Deborah S.K. Thomas

Arriving at estimates for the frequency of hazard events or assessing loss is entirely data dependent. Unfortunately, no comprehensive database or inventory currently exists that provides an account of hazard events and losses throughout the world, much less in the United States (NRC 1999a, b). True, a plethora of data can be found, sometimes as easily as downloading it from or viewing it on the Internet. Some agencies collect data about the characteristics of specific hazards, whereas others are concerned with the effects of hazard events on people. Private companies collect data on insured losses, but these are not readily available to public agencies or researchers, nor are they necessarily geographic in nature. Natural hazards are catalogued separately from technological hazards; the same is true of climatic versus geologic events. The result is a fragmented and questionable data inventory on the nature of hazard events and the impacts that they have at the local, state, and national levels.

Using data throughout the disaster cycle to improve planning, response, relief, and recovery so that losses are minimized translates into two broad issues: (1) a seeming lack of coordination on what information is collected and (2) the need to transform data into useful and timely infor-

mation that users (including those in decision-making positions) can access easily. Even though data collection efforts, databases, and data integration may not be the most fascinating subjects, a thorough appraisal of data reveals what we currently can and cannot say about hazard events and losses. This chapter explores how hazard data are collected, offers a critique of these sources, and suggests future directions for improvement.

The interpretation of hazard events and loss measurements is problematic because the data are collected at a variety of scales and must be considered within the context of their use (Mileti 1999). The methods used by different agencies and organizations for collecting loss data vary widely, consider different types of costs and losses, and frequently change, making geographic comparisons over time challenging, if not impossible. Agencies have different missions, which are reflected in the type of data they collect, compile, and disseminate. Some organizations are more interested in event information for forecasting and modeling, and thus do not even collect figures on damages, economic losses, or human casualties. On the other hand, some agencies are concerned primarily with loss information but do not focus on event frequency or risk estimation. When it is included, damage information is reported in various ways as well. For example, some databases, such as those for tornadoes or lightning, report damage by category in earlier years and as actual dollar amounts beginning in the 1990s. The historical time frame for most data sets also varies substantially, limiting evaluations over time. The ability to compare geographic regions is virtually impossible because of the spatial scale at which data are collected (precise location using latitude/longitude, census tracts, counties, or state).

The need for coordinating disaster data collection and dissemination efforts has not gone unnoticed. The late 1990s saw a coordinated effort to establish a disaster information network when the Disaster Information Task Force (DITF 1997) recommended that the United States take the lead in creating the Global Disaster Information Network (GDIN). The President signed an executive order in April 2000 officially creating GDIN. Efforts are currently under way to formulate future directions for this network. Fortunately, this challenging task has many resources from which to draw. Most relevant, the Federal Geographic Data Committee is a forum in which many federal agencies have worked together since

1990 to establish the best and most efficient means for data transfer between organizations and individuals at the local, state, and national levels (FGDC 1997). The National Academy of Public Administration (NAPA 1998) endorsed the need for data development and sharing now, with more concerted efforts in the future. Of course, hazards data are only one small category of the geographic data being discussed in these forums, but the issues are no less relevant to the hazards community. Finally, consistently collecting loss data is only one component of a much broader endeavor for improved information gathering and distribution for the purpose of reducing the impacts of environmental hazards. Only time will reveal how successful GDIN will be and how the efforts already under way will contribute to its success or failure.

Many organizations have experience integrating hazard-specific information for single or multiple events, although the way they go about it can be very different. For example, a number of agencies collect and disseminate raw data (e.g., National Aeronautics and Space Administration), whereas others (e.g., National Climatic Data Center) process data and distribute summary findings. Sometimes an agency, such as the U.S. Geological Survey (USGS), fills both roles. Another approach is the idea of a data clearinghouse where data are disseminated either through downloaded files or via a series of links to the agency that actually maintains the information. This form of a distributed information system allows the clearinghouse to become a window or portal to raw data stored elsewhere. The Center for Integration of Natural Disaster Information (CINDI) at the USGS is one example. CINDI currently functions as a clearinghouse for disaster information, organizing references to data maintained by others, not only the USGS.

