Challenging the Odds of Hurricane Damage Data Collection
A Detailed Account from a First Responder
|Dr. Beverley Adams
Remote Sensing Technical Director
Dr. J. Arn Womble
Long Beach, Calif.
|Shubharoop Ghosh of ImageCat, Inc. and Carol J. Friedland of the Louisiana State University Hurricane Center also contributed to this article. |
Even though it may be the most well-known hurricane of recent history, Hurricane Kat-rina, which struck the U.S. Gulf Coast states on Aug. 29, 2005, was only the third most powerful windstorm of the 2005 season. It was not Katrina’s power, but her storm surge that caused catastrophic damage along the coastlines of Louisiana, Mississippi and Alabama. After the levees separating Lake Pontchartrain from New Orleans were breached by the surge, about 80 percent of the city was left flooded. All told, Hurricane Katrina was responsible for an estimated $75 billion in damages, and the deaths of 1,418 people.
Countless relief organizations made themselves available to help the impacted regions using an arsenal of advanced technologies. We at ImageCat have spent the past few years refining our portable VIEWS® (Visualizing Impacts of Earthquakes with Satellites) reconnaissance and visualization system to map and assess damage in the aftermath of major natural disasters such as Hurricane Charley, the Bam Iran earthquake and the Southeast Asian tsunami.
By the time Hurricane Katrina hit, we could draw upon our collective experience and were well equipped to respond. We quickly realized that Katrina’s unique combination of certain circumstances created dramatic — but not insurmountable — challenges for mapping and assessing damage. See Figure 1.
Hurricane Katrina: Unique Among Storms
In terms of windspeed, Hurricane Katrina was not a major event. Prior to reaching land, Katrina’s windspeed measurements indicated that maximum sustained winds exceeded 170 miles per hour, making it a Category 5 storm. But the storm weakened significantly in the 18 hours preceding landfall, and the storm had decreased to a Category 3, based on sustained windspeeds. However, tremendous storm-surge capabilities persisted from the storm’s prior Category 5 status.
Hurricane Katrina’s landfall near the Louisiana-Mississippi border placed the Mississippi Coast in the classically vulnerable right side of the hurricane. Windspeeds in New Orleans likely did not exceed Category 1 or 2 levels. While the west side of the hurricane did experience lower windspeeds compared to the east, the particular location of Lake Pontchartrain resulted in water being pushed towards New Orleans, stressing the miles of levees constructed to hold back the lake and the network of canals. Coupled with the below-sea-level elevation of the city, this situation proved disastrous as floodwaters poured into New Orleans.
As a historic event, Hurricane Katrina was devastatingly record-breaking. Certainly, it was quickly acknowledged as the costliest U.S. hurricane on record, causing billions of dollars worth of damage. But Katrina was also characterized by the convergence of multiple hazards within a single event — including flooding, windstorm, storm surge and levee breach. The unusual combination of these hazards created challenges for performing rapid and widespread assessments of the damage, and for preserving the perishable damage characteristics for conducting future research activities. Physical access to the area was severely limited due to evacuation orders or the obstruction of roads by fallen trees, high flood waters or collapsed bridges.
In the wake of disasters such as Hurricane Katrina, the gathering of situational information is a critical priority, since it provides decision support to guide immediate response and long-term recovery activities. To conduct any accurate mapping efforts, we at ImageCat knew we would have to find a way either to clear a path to the scene, or to gather information about the damage using other means.
The Importance of Post-Disaster Reconnaissance
Reconnaissance activities aimed at uniformly and consistently documenting damage conditions following windstorms are important for rapidly assessing damage and losses, understanding the effect of these storms on the built environment, evaluating the effectiveness of mitigation measures and gauging the progress of windstorm mitigation over time. With proper analysis and implementation, such studies can help to lessen or eliminate the tragic consequences of future windstorms and help with construction and reconstruction efforts.
However, a number of barriers to conducting complete and uniform surveys have emerged in engineering-oriented investigations of post-windstorm damage areas.
