New double decomposition deep learning methods for river water level forecasting.

Area of Science:

• Hydrology and Water Resource Management
• Artificial Intelligence in Environmental Science
• Deep Learning for Time Series Forecasting

Background:

• Optimizing water resource use requires accurate streamflow water level (SWL) forecasting.
• Existing methods face challenges in handling complex hydrological dynamics and diverse data sources.

Purpose of the Study:

• To develop and evaluate a novel hybrid deep learning model for enhanced SWL forecasting.
• To integrate multi-source data for improved prediction accuracy across various forecast horizons.

Main Methods:

• A hybrid model combining Convolutional Neural Networks (CNN), Bi-directional Long-Short Term Memory (BiLSTM), and Ant Colony Optimization (ACO) was developed.
• The model employed Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Variational Mode Decomposition (VMD) for feature extraction.
• Integration of satellite-derived, climate indices, and ground-based meteorological data was performed.

Main Results:

• The proposed CVMD-CBiLSTM model achieved high accuracy, with ~98% of prediction errors within ±0.020 m.
• A low relative root mean square error of ~0.08% was recorded, indicating superior performance.
• The model demonstrated significant improvements over benchmark models in forecasting SWL at 19 gauging stations.

Conclusions:

• The hybrid deep learning approach with ACO feature selection substantially enhances river water level forecasting accuracy.
• The model's effectiveness in integrating remote sensing and ground-based data supports strategic water management and climate change adaptation.
• This method offers a promising tool for addressing extreme events like droughts and optimizing water resource planning.