WMOD00528 - Laserfiche WebLink

Home Browse Search

7. DESIGN OF A STATISTICAUPHYSICAL EXPERIMENT FOR THE WASATCH PLATEAU Near the end of the physical experiment phase, a randomized seeding program would be designed which, by present estimates, would require about three to four winters to conduct. While many of the specifics of the statistical/physical design must await analysis of the comprehensive physical cloud seeding experiments, several general concepts can be stated now. 7.1 General Design Considerations The design would attempt to seed those stonn conditions expected to result in enhanced snowfall according to results of physical experiments and numerical model runs. The statistical experiment would incorporate randomized seeding of either a single target area, likely with an upwind control (target-control design), or one of two target areas (crossover design). The long north-south extent and relative unifonnity of terrain should make practical the crossover design (Schickedanz and Huff, 1971) for the Wasatch Plateau. The effectiveness of the crossover design increases as the correlation between the two areas increases. Even with a correlation coefficient of 0.5, the crossover can be 3 times as effective as a target- control design (Dennis, 1980). However, the crossover requires minimal contamination of the untreated area, which usually requires a buffer zone between the target areas. Observations from the physical experiment phase will indicate whether a crossover design is practical for the plateau. Should the crossover design prove impractical a single target area would be seeded or not according to a random decision with approximately half the experimental units treated. Blocking would be used to guard against long strings of the same random decision. The use of covariates would lessen the impact of the natural variability. Covariates would allow estimation of departures (residuals) from the precipitation that would have been expected without seeding. These departures would be statistically tested similar to the approach used in the Super and Heimbach (1983) exploratory analysis of the Bridger Range Experiment, which was based on the Mielke et al. (1981) and Mielke et al. (1982) reanalysis of the Climax Experiments. Upwind and crosswind precipitation observations have been used as covariates in earlier experiments. However, improved covariates may result from combining additional routine observations such as radiometer SLW, radar cloud tops, and rawinsonde data, possibly with the aid of a numerical model. It is important to identify highly correlated covariates, that would not be affected by seeding, to maximize the power of the statistical design which minimizes the duration of the experiment. 7.2 Experimental Units A number of past programs have used the 24 hour day as their experimental unit. However, stonn conditions change dramatically in that long an inteIVal. It is anticipated that experimental units for the new statistical experiment would be much shorter, perhaps 6 hours. The briefer units would be more homogeneous and, therefore, better represented by obseIVations made for partitioning experimental units into similar populations. It is now recognized that a single value may only coarsely represent an entire 24 hour period (e.g., for cloud top temperature). Use of shorter units would provide the opportunity for obtaining more cases (experimental units) per winter, increasing the power of the experiment. Experimental units would be separated by a buffer period empirically detennined during the physical experiment phase as long enough for the seeding agent to essentially leave the area. The statistical design should declare experimental units by objective observations, not forecasts. In general, winter stonn forecasts in mountains have modest accuracy. Their use in declaring experimental units results in many cases with unsuitable conditions, while other periods with suitable conditions are 53