Reexamination of Historical Regression Analysis Applied to a Recent Idaho Cloud Seeding Project

Authors

  • Arlin B Super St. Cloud, MN
  • James A Heimbach Springvale, ME

Abstract

A recently reported analysis of an Idaho operational winter cloud seeding project is examined in detail. The analysis used the traditional historical regression method. It appeared to provide impressive evidence that seeding was effective in increasing seasonal snowpack accumulation during each of four winters with a mean increase near 12%. The analysis was based on a strong relationship with the April 1st control station mean explaining 96% of the target area variance. However, frequent snow melt prior to April 1st was discovered at 4 of 7 control stations and 3 of 10 target stations. Snow melt is shown to have introduced an important but apparently unrecognized variable into the target-control relationship, making it inappropriate for evaluation of seeding effectiveness. Repeating the analysis procedures with March 1st observations from the same stations reduced the "seeding signal" to under 4%. Additional target and control stations were identified and used in comprehensive historical regression analyses. Once a limited number of stations with February melt were discarded, all possible combinations of available control stations were tested to detect target-control relationships which explained the most variance. These relationships did not support the hypothesis that cloud seeding significantly enhanced the seasonal snow water content. Additional testing was done with April 1st observations to demonstrate that a wide range of results can emerge from a large database even with the requirement of a strong target-control association. Recommendations are made for future application of this statistical approach.

Downloads

Issue

Section

Scientific Papers