A novel coupled rainfall prediction model based on stepwise decomposition technique.

Area of Science:

• Environmental Science
• Data Science
• Hydrology

Background:

• Traditional ensemble models for rainfall forecasting often suffer from future information leakage during training.
• This limitation hinders their practical application and accuracy in real-world scenarios.

Purpose of the Study:

• To propose and evaluate a novel stepwise decomposed ensemble coupling model for improved rainfall forecasting.
• To address the issue of future information leakage in existing ensemble prediction methods.

Main Methods:

• The proposed model utilizes Variational Mode Decomposition (VMD) for signal decomposition.
• Bidirectional Long Short-Term Memory (BiLSTM) neural networks are employed for sequence modeling.
• Model parameters are optimized using an Improved Particle Swarm Optimization (IPSO) algorithm.

Main Results:

• The IPSO algorithm demonstrated superior performance over the standard PSO algorithm, reducing Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).
• The VMD-IPSO-BiLSTM model, incorporating stepwise decomposition, significantly outperformed the IPSO-BiLSTM model, showing substantial reductions in MAE and RMSE.
• The Nash-Sutcliffe Efficiency (NSE) for the VMD-IPSO-BiLSTM model consistently exceeded 0.88 across all test cities, indicating high prediction accuracy.

Conclusions:

• The novel VMD-IPSO-BiLSTM model effectively overcomes the limitations of traditional methods by preventing future data leakage.
• The enhanced optimization through IPSO and the stepwise decomposition technique lead to significant improvements in rainfall forecasting accuracy.
• This research offers valuable insights for developing more reliable and accurate rainfall prediction systems.