The purpose of the NFF mailing list is to provide information on the U.S. Geological Survey (USGS) National Flood Frequency (NFF) program. NFF is a computer application that is used to estimate peak discharges for unregulated streams. Values (discharges) derived by the program often serve as input for other applications (such as hydraulic computer models) that are used to technically support end-products (map revisions) of the National Flood Insurance Program (NFIP).
Discussion on USGS Regression Equations and the NFF Program
The USGS developed a computer program titled "National Flood Frequency" or "NFF" that estimates the flood frequency and magnitude for ungaged sites through the application of the appropriate regional regression equations. NFF was released in 1993 and does not incorporate any revisions to regional regression equations that occurred after September 30, 1993. Since 1993 a significant number of the regression equations have been revised. The USGS is in the process of revising the NFF computer program to incorporate the updated regression equations. The revised version of NFF will be released soon.
The regional regression equations are currently being used for National Flood Insurance Program (NFIP) purposes. Therefore, FEMA would like to continue with this listserv and discuss issues of interest in the application of the regional regression equations for NFIP purposes. Upon the release of the revised NFF program, the focus of this listserv will shift to assist users in becoming familiar with the revised NFF program and its application for NFIP purposes.
Measures of Accuracy in NFF
Every USGS regional flood report contains a discussion of the accuracy achieved in the resulting regression equations. The most frequently used value is the standard error of the estimate in percent. Below is an example from the South Carolina equations for a tributary to Little Beaverdam Creek with a drainage area of 1.039 square miles in the Piedmont region of the State.
These equations report the standard error of prediction, which is being used more frequently by analysts. The standard error of estimate is a measure of the variation between the estimates from the equations and the analyses of the gage station data from those stations used to derive the equations. The standard error of prediction is a measure of the accuracy of the equations when compared with the results from station data not used in developing the equations. NFF provides the one used in the individual State report because that is the only one available.
The prediction error for ungaged sites is made up of modeling error and sampling error. The modeling error is that portion of the error that cannot be reduced by additional data collection. The sampling error can be reduced by operating the stations longer, by adding sampling stations to the network, or by a combination of both.
Often, the standard errors are converted to equivalent years of record, as seen above. For example, the 50-year recurrence interval discharge has a standard error of 28% and equivalent years of record of 15 years. Provide a range in discharge which represents this 28% error. For example, does this mean the error is +- 28% or (420*.28=117.6, therefore the error band is 302.4cfs to 537.6 cfs) ? This means that to obtain a result with the same accuracy as the equation for this location, one would need 15 years of station data to analyze. As one would expect, the longer the return period, the more years of equivalent record (and lower standard error) are needed to reproduce the regression results with the same accuracy.
Standard errors of estimate or prediction generally range from 30 to 60%. The Commonwealth of Puerto Rico and 21 States have equations in this range. In 14 States, there is at least one hydrologic region having a standard error less than 30%. However, in 15 States, there is at least one hydrologic region having a standard error greater than 60%.
The factors that may cause large standard errors are generally a large variability of flood records at a site; the unregulated gaging station network being sparse, thereby making regionalizing flood characteristics more difficult; and short flood records. These conditions all exist in the western portion of the U.S. Conversely, smaller standard errors are generally found in the eastern U.S., where longer periods of records, denser sampling networks, and less variability among station events exist.
Previous Bulletin Topics
- Introduction to the NFF Program and USGS regression equations, the applicability of the regression equations, and the advantages and limitations of the regression equations
- Use of USGS regression equations in the NFIP, and criteria for using USGS regression equations in the NFIP
- Revisions to the USGS regression equations since the NFF software was released
- Unusual parameters of USGS regression equations and how to obtain them (Parts 1, 2, and 3)
- Examples in which USGS regression equations are used for NFIP purposes
- How to treat State Line faults (basins lying in more than one state)
- Estimating drainage area and cross sections from USGS topographic maps
Upcoming Bulletin Topics
- Estimation of extreme floods
- Weighting NFF results with observed data
View the archive page for all Flood Hazard Mapping listservs.
Last Modified: Monday, 25-Jun-2007 11:57:20 EDT