Laserfiche WebLink
<br />assessmcnt and generator placement as well as evaluation of secding effects, The output of such <br />models is exemplified in Figure 5. Hydrologic models have been uscd to estimatc streamtlows <br />resulting from assumed secding-induced snowpack increases. Modeling has seen considerable <br />improvemcnt in physical simulation, theory, speed, sophistication, and accuracy, as <br />acknowledged in the NRC report. Nevertheless, model simulations arc not presently accurate <br />enough to distinguish seeded from natural precipitation or strcamtlow, and therefore models are <br />generally used for guidance purposes only, Use of models in conjunction with physical <br />mcasurements and statistical analyses can be a vcry useful integrated approach, however. There <br />are some instances of mod cling results comparing favorably with physical sampling7o. <br /> <br />Statistical Techniques, Statistics have bccn the most common and long-standing tools to <br />asscss seeding elTects, having been used almost since the inccption of wcather modification <br />itself. The task has proven tomlidable, sincc precipitation augmentation from seeding is small <br />compared to the natural variability of precipitation, The problem is exaccrbated because it is <br />ditlicult even to prcdict thc bchavior of natural clouds, The statistical approach has largcly <br />consisted of two types - historical target-control regrcssion and randomizcd seeding trials, Thc <br />former attempts to comparc precipitation from an area assumed to bc targetcd by sccding and <br />from a nearby but similar area unaffectcd by seeding (similar in geography, altitude etc.). This <br />approach requircs a suitably long duration of observations in both the scedcd and non-sceded <br />arcas during the historical period, to establish a rclationship for predicting natural target <br />precipitation during the operational sccding period. Departurcs between prcdicted and obscrved <br />target amounts can then be statistically tested. The comparison can be between variables such as <br />snow watcr and runoll as well as precipitation. A long duration, perhaps 10 ycars or more, is <br />required to achieve stable. statistically significant results (as cxemplified by some Kings River <br />invcstigationsI9'1o), The main assumption hcrc is that the relationship between natural <br />prccipitation in thc target and control arcas is stablc with time, therefore littlc c1imatc change, <br />The validity of this assumption and other limitations of target-control regression have been <br />described by Silvcmlan (Appendix A) and others18. <br /> <br />The "gold standard" of statistical techniques for cvaluation of sceding effccts is the <br />randomized experiment, and is encouraged by the weather modification opcrational and research <br />communities3!. This approach rcquires a careful (l priori design. unlike many regrcssion <br />analyses that have bcen done post hoc. This dcsign would bc for an exploratory or confimlatory <br />cxperiment that is bascd on findings from a prcceding modcl or cxploratory experiment. <br />Experimcntal units of a fixed duration are eithcr seeded or unseedcd (placebo) and variables <br />(usually prccipitation) from the two pcriods are compared. It is essential that natural <br />precipitation in one o,r morc ncarby c.ontrol.arcas be measured.. tOAf~ard against, statistical.crrors <br />and to allow completIOn of the cxpenment In a reasonablc penod ' . Randomized expcnments <br />require numerous. precise measurement of EU response variables and typically live or more <br />years of data to achieve statistically significant results. Since a portion of the EUs in randomized <br />expcriments must be unsecded. they are more costly and arc thercfore usually auemptcd only <br />\....ithin rcsearch projects. There have bccn relativdy few such experiments in the Western United <br />States. Moreover. thesc experiments have not always adequately studied rele....ant physical <br />processes and T&D. leading somc to question their conclusions, The reccnt Utah \VD~tp <br />randomizcd cxperiment used high-resolution crosswind control and target area snow gauges, <br />short duration EUs. and thrce ditTerent statistical tests. These capabilities Icd to strongly <br />suggestivc positive seeding etTects o....er just one winter.J7;.I9, Whilc this experiment was <br /> <br />23 <br />