Model Prediction of Land Surface Temperature from Satellite Data with Machine Learning

Abstract

BACKGROUND AND OBJECTIVES: Urbanization leads to increased building construction, forming Urban Heat Island (UHI). UHI causes heat accumulation in urban areas, making it difficult to ventilate the area. This study aims to analyze Land Surface Temperatures (LST) using Remote Sensing (RS) data to predict UHI development in urban localities in Chiang Mai, Thailand. The study aims to compare the performance of different machine learning (ML) algorithms in predicting LST and assess their potential for future use in mitigating UHI consequences in urban areas.

METHODS: RS data from Landsat 8 and Sentinel-2 satellites were used to analyze LST from 2016 to 2022. Five different ML algorithms were employed in this study: Random Forest (RF), AdaBoost Regressor (ABR), Artificial Neural Network (ANN), Linear Regression (LR), and Gradient Boosting (GB). The performance of these algorithms was evaluated using statistical variables.

FINDINGS: The study found that the RF model had the highest precision in predicting LST, with the lowest Mean Absolute Error (MAEvalues among the models. However, all models had relatively low R-squared (R²)values, indicating room for developing the accuracy of the predictions.

CONCLUSION: The study demonstrates the feasibility of using RS data and ML algorithms to predict LST and comprehend UHI development in urban localities. The study also emphasizes the importance of using ML techniques to address UHI consequences in urban areas and can apply these data to urban planning to promote sustainable urban development. Further investigation is necessary to improve the accuracy of the models and determine effective strategies to mitigate UHI effects in urban areas.