Prediction of land surface temperature using spectral indices, air pollutants, and urbanization parameters for Hyderabad city of India using six machine learning approaches

Abstract

Urban areas face rising land surface temperatures (LST) and increased air pollution due to urbanization and industrialization. The consistent rise in LST impacts lives of the residents. Thus, LST forecasts can help manage activities more comfortably. The present study aims to predict LST for Hyderabad city, India using five-year (2018–2022) data on air pollution and meteorological parameters (from ambient air quality monitoring stations) and MODIS LST data. Six machine learning models i.e., Multiple Linear Regression (MLR), Support Vector Regression (SVR), Random Forest (RF), Artificial Neural Networks (ANN), XGBoost, and Long Short-Term Memory (LSTM), were used for LST prediction. Considerable influence of PM2.5 and CO (during summer) and SO2 (during winter) on LST was observed which demonstrated high sensitivity of these parameters on LST. LST exhibited a weak correlation with individual air pollutants while strong relationship of LST with all the study variables, when considered simultaneously, was observed. ANN method demonstrated better accuracy with lower error metrics, comprising of Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Squared Error (MSE), compared to the other approaches with ranking in the order ANN > RF > SVR > XGBoost > LSTM > MLR. ANN model for Hyderabad city was validated by using it for prediction of LST of another geographical area i.e. Bengaluru city, India. The result of this study can provide insights for policymakers, urban planners, and environmental agencies for targeted interventions for temperature regulations and to mitigate urbanization's impact.