WMOD00464 - Laserfiche WebLink

Home Browse Search

A-I A. SELECTED ISSUES The issues Wl~ are about to discuss are important for at least one (more usually both) of (i) how we assess Jrecent experiments and present knowledge and (iO how experiments should be designed, conducted, and analyzed (beginning at once). 1. Basic Ideas * randomization * The only reliable cure to many problems of experimentation is randomization t, the use of sequences of application of treatments -- here "seeding" or "not seeding" -.. that are not merely seemingly unpredictable but have known (and carefully chosen) probabilities of occurrence. In experiments in which people play important roles in either choice or assess- ment, randomization often plays an extra role. (For instances in medicine and surgery, see Gil- bert, McPeek and Mosteller, 1977 J Over the last decade or so, randomization has come to be recognized as an essential part of gathering trustworthy data about weather modification.. It will remain absolutely essential for the foreseeable future tl: * exploratory/confirmatory/piggyback * Today we realize that experiments focused on weather modification will almost inevitably have to have two separate phases, each of which is likely to require several years. This is because we just do not know enough, when moving from one geographic area to another, to forecast reliably which weather situations are most Iik(~ly to be susceptible to seeding. Without an exploratory phase, in which we obtain reasonable identification of the kind(s) of situations for which seeding seems likely to work in this place and season -- and, if we are fortunate, we learn how to adjust observed rainfall for other observed variables -- conducting a serious confirmatory trial will usually be wasteful of both money and time. Without a confirmatory phase, in which seeding under apparently favorable situations is tested under the high1est stan- dards, we will never have the hard evidence that we need. Both phases must be extensive enough to giv,e reasonable precision of measun:ment, something usually not available in a year or two. Should we continue to learn more and more about which situations are most su:scl~ptible to seeding -- at least in familiar combinations of land conformation, weather patterns, and sea- son -- we may come to have less need for exploratory phases. An optimist might hope that this will begin to occur in five to ten years. There is no indication that this is the case today. Finally, then: is the important possibility of "piggyback" experimentation, where cloud seeding operations can be modified enough to allow them to also serve an experimental purpose. Here the most difficult aspects lie in finding modifications that both (i) are acceptable to those paying for and conducting the operations and (iO meet the necessarily high standards, including randomization, that are needed if we are to be able to use the results as solid evidence. The tWe admit there are occasions where randomization is not feasible, but we assert confidently (a) that thiil happens less frequently than is generally believed and (b) that such exceptions do not arise in experimental weather modification. ,tThere may be rare opportunities where natural controls or other concomitant observations (unaffected by seeding) are so effectiv,e, usually under unusual circumstances and only when long periods, such as years, am to be treated as wholes, that unrandomized trials make some sense. If such opportunities are found, particu- lar attention should be: given to incorporating, in the planned analysis, measurable aspects of the weather over both the long periods involved and large areas (such as whole or half hemispheres).