WMOD00444

• Title: Evaluation of Idaho's 1992-1993 Winter Cloud Seeding Program
• Date: 10/1/1993
• Weather Modification - Doc Type: Report

~

3.3 Applying The Target-Control Approach

Figure 1, which illustrates the target-control approach, is a plot of the April 1 SWE (snow
water equivalent) measured at two SOS (Soil Conservation Service) snow measuring sites:
the Atlanta Summit site east of Boise (target) and the Bear Saddle site northwest of Boise
(control). The pairs of April 1 observations are plotted as a point for each of 30 years as
labeled. The straight line through the middle of the points is the "linear regression"
equation, or the line which is mathematically calculated as the best "fit" to the 30 data
points.

Figure 1 shows considerable scatter of points representing individual years. In general, when
the SWE is high (low) at the control site, it is high (low) at the target site. The linear
regression line provides the best estimate of target SWE for any given control SWE. But the
figure shows that measurements in individual years often depart from the line by a few to
several inches.

Another mathematical calculation provides the "correlation coefficient," represented by "R"
or "R-value", which is a measure of the scatter of the points. The higher the value of R, the
less the scatter, and the better the predictability of natural target SWE. If all the points fit
exactly on a straight line, R would equal 1.0, indicating a perfect correlation. If the points
had no correlation, forming a pattern similar to shotgun pellet holes in a paper target, the
value of R would equal 0.0. In the case of figure 1, R = 0.84, a low value for attempting a
target-control analysis with 30 data points. Because the nonseeded 1965 winter (the point
labeled "65") is plotted over 20 inches above the regression line, seeding would have to result
in a similar departure to demonstrate effectiveness. But such an indication would require
over a 60-percent increase in snowfall, several times greater than the approximately 10- to
15-percent increases suggested by some well-designed and well-operated projects. Therefore,
no reasonable chance exists to detect a real seeding effect with the target-control relationship
shown on figure 1. But figure 1 was chosen to illustrate the points made, not because it
indicated a high target-control relationship. Fortunately, much better relationships exist, as
will be shown.

Averaging observations from a number of snow measurement sites often results in higher R-
values and more stable results. The appropriate approach for this, type of target-control
analysis is to select as many observing sites for averaging as possible so long as the selection
process is physically justified. At the same time, a high R-value should be sought, although
attempting to increase it from, say, 0.955 to 0.960 has little justification because of natural
variability. Calculating the highest possible R-value by selectively choosing particular sets
of target and control sites offers no guarantee that a similarly high value will result from a
different time period. Most likely, a lower R-value would be calculated from another period
of observations. A more stable result would be expected from averages of more measurement
sites, even if the R-value is slightly reduced.

4