Flood Damage in the United States, 1926-2003
A Reanalysis of National Weather Service Estimates

3. Development of the Data Sets

The national data obtained from NWS consisted of annual total damage estimates for the U.S., including three territories: Puerto Rico (since 1975), the Virgin Islands (since mid-1980s), and Guam (since 1994). The state data contained annual damage estimates for each state and, in recent years, the three territories. In the national data, we subtracted estimates for the three territories from the U.S. totals to create a more uniform time series representing only the 50 states.

NWS estimates were spot-checked against those from other agencies. Estimates that appeared to be extremely large or small compared to published accounts of events were examined especially closely. In individual events that received follow-up study by the USACE, more accurate estimates were sometimes available. However, except during 1976-1982, there exists no compelling reason to change the NWS estimates or defer to another agency’s estimates. Section 5 provides a quantitative assessment of uncertainty in the estimates and the implications for their effective use.

With a few important exceptions, the estimates presented as a result of this project have their origins in published NWS data. Obvious clerical errors have been corrected (see Section 4).

A. Resolving the Data Gap, 1976-1982

To compile a complete time series of annual estimates required finding additional flood damage estimates for the years 1976-1982. As explained in Section 2, NWS ceased publication of annual flood damage summaries after 1975. Publication of comparable damage estimates did not resume until 1983, when USACE reports made damage estimates available again at the state and national levels, but not at the river basin level.

To make the state and national data sets as complete as possible, we focused on obtaining and evaluating estimates for 1976 through 1982. The NWS website (NWS-HIC 2001) included previously unpublished national flood damage estimates for 1976-1982, and an NWS spreadsheet included unpublished state estimates for that period. However, the national estimates and the state total estimates differed by large margins. An old, undocumented NWS computer printout tallied individual floods, by state, in the years 1976-1988, but we found it to be filled with errors and inconsistencies.

Despite a curtailment of effort, the NWS continued to compile some damage estimates during 1976-1979, which served as a starting point for our reconstruction attempts. We were able to develop estimates for 1976-1979 based on information in the NWS files and reports from other sources, as described in Appendix A.

Although we tried to reconstruct estimates for 1980-1982, there were not enough sources of information, either from NWS or other agency publications, to provide estimates for those years comparable to the data in the overall data set. Furthermore, there were some large disparities between estimates found in the NWS-HIC archives for the period 1980-1982 and damage estimates provided by states, leading us to conclude that some of the damage estimates for this time period are highly unreliable (see Section 5). Therefore, estimates for 1980-1982 are not included in the reanalyzed data sets, and we judge that data published by NWS for this period is of consistently lower quality than in other years.

A few general comments can be made about 1980-1982. Flood damage descriptions in Storm Data, which were sparse in previous years, became even rarer in 1980-1981. The information that does exist for the period suggests that 1980 and 1981 were extremely dry years in most parts of the country, so flood damage was probably small compared to other years (Wagner 1982, USGS 1991, notes in NWS files). On the other hand, descriptions in Storm Data suggest that flood damage rose to a higher level in 1982, perhaps close to the average level of that time.

B. Annual National Flood Damage Estimates (1926-1979, 1983-2000)

Since flood damage estimates for 1983 through 2000 are available only for fiscal years (October-September), it is desirable to compile the entire national flood damage data set using fiscal years. Fortunately, in its annual flood damage summary for 1975, Climatological Data National Summary (NWS 1977, vol. 13, p. 117) published national flood damage estimates by month for the years 1925 to 1975. Therefore, we were able to calculate national annual damage totals based on fiscal years for 1926-1979, creating a consistent form for the full national data set.

Table 3-1 shows annual damage estimates for the United States, by fiscal year, in millions of current dollars and in millions of inflation-adjusted 1995 dollars. The implicit price deflator used to adjust for inflation is also shown in the table.

C. Annual Flood Damage Estimates for the States (1955-1979, 1983-2000)

Annual damage estimates for each of the 50 states are given in Appendix B. The estimates for 1955 through 1975 are taken from Climatological Data National Summary (NWS 1977, vol. 13, p. 121), and are based on calendar years. Estimates for 1976-1979 are based on our reanalysis of available data (described above), and are presented by calendar year to be consistent with the earlier data. The estimates for 1983-2000 are taken from Army Corps of Engineers Annual Damage Report to Congress (1993, 2001), and are based on fiscal years (October-September).

D. Annual Flood Damage Estimates in River Basins (1933-1975)

The NWS and U.S. Weather Bureau compiled annual damage estimates by river basin from 1933 through 1975, publishing them first in the Monthly Weather Review (1933-1947) and later in Climatological Data National Summary (1948-1975). To make these estimates accessible to users, we organized them by large river drainages in a uniform format for the full time period.

Table 3-1. Estimated U.S. Flood Damage, by Fiscal Year (Oct-Sep).

* Source: U.S. Bureau of Economic Analysis, 2001.
— Data unavailable, see text for discussion.

The basin-level damage estimates are available in spreadsheet form from our website. Estimates are presented by calendar year. The grouping of basins within drainages is somewhat different from that commonly used to define water resources regions (e.g., U.S. Dept. of Commerce, 1978 Census of Agriculture) because, over the years, the NWS sometimes changed its groupings. We developed uniform basin definitions for the full time period by using the following organizational system:

1. Damages are grouped by drainage (e.g, St. Lawrence Drainage, Upper Mississippi, Great Basin) starting in the eastern part of the United States and moving towards the west coast, and then alphabetically by individual or grouped river basin(s).
2. Often, the NWS grouped individual rivers together in annual summaries. For example, damage on the White and Wabash Rivers were usually included together as one estimate. If the published sources of flood data included damage for two river basins together in one year, then data for these two (or more) rivers were added together for all other years. This was the simplest way to produce a coherent data set that could be searched and produce just one row of data for one river basin.
3. In many of the years, damage on unnamed streams was included. If the publication did not give a stream name, damage was included in a row for the drainage called "small streams."
4. Sometimes the publications would include a river and its small tributaries together, by saying "X River and tributaries." When damage was published in this format, it was entered into the database under the river itself. So, damage listed for some rivers in some years may include not just the river, but its small tributaries (such as creeks).
5. Creeks that were included separately in NWS publications from the rivers to which they are tributaries were entered into the database separately. Creeks can be differentiated from rivers in the database because they are labeled "Cr.," whereas rivers are entered with the river name only. An exception to this rule is for rivers with Spanish names, such as the Rio Hondo and Rio Grande. Since users may want to search for "Rio Hondo" rather than "Hondo," "Rio" is included in the database.
6. Users looking for damage information on rivers with branches (such as North Platte, South Platte, and Platte) should look for each of these branches. In some cases, all of the branches of one stream are included together, and in some cases they are not.
7. Several of the streams in the data set cross drainage boundaries. If there is a question about which drainage a stream is in, a user should look in both drainages.

E. Use of the Damage Estimates

Users of these data sets should be aware that there is uncertainty in the damage estimates, with a likelihood of large errors in some estimates. Types of inaccuracy are described in Section 4, and the magnitude of errors is analyzed in Section 5. In consideration of uncertainty, recommendations regarding appropriate uses of the data are offered in Sections 6 and 7.

[ Back ]   [ Next ]