FLOOD00005 - Laserfiche WebLink

Home Browse Search

The log-Pearson Type III distribution is not recommended for analysis of precipitation data by the National Weather Service. The National Weather Service position is that skewness is too susceptible to influence of sampling error to be a reliable parameter in describing storm data (Frank Richards, National Weather Service, oral commun., 1987); however, the like- lihood of sampling error having a significant effect on the calculated value of skewness decreases with the length of the record being considered. There is no theoretical reason (other than the previously described sensitivity of the skewness to sampling error) for not using the log-Pearson Type III distribution. In fact, the State of California Department of Water Resources uses the log-Pearson Type III distribution for analysis of precip- itation data, rather than the Gumbel distribution, in order to incorporate the effects of the skewness of the data into their analyses. Results Ranked precipitation data for 5-, 10-, 15-, and 3D-minute, and 1-, 2-, and 24-hour durations are plotted against recurrence interval in the top half of figures 2a-2g. Data points representing the three or four largest storms in the historical record are identified by year on the figures. The updated precipitation frequency curve and associated 95 percent confidence interval are superimposed over the data points so outliers may be identi- fied. The horizontal axis on these figures is distorted such that data that is described by a Gumbel distribution will plot as a straight line. The same data, logarithmically transformed, is plotted in figures 3a-3g. Two computer-fitted lines with associated 95 percent confidence intervals--the previously defined and updated frequency curves--are presented in the bottom halves of figures 2a-2g and 3a-3g. The curves show how the estimate of the magnitude of a storm for a given recurrence interval would be modified by inclusion of the August I, 1985, storm. Two different straight lines can be fitted through the annual maximum precipitation data points plotted on figures 2d-2f and 3e-3f. One line fits the upper few data points, representing the more extreme events in the record; the other line describes the remaining data points. The calculated lines shown on each figure lie between these two lines. Two types of storms occur in Wyoming, thunderstorms and general storms; however, no distinction was made between the two types in creating figures 2 and 3. The data were re-evaluated by identifying those storms that occurred during the maj or thunderstorm months of July and August of each year. The two possible fitted lines are quite distinct on figure 3e; therefore, these data were chosen as a test ease. The data were divided into three groups: (1) Storms occurring during May and June; (2) storms occurring during July and August; and (3) storms occurring during September. (None of the I-hour annual maximum precipitation occur,red during the period October through April.) Each of the data groups was identified by a different symbol and was plotted again on a copy of figure 3e. Except for the observation that the two largest storms were thunderstorms that occurred during, the months of July and August, no consistent relationship between occurrence and month could be determined. Records were researched to determine whether the years during which the largest storms occurred coincided with possible changes in either the location, or construction, or both, of the precipitation gage. No pat- tern with time could be established. 14