Enhancing urban flood forecasting in drainage systems using dynamic ensemble-based data mining.

Area of Science:

• Environmental Science
• Data Mining
• Hydrology

Background:

• Urban drainage systems face increasing challenges from flooding due to climate change and urbanization.
• Accurate and timely flood forecasting is crucial for effective risk management and mitigation strategies.

Purpose of the Study:

• To develop and evaluate a novel dynamic ensemble-based data mining model for urban flood forecasting in drainage systems.
• To improve the accuracy and reduce false alarms in real-time flood early warning systems.

Main Methods:

• Developed a dynamic ensemble data mining model incorporating event identification and rainfall feature extraction.
• Utilized weak learner data mining models stacked using a decision tree algorithm and confusion matrix-based blending.
• Compared the proposed model against commonly used ensemble models in a UK urban drainage system.

Main Results:

• The proposed model achieved a higher hit rate (approx. 85%) compared to benchmark models (approx. 70%) for 3-hour ahead flood forecasting.
• The model accurately classifies flood and non-flood events with minimal lag time, reducing false alarms.
• Identified "antecedent precipitation history" and "seasonal time occurrence of rainfall" as key features enhancing forecasting accuracy.

Conclusions:

• The dynamic ensemble data mining model offers a significant advancement in urban flood forecasting accuracy and reliability.
• This approach enhances the effectiveness of real-time early warning systems, leading to reduced risks and costs.
• Feature engineering, specifically antecedent precipitation and rainfall timing, is vital for improving flood prediction models.