Exploring machine learning algorithms for accurate water level forecasting in Muda river, Malaysia.

Area of Science:

• Environmental science
• Hydrology
• Data science

Background:

• Accurate water level prediction is vital for effective flood warning systems and sustainable freshwater resource management.
• Forecasting water levels in rivers and lakes requires robust modeling techniques to handle complex hydrological dynamics.

Purpose of the Study:

• To develop and compare the performance of three machine learning algorithms for water level forecasting in the Muda River, Malaysia.
• To evaluate the impact of meteorological data integration and different time horizons on model accuracy.

Main Methods:

• Applied multi-layer perceptron neural network (MLP-NN), long short-term memory neural network (LSTM), and extreme gradient boosting (XGBoost) models.
• Utilized daily water level and meteorological data from 2016 to 2018 for model development and testing.
• Assessed model performance using accuracy scores and evaluated prediction capabilities across different time horizons.

Main Results:

• The MLP-NN model achieved the highest overall accuracy (0.871), surpassing LSTM (0.865) and XGBoost (0.831).
• Incorporating meteorological data did not significantly improve the prediction accuracy of any tested model.
• The LSTM model demonstrated superior performance for 7-day ahead forecasting, highlighting its strength in capturing long-term dependencies.

Conclusions:

• Each machine learning model exhibits unique strengths and weaknesses, with performance being highly dependent on data quantity and quality.
• MLP-NN is effective for short-term water level prediction, while LSTM excels in longer-term forecasting.
• The choice of machine learning model for water level prediction should be tailored to specific project requirements and data availability.