ABTCN: an efficient hybrid deep learning approach for atmospheric temperature prediction

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Accurate temperature prediction is crucial for monitoring global warming and environmental changes.
• Data-driven models excel at time-series forecasting but struggle with missing or erroneous data due to sensor failures or natural disasters.

Purpose of the Study:

• To propose an efficient hybrid model, the attention-based bidirectional long short term memory temporal convolution network (ABTCN), to address limitations in current data-driven models for climate parameter prediction.
• To enhance the prediction accuracy of climatology parameters by effectively handling missing and erroneous data.

Main Methods:

• The proposed ABTCN architecture integrates the k-nearest neighbor (KNN) imputation method for missing data handling.
• It employs a bidirectional long short term memory (Bi-LSTM) network with a self-attention mechanism for feature extraction.
• A temporal convolutional network (TCN) is utilized for predicting long data sequences.

Main Results:

• The ABTCN model demonstrated superior performance compared to state-of-the-art deep learning models.
• Evaluation using metrics like MAE, MSE, RMSE, and R² score confirmed the model's high accuracy in temperature prediction.
• The hybrid approach effectively managed missing values, a common challenge in environmental time-series data.

Conclusions:

• The developed ABTCN model offers a robust and accurate solution for temperature prediction, outperforming existing deep learning approaches.
• This hybrid model provides a significant advancement in environmental monitoring and global warming studies by reliably handling data imperfections.
• The ABTCN architecture presents a promising direction for future research in time-series forecasting for climate science.