These most notably include the inability to:
- Document the damage states of all buildings in large areas, both rapidly and in detail, before clean-up efforts commence;
- Examine and document the “pristine” spread of windborne debris, which is frequently moved before investigators arrive at the damage scene;
- Access damaged areas isolated by either law-enforcement agencies or by natural causes, such as fallen trees, washed-out roadways and collapsed bridges;
- Rapidly screen large areas for relative levels of damage, for information to use in the strategic planning of statistically sampled damage surveys, especially in unfamiliar areas; and
- Assign consistent damage characteristics to structures due to the biases among investigators.
These types of barriers can lead to costly time delays in assessing damage across a wide area and in strategically deploying emergency-response personnel and supplies. Additionally, there is inevitably a loss of valuable information for long-term research that could otherwise lead to improved understanding of severe wind effects on the built environment.
Carol J. Friedland, a graduate research assistant with the Louisiana State University (LSU) Hurricane Center, agrees that the ability to gather damage data uniformly is critical — but, she admits, it is often difficult to achieve accurate data from ground surveys due to human error and natural biases. “Collecting data hour after hour in the aftermath of any major disaster creates significant ‘damage fatigue.’ Seeing the same damage over and over, we as humans are bound to make mistakes or miss critical damage characteristics. Also, wind and water experts ‘see’ things differently and may catalog data with biases toward their specific expertise.”
The benefits of augmenting the conventional ground survey approach with advanced technology, such as remote sensing, cannot be denied. The emergence of remote sensing as an accepted source of post-disaster information can be traced back to the Sept. 11, 2001 attacks on the World Trade Center towers, where remote sensing was widely deployed at Ground Zero for situation assessment.
Within the wind-engineering arena, Hurricane Charley, which struck the Gulf Coast of Florida in August 2004, marked the initial deployment of satellite remote sensing combined with ground truthing for perishable damage assessment. In this instance, pre- and post-hurricane satellite imagery was integrated within the technology-driven VIEWS reconnaissance system and used to record the damage state of more than 10,000 buildings. The 2005 hurricane season saw the implementation of remote sensing techniques by risk modeling companies to help generate initial loss estimates.
Leveraging Reconnaissance from Post-Hurricane Katrina
On Aug. 30, one day after Hurricane Katrina made landfall in Louisiana, we deployed a response team to the Gulf Coast region, consisting of our own staff and researchers from the Multidisciplinary Center for Earthquake Engineering Research (MCEER), with backing from Risk Management Solutions (RMS). Our objectives were to perform rapid and widespread assessments of damage at a per-building scale, and to preserve the perishable damage characteristics of this unique multi-hazard event for future research activities.
We overcame the challenges associated with Katrina’s multiple hazards by leveraging a variety of satellite and aerial imaging platforms that enabled us to conduct thorough and effective analysis of the damage.
|Figure 2 - Radarsat-1 SAR image of Hurricane Katrina approaching Louisiana and Mississippi on August 28, 2005. This image shows the surface of the sea within the hurricane (SAR imagery from MDA Corporation).|
Our Post-disaster Damage Verification (PDV) team used an airplane to fly over the Biloxi, Miss. and New Orleans, La. regions and record the devastation. Within 48 hours of landfall, GPS-referenced, five- to 15-centimeter resolution imagery was acquired by airborne deployment of our VIEWS field data collection and visualization system. VIEWS integrates pre- and post-disaster remote sensing data with in-field observations and provides a permanent, geographically referenced record of the post-disaster scene.
Geo-referenced high-definition video and still photographs recorded the initial storm flooding and subsequent overtopping of the levees surrounding New Orleans. Captured on Aug. 30, prior to either cloud-free post-storm satellite or NOAA airborne coverage, the information provided the earliest known indication of damage within New Orleans.
In the days following the hurricane, remote sensing data from optical and radar sensors were analyzed in detail, providing response teams with a rapid and synoptic view of the affected areas at the regional, neighborhood and individual-building level. Through expert interpretation, these data-sets yielded important information for disaster managers concerning the nature, severity and extent of building damage at regional and neighborhood scales throughout the Gulf Coast region.
