Document Type : Original research article

Authors

• Tohid Bagherpoor ¹
• Somayeh Kazemi Sormoli ²

¹ Department of Agronomy, College of Agriculture, University of Zabol, Zabol, Iran

² Department of Computer, College of Computer Science, China University of Geoscience (CUG) , Wuhan, China

Abstract

This study compares three models, Back -Propagation Feed-Forward Neural Networks (BP-FFNN), Recurrent Neural Networks with Long Short-Term Memory (RNN-LSTM), and Seasonal Autoregressive Integrated Moving Average (SARIMA), for temperature prediction using historical air temperature data from Zabol City, Iran. The dataset consists of daily average air temperature observations, and the models were evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Median Absolute Percentage Error (MDAPE), and Coefficient of Determination ( ) metrics. The BP-FFNN model outperformed the RNN-LSTM and SARIMA models, achieving the lowest values for RMSE (0.018), MAE (0.013), and MDAPE (1.59%). It demonstrated accurate temperature predictions with a strong correlation between predicted and actual values ( = 0.99). The RNN-LSTM model showed comparable results, capturing long-term patterns with RMSE of 0.042, MAE of 0.031, and MDAPE of 3.53%. The SARIMA model provided insights into seasonality and autocorrelation, achieving RMSE of 0.042, MAE of 0.03, and MDAPE of 3.65%.The study's findings have implications for weather forecasting, climate research, and energy management systems. The superior performance of the BP-FFNN model suggests its reliability for accurate temperature prediction, while the RNN-LSTM model offers an alternative approach for capturing long-term patterns. The SARIMA model contributes insights into seasonality and autocorrelation. The study highlights the strengths and limitations of each model and their practical applications in temperature forecasting. In conclusion, the BP-FFNN model effectively predicts temperatures in Zabol City while the RNN-LSTM and SARIMA models provide alternative approaches for capturing long-term patterns and understanding seasonality. The study's results advance temperature prediction techniques and have practical implications for various fields reliant on accurate temperature forecasting.

Keywords

• Air temperature forecasting
• Machine learning techniques
• Deep learning models
• Statistical approach
• Time series analysis