A larger issue is that some agencies or organizations focus on the collection and dissemination only of specific categories of hazard data, rather than all types of hazards. This is to be expected, given the missions of various agencies. For example, the National Climate Data Center (NCDC) is the repository for all types of data related to weather and climate in the United States and worldwide. Geophysical hazards, however, are not represented. Data on geophysical and geohydrological hazards are housed at the USGS. For flood events, one must go to a number of different sources to obtain flood hazard events and losses (NCDC,

USGS, Federal Emergency Management Agency [FEMA]). Technological hazards data are found predominately with the U.S. Environmental Protection Agency (USEPA) with the exception of nuclear hazards, which can be found in the U.S. Department of Energy and the U.S. Nuclear Regulatory Commission (USNRC).

Although there may be integration of broad categories of data (e.g., weather-related hazards) within one agency, a true systematic integration of multiple types of hazard data currently does not exist. Further, loss information is not always included in many of these data sets. The Center for Research on the Epidemiology of Disasters and the Office of U.S. Foreign Disaster Assistance’s EM-DAT is one attempt to create an international database on disaster impacts at the country level (CRED 2000). The scale is coarser than that required for local-level assessments, but it is a start toward a spatially integrated disaster event and loss database.

Efforts are under way by some groups to pull data together on multiple hazards from various sources through geographic referencing and mapping. In fact, hazards data are increasingly available in Geographic Information Systems (GIS) formats, as noted in the previous chapter. However, most of those efforts focus on risks from single hazards through the delineation of risk zones. There is little data on multiple hazards (and thus overlapping risk zones), or on the losses attributed to specific hazard events.

The insurance industry has recognized the need for more comprehensive hazards data. At least two major efforts are under way to aggregate data on insured losses from all types of hazards. The Institute of Business and Home Safety began development of a Catastrophe Paid Loss Database in 1994 to establish loss estimates throughout the insurance industry (IBHS 1998). The Munich Insurance Group aggregates estimates on the cost of disasters throughout the world, which is summarized in a world map of natural disasters (Munich Insurance Group 1998).

Although many attempts to integrate hazards data exist, the data sources and applications do not always present a comprehensive picture of damages or loss of life. Simply put, measuring losses for hazards and disasters is a difficult proposition; and no widely accepted framework or

formula for estimating losses exists (NRC 1999b). Therefore, the losses from natural and technological hazards in the United States are not known with much certainty at all.

At least four challenges are presented when trying to identify and quantify hazard losses: hazards produce multiple-order impacts, costs may be direct or indirect, quantifying intangibles is difficult, and tallying up losses against benefits is not so clear-cut (Pielke 1997). As these four problems attest, hazard loss is a much broader concept than costs or expenditures. Costs generally refer to those losses reimbursed by government or insurance, whereas expenditures involve up-front investments associated with mitigation efforts (NRC 1999b, Godschalk et al. 1999). Clearly, there are many aspects of loss that do not meet these criteria. As a result, the compilation of loss data using bits and pieces of disparate information may not fully reveal the true character of loss.

Most reported loss estimates comprise direct losses in the form of costs, rather than expenditures. Direct loss corresponds closely to the actual event where loss of life or the physical loss of property is apparent (Heinz Center 2000a). These losses may or may not be insured. On the other hand, indirect losses are often referred to as hidden because they appear later than the initial event and involve other social and economic sectors not immediately associated with the direct impact of the disaster. Examples of indirect losses include the disruption of tourism activity in a hurricane-devastated community or the loss of income to local workers because a manufacturing plant was damaged by a tornado. Another form of indirect economic loss that is rarely, if ever, included in databases is that from the informal economic sector. For example, the income loss of illegal immigrants who harvest crops is not normally calculated in the overall cost of a disaster (by indirect loss indicators), but the damage to crops is included. As a general statement, indirect losses are uninsured, a condition that illustrates the importance of looking beyond a single measure (such as direct loss) for calculating losses.

