Designing empirical fourier decomposition reinforced with multiscale increment entropy and deep learning to forecast dry bulb air temperature

Accurate prediction of dry bulb air temperature (DBTair) is significant to determine the state of humid air and supporting experts in the environmental sector. Traditional machine learning based approaches struggle to deliver accurate predictions when temperature is suddenly fluctuated during extreme weather conditions. This paper aims to design an intelligent model namely MEFD-MSIE-FCNN to forecast DBTair which integrates multivariate empirical Fourier decomposition (MEFD), multiscale increment entropy (MSIE), and FCSM model that integrates a fully connected neural network FCNN with long short-term memory (LSTM) to forecast DBTair. The multivariant time series of each predictor variable is passed through the MEFD to extract mutual features across multivariant time series and deliver multivariable-aligned modes. Then, the MSIE is extracted to form a feature final matrix to represent mutual information from multivariant time series. Finally, the features set is sent to the FCSM to forecast multistep ahead DBTair using goodness-of-fit statistical metrics for two regions in Saudi Arabia. The proposed model showed highest accuracy for Jazan station (RMSE=2.120, MAE=2.912, RSE=0.123, ECC=0.971, WIA=0.981, CC=0.982), and Jeddah station (RMSE=2.131, MAE=2.921, RSE=0.113, ECC=0.969, WIA=0.979, CC=0.980). A comprehensive comparison is made against state-of-the art benchmarking models, concluding that there is a noticeable improvement in model's performance in terms of AME, ECC, CC, WIA, RMSE and correlation coefficient. The proposed FCSM can be helpful for many applications such as improving weather prediction, preventing climate risks, energy consumption, water resources management and agricultural industry. Additionally, the proposed model can support decision makers and industries in the environmental sector to make informed decisions to mitigate the effects of climate change.