Univariate Deep Learning models for prediction of daily average temperature and Relative Humidity: The case study of Chennai, India

Accurate weather forecasting plays a crucial component in everyday human life, especially in the agricultural and industrial sectors around the world. The time series prediction of meteorological variables by organizations or government agencies may assist in the decision-making process to plan their daily activities, safeguarding agricultural and water resources. However, the models used by them are computationally expensive. With the rapid development of Artificial Intelligence (AI), a univariate time series model for forecasting average daily temperature and Relative Humidity prediction was developed using Deep Learning (DL) models. Feed-Forward Neural Networks (FFNN), Long Short-Term Memory (LSTM), Gated Recurrent units, Bidirectional LSTM, Convolutional Neural Networks combined with LSTM (CNN+LSTM) and Convolutional LSTM (ConvLSTM) are the DL models considered for this study. In addition, the proposed models are trained and tested with different optimizers (Stochastic Gradient Descent, Root Mean Square Propagation, and Adaptive Moment Estimation (ADAM)). The experimental results show that the CNN+LSTM with ADAM optimizer is the best forecasting model for average temperature and Relative Humidity prediction, followed by ConvLSTM with ADAM optimizer compared to the other DL models. Highlights Accurate Temperature and Relative Humidity forecasting plays an important role in planning everyday human activities in the agricultural and industrial sectors. Univariate deep learning forecasting models are designed and implemented to predict the future trends of these climatic variables. Experimental results show the CNN+LSTM outperforms other deep learning forecasting models.