Laserfiche WebLink
<br />DECEMBER 1981 <br /> <br />SILVERMAN, ROGERS AND DAHL <br /> <br />1473 <br /> <br />'.. <br /> <br />limitations in estimating raincell area by summing <br />grid subareas with nonzero gage precipitation and <br />2) the inability to determine and account for the <br />precipitation gradient in estimating the subarea pre- <br />cipitation amounts. The effects of both shortcomings <br />tend to be amplified as the gage density decreases. <br />When the precipitation gradient is approximately <br />linear, the fit of the distribution of total precipitation <br />estimates in' a sample set to a normal distribution is <br />quite good. When the precipitation gradient is non- <br />linear, the distribution is bimodal with about equal <br />probability of over- and underestimating the true <br />precipitation amount, depending on whether the par- <br />ticular gage configuration locates a gage(s) in the <br />high-precipitation region of the pattern or not, re- <br />spectively. <br />The sampling variance numerical experiment gen- <br />erated a total of 530 data points-five gage densities <br />for each of 100 isohyetal patterns and two gage den- <br />sities for each of 15 isohyetal patterns. Before re- <br />lating the sampling variance to the isohyetal pattern <br />characteristics, the input parameters were general- <br />ized by removing the scale factors. Thus, the pattern <br />shape E, which implicitly defines the raincell area <br />in the simulations, was combined with the appro- <br />priately scaled gage network densities to form the <br />more general variable called GPR, the number of <br />gages per raincell. Thus, for example, a raincell that <br />has an area of 100 km2 (E = 2) and is sampled by <br />a gage network with a density of 10 km2 per gage <br />(approximately 4 mi2 per gage) will have an average <br />GPR = 10. The range of variation of GPR corre- <br />sponding to the range of variation of E and gage <br />density used in this study is 0.157 to 714.0. In ad- <br />dition, the maximum precipitation point offset from <br />the geometrical center, originally expressed as X off- <br />set and Y offset, was vectorially combined into a <br />single variable which was called F. <br />Using the coefficient of variation six as the de- <br />pendent variable, a linear regression model was fitted <br />by a stepwise routine to the independent variables. <br />Since six -+ 0 as GPR -+ co, regression was forced <br />through the origin and the independent variables <br />used in the regression were GPR, powers and roots <br />of GPR, and combinations of the products of GPR <br />and its power functions with G and F and their power <br />functions. The following regression equation was ob- <br />tained as the best fit over the range of variation of <br />, the input data: <br /> <br />six = 0.71051GPR + 0.5079GIGPR <br /> <br />- 0.1381G/GPR2 + 0.012IG/GPR3 <br /> <br />- 0.0531/GPR2. (2) <br /> <br />The terms in Eq. (2) are presented in the order with <br />which they contribute to the explanation of the vari- <br />ance of six, The multiple correlation coefficient for <br />this equation is 0.992 such that 98.4% of the variance <br /> <br /> <br />of six is accounted for by the regression equation. <br />The first term in the equation alone accounts for 80% <br />of the variance. Thus, it can be seen that the data <br />are very well fitted by Eq. (2). The coefficient of <br />variation is primarily a function of the number of <br />gages per raincell and secondarily, but importantly, <br />a function of the spatial precipitation gradient. It <br />should be noted that the coefficient of variation is <br />independent of the location of the precipitation max- <br />imum in the isohyetal pattern. <br />According to random sampling theory, the stan- <br />dard deviation (and coefficient of variation) should <br />be inversely proportional to the square root of GPR. <br />Independent variables involving this term were, <br />therefore, included in the search for an appropriate <br />regression equation. The term GPR-O.5 was not se- <br />lected early in the stepwise regression although it <br />alone (six = L429GPR-o.5) had a correlation coef- <br />ficient of 0.916. However, the first term selected, <br />(GIGPR)O,5, included this functional relationship [sf <br />i = 1.119(GIGPR)O,5] and had a correlation coef- <br />ficent of 0.978. ,In fact, the regression equation six <br />= 0.8659(G/GPR)O,5 + 0.2549/GPR fit the data <br />quite well and had a multiple correlation coefficient <br />of 0,990. This equation was not selected as the best <br />fit of the data because it had a higher standard error <br />of estimate than Eq. (2) and it systematically over- <br />estimated the values of s / x at very high values of <br />GPR. Attempts to improve the fit by the inclusion <br />of additional terms resulted in an equation which <br />predicted negative values of s / x for some of the high <br />values of GPR. Gabriel (1981) evaluated the relative <br />variability of simple random sampling from single <br />cell storms defined by Eq. (I) and indicated that it <br />provided an upper bound to the corresponding rel- <br />ative variabilities from systematic uniform grid sam- <br />pling given by Eq. (2). His comparison of the relative <br />variabilities from the two network sampling methods <br />revealed that Eq. (2) did, indeed, give smaller values <br />of the network sampling standard deviation for all <br />values of GPR ~ 1 but gave higher values for some <br />values of GPR < L The reason for the relative be- <br />havior of the results from the two sampling methods <br />is probably due to the way the area of the raincell <br />is treate:d. In Gabriel's evaluation of simple random <br />sampling, the area of the raincell is explicitly taken <br />into account in his analytical solution. His equation <br />for the network sampling standard deviation is a <br />function of his variable b, the proportion of the region <br />gaged that is covered by the area of the rain cell. In <br />this study, on the other hand, all information on the <br />characteristics of the raincell are deduced from only <br />the gage catches in accordance with usual observa- <br />tional procedure. As was mentioned earlier in this <br />section, the network sampling standard deviation is <br />caused, in part, by the limitations in estimating the <br />raincell area from the gage network data. The esti- <br />mates of raincell area become increasingly less ace <br />