Another continually vexing question involves how to count death, injury, and damage, and when to stop counting. For example, insured losses from any single disaster event remained below $1 billion prior to 1989’s Hurricane Hugo and the Loma Prieta earthquake (IBHS 1998). Since then, insured losses from Hurricane Andrew and the Northridge earthquake have exceeded $15 billion (Munich Re 2000). If indirect, direct, insured, and uninsured losses were all considered, that threshold (greater than $1 billion in loss) would have been reached sooner. Even with these factors taken into consideration, the question of how to assign

value and then measure the value of losses—especially cultural or similar assets—still remains. The interpretation of those measurements is problematic as well since the data are collected at a variety of scales and for many different purposes.

Less tangible are the losses that are not so easily quantified. How do we adequately measure the loss of human life or the losses due to injuries in a disaster? What constitutes a loss for some, is an opportunity for another. For example, construction laborers see an increase in available work and outdated infrastructure is replaced to the benefit of an industry’s productivity and market competitiveness. Furthermore, assessing loss that is emotional, psychological, ecological, or cultural often does not involve a simple count as one might undertake for the number of damaged structures. We need better tools and methods for estimating indirect losses.

Loss estimates also assume that people are impacted equally. In fact, many studies suggest that women, children, elderly, minority, and low-income populations are disproportionately affected by disasters (Cutter 1995, Enarson and Morrow 1998, Mileti 1999). Yet, of the loss databases that do exist, few, if any, systematically collect data on these vulnerable groups. Those who often endure the greatest burden of disasters are simply not distinguished, or become silent victims. If we are to understand all facets of loss so that we can reduce it in the future, then we must quantify the impacts on diverse subpopulations in order to better understand the differential burdens of loss.

The potential for sharing data is immense and has many benefits. More and more agencies and institutions are placing data on the Internet, either in the form of raw data for download or as informational maps. Access to data is easier now, more than ever before. Clearinghouses act as an informational conduit between departments in the same agency or between agencies.

Despite the improved availability, all data have their limitations and people who use some of the national data for loss assessment should be cognizant of these caveats. All data are not of equal quality. Particularly with maps and data in digital form, people tend to lose sight of the fact that data can contain geographic or spatial errors as discussed in the previous chapter. For example, if the coordinates (longitude/latitude) for a tornado were entered incorrectly in 1965, this error now appears as a

point location in the historical database, just like all other tornadoes. There is no real way to verify the location and it now appears “correct.” The point location also may be in the wrong place because the instrumentation in the mid-1960s was not nearly as precise as current technologies. Misunderstandings can occur because people do not completely comprehend the nuances in the data being viewed. For instance, the designation of the 100-year floodplain may be misleading to a homeowner because it implies that by being just outside the “line” drawn on the map, there is no flood risk. Just because data are mapped does not make them any more correct.

Along with issues of accuracy and precision, organizations have other considerations where data are concerned. Data collection, maintenance, revision, and archiving are a significant financial investment. As such, data become a commodity, have a monetary value, and can be sold or purchased. The same data set can also serve very different purposes, depending on the person or agency using it. Whereas the emergency planner in a municipality may intend to use a risk map for mitigation purposes, an insurance company may be interested in raising or lowering insurance rates. Many of these issues can affect the availability of data since agencies must take them into consideration when disseminating information. There are also important legal issues and concerns (liability) about data use and interpretation that are discussed elsewhere (NAPA 1999).

Any attempt at loss estimation, regardless of whether it is local or national, requires a thorough understanding of data nuances. Many of the general caveats regarding hazards data have been mentioned already. Because this book is focused on the availability of loss data associated with hazards, not with warning, prediction, or event information, some of the specific data sources that were used to describe the temporal and spatial trends in hazards losses (Chapters 5 and 6) are highlighted next. Table 4-1 provides an overview of these sources and their limitations.

The NCDC’s Storm Data and Unusual Weather Phenomena is one of the primary data sources that compiles death and damage estimates. This publication is a fairly comprehensive source for information on

meteorological hazards, including severe storms, tornadoes, wind, hail, ice, extreme cold, extreme heat, drought, and all other climate-related events. Each chronological entry listed by state contains information on the hazard type, death, injury, property damage, crop damage, description of the event, and an inventory of affected counties or regions within the state.

