum cooling with just 57% of full water use. This is a particularly pertinent finding for areas with limited water supplies. Although transferability is influenced by variations in urban form, local climate variability, and governance context, a comparative study with parks in Abu Dhabi, Dubai, and Phoenix reveals consistent threshold patterns across BWh climates. In order to maximize green infrastructure in arid cities, the framework provides urban planners with evidence-based tools.