Suspended sediment load prediction modelling based on artificial intelligence methods: The tropical region as a case...

Area of Science:

• Environmental Science
• Hydrology
• Water Resource Management

Background:

• Suspended sediment load (SSL) significantly impacts environmental health, agriculture, and water resource planning.
• SSL deposition alters streamflow, affects aquatic ecosystems, and can cause river course shifts, necessitating accurate prediction.
• Existing SSL prediction models face challenges due to site-specific data, model complexity, and data limitations, leading to suboptimal accuracy.

Purpose of the Study:

• To develop and evaluate an improved model for predicting suspended sediment load (SSL).
• To address the limitations of previous machine learning models in SSL river prediction.
• To enhance the accuracy and reliability of SSL forecasting for water resource decision-making.

Main Methods:

• A Long Short-Term Memory (LSTM) model was proposed for SSL prediction.
• The LSTM model was compared against Multi-Layer Perceptron (MLP), Support Vector Regression (SVR), and Random Forest (RF) models.
• Data from the Johor River, Malaysia, spanning 2010-2020, including suspended sediment load and river flow, were utilized.

Main Results:

• The proposed LSTM model achieved a high correlation coefficient (0.97) between predicted and actual SSL.
• The LSTM model demonstrated superior performance with a minimum Root Mean Square Error (RMSE) of 148.4 ton/day.
• The model also achieved a minimum Mean Absolute Error (MAE) of 33.43 ton/day, indicating high prediction accuracy.

Conclusions:

• The Long Short-Term Memory (LSTM) model offers a highly accurate and reliable method for suspended sediment load prediction.
• The model's performance suggests its generalizability for application in similar river systems globally.
• Accurate SSL prediction using the LSTM model can significantly aid water resource managers in planning and environmental protection efforts.