The use of Storm Data for hazard loss assessment has several limitations, however. First, because the data are not directly linked to any geographic database, the capability for geographic analysis is severely restricted. Hail, thunderstorm wind, and tornadoes all have latitude/longitude locations, but no other hazard events have such a precise location associated with them. Instead, a single county, multiple counties, or region of a state (e.g., northwest Kansas) is listed. With some modification, those events for which counties are listed could be mapped, but if this is not listed, only state-level mapping is possible. Also, many events transcend a political boundary, creating other difficulties in distributing damages across counties. Another limitation is the inconsistency of hazard classification. For example, snow-related events have as many as 15 different listings, ranging from blizzard/heavy snow to heavy snow to heavy blowing snow. Further complicating its use for damage assessment is the fact that, until recently, property and crop damage was entered in broadrange categories rather than actual amounts. Although Storm Data has many limitations when it is used for hazards assessment, it is often the only source for event and loss estimations collected in any systematic fashion at the national scale for many hazards.

Although Storm Data began in 1959, most of the pre-1993 information is in hard copy only. The NCDC Storm Event Database is a digital version of the Storm Data that began in 1993. As long as the digital version continues, it will be an incredibly useful source into the future and will enable users to develop more place-based studies of weather events and losses.

The National Hurricane Center (NHC) rigorously collects data on hurricanes, tropical storms, and tropical depressions in order to improve warnings and reduce the loss of life. The event information contained in the databases that the NHC maintains is extensive. Data, such as central pressure, wind speed, and category, are collected for each named storm at 6-hour intervals along its path. This event data set does not, however,

contain information on losses, but the NHC does release preliminary reports on each hurricane that does contain overall death and damage estimates. An annual article in the Monthly Weather Review describes the previous year’s hurricane season. This publication summarizes the past season and then provides accounts of individual storms, including information on death, injury, and property damage for each event. This data set is particularly good for looking at yearly trends. Unfortunately, these numbers cannot always be assigned to a given state. Consequently, loss information on hurricanes by state must be culled partially from Storm Data, which has all of the same limitations for hurricanes as for other data sets, described previously. Still, it does provide some idea of the geographic distribution of the losses by state.

Flooding has the most diverse set of data sources available. However, these are consistently ineffectual in trying to understand the magnitude of hazard losses and the geographic distributions of events and losses. For example, the USGS maintains stream gauge readings (in near real time) that can be obtained in digital form (http://water.usgs.gov) but flood stage by stream is not as readily available. The USGS is not responsible for declaring flood watches and warnings. Instead, the information is passed on to the National Weather Service (NWS) and the interpretation of stream gauge data occurs there. The NWS does keep a record of associated losses with flood events, although it is not mandated to do so. Unfortunately, estimates are only maintained at the state level by year back to 1955, although national estimates go back to 1903. The U.S. Army Corps of Engineers also collects loss information, but for a much more limited time period. Further, these estimates are reported by federal fiscal year (October 1-September 30), making calendar-year (the way the vast majority of data are reported) comparisons difficult, if not impossible. As discussed previously, Storm Data does include death, injury, and damage information for all types of flooding, and so, information obtained from the NWS was supplemented for the estimates presented in the following chapters.

With the increased installation of seismographs in all parts of the world, the quality and amount of data collected for earthquake events

has improved greatly (NEIC 1999). Several data sets contain variables such as epicenter location, magnitude, intensity, and/or losses. Examples include the Catalog of Significant Earthquakes from the National Geophysical Data Center (NGDC) and the Centroid Moment Tensor Catalog from Harvard Seismology. The Council of National Seismic Systems (CNSS) is a composite catalog of earthquakes created by merging earthquake catalogs of member institutions and removing duplicate events. Thirty member institutions comprise the CNSS, including public and private universities as well as state and federal agencies. This data set includes the event’s time of occurrence, location in latitude and longitude, and Richter magnitude. Although this is a good database to illustrate event distributions, no deaths, injuries, or damages are available to chart or map loss trends.

The Catalog of Significant Earthquakes is a selective database containing latitude/longitude locations for earthquakes meeting one of the following conditions: damage greater than $1 million, ten or more deaths, magnitude (Richter) 7.5 or greater or Mercalli intensity X (10) or greater. Because of the subset of events, it does not represent the true number of earthquake occurrences, nor does it portray all events that caused any type of loss. The time period is quite extensive, covering centuries. However, one important limitation in the database (similar to Storm Data) is the reporting of economic losses by damage category rather than dollar amount (Table 4-2).

