WMOD00444

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

An attempt was made to use the differences between April 1 and February 1 SWE
measurements because the entire seeding generator network in the upper Snake was not
operational until about February 1, 1993. Only four ofthe target sites and five ofthe control
sites shown on figure 8 had the necessary observations. The resulting R-value was a low
0.798. The results indicated that the 1993 data point was 11 percent below the regression
line, but the departure had much less than 1 standard error of estimate. Consequently, this
attempt produced nothing significant.

One additional analysis was done with this data set to illustrate a point. A search was made
for a combination of target and control sites that would make the 1993 seeded winter "look
good," regardless of how high or low the degree of association might be or how physically
reasonable it was to exclude particular sites. This analysis is not valid, but shows that
indiscriminate searching for "good" statistical results sometimes can produce apparently
impressive outcomes. Figure 11 shows the highest positive departure obtained for the 1993
data point from many (but not all possible) attempts ,'lith various combinations of target and
control sites. The 1993 value lies beyond 2 standard errors of estimate; only 1989 has a
greater positive departure. One might suggest that the 1993 point is significant at about the
2-percent level, although that suggestion would also be misleading because no randomization
was involved.

The target sites used to produce figure 11 are 28, 56, and 60, clustered at the extreme east
end of the target area. The control sites are 40, 63, 66, and 67, clustered on the northwest
end of the control zone. The target and control sites are about as distant as possible from
the selection in table 2. This particular combination of target and control sites probably
would not be chosen on any physical basis.

Figure 11 demonstrates that misused statistics can produce misleading results from cloud
seeding programs. It would be prudent for cloud seeding sponsors to request the details of
any evaluation scheme before a seeding program is initiated. These details should include
the specific analysis approaches and the specific data to be used in later statistical analysis.

In summary, the analysis presented provides no evidence that the 1993 cloud seeding in the
northern upper Snake target had any significant effect on snowfall. The best target-control
relationship that was found, shown as figure 10, suggested no more than a 5-percent
increase. However, many nonseeded years had equal and greater departures from the
regression equation, and the suggested increase has no statistical significance.

25