Classifying Building Roof Damage Using High Resolution Imagery for Disaster Recovery

Area of Science:

• Remote Sensing
• Disaster Management
• Geospatial Analysis

Background:

• Traditional post-hurricane damage assessments are resource-intensive and time-consuming.
• Remotely sensed data offers a rapid and cost-effective alternative for data collection.
• Hurricane Michael (2018) caused significant damage in 15 Florida counties.

Purpose of the Study:

• To evaluate the cost-effectiveness of aerial imagery for measuring blue tarps on buildings post-disaster.
• To assess the utility of a support vector machine model for identifying blue tarps and indicating building damage.
• To provide a scalable damage assessment method for resource-limited jurisdictions.

Main Methods:

• Utilized aerial imagery from 15 Florida counties impacted by Hurricane Michael.
• Developed and applied a support vector machine model to classify blue tarps on building parcels.
• Assigned a damage indicator to parcels based on the model's blue tarp detection.

Main Results:

• The support vector machine model achieved an overall accuracy of 85.3% (74% sensitivity, 96.7% specificity).
• Approximately 7% of all parcels (32,357 total) in the study area were identified as having blue tarps.
• Detected blue tarps indicate areas requiring disaster impact and recovery attention.

Conclusions:

• Aerial imagery combined with machine learning provides a cost-effective method for post-hurricane damage assessment.
• This approach can significantly aid jurisdictions with limited financial resources for on-the-ground evaluations.
• The detection of blue tarps serves as a valuable metric for disaster impact and recovery monitoring.