Flood Mapping | Hadi Shokati

This project develops efficient deep learning models for rapid flood mapping from aerial imagery, enabling quick identification of flood-affected areas for emergency response and damage assessment.

Overview

Flooding is a major natural hazard that requires rapid response to minimize loss of life and property and to facilitate damage assessment. Aerial imagery, especially from unmanned aerial vehicles (UAVs) and helicopters, plays a crucial role in identifying flood-affected areas. This study presents two state-of-the-art segmentation approaches for automated flood mapping from aerial imagery.

Dataset and Study Area

We used the Flood Area Dataset Flood Area Dataset comprising 290 images with their corresponding masks. The images were captured after some flood events in different regions using UAVs and helicopters with optical sensors.

Methodology

We developed and compared two complementary approaches for flood segmentation:

Approach 1: Fine-Tuned Segment Anything Model (SAM)

Building on the success of SAM for image segmentation, we fine-tuned the model specifically for flood detection from aerial imagery. We compared two prompting strategies:

Point Prompts: Users click on flooded areas to guide segmentation

More flexible and intuitive for complex flood boundaries
Better captures irregular flood patterns
Performance: Accuracy 0.96, IoU 0.90

Bounding Box (Bbox) Prompts: Users draw boxes around flooded regions

Simpler input method
Less effective when floods extend to image edges
Performance: Accuracy 0.82, IoU 0.67

Key Finding: Point prompts significantly outperformed Bbox prompts because flood imagery often shows water extending from edge to edge, making bounding boxes less effective at capturing boundary details.

Approach 2: U-Net with ResNet Backbones

We implemented U-Net architectures with pre-trained ResNet backbones for feature extraction:

U-Net + ResNet-50:

Standard deep architecture
Performance: Accuracy 0.87, IoU 0.72

U-Net + ResNet-101:

Deeper architecture for more complex feature extraction
Better segmentation of detailed flood boundaries
Performance: Accuracy 0.88, IoU 0.74

Key Finding: The deeper ResNet-101 backbone improved performance by extracting more complex and detailed features, enhancing U-Net’s ability to segment flood-affected areas accurately.

Results

Performance Comparison

Performance comparison of fine-tuned SAM (with point and Bbox prompts) and U-Net models (with ResNet-50 and ResNet-101 backbones) for flood segmentation.

Best Overall Performance: Fine-tuned SAM with point prompts

Highest accuracy (0.96) and IoU (0.90)
Most suitable for operational flood mapping
Requires minimal user interaction

Best Fully Automated Approach: U-Net + ResNet-101

No prompts required
Good performance (Accuracy 0.88, IoU 0.74)
Suitable for batch processing of large image datasets

Segmentation Examples

Examples of flood segmentation results showing original aerial imagery, ground truth, and model predictions across different flood scenarios.

Applications

The developed models provide valuable tools for multiple stakeholders:

Emergency Response Teams

Insurance Companies

Urban Planning and Management

Research and Model Development:

Advantages Over Traditional Methods

Speed: Near real-time processing compared to hours of manual interpretation
Consistency: Objective, reproducible results across different analysts
Scalability: Can process large numbers of images automatically
Accuracy: Surpasses manual interpretation in challenging conditions
Flexibility: Adaptable to different types of aerial imagery (UAV, helicopter, satellite)

Publication

Citation:

Shokati, H., Seufferheld, K. D., Fiener, P., & Scholten, T. (2026). Rapid flood mapping from aerial imagery using fine-tuned SAM and ResNet-backboned U-Net. Hydrology and Earth System Sciences, 30(5), 743-756.

DOI: 10.5194/hess-30-743-2026

Status: Published in Hydrology and Earth System Sciences (HESS) (Shokati et al., 2026)

Flooding is a major natural hazard that requires a rapid response to minimize the loss of life and property and to facilitate damage assessment. Aerial imagery, especially images from unmanned aerial vehicles (UAVs) and helicopters, plays a crucial role in identifying areas affected by flooding. Therefore, developing an efficient model for rapid flood mapping is essential. In this study, we present two segmentation approaches for the mapping of flood-affected areas: (1) a fine-tuned Segment Anything Model (SAM), comparing the performance of point prompts versus bounding box prompts, and (2) a U-Net model with ResNet-50 and ResNet-101 as pre-trained backbones. Our results showed that the fine-tuned SAM performed best in segmenting floods with point prompts (Accuracy: 0.96, IoU: 0.90), while bounding box prompts led to a significant drop (Accuracy: 0.82, IoU: 0.67). This is because flood images often cover the image from edge to edge, making bounding box prompts less effective at capturing boundary details. For the U-Net model, the ResNet-50 backbone yielded an accuracy of 0.87 and an IoU of 0.72. Performance improved slightly with the ResNet-101 backbone, achieving an accuracy of 0.88 and an IoU of 0.74. This improvement can be attributed to the deeper architecture of ResNet-101, which allows more complex and detailed features to be extracted, improving U-Net’s ability to segment flood-affected areas accurately. The results of this study will help emergency response teams identify flood-affected areas more quickly and accurately. In addition, these models could serve as valuable tools for insurance companies when assessing damage. Moreover, the segmented flood images generated by these models can serve as training data for other machine learning models, creating a pipeline for more advanced flood analysis and prediction systems.