Data and Code for: Global Drivers of Urban Thermal Environments: Integrating Geophysical Surface Dynamics and Explainable Machine Learning Across 30 Diverse Cities

OverviewThis repository contains the harmonized multi-city dataset and integrated computational scripts used in the study "Global Drivers of Urban Thermal Environments: Integrating Geophysical Surface Dynamics and Explainable Machine Learning Across 30 Diverse Cities." The research establishes a planetary-scale framework to quantify the drivers of Land Surface Temperature (LST) across 30 major cities spanning Tropical, Arid, and Temperate climate zones.

Dataset ContentThe dataset consists of 30 CSV files (one per city), totaling 90,000 stratified random samples (

N=3,000N=3,000

 per city). Each file includes the following eight variables:

LST: Land Surface Temperature (Target Variable)

NDVI: Normalized Difference Vegetation Index

NDBI: Normalized Difference Built-up Index

MNDWI: Modified Normalized Difference Water Index

Albedo: Surface Solar Reflectance (Clamped to 0-1 range)

DEM: Elevation from SRTM 30m

Pop_Density: Population density from GHSL

Built_Fraction: Built-up fraction from GHSL

Code ContentThis repository includes:

Google Earth Engine (GEE) Script: JavaScript code used for multi-source data fusion (Landsat 8/9, SRTM, GHSL), data cleaning, albedo correction, and stratified sampling.

Python (Google Colab) Notebook: Complete machine learning workflow including Random Forest regression, 5-fold spatial cross-validation, SHAP (SHapley Additive exPlanations) analysis, and Partial Dependence Plot (PDP) generation.

Source Data CreditsRaw data used to generate these files are sourced from:

USGS/NASA Landsat 8-9 Collection 2 Level 2.

NASA Shuttle Radar Topography Mission (SRTM) 30m.

European Commission Joint Research Centre (JRC) Global Human Settlement Layer (GHSL).