Abstract
After four summer periods of randomized experimentation with dynamic cumulus seeding in a 1.3 × 104km2 target area in south Florida, 14 seed and 23 control cases are available, with increased documentation of radar measurement accuracy.Seed-control rainfall comparisons are made for “floating” and total target for the 6 h period following the first seeding. On days screened as suitable for the experiments, natural rain volume varied by a factor of 62 for floating target and by a factor of 25 for total target. Area seed-control rainfall differences are not significant with six classical tests, nor is the difference between random and non-random controls.Analysis of isolated experimental clouds obtained on days of multiple cloud seeding produced significant findings. Results were stratified depending on whether the single clouds dissipated in the target area without merger or whether they merged with a neighbor. With the former stratification, the mean seeded rainfall exceeded the mean control rainfall by a factor of 2, a result (one-tailed significance of 3%) that is consistent with earlier single cloud studies. No meaningful rainfall comparison was possible with the latter stratification because, on the average, the seeded clouds merged (and were dropped) 13 min earlier than the controls. This disparity in mean lifetimes before merger (two-tailed significance level of 0.5%) suggests that seeding is promoting merger in FACE as intended.Several Bayesian approaches are used to estimate a probability distribution of a multiplicative seeding factor, based on gamma rainfall distributions, with the same shape parameter for seeded and control populations. The most general treatment assigns prior probabilities to three variables, the common shape parameters, the mean of the control distribution, and the multiplicative seeding factor. With existing data, 95% of the area under the marginal density of the seeding factor lies between about 0.7 and 1.7, with a mean just above and a mode just below 1.After extensive search for physically meaningful covariates or predictors, radar echo motions in or near the target related to two distinct rainfall populations. Category 1 comprised those cases where echoes were “marching” across the area. Category 2 comprised those cases with growth and dissipation virtually without motion. Echo motion is shown to be a statistically significant covariate, accounting for 30% of the variation in the total rainfall. For the afternoon measurement period, the mean target rainfall in Category 2 cases exceeded that in Category 1 cases by a factor of 2.5.Separate seed-control comparisons in the two categories indicate that different effects of seeding might be sought in continued experimentation. Although the existing sample is small, there is evidence that in Category 1 (marching) the seeding effect is probably not multiplicative.Attempts are in progress to estimate the number of further cases required to resolve a range of postulated seeding effects in this experimental context.
