Runoff Forecasting Using Machine-Learning Methods: Case Study in the Middle Reaches of Xijiang River

Area of Science:

• Hydrology and Water Resources
• Artificial Intelligence in Environmental Science
• Machine Learning Applications

Background:

• Effective runoff forecasting is crucial for flood early warning systems and efficient water resource management.
• Accurate streamflow and water level predictions are essential for mitigating flood risks and optimizing water allocation.
• Previous studies have explored various hydrological modeling techniques with varying degrees of success.

Purpose of the Study:

• To develop and evaluate high-accuracy runoff forecasting models using different machine learning approaches.
• To identify the optimal model for predicting streamflow and water level at Wuzhou station in the Xijiang River.
• To assess the performance of selected models, particularly the generalized regression neural network (GRNN), in forecasting with lead times and considering flood propagation.

Main Methods:

• Employed and compared four machine learning models: backpropagation (BP) neural network, generalized regression neural network (GRNN), extreme learning machine (ELM), and wavelet neural network (WNN).
• Developed a runoff forecasting model at Wuzhou station, focusing on streamflow and water level prediction.
• Utilized GRNN for streamflow and water level forecasting, incorporating flood propagation time to improve accuracy.

Main Results:

• The GRNN model exhibited the best performance for 7-day lead time streamflow forecasting.
• The wavelet neural network (WNN) model achieved the highest accuracy for 7-day lead time water level forecasting.
• GRNN significantly improved runoff forecasting by considering flood propagation time, increasing the Qualification Rate (QR) for mean streamflow and water level to 98.36% and 82.74%, respectively, while addressing peak underestimation.

Conclusions:

• The generalized regression neural network (GRNN) is identified as the optimal model for accurate runoff forecasting, especially when considering flood propagation time.
• The developed machine learning-based runoff forecasting models enhance early warning capabilities for floods and droughts.
• This research provides a strong foundation for mid-to-long-term runoff forecasting and improved water resource management.