Enhanced large-scale flood mapping using data-efficient unsupervised framework based on morphological active contour...

Area of Science:

• Hydrology and Remote Sensing
• Computer Vision and Image Processing
• Environmental Monitoring

Background:

• Floods are significant hydrological extremes causing extensive environmental damage.
• Synthetic Aperture Radar (SAR) derived flood maps are vital for damage assessment and management.
• Existing SAR flood mapping often relies on supervised methods, facing challenges in data generalization, labeling, and model transferability.

Purpose of the Study:

• To develop an unsupervised computer vision framework for accurate flood extent mapping using unitemporal SAR images.
• To address the limitations of supervised SAR flood mapping, particularly regarding data requirements and model transferability.
• To enhance the accuracy and efficiency of flood mapping for improved disaster management.

Main Methods:

• Proposed an unsupervised framework utilizing the Morphological Chan-Vese Active Contour Model (Morph CV ACM) with unitemporal SAR imagery.
• Conducted sensitivity analysis of model parameters for clustered and scattered flooding patterns.
• Developed a localized Morph CV ACM version with adaptive parameter adjustment based on empirical formulas.

Main Results:

• The proposed framework achieved a high F1 score of 0.935 in mapping flood extents in North India.
• Demonstrated significant improvement over the Otsu segmentation method in detecting flood extents.
• Showcased enhanced mapping accuracy in built-up and agricultural areas, crucial for damage assessment.

Conclusions:

• The novel unsupervised Morph CV ACM framework provides accurate and resilient flood extent mapping from SAR data.
• The framework's automation and minimal data requirements facilitate near-real-time, large-scale flood mapping.
• This approach offers a valuable tool for improving flood damage assessment and informing effective flood management strategies.