WMOD00444

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

~.

A high correlation coefficient (R-value) of 0.971 resulted from the data of figure 3. Therefore,
a reasonably high degree of predictability exists for natural (nonseeded) target snowpack
SWE on April 1 using the average of the control observations and the regression equation.

The upper and lower lines parallel to the linear regression equation represent :t 2 standard
errors of estimate. For a nomal distribution of data points, 95 percent of all points would
be expected to lie within the two outer lines. Therefore, a point above (or below) the line
would have approximately a 5 percent probability of being from the same population as the
points between the lines. In other words, such an outlying point would be suspected of being
significantly "different" than the other points. Such a difference could raise the suspicion
that it was caused by seeding. In fact, the only year outside the standard error of estimate
lines is 1965, a nonseeded year. Having 1 point in 301 outside the lines would be about what
might be expected as 1/30 = 0.033 or 3.3 percent, close to the 5 percent typical for a normal
distribution.

It should be understood that the 1993 observations do not enter into any of the calculations
of the regression equation, standard errors of estimate, or correlation coefficient on figure 3
or similar figures to follow. Such calculations are based only on the 30 nonseeded winters
1963-1993.

The 1993 point for the average target and control data is plotted as a star labeled "93." The
1993 departure lies 2.9 inches of SWE above the regression line fitted to the 30 nonseeded
winters. This departure is equivalent to 9.6 percent more snowfall than predicted. But the
1993 observation lies about one standard errors of estimate above the regression line; that
is, about midway between the regression line and the upper two standard errors line. For
a normal distribution, {iB percent of all points can be expected to fall within :t1 standard error
of estimate. Consequently, although the 1993 departure deviated in the desired direction,
it has no statistical significance.

Figure 3 illustrates one of the problems of attempting to evaluate a single winter's seeding.
Even though the 1993 winter's departure is near 10 percent, a figure often quoted as about
what successful winter orographic cloud seeding might produce, the departure is well within
the natural winter-to-winter variation. Moreover, the R-value calculated for the particular
data set is quite respectable. Correlation coefficients based on April 1 SWE observations are
seldom higher than 0.9'7 unless the measurement sites are very near one another. Therefore,
one cannot expect much better predictability than provided by the data of figure 3. A number
of winters with consistent departures well above the regression line (SWE increases) would
be needed before the evidence would strongly suggest a seeding effect.

Had only the year 1965 been seeded, one would might have concluded from figure 3 that
seeding had been very effective. Conversely, had only 1985 been seeded, one might have
concluded that seeding decreased the snowpack. Of course, both conclusions would have been

13