Novel optimized coupled rainfall model simulation based on stepwise decomposition technique.

Area of Science:

• Hydrology and Water Resources
• Artificial Intelligence in Environmental Science
• Climate Modeling and Prediction

Background:

• Effective precipitation forecasting is vital for regional water resource management, drought, and flood warnings.
• Developing robust monthly precipitation simulation models is a significant challenge in practical prediction.
• Existing machine learning models require optimized hyperparameters for accurate forecasting.

Purpose of the Study:

• To introduce modified African vulture optimization algorithms (MAVOA1 and MAVOA2) for hyperparameter optimization.
• To construct monthly precipitation simulation models using algorithmic optimization coupled with variational mode decomposition.
• To evaluate the performance of optimized machine learning models for precipitation prediction.

Main Methods:

• Implemented two modified African vulture optimization algorithms (MAVOA1, MAVOA2).
• Applied hyperparameter optimization to Least Squares Support Vector Machine (LSSVM), Long Short-Term Memory (LSTM), and Random Forest (RF) models.
• Utilized variational mode decomposition for full decomposition in model construction.

Main Results:

• The LSSVM model outperformed LSTM and RF models in precipitation simulation.
• The MAVOA2-LSSVM model achieved the best overall performance with RMSE = 17.50 mm/month, MRE = 117.25%, and NSE = 0.90.
• MAVOA2 demonstrated suitability for optimizing machine learning models with numerous hyperparameters.

Conclusions:

• The MAVOA2-LSSVM model significantly improves monthly precipitation prediction accuracy.
• Optimized machine learning models, particularly MAVOA2-LSSVM, offer superior performance for practical applications.
• The developed hyperparameter optimization approach provides a valuable reference for other fields.