Catastrophic Risk Management

Citizen Reporting in Disasters
By Charles K. Huyck | Published January 13, 2010


Hurricanes, tornadoes, blizzards, earthquakes, floods and terrorism – the catastrophic disasters that remind us of the fragility of life and the uncertainty of investments in the future are subject to sophisticated risk modeling. Risk management is both an actuarial and a spatial problem, and the ability to diversify risk is critical to a healthy insurance industry. As computerized risk models have emerged over the past 20 years, GIS (geographic information system) has become a central hub in the risk management process, enabling risk managers to diversify risk more effectively by linking disparate datasets representing expertise from a variety of disciplines.

Risk assessment is an inherently spatial process that complements and enhances actuarial analysis, which is resulting in growing market demand by catastrophic risk managers for GIS analysts. Two trends in the use of GIS for risk management by government and insurers promise to result in better underwriting, pricing and management for insurance companies, and in better government regulations, emergency response, and allocation of resources.

These trends are:

  1. the migration of risk models to an online environment, and
  2. integration of post-event damage observations into online risk models.

CAT Models

Catastrophe models, or CAT models, are computerized models used to assess the impacts of disasters such as hurricanes, earthquakes, floods, or terrorist attacks. CAT models can be used either to evaluate a single event or to analyze thousands of events and determine probable outcomes. Governments use similar tools, typically called loss estimation models, to assess likely impacts for emergency management purposes. Loss estimation programs typically examine only a single event at a time and may include modules to assess impact to infrastructure, such as transportation, pipelines, or electric utilities.

In both types of models, there are three major components:

  1. the hazard component, where data such as faults, topography, and soil type are used to assess occurrence of hazard at a given location;
  2. the damage component, where building height, age, construction, occupancy, and building codes are used to assess probable damage; and
  3. the loss component, where the financial impact of damage is assessed.

These risk models are used by a wide array of users to support critical decisions:

  • Governments use loss estimates to justify building codes, assess mitigation strategies, stage resources, assess readiness, enact elaborate drills, declare emergencies, plan evacuations, assess post-event loss claims, enhance outreach, perform cost/benefit analysis, and supplement a variety of other activities touching almost every aspect of emergency response. Insurers use CAT models to assess probable losses, inform underwriting strategies, set premiums, and price reinsurance. In addition, insurers, reinsurers, and reinsurance brokers use CAT models to negotiate the price of reinsurance.
  • Agencies such as AM Best and Standard and Poor’s use CAT models to assess the solvency of insurers, particularly in the weeks following a major event.
  • Large corporations with significant real estate may use CAT models to examine business continuity, negotiate with insurers, or assess mitigation opportunities.
  • Commodity traders may also use CAT models to assess the relative impact of large events on world markets.

CAT Models Jump Online

At a basic level, CAT models are desktop applications that provide risk analysts with an interface to import their portfolio of properties in a standardized format, geocode them, and run an analysis. The analysis, which may take a few days when millions of locations are involved, runs a series of simulated events against the portfolio of properties. Such models may not seem suitable for web 2.0 applications, as most users would not want to wait while billions of calculations assess their risk. But with the advent of cloud computing and virtualization, online CAT models stand to change the way in which these applications are used.

Figure 1. VDV (Virtual Disaster Viewer) is a social networking platform for disaster research. Users upload photos and experts interpret them in the context of spatial data. In this image, the Pictometry data in Bing provides users with information that this collapsed building in La’ Aquila, Italy, had a “soft story,” or parking under the structure. This would be difficult to determine from the damage data alone.

Despite the wide pool of potential users discussed above, the loss modeling community is quite small. A handful of experts within a given organization is responsible for massaging the data, adjusting input parameters, and interpreting results. Typically, results are distributed in the form of tables presenting losses by zip code, county, or state, and are used only in tabular form. Models are run infrequently due to time and resource limitations, resulting in significant lag time between changes in a book of business, changes in the underlying science, and changes in behavior at the company level.

Much as GIS was democratized through the advent of Google Earth, CAT models stand to be democratized through Web 2.0.

