Large-scale planning for improved flood-plain management and expanding water-resources development has made it increasingly important that a consistent approach be adopted for estimating flood frequencies, a major analytical component in studies required in flood-plain management and, in a larger sense, in river-basin management. A Federal interagency group has studied the most commonly used methods of flood-frequency analysis and has compared the results of applying these methods to a selected group of long-record representative sites in different parts of the country. Based on these comparisons and on other considerations, it is recommended that all government agencies adopt a uniform procedure for flood-frequency analysis at sites where records are available. The log-Pearson Type III distribution has been selected as the base method, with provisions for departures from the base method where justified. Continuing study leading toward improvement or revision of methods is recommended. (Key words: Floods; rivers; statistics) •A•V• A. S•SO• These circumstances were recognized by a Bureau of Standards and Geoffrey S. Watson Task Force on Federal Flood Control Policy of The Johns Hopkins University. which, in August 1966, transmitted a report to the President entitled 'A Uniform National INVESTIGATIONS Program for Managing Flood Losses.' This re-The Work Group decided that several methport was subsequently submitted to the Con-ods of flood-frequency analysis in common use gress [House Doc. •,65, 1'966]. In the report among Federal agencies and elsewhere would the following statements are included relating be applied to a group of 10 long-term records to flood-frequency methods: of annual flood peaks at selected locations in Techniques for determining and reporting the frequency of floods used by the several Federal agencies are not now in consistent form. This results in misunderstanding and confusion of interpretation by State and local authorities who use the published information. Inasmuch as wider, discerning use of flood information is essential to mitigation of flood losses, the techniques for reporting flood frequencies should be resolved.

The concept of dominant (stream) discharge in geomorphology has not been firmly defined. By defining it, on rational grounds, as that discharge at which, over a long period of time, most sediment has been transported, a definite value can be computed for a stream for which there are available durationcurve values of daily discharge and a sediment-rating curve. This has been done for 5 eastern and 4 western rivers. The dominant discharge, when expressed in terms of percentage duration, is remarkably consistent, and is found to be a lower discharge than has generally been considered as dominant.

The generation of a long series of hydrologic events by use of the statistical parameters based on a short sample has come to be known as synthetic hydrology. In present practice, there are two major deficiencies in the method. There are large errors due to the sampling errors of the original sample, and there is no way to generate a series for an ungaged location. The authors propose a method to correct for these deficiencies.

A flood-frequency study for the Susquehanna River at Harrisburg. Pa., involved the use of both historical and recent flood data. The historical data started in 1786 and covered all flood peaks above a stage of 18.0 ft or approximately 340,000 cfs. The continuous record began in 1874. The use of hitherto available techniques for combining the historical with recent data was found to give inconsistent results. Analysis showed that in plotting points for the purpose of drawing a frequency curve, consistent results could be obtained only if all points were based on the same period of record. It was known that all the missing years of record since 1786 had annual peak floods of less than 340,000 cfs. It was assumed that the same distribution of peak floods below 340,000 cfs occurred during these years as is known to have occurred during those years in the recent period. The method of computation involved an ad justing of order numbers for all items below 340,000 cfs. The result is a consistent set of points and a well-defined frequency curve.A study of flood frequencies for the Susquehanna River at Harrisburg, Pa., has led to a recon sideration of the method for computing plotting positions when combining historical floods with a recent period of record. The study was made on the basis of annual peak discharges. Probability charts based on the theory of maximum values [GUMBEL, 1945] were used for plotting purposes, although any other standard plotting procedure would have given like results.

In fitting a theoretical frequency distribution to a set of data, a problem arises if the series contains a number of zero values, as may occur in annual flood peak data for small, arid-region streams. The problem is twofold: first, commonly used distributions do not fit such a set of data; second, if a logarithmic transformation of the data is being used, logarithms of zero flows are not usable in a computation. To overcome the difficulties, a theorem of conditional probability is used. The probability of occurrence of a nonzero peak is combined with the conditional probability of exceeding a given flood magnitude, given that a nonzero peak has occurred. The method has been found useful also for fitting flood series in which information of peak annual floods below a specific stage is lacking.