Laserfiche WebLink
<br />30 <br /> 18 <br /> 16 <br /> i4 <br /> "- <br /> 12 <br /> 10 <br />Y <br /> 8 <br /> t <br /> 6 <br /> 4 <br /> <br />, <br /> <br />TECHNIQUES OF WATER-RESOURCES INVESTIGATIONS <br /> <br /> <br />- ~ :!- <br /> <br />U' <br /> <br />" <br />1 j-+-'- <br /> <br />, <br />-t+=" <br /> <br />-':fu , r r;r -~ <br />r:, ~ -'- <br /> <br />,.... ,t-r- <br /> <br />,-,-~ -r <br />::.:-:.:+ <br /> <br />--.--g:- <br /> <br />rrll;;:: <br />! '.r- <br /> <br />'1,":' <br /> <br />'O~ ... c <br />2 <br /> 'r. <br /> , , , <br />0 <br />0 2 4 6 8 <br /> X, <br /> <br />+-.- -~ D6 <br /> <br />10 <br /> <br />8 <br /> <br />. <br /> <br />4 <br /> <br /> <br />6 <br /> <br />f- <br />::, 2 <br />~ <br />f- <br /><{ <br />~ 0 <br />0: <br />::> <br />U <br /> <br />~ -2 _..:;~ ~_ <br />~ <br />~ <br />~ -4 <br />Cl <br />U5 <br />W <br />0: <br /> <br />-6 <br /> <br />-8 <br /> <br />-10 <br /> <br />12 <br /> <br />14 <br /> <br />o <br /> <br />2 <br /> <br />4 <br />X, <br /> <br />6 <br /> <br />8 <br /> <br />. <br /> <br />Figure 21.-Multiple linear regression bv the method of residue Is. <br /> <br />4. Lines of equal S are logarithmically spaced <br />at twice the logarithmic scale of the paper <br />and are parallel. <br />To solve, separate the regression into two <br />parts by introducing an intermediate variable <br />Q"j (to an arbitrary scale) so that <br /> <br />Q'dj j(A, P) <br /> <br />and <br /> <br />Q,dj~j(S, QIO)' <br /> <br />Consider the first relation. For a fixed P, the <br />model would be <br /> <br />QIJ(lj=KAn, <br /> <br />in which K is the intercept on the Q'dj scale <br />(at A=I) and n is the slope of the line. In this <br />example, n= 1.28, which is the ratio of the <br />linear vertical to horizontal lengths. When <br />P= 1.20, the intercept K is 78. To obtain this <br />intercept graphically requires a long curve <br />extension. It is simpler to compute the intercept <br />from some other value of A than one. For <br /> <br />instance, for Qodj=I,OOO, A=7.3. Then <br /> <br />1,000=K(7.3)1.28, from which K=78. <br /> <br />Similarly, for Qodj= 10,000, A=44.5 and K=78. <br />Introducing P as a variable makes it neces- <br />sary to define the intercept K in terms of P <br />(because the intercept is d:ifl'erent for each <br />value of P). The interval per tenth difference <br />in P when projected on the Q'dj scale is 1.36, <br />that is, for each tenth increase in P, the inter- <br />cept increases 1.36 times (the intercept is on a <br />logarithmic scale). This increase can be meas- <br />ured for individual intervals or computed from <br />the total increase: For instance, for A= 10: <br /> <br />Qodj=I,480 for P=1.2, <br /> <br />and <br /> <br />Qodj=17,000 for P=2.0. <br /> <br />The increase for eight intervals IS 17,000; <br />1,480=11.5, and (1.356)'=11.5. <br />Then <br /> <br />. <br /> <br />K=78(1.36)1O(P-l.2) , <br />