LSU’s Friedland accompanied the ImageCat team to conduct surveillance and collect damage information in New Orleans. According to her, VIEWS provided a critical ability to gather Hurricane Kat-rina damage data because the system is always on. “Because VIEWS is continuously and systematically collecting data, it eases the strain of manually cataloging data and analyzing it as we go along.”
Also, Friedland says, VIEWS creates an accurate, unbiased record of the physical situation that can be referenced on a computer screen hours, weeks — or even years — afterward. “The ability to digitally catalog data as a permanent record will no doubt lead to more accurate damage assessments for many types of disasters, and will help speed the process of categorizing damage data for analysis and reporting purposes.” See Figures 2-3.
|Figure 3 - Expert-interpretation-based flood limit (delineated in yellow) overlaid with the QuickBird false-color composite image of New Orleans for September 3, 2005 (from which the flood zone was derived), comprising enhanced and mosaiced tiles (QuickBird Imagery from DigitalGlobe).|
Data Variety is Fundamental
The complementary characteristics of the different remote sensing platforms, which included optical, radar and LiDAR, afforded us the ability to view the damage from multiple vantage points: optical imagery allowed us to view the landscape as it naturally appears to the human eye, radar gave us the magical ability to “see” through clouds and darkness, and LiDAR helped us visualize the varying heights of terrain and water. Suddenly, our picture was complete and led to the rapid, thorough, and consistent assessment of the situation.
Our team used post-Hurricane Katrina remote sensing images that span a wide variety of spatial resolutions. The spatial resolution of an image determines the ability to view individual features such as buildings and bridges. It also affects the ability to monitor and assess damage conditions, and depends on the nature of the hazard itself — for example, flooding, wind pressure, and storm surge.
Pixel sizes of approximately 10 meters or smaller are necessary to discern the presence and location of individual buildings, while much smaller pixels, on the order of one meter or less, can distinguish damage conditions of individual buildings, such as damage to roofs caused by wind pressure. Widespread flooding can be detected and monitored using less-detailed moderate-resolution imagery.
In addition to spatial factors, spectral resolution influences the use and usefulness of the data. Physical materials have different reflectance values in different portions of the electromagnetic spectrum, and so features of interest, such as construction materials, water, and vegetation, can be identified by unique and distinguishing characteristics. The use of multispectral remote sensing systems is therefore critical for the separation of constituent materials within an image, and for the interpretation of images for damage assessment.
While post-disaster remote sensing images accurately capture damage caused by a disaster, pre-disaster coverage is extremely useful for establishing the “normal” situation. Pre-storm images provide a benchmark “no-damage” baseline for change-detection operations, comparing pre- and post-storm images on a regional and/or per-building basis, which creates important validation of damage-assessment results, particularly for moderate damage levels.
|Figure 4a and Figure 4b --Pre- and post-Hurricane Katrina Landsat-5 TM imagery was used to detect storm surge damage in Mississippi (Landsat imagery from USGS and NASA). |
|Figure 5a and Figure 5b --NASA’s Landsat-7 30-meter imagery reveals flooding extent in New Orleans on August 30, 2005. (NASA imagery from Jesse Allen with data provided by USGS EROS Data Center and Landsat Project Science Office at Goddard Space Flight Center).|
For the investigation of Hurricane Kat-rina damage, pre-storm satellite imagery from Landsat-5 and QuickBird provided this important comparative baseline. Recent Landsat-5 imagery for the New Orleans area and Mississippi Coast was accessed through the United States Geological Survey (USGS) disaster Web site, and an older scene collected in 2000 was obtained from the Stennis Space Center Web site. DigitalGlobe’s QuickBird imagery, which had been collected over the area in March 2004, created critical capabilities for comparing the landscape and infrastructure of the Gulf Coast region before and after the disaster occurred.
