WMOD00277 - Laserfiche WebLink

Home Browse Search

winter season of 1964-65, and by Colorado State University personnel during the winter season of 1967-68. An atterr~pt was made to have the appropriate ground-based seeding generators running at all times v'.Then precipita- tion was occurring or was imminent. 2. Sta tistical Evaluation Procedures Three statistical methods that evaluate differences in precipitation between seeded and non- seeded periods have been discussed by Grant and Mielke (1967) so only a brier synopsis is repeated here. The first two methods apply nonparametric procedures to all observations. The third method is based on a parametric technique introduced by Thorn (1957). The rirst nonparametric technique (NP1) employs a two-sample Wilcoxon test while the second nonparametric procedure (NP2) utilizes a two- sample sum of squared ranks test. The underlying assumption in both techniqu es is that if seeding has no effect on the amount of precipitation, then the target precipitation for seeded and non-seeded days represent observations rrom identical distributions. When analyzed with controls the assumption is that the difference between the control and target precipita- tion for seeded and non-seeded days represents obs ervations from identical distributions. The parametric approach (PAR) assumes that precipitation data may be approximated by a gamma distribution. The raw data is trans- formed into normalized data that is suitable for the application of a simple regression analysis. The basic inform8tion for both seeded and non-seeded periods consists of non-zero paired observations (target and control). The expectation or the resulting normalized test statistic is taken in terms of the assumed under- lying gamma distributed variables. Then a point estimate of a scale change during the seeded period with respect to the non-seeded period can be obtained. The utilization of three different methods to statistically evaluate the experimental data has certain advantages for interpreting results. The two-sample Wilcoxon test (NP1) gives equal emphasis to all observations while the two-sample sum of squared ranks test (NP2) places greater emphasis on the larger precipitation amounts. The nonparametric methods (NP1 and NP2) have the ad- vantage of being applicable to all forms of data. However, the parametric method (PAR) is seriously deficient in bebg able to only cope with reduced data amounts involvbe: days in which both target and control precipitafion "amounts are not zero. Analysis of the Climax and Wolf Creek experiments is shown with and without the inclusion of control precipita,tion in the computations. The large variability in the experimental data combined with the relatively small sample sizes resulting from partitioning, make it impossible to derive tight confidence intervals orscale estimators. In order to supply some meaningful interpretation or the results, the probability of the scale change bei!1g exceeded in the same sense by chance (p-value) is included in the final tabulat'ed surnrn.8..ries. Results of three different and independent experimental samples are depicted in the summary tables of final results. Since the emphasis in this paper is on the distribution of seeding effects under specified meteorological criteria, a