Laserfiche WebLink
<br />The accuracy of an estimate made using a statistically derived equation <br />is a function of the accuracy described above and also random variations <br />produced by different users of that equation. Each user or planner will make <br />certain decisions based on his or her best judgment about the actual outline <br />of the drainage basin, what constitutes storage, tracing the path of the main <br />channel, and interpolating values from contour lines. These decisions intro- <br />duce additional random variations into the model that are not accounted for by <br />the statistical techniques used. The variations introduced in this way differ <br />from user to user and basin to basin. For example, the drainage-basin <br />outline generally is much easier to define in hilly terrain than in flat <br />terrain because the more closely spaced contours reduce ambiguity in the <br />location of the basin divide. Therefore, the random errors introduced by the <br />hydrologist generally will be smaller where the topography is hilly because <br />the judgments regarding drawing of the drainage outline will be easier and <br />will be more consistent between users than in areas of flat topography. <br /> <br />In general, estimates of the probability of future flood occurrences <br />become more accurate with greater length of record (Hardison, 1969). The <br />standard error of the estimate of a predicted flood magnitude for a given <br />recurrence interval decreases approximately proportionally to the square root <br />of the length of gaging station record. At or near gage sites, flood charac- <br />teristics may be based on analysis of actual records collected at the gage <br />(from tables 2 to 5) or may be computed from regional estimating relations. <br />Weighted averages of flood estimates by regression equations and by transfer <br />relations generally are used when the percent change in any basin character- <br />istic exceeds 10 percent or when the period of gaging is less than 20 years. <br /> <br />The standard error of the estimate is a measure of the dist::ibution of <br />the observed data about the regression surface. The standard error, reported <br />in percent of estimated flow, is the range of deviations from the regression <br />surface to be expected approximately two-thirds of the time. Because the <br />variables used in these analyses are expressed in logarithmic form, the <br />percent standard errors are larger in the positive direction. The values of <br />percent standard error reported in table 1 are the average values. The <br />equivalent number of years of record is an estimate of the information <br />obtained from the regression equation when applied to an ungaged site. In <br />other words, an estimate of a flood based on a regression equation is <br />approximately as good as that obtained from a gaged site operated for that <br />period of time. Hardison (1971) presents an equation that defines the <br />equivalent years of record represented by a regression equation. The <br />equivalent number of years of record is computed from the ratio of the mean <br />variance of the logs of the annual peaks to the mean square error of the <br />regression, multiplied by a factor dependent on the return period and mean <br />coefficient of skewness. <br /> <br />17 <br />