WMOD00478 - Laserfiche WebLink

Home Browse Search

i )060 manner. At' the start of each field season, Professors William Kruskal and K. Alexander Brownlee prepared a set of sealed and . dated opaque envelopes containing randomized instructions to seed or not to seed-one envelope for each day of the season. A record of the instructions was retained for later verification. In the field on an experimental day the envelope was opened only after we had completed all actions and decisions re- lating to designation of an experimental day, forecasting mean winds, selection of the position of the seeding line, designation of seeding altitude, determination that seeder planes and radar were operational and that weather conditions would permit operation of the seeder planes. Envelopes for days declared nonexperimental (nonoperational) were returned to the University unopened. . Local winds were measured every two hours at the West Plains Airport and were used to define a wind-borne target area (plume) where measured winds could have transported the seeding material. The rem~inder of the exp~rimental area was designated as the control area (nbnplume) and was considered to be that area within the experimental circle where winds and turb~ence were not likely to have carried silver iodide. Target and con- trol areas were constructed in an identical fashion on all experimental days. In the design of the experiment the non plume control area was conceived as a within-day covariate to help allow for meteorological and topographical differences not balanced out by randomization. Differences between target and control precipitation (rainfall and radar echo cover) are regarded as the primary dimensions for testing seeding effects. 'Recognizing the data-reduction phase as one which had been severely criticized on earlier projects, we took a number of steps to isolate it from all knowledge about seeding. In the case of the surface rainfall data, rain gage charts were mailed directly to Professor Wayne Decker, University of Missouri, who was supported under a separate NSF grant for rainfall data reduction. Only after he had reduced the rainfall data to a set of tabulated hourly rain amounts for each station, and reported them to NSF, was he supplied with information about when and where seeding took place. . Integrity of radar data reduction was maintained by placing a randomized identifier number inside the radar camera in lieu of date identification. These numbers were provided to the radar operator by Professors Brownlee and Kruskal and were not available to any other person. Only after the radar echo data had been extracted and tabulated from the coded films was the calendar date of each film identified. With the exception of dispatching the seeder planes, all activities on the project proceeded identically on seeded and not-seeded days. Once designated for seeding, a day remained in the seeded group regardless of the amount of seeding material dispensed. . -.!'C'-i~,:',4-~~~~_~_'~.;';;" Joumal of the American Statistical Association, March 1979 Whitetop was a pioneering experiment. It was the first large-scale randomized experiment to seed nonorQgraphic swnmer convective clouds. A large fraction of the basic dal~a, including all the rainfall data, was published at about the sam(' time that the data became available to project scientists. It was the first large seeding experi- ment to be conducted in parallel with extensive physical st\lldies of clouds on the seeded and not-seeded days. 'c I 3.1 Whitetop Results. Analyses of the seeding results were conducted by Professor Decker at the University of Missouri (Decker and Schickedanz 1966; Decker, Chang, and Krause 19n), by Professor Neyman and associates at Berkeley (Neyman, Scott; and Wells 1969; Lovasich et al. 1969, 19na, b), and by Whitetop scientists (Flueck .1968, 19n; Braham 1965; Braham, McCarthy, and Flueck 19n). . Preliminary analyses of Whitetop data were conducted by Mr. Michael Hoyle, under the supervision of Profes- sors Brownlee and Kruskal. These analyses contributed to papers presented at scientific meetings, but were not publishe~. The main project analysis of statistical aspects of lOur seeding trials was conducted, under the guidance of :Professors Brownlee and Kruskal, by John A. Flueck, who was able to draw on training in both physics and statistics. Dr. Flueck took a data analysis point uf view, placing' emphasis on "discovery, exposure and sum- marization of the treatment effect relationships." His published findings (Flueck 1971) consider both rain and echo in the six hours reserved for seeding, the five post- seeding hours, and the eleven hours combined. He also considered several post-factum data partitions based on meteorological factors and on distance downwind from the seeding line. Complete tabulated statistical data for all tests used (both' parametric and nonparametric), by yes,r and across years, and for all partitions considered, are given by Flueck (1971). . Because of space limitations, this article only sum- marizes the highlights of Whitetop results. We stress the use of statistics to assess the strengths of "structural re- lationships" anticipated from a knowledge of the physical processes governing movement of seeding materials, in- gestion of materials into cumulus clouds, and the re- sponse of clouds to seeding. We also take advantage of the various physical measurements and meteorological studies built into the project and of subsequent findings by other groups. We begin by accept~ng that the effects of seeding should be most pronounced in the wind-advected target are:!!., which was constructed to include all the" area where winds would have been likely to carry the seeding mate- rial. We further accept that the control area should give a measure of protection against the effects of uncontrolled factors not balanced out by randomization, even if there were some seeding spillover into it. Finally, we accept that target-control differences should give the correct ....i.Iit...~.'z~:_..~