Exploring urban heat dynamics through multi-model machine learning analysis of land surface temperature in Bengaluru India

Urban heat regulation is key to sustainable city planning and climate adaptation. Predicting land surface temperature (LST) helps understand how environmental and urban factors influence temperature. This study utilizes a combination of air pollutant data, meteorological parameters, urbanization indicators, spectral indices, and elevation data over a five-year period to predict LST in Bengaluru. Advanced machine learning models, including convolutional neural networks (CNN), K-nearest neighbors (KNN), gated recurrent units (GRU), recurrent neural networks (RNN) and artificial neural networks (ANN) were utilized to capture these complex interactions. NH3 is the most influential factor affecting LST in both summer and winter. The results provided deeper insights into the interactions of pollutants as well as urbanization indicators with LST across seasons using different impact analysis. Among the machine learning approaches, CNNs showed superior predictive accuracy, effectively capturing spatial patterns and minimizing prediction errors better than the other approaches. The performance was ranked as high in CNN (R2 = 0.947 in summer, 0.913 in winter), followed by KNN, GRU, ANN, RNN, with CNN being the most reliable due to its lowest errors. This study highlighted the value of advanced models in identifying urban temperature dynamics, aiding better urban planning and climate adaptation.