Abstract
Rainfall data collected at 67 gages in a 2750 mi2 target area during a four-year randomized cloud seeding experiment in North Dakota have been stratified in a variety of ways and subjected to several kinds of statistical tests. Some stratifications related to cloud model predictions were possible for only the last two years when a rawinsonde station was operated as part of the project. Monte Carlo experiments simulating 500 reruns of the four-year experiment have been used to establish significance levels for the tests within each data stratification.
The analysis provides significant evidence that seeding convective clouds on a determinate set of days leads to 1) an increase in the frequency of rainfall events at the individual target gages, 2) an increase in the average rainfall recorded per rainfall event, and 3) an increase in total rainfall on the target. The set of days to which this evidence applies is those days with dynamic seedability; that is, days for which a cloud model predicted an increase in cloud top height under the influence of silver iodide seeding. Rainfall observations on days when the cloud model predicted no increase in cloud height show no significant differences between seed and no-seed days.
The possibility of bias has been checked by comparing the frequencies of wet and dry days and the averages of several meteorological variables for seed and no-seed days within each stratification, by cross-checking the stratifications, and by comparing rainfall on seed and no-seed days over an area of roughly 50,000 square miles surrounding the target area. There is no obvious bias to account for the significant differences between seed and no-seed days with dynamic seedability.
It is tentatively concluded that dynamic effects, including rainfall increases, were produced by light to moderate silver iodide seeding from below cloud base. The potential rainfall increase resulting from seeding below selected clouds on days with dynamic seedability is estimated at one inch per growing season.