Multi-Model Ensemble Learning Prediction of Surface Air Temperature in Four Major Metro Cities: A Stacking-Based Approach

Surface air temperature prediction research within four major metro Indian cities: Delhi, Chennai, Kolkata, and Mumbai employ stacked ensemble models through ERA5 reanalysis data from 1979 to 2023. The stacking approach uses five machine learning models including Decision Trees, Linear Regression and Ridge Regression, Lasso and Support Vector Regression. The meta-learner is Linear Regression. Training data represents 80% of the total while validation uses 10% and testing contains the remaining 10%. The model assessment utilizes Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and the coefficient of determination (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>). The research focuses on extracting the best model combination for each desired city on basis of their R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> reading. Like here for Chennai R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>=0.85, Kolkata R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>=0.83, Delhi R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>=0.80 and for Mumbai its R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>=0.75. The stacking ensemble provides effective long-range temperature pattern forecasting which helps researchers build better knowledge about ensemble learning approaches for temperature prediction.