Automated Weather Damage Assessment for Event Response from Overhead Imagery

Every year, at least one severe weather event such as a hurricane or tornado impacts a populated area, causing widespread and extensive damage to infrastructure. The number one priority during the critical 24-hour period immediately following a disaster is situational awareness: Which areas have been damaged and to what extent? Are damaged areas accessible and by which routes? The current method to answer these questions is to send crews out for visual surveys, on the ground and in the air, if possible. But surveys can be impeded by weather conditions and by limited ground accessibility due to flooding, bridge collapses, downed vegetation, landslides and other damage from the event. And the resulting information is not readily disseminated.

Automated damage assessment from overhead imagery has the potential to improve the efficiency and effectiveness of response operations. Advances in remote sensing technology – both airborne and space-based platforms – have led to more and better quality imagery and better availability, and this trend is likely to continue. In order to utilize available imagery, automated processing is needed to quickly distill useful information, reducing the need for human analysis.

We are developing algorithms for automated damage assessment, and a framework for delivering geospatial situational intelligence to support response operations, primarily for electrical utilities. Electrical power is critical to the overall response to an event for communications, and to human safety for heat, light, food preservation and clean water. Electrical utilities are responsible for bringing their services back online as quickly as possible and they could restore power more quickly if accurate damage assessments were provided within 24 hours of an event.

The initial algorithms developed to date are a change-based damage assessment algorithm and a rubble-based damage assessment algorithm. The former is applied to two images, one pre-event and one post-event. The algorithm was designed to be flexible enough to work with two images from different sensors and with different resolutions, so that whatever imagery is available can be utilized. The algorithm was applied to satellite imagery for eight different historical tornado events and the resulting damage assessment agreed well with human analysis. The rubble-based algorithm is applied to a single, post-event image and uses the presence of rubble and debris as an indicator of damage. This algorithm requires sub-meter resolution and works well with the imagery acquired by NOAA's Remote Sensing Division to support emergency response using their airborne sensor. The algorithm was applied to the imagery collected after Hurricane Ike and Hurricane Sandy. The results show that the effectiveness of this approach will vary by event. In the case of Hurricane Ike, there was a significant amount of wind damage to structures that produced detectable rubble. In the case of Hurricane Sandy, however, much of the damage was due to flooding and storm surge and so a different algorithm would be more appropriate, such as flood mapping from synthetic aperture radar imagery.

As we continue to develop algorithms and the framework for information delivery, we invite collaboration from the meteorological and emergency response communities, electrical utilities, and imagery providers. Our long-term vision for a weather-event decision-support system includes the integration of data such as storm tracks, localized wind speed and direction, stream gauge levels, land cover and infrastructure locations to enhance the accuracy of automated image-based damage assessment.