Laserfiche WebLink
<br />is selected yields a value of six from Eq. (2) which is then inserted <br />into Eq. (3) to yield the required Nsv. A one-sided test with a = 0.05 (or <br />two-sided test with a = 0.10) and B = 0.10 were assumed. Table 3 presents <br />the sample size, Nsv, as a function of GPR and G, required to detect a <br />25-percent change in precipitation means. <br /> <br /> Table 3. - Sample size, Nsv, due to sampling variance alone <br /> requi red to dete!ct a 25-percent change in mean <br /> precipitation amount due to seeding <br /> GPR <br /> G (A50) 100 50 10 5 2 1 0.5 <br /> 2.25 (10) 1 1 18 70 389 1 , 270 3,368 <br /> 1.40 (15 ) 1 1 10 39 226 769 2,172 <br /> .90 (20) 1 1 7 28 160 550 1,617 <br /> .60 (25) 1 1 6 23 126 438 1 , 3 27 <br />It cain be seen from Table 3 that the sampl e size increases as the gage <br /> <br />density (GPR) decreases and the precipitation gradient increases (increasing <br />G or decreasing A50). For reasonable gage densities and nonuniform precipi- <br />tation gradients, the sample size due to sampling variance alone is appreciable. <br /> <br />Convective rain showers of short duration like those reported for Florida, <br />Arizona, and Montana (see Table 3) appear to be most problemmatical because <br /> <br />they are relatively small in arE!a, implying a high gage density for a given <br />value of GPR, and they are charalcterized by high spatial precipitation <br />gradients. The convective rains in the Midwest appear to be associated with <br />large~r scale systems that have more uniform precipitation gradients. This <br />type of weather system poses less of a problem due to sampling variance. <br /> <br />We shall now place the sample sizes due to sampling variance in perspective <br />by compari ng them to the sampl e si zes requi red as a result of natural <br />storm variability. To do this \I,e consider the statistical characteristics <br /> <br />19 <br />