Temperature Prediction Based on Bidirectional Long Short-Term Memory and Convolutional Neural Network Combining...

Area of Science:

• Meteorology
• Artificial Intelligence
• Data Science

Background:

• Accurate temperature prediction is crucial for various applications.
• Weather forecasting relies on diverse data sources, including time-series observations and spatial data.
• Integrating heterogeneous data sources presents a significant challenge in meteorological modeling.

Purpose of the Study:

• To develop and evaluate a deep neural network model for predicting temperature in Korea.
• To effectively combine time-series weather data from automatic weather stations and image data from the Regional Data Assimilation and Prediction System (RDAPS).
• To enhance the accuracy and lead time of future temperature predictions.

Main Methods:

• Utilized a deep neural network architecture combining Bidirectional Long Short-Term Memory (BLSTM) for time-series data and Convolutional Neural Network (CNN) for image data.
• Integrated observed weather data and RDAPS image data into a single predictive model.
• Evaluated model performance using Root Mean Squared Error (RMSE) and Mean Bias Error (MBE) for predictions up to 14 days.

Main Results:

• The proposed model, integrating both observed and RDAPS data, outperformed models using only one data type.
• Demonstrated superior performance across all objective measures (RMSE, MBE) for all prediction lead times.
• Achieved enhanced accuracy in temperature predictions compared to baseline models.

Conclusions:

• Combining diverse data sources with advanced deep learning techniques significantly improves temperature forecasting accuracy.
• The proposed hybrid BLSTM-CNN model offers a robust solution for medium-range temperature prediction.
• This approach provides a valuable tool for meteorological agencies and stakeholders requiring reliable weather forecasts.