A coupled CEEMD-BiLSTM model for regional monthly temperature prediction.

Area of Science:

• Environmental Science
• Climate Science
• Data Science

Background:

• Global warming necessitates accurate climate change indicators like temperature.
• Monthly temperature prediction is complex due to solar activity, monsoons, and inherent randomness.
• Improved temperature forecasting aids disaster prevention and economic planning.

Purpose of the Study:

• To develop and validate a novel hybrid model for enhanced monthly temperature prediction.
• To assess the model's performance against existing methods for climate forecasting.
• To improve the accuracy and reliability of temperature predictions in Zhengzhou City.

Main Methods:

• Utilized the Complementary Ensemble Empirical Mode Decomposition (CEEMD) for time series decomposition and reconstruction.
• Employed a Bidirectional Long Short-Term Memory (BiLSTM) network for stochastic prediction of temperature data.
• Coupled CEEMD with BiLSTM (CEEMD-BiLSTM) to leverage strengths in handling uncertainty and time-series forecasting.

Main Results:

• The CEEMD-BiLSTM model achieved a minimum relative error of 0.01% and an average relative error of 0.22%.
• Demonstrated significantly higher prediction accuracy compared to CEEMD-LSTM, EEMD-BiLSTM, and BP neural network models.
• Exhibited superior stability and adaptability in monthly temperature predictions for Zhengzhou City.

Conclusions:

• The coupled CEEMD-BiLSTM model offers a robust and accurate approach for monthly temperature forecasting.
• This hybrid model effectively addresses the uncertainties inherent in climate data.
• The findings support the application of advanced machine learning techniques for climate change mitigation and economic development.