From Data to Insights: Modeling Urban Land Surface Temperature Using Geospatial Analysis and Interpretable Machine...

Area of Science:

• Environmental Science
• Geospatial Analysis
• Urban Planning

Background:

• Urban heat islands significantly impact city environments.
• Accurate modeling of land surface temperature (LST) is crucial for understanding urban heat stress.
• Da Nang, Vietnam, faces increasing urban development and associated thermal challenges.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting urban land surface temperature (LST) in Da Nang.
• To identify key factors influencing spatial LST variations.
• To provide insights for sustainable urban planning and heat stress mitigation.

Main Methods:

• Employed Light Gradient Boosting Machine (LightGBM), Support Vector Machine, Random Forest, and Deep Neural Network.
• Utilized remote sensing data from 2014, 2019, and 2024 for model training and validation.
• Applied Shapley Additive Explanations to interpret model results and identify influential factors.

Main Results:

• LightGBM demonstrated superior performance compared to other benchmark machine learning methods.
• Urban density and greenspace density were consistently identified as the most influential factors affecting LST.
• Achieved high R-squared values (0.85, 0.92, 0.91) for 2014, 2019, and 2024, respectively.

Conclusions:

• Machine learning, particularly LightGBM, effectively models spatial LST variations in urban areas.
• Urban form characteristics, specifically density and green space, are critical drivers of LST.
• Findings support evidence-based urban planning for mitigating heat stress and enhancing urban resilience.