Streamflow classification by employing various machine learning models for peninsular Malaysia.

Area of Science:

• Hydrology
• Environmental Science
• Machine Learning

Background:

• Peninsular Malaysia faces significant flood and drought risks due to extreme streamflow.
• Accurate streamflow forecasting is crucial for mitigating environmental and municipal damage.
• Predicting continuous streamflow values presents challenges due to inherent uncertainties.

Purpose of the Study:

• To formulate streamflow prediction as a time series classification problem.
• To classify streamflow into discrete categories (5 or 10 classes) for improved uncertainty management.
• To evaluate machine learning models for streamflow category prediction across Malaysian rivers.

Main Methods:

• Time series classification approach was employed.
• Machine learning models including Long Short-Term Memory (LSTM), Support Vector Machine (SVM), and Gradient Boosting (GB) were utilized.
• Ensemble stacking of SVM and GB models was investigated.

Main Results:

• LSTM models demonstrated superior performance in predicting streamflow categories 2-3 days ahead compared to SVM and GB.
• LSTM achieved higher F1 scores across various Malaysian rivers, indicating improved prediction accuracy.
• Ensemble stacking of SVM and GB models also yielded high performance, with a notable improvement in F1 score for the Perak River.

Conclusions:

• Streamflow category prediction offers an advantageous approach to managing uncertainty in forecasting.
• LSTM is a highly effective model for short-term streamflow category prediction.
• Ensemble methods provide a robust alternative for enhancing streamflow prediction accuracy.