Assessment of urban flood susceptibility using semi-supervised machine learning model

Area of Science:

• Environmental Science
• Geospatial Analysis
• Machine Learning

Background:

• Urban flood susceptibility assessment is crucial for disaster risk reduction.
• Limited flood inventory data poses a challenge for traditional modeling methods.
• Accurate flood susceptibility mapping aids in effective urban planning and management.

Purpose of the Study:

• To develop and evaluate a semi-supervised machine learning model for urban flood susceptibility.
• To assess the performance of the weakly labeled support vector machine (WELLSVM) against other models.
• To create a high-quality flood susceptibility map for Beijing's metropolitan area.

Main Methods:

• Utilized a spatial database including flood inventories (2004-2014) and nine explanatory factors (meteorological, geographical, anthropogenic).
• Employed a weakly labeled support vector machine (WELLSVM) for semi-supervised learning.
• Compared WELLSVM with logistic regression, artificial neural networks, and standard support vector machine.

Main Results:

• WELLSVM demonstrated superior performance by effectively utilizing spatial information from unlabeled data.
• Model performance was evaluated using accuracy, precision, recall, F-score, and receiver operating characteristic curves.
• The WELLSVM model outperformed all comparison models in flood susceptibility mapping.

Conclusions:

• The weakly labeled support vector machine (WELLSVM) is a robust method for urban flood susceptibility assessment, especially with limited flood inventory data.
• The developed flood susceptibility map provides valuable insights for efficient urban flood risk management.
• This study highlights the potential of semi-supervised learning in environmental modeling and disaster management.