Statistical evaluations of a cloud-seeding program at the streamflow control level

Abstract

During the winters of 1962–1963 and 1963–1964 cloud-seeding operations were undertaken on the Skagit River (Washington) watershed. Conditions required that the statistical evaluation be made at the streamflow control level. A combination of least squares regression with principal components analysis is developed. This requires preliminary screening of variables using a liberal (0.10 level of significance) critical region, followed by the orthogonalization procedure of principal components. Final stepwise regression analysis using the principal component scores and the 0.05 level of significance follows. Tests are consistent with treating the residuals as Gaussian, homoscedastic, and non-autocorrelated. The analysis is replicated in various ways. Without exception, the excess of actual over predicted flows for the full period of the experiment was significant at 0.01 “or higher” level. However, whereas the excess in 1964 is invariably significant at the 0.005 “or higher” level, the 1963 excess would not be judged significant by itself at the 0.05 level. This is explained by a five-fold increase in seeding between 1963 and 1964. Two years subsequent to the period of seeding are analyzed: observed flows fail to show significant deviations from expected flows. Estimates indicate at least 15 percent runoff increase from seeding.