Laserfiche WebLink
<br />12 <br /> <br />ADDITIONAL CONSIDERATIONS IN FLOW-FREQUENCY ANALYSIS <br /> <br />There are important considerations in flow-frequency analysis <br />beyond the type of distribution that may be used to fit the data. Some <br />of these considerations are discussed below. <br /> <br />A short record of flood flows may contain large sampling errors <br />because of chance geographical or temporal variations in rainfall <br />during the period of record. Therefore a short record may be a <br />poor indicator of the basic long-time distribution of floods at the <br />site. <br /> <br />1 <br /> <br />Flood flow frequencies often need to be determined for sites where <br />there are no streamflow data. It is possible to examine the individual <br />flood records within the region as a unit and to develop generalized <br />relationships that apply anywhere in the region including ungaged sites. <br />This approach often overcomes many of the uncertainties due to sam- <br />pling error at individual sites. Several methods of regionalization <br />have been developed to date; a study of such methods, and recommen- <br />dations for their use, should be part of the continuing program of the <br />Committee. <br /> <br />Another problem is the treatment of a record that contains one or <br />more events of rare frequency--the so-called outlier problem. By <br />using historical information at the site or at nearby sites it is often <br />possible to assign a realistic recurrence interval to each outlier. <br />This information is incorporated into the set of data to define the <br />overall distribution. An alternative is to compute the frequency <br />distribution omitting the rare events, plot the frequency line, and <br />then to adjust the line to conform to the historical information. <br />Where no historical information is available, an obviously very rare <br />event may be excluded from the computations. The specific treat- <br />ment that is used to handle outliers should become a matter of record. <br /> <br />Where streamflow data are lacking at the site or where a regional <br />analysis is not justified, the use of hydrologic methods, such as <br />rainfall-runoff relationships and unit hydrograph studies, may be <br />the only feasible approach. <br /> <br />In the flood series for some streams in arid regions, it is not <br />unusual for one or more of the flood values to be zero. This poses <br />a difficulty when using a logarithmic transformation because the <br />logarithm of zero is minus infinity. One way around the difficulty <br />is to add a small constant to all the items of data. A second is to <br />determine the frequency relation from only non-zero items and <br />afterwards to adjust the relation to the full period of record. This <br />