Laserfiche WebLink
Reclamation (USBR), were contacted. Mr. Doeskin indicated there are no standards used by the <br />National Weather Service (NWS) for classifying hydrologic conditions. He mentioned that the <br />Natural Resources Conservation Service, however, uses five classifications in their streamflow <br />forecasts, based upon percent of average. It is Mr. Doeskin's opinion that this is a poor method <br />because it only takes into account the average, and not the variance, of the flows. Failing to take <br />the variance into account causes a basin with very little variance to always be indicated as near <br />average, while a basin with very high variance is always indicated as extremely wet, or <br />extremely dry. Mr. Doeskin said that the State Climatologist's Office is finishing a project that <br />has developed a drought index which uses a ratio of the variance of the historic data as an <br />indicator of drought severity. For example, an index of -1 indicates avalue -1 standard deviation <br />from average conditions. <br />Randy Peterson indicated that the USBR has addressed the characterization of flows in two <br />ways. One uses a measure of the quintiles of the distribution of the flows, with 25% and 75% as <br />cutoff limits. These limits represent the 25% and 75% non-exceedence probabilities. For <br />example, on average, 1 out of every 4 years would be classified as wet, 1 would be classified as <br />dry, and 2 would be classified as average. The second method is applied for the U.S. Fish and <br />Wildlife Service in which their operation requires the use of specific values for average, wet, and <br />dry conditions. These values are taken as the 10%, 50%, and 90% quintiles, and represent <br />"average" average, wet, and dry conditions. Utilizing cutoff values represented by the 25% and <br />75% quintiles would be consistent with the USBR. <br />Terms and Defiizit~ons <br />Before quintile analysis discussion, a brief review of basic statistical terms and principles may <br />be beneficial. Specifically, populations, population samples, and empirical and fitted frequency <br />distributions will be touched upon. First, a population refers to an entire set of data, regardless <br />of whether or not it has been observed or recorded. A population sample is a finite sampling of <br />data from the population. Thus, a gaging station record is a sample of the entire population. A <br />sample may, or may not be a good representation of the actual population, depending on the <br />particular time period, or window, the sample covers, and the sample size (i.e. length of period of <br />record). The sample will generally be a good representation of the population if it contains a <br />well-distributed sampling of frequent, non-frequent, and average events. When this is the case, a <br />frequency analysis and quintile estimation can be performed directly on the data set with <br />reasonably good results. The resulting frequency distribution is called the empirical frequency <br />dZSCl'ZI71[t1011, since it is based upon actual data. Often times, however, data is scattered or <br />incomplete and may not provide a good representation of the population. The statistical <br />approach to handling this is to fit a known distribution to the data. This enables one to model the <br />distribution of the population with a curve (expressed as a mathematical equation), and calculate <br />specific quintile values. The fitted distribution has the effect of smoothing over gaps, or <br />unevenly distributed areas, that may appear as a result of the data set being a sample of the <br />population. <br />Appendix E E-69 <br />