Enhancing flood risk mitigation by advanced data-driven approach

Area of Science:

• Hydrology
• Machine Learning
• Environmental Science

Background:

• Flood events in the Sefidrud River basin cause substantial damage.
• Traditional hydrological models struggle with complex flood dynamics.

Purpose of the Study:

• To develop accurate flood estimation models using machine learning.
• To compare the performance of various machine learning algorithms for flood prediction.

Main Methods:

• Applied Random Forest (RF), Bagging, SMOreg, Multilayer Perceptron (MLP), and Adaptive Neuro-Fuzzy Inference System (ANFIS) models.
• Utilized 50 years of historical hydrological data, split into training (50-70%) and validation sets.
• Processed data using WEKA 3.9 software.

Main Results:

• The nonlinear ensemble RF model achieved the highest accuracy (correlation=0.868, RMSE=0.104).
• RF and MLP models significantly outperformed the linear SMOreg approach.
• The Adaptive Neuro-Fuzzy Inference System (ANFIS) model demonstrated exceptional R-squared accuracy (0.99).

Conclusions:

• Data-driven machine learning models offer accurate flood estimation capabilities.
• Findings provide a benchmark for selecting algorithms in flood risk management.
• Advanced ML techniques are suitable for capturing complex flood dynamics.