LiDAR airborne imagery was available from the NOAA and the USGS disaster Web sites for use in flood-elevation mapping. NOAA aerial images acquired in 2004 following Hurricane Ivan were available for such areas as Gulfport and Biloxi, Miss. Because the effects of Hurricane Ivan on buildings in these locations had been so minimal, the post-event images also served as a set of pre-storm data for Hurricane Katrina. See Figures 4 and 5.
Neighborhood Damage Detection
While various sources of imagery have been used for decades to evaluate the environmental impacts of storms, the newest player in the field of post-disaster reconnaissance—high-resolution satellite imagery—has raised the bar in terms of the amount of visual detail typically available in non-airborne collected datasets. While providing visual coverage over large expanses of area, this imagery is fine enough in detail to depict damage down to the neighborhood and individual-building level.
High-resolution satellite imagery of New Orleans, acquired by DigitalGlobe, became available for analysis within days of Hurricane Katrina. The imagery was collected on Sept. 3 and distributed to MCEER researchers at ImageCat on Sept. 4. A neighborhood-based high-velocity flood-damage assessment for areas near the levee breaches was performed using this dataset. Each of the levee-failure locations was identified using expert interpretation of a pan-sharpened natural-color composite of the imagery. Areas surrounding the levee failures were inspected using the composite image to extract information concerning the approximate length of the breaches, and the locations of flooded areas.
For example, the east side of the 17th Street Canal bordering Jefferson and Orleans Parishes failed, causing high-velocity flooding in Orleans Parish on the east side of the canal, while Jefferson Parish to the west was not affected. The imagery also revealed information about changes in urban vegetation, and about major changes in building configurations due to high-velocity floodwaters, which caused the displacement of entire buildings. See Figure 6.
|Figure 6 - Example of per-building visual assessment of wind-pressure damage. Colored symbols are superimposed on a NOAA aerial image, indicating the damage states of individual roof facets according to the Remote-Sensing Damage Scale (base image from NOAA).|
From Satellites to Google to 3D Cities
Recognizing the importance of promoting rapid and accurate post-disaster reconnaissance, MCEER and ImageCat researchers have, for a number of years, worked to streamline information-gathering activities through the implementation of advanced technologies such as remote sensing. Hurricane Katrina heralded the benchmark deployment of remote sensing for reconnaissance planning and damage-assessment support in a unique multi-hazard context.
Taking remote sensing technology leaps and bounds further were the launches of Internet-based Google’s Google Earth and Microsoft’s Virtual Earth. Immediately after Hurricane Katrina, Google Earth provided a publicly accessible vehicle for distributing freshly collected satellite imagery as well as VIEWS ground-reconnaissance information. More than 18,000 VIEWS images illustrating Katrina’s damage along the Mississippi Coast, and more than 27,000 images from New Orleans, were integrated into Google Earth.
Evacuated residents with access to an Internet connection were able to view post-event images showing damage to their neighborhood — offering an unbeatable mechanism for helping residents understand the damage sustained by their properties before trying to return home.
|Figure 7 - VIEWS field-reconnaissance data accessible via Google Earth. This image shows storm-surge damage in Biloxi, Miss.|
|Figure 8 - This image depicting 3D-LondonTM provides detailed information about building shape, height, number of stories and footprint area. Courtesy of ImageCat.|
What’s next for the field of post-disaster reconnaissance? Our work is now emphasizing pre-disaster planning with the release of 3D-City, a series of 3D city models serving as tools for managing corporate risks within vulnerable urban and industrial environments.
Offering a detailed structural rendering of buildings in a given city, 3D-City provides attribute information about building height, number of stories, and footprint area. The model can be visualized using popular online and off-the-shelf software packages including Google Earth (see Figure 7), MSN Virtual Earth, VRML, ESRI’s ArcGIS 3D Analyst and other common GIS applications, putting critical geotechnology data at the fingertips of decision makers. First issued in April of this year representing the city of London (see Figure 8), 3D-City takes a proactive approach to planning and preparing for potential disasters or other events impacting a city.