|
<br />325
<br />
<br />JOURNAL OF APPLIED METEOROLOGY
<br />
<br />VOLUME 12
<br />
<br />TABLE 1. Names and characteristics of the runoff stations in the target area and number of years needed for evaluation,
<br />as calculated from (1),
<br />
<br /> Number of years
<br /> needed for significance
<br /> Seasonal at 95% level (two-
<br /> Mean runoff tailed) and 50% power,
<br /> Drainage seasonal standard Coefficient assuming a 10%
<br />CSU area Elevation runoff deviation of variation runoff increase
<br />number Station name* (mi') (ft) (acre-ft) (acre-ft) (%) (and no control)
<br />1073080 La Plata River at 37 8105 26,810 12,750 48 87
<br /> Hesperus
<br />1073408 Animas River near 1090 5960 525,100 221,700 42 69
<br /> Cedarhill
<br />1073436 Animas River at 692 6502 454,200 173,500 38 56
<br /> Durango
<br />1073448 Hermosa Creek near 172 6705 75,350 44,270 59 133
<br /> Hermosa
<br />1073480 Animas River at 55.9 9617 64,460 17,920 28 30
<br /> Howardsville Park
<br />1075830 Los Pinos River near 284 7515 210,500 84,070 40 62
<br /> Bayfield
<br />1076420 Piedra River near 371 6530 191,500 96,920 51 99
<br /> Piedra
<br />1077090 Navajo River at Banded Peak 69,8 7941 61,410 23,160 38 55
<br /> Ranch near Chromo
<br />1077250 Rio Blanco near 58 7950 51,190 21,170 41 66
<br /> Pagosa Springs
<br />1077400 San Juan River at 298 7052 211,000 116,900 55 118
<br /> Pagosa Springs
<br />1078000 East Fork San Juan River 86.9 7598 74,730 32,770 44 74
<br /> near Pagosa Springs
<br />1272445 San Miguel River near 308 7096 132,400 58,210 44 75
<br /> Placerville
<br />1277200 Dolores River at 556 6925 265,100 114,100 43 72
<br /> Dolores
<br />1371520 West Fork Dallas Creek near 437 8400 145,600 55,620 38 56
<br /> Ridgeway
<br />1375400 Lake Fork at 388 7828 147,800 49,390 33 43
<br /> Gateview
<br />
<br />* U, S. Geological Survey stations in Colorado.
<br />
<br />in zones 1 and 2, using basins in zones 3 and 4 as
<br />controls. The results are shown in Table 2. The improve-
<br />ment over the results of Table 1 is striking but not
<br />sufficient. Additional results can be obtained by
<br />combining data of Tables 1 and 3. For example, for
<br />station 1077250 in zone 2 without control, the number
<br />of years is 66. The highest correlation with any other
<br />station in the target area is with station 1077090 and
<br />p=0.95 (see Table 3). With that control the value of iV
<br />is 66[1- (0.95)2J = 6.6, which exceeds 5; moreover,
<br />that control station (Navajo River at Banded Peak
<br />Ranch) is a neighbor in zone 2 (as could be expected
<br />with such a high correlation) and therefore not accept-
<br />able as a control. What then can be done to reduce the
<br />number of years needed to obtain significance?
<br />
<br />3. The concept of grouping of observations
<br />
<br />Because of the many interacting complex parameters
<br />and variables involved, local weather variability, etc.,
<br />the runoff of a watershed can be viewed as a random
<br />variable. This point of view is rather old in hydrology
<br />but nevertheless still gaining ground (Yevjevich, 1972).
<br />Fundamentally, a hydrologic change in the course of
<br />time is detected when the recent observations can no
<br />
<br />longer be viewed as coming from the statistical distribu-
<br />tion that prevailed over the long past. Let us deno1:e
<br />the (seasonal) runoff-random variable from watershed
<br />i by Qi. An observation (realization) of this random
<br />variable (LV.) in year j is denoted qij. There are many
<br />such r.v. in the target as well as in the control area.
<br />Up to this point all techniques involved either one
<br />target r.v. (two-sample test) or a pair of LV. (two-
<br />sample target-control test). Of course, for the purpose
<br />of rapid detection, if the percentage increase in runoff
<br />was the same in all watersheds one would pick up the
<br />"best" pair of target control watersheds in Table 2,
<br />namely the pair with lowest value of iV in the last
<br />column of Table 2 (the pair with iV =5.9). The criterion
<br />for the selection was the minimization of iV. Looking
<br />back at (2) one realizes that there are at least two ways
<br />of minimizing iV, namely minimize C, i.e., select the
<br />watershed with minimum natural variability in runoff,
<br />or maximize p, i.e., select the pair of target and control
<br />watersheds which are most highly correlated. It is well
<br />known in hydrometeorology that the natural variability
<br />tends to decrease when the hydrologic variables are
<br />averaged over a large region. Thus, another possibility
<br />for decreasing N is to consider not pairs of elemental
<br />
<br />I
<br />~
<br />
|