ImageCat is actively involved in the development of two models, one for government and one for insurance, that are bringing the power of CAT models to an online environment. A primary concern to insurers is the vulnerability of portfolios to catastrophic losses from a single event. This risk can be discerned visually through examination of exposure at a regional level. Furthermore, an online application that integrates remotely sensed imagery can be used to assess not only risk, but also opportunity. Remotely sensed images are used to assess land use and building height, age, or construction type. This is particularly important as insurers begin to examine building stock that has not been exploited by underwriters. And since in a cloud computing environment it becomes much easier to run CAT models frequently, underwriters and upper management can monitor changes in their book of business by drilling back through time. Much as GIS was democratized through the advent of Google Earth, CAT models stand to be democratized through Web 2.0.

Post Event: What Just Happened?

CAT models typically forecast probable future losses, but after an event, they can answer the question “What just happened?”

It generally takes days or weeks for insurers and government agencies to understand the impact of an event after it has occurred. Online loss estimation tools such as InLET can connect to real-time data feeds from the USGS and NOAA to provide a first-order assessment of what is likely to have happened (see sidebar on this page). Such estimates are used for presidential disaster declarations and mobilization of adjusters. The days immediately after an event are when search and rescue efforts are most effective, and when the mobilization of resources has the greatest impact on people’s lives. This is also when the solvency of insurers is brought into question.

Figure 2. The Baseline Risk Platform (BRP) is a web-based CAT model that includes Microsoft Bing maps developed by ImageCat, Inc. This interface displays a sample insurance portfolio that allows users to view Average Annualized Losses aggregated by state, county, or zip code.

The days or weeks normally required for understanding what happened are often shortened by recent technological phenomena widely available to the general public. Thus, a sensor network is emerging with the world’s most sophisticated data processing engine: people. Historically, people have not made good sensors in large-scale disasters, as phone lines are typically overloaded and prioritization of response based on first-hand accounts is highly problematic. However, with digital cameras and GPS-enabled cell phones, geotagged photos are easy to distribute. These data can be interpreted by experts and translated into actionable information.

Figure 3. InLET is a real-time earthquake loss estimation tool for the State of California. In an actual event, USGS Shakecast data is used to predict damage to buildings in less than a minute. This map displays Inglewood building damage at the parcel level for The Great California ShakeOut planning scenario.

ImageCat has deployed the “Virtual Disaster Viewer” (, a system for georeferenced video and GPS-enabled video for two major earthquakes: the L’Aquila earthquake in Italy, and the Sichuan earthquake in China. A social network of engineers interpreting the imagery is used to assess damage. Figure 1 illustrates how analyzing field data within Bing maps reveals not only the spatial distribution of damage, but important information for response, such as how certain buildings looked before they fell down. Expanding these systems to harness users as sensors can capture damage missed by CAT models, and can be used to inform search and rescue efforts and mobilize resources. ImageCat has developed a centralized process for distributing these user-generated disaster data to government agencies.

Real-time information on disasters through such social networks as Twitter, Facebook, and YouTube provide more immediate awareness of the event. However, because they are not integrated into a coordinated 911 system, they leave government and society in an often untenable situation – government cannot respond to emergencies based on ad hoc information. Government needs centralized data and processes in order to dispatch emergency response services, resources, and funds.

A World of Risk Assessors

IT experts and analysts are in a race against time to transform massive amounts of incoming data into actionable business intelligence before it becomes obsolete. The IT challenge is daunting, but with GIS as the basis for Visual Business Intelligence, this burden can be reduced. Easy-to-use mapping tools have the power to turn anyone into a risk analyst. In much the same way that Google Earth turned laymen into image analysts, online CAT modeling tools and real-time data will drastically increase the number of planners, engineers, emergency responders, and citizens with the ability to assess risk.

Worldwide, populations are booming in megacities with substandard infrastructure and building practices that are vulnerable to natural disasters, but despite this escalating risk, compelling disaster mitigation options such as an Indian Ocean early warning system or strengthening of the levees around New Orleans are neglected until it is too late. Mitigation strategies and recommendations sit on bookshelves, while the public feels largely powerless. Online risk models promise not only to expand the capabilities of planners and responders, but also, if implemented correctly, to empower citizens to understand risk and face the political challenges necessary to build safer communities.