This database from the Smithsonian Institution’s Global Volcanism Program (1999) contains only location and descriptive geologic information for volcanoes active in the past 10,000 years. No deaths, injuries, or damages are available to chart or map loss trends over time or geographically. In the United States, the deaths and damages from Mt. St. Helens were included in NGDC’s Catalog of Significant Earthquakes.

This is one of the few technological hazards databases that includes information on death, injury, and damage. The U.S. Department of Transportation requires that any transportation incident involving hazardous materials be reported. These data are maintained in the Hazardous Materials Information System and include transportation type, death, injury, and damage summaries by state for each year beginning in 1971 (USDOT 2000). Unfortunately, the data are self-reported and are not verified by independent sources. Consequently, the numbers are somewhat questionable, but they at least provide some basis for loss estimation comparisons over time and geographically by state.

Like much of the data on chronic technological hazards, this database also does not contain information about deaths, injuries, or damages (USEPA 2000f). The data only contain locational (latitude/ longitude) information. They can be used to evaluate those who are potentially impacted by these sites, but the specific threat posed by most hazardous sites is not well documented or understood at the national level. We do know, however, of the relative risk associated with these sites based on their listing on the National Priority List. Data on the costs of remediation are available, but the temporal and spatial coverage is often limited.

The Emergency Planning and Community Right-To-Know Act (EPCRA) of 1986 (also known as Title III of the Superfund Amendments and Reauthorization Act) provides for the collection and public release

of information about the presence and release of hazardous chemicals. Through EPCRA, Congress mandated that a Toxic Release Inventory (TRI) be made public to help citizens and community leaders be better informed about hazardous materials in their communities. Facilities are required to report to the USEPA and state governments on releases into the air, water, and land; they must also report the transfer of wastes for treatment or disposal at a separate facility. Note that the TRI reports reflect releases of chemicals, not exposures to people. These estimates alone cannot be used to determine potential adverse effects on human health and the environment. No information about deaths, injuries, or damages are reported (Table 4-1).

Some controversy exists over the reporting accuracy of industrial emissions generally (AWMA 1997) and TRI specifically (Lynn and Kartez 1994). The releases are self-reported annual estimates, and facilities may either over- or underestimate their releases, depending upon their estimation methodology. Some facilities fail to report at all or report only some of their covered chemicals. Additionally, the amounts reported could be the result of a single release or may have been released evenly throughout the year. Still, this particular database is commonly used to assess the differential exposures by various subpopulations, especially in environmental equity studies.

In the absence of a historical record of nuclear incidents comparable to other threats, such as hurricanes or earthquakes, another measure is needed. Equipment failure and human error both contribute to increasing the likelihood of an incident, and so, a review of safety records for nuclear facilities provides a picture of the relative level of risk posed by one facility over another. The Systematic Assessment of Licensee Performance (SALP) by the USNRC is one measure of plant safety that may be used to assess the relative risk from a particular facility (USNRC 1996).

Conducted roughly every 18 to 24 months, the SALP reports grade several operational categories at each commercial nuclear facility (Table 4-3). The USNRC has recently discontinued the SALP review as part of its development of a new reactor oversight and assessment program. The new program began on a pilot basis at nine plants in June 1999 and was adopted for all facilities early in 2000 (USNRC 2000).

Recently, many have recognized the distinct need to institutionalize the collection of compatible data on hazard events and losses. The extent to which losses may be mitigated effectively depends upon an understanding of current and historic trends. Losses stemming from natural and technological sources in the United States, however, are not known with certainty. The plain fact is that, unless loss assessments are based on reliable, comprehensive data collected in a systematic fashion, their value is limited.

We still do not know the true extent of hazard losses in this country. If we simply look at natural hazard events, we can speculate on the economic value of direct losses for a given year, but this is an educated guess, not based on strong empirical proof. We are less able to determine what natural hazards cost at the county or community level and their local economic impact. Only through a systematic effort of loss estimation procedures, collection of comparable data across hazards, georeferencing of all data, and the archiving of the resulting databases will we be able to assess our collective progress toward hazard loss reductions. The time for a multihazard national inventory of hazard events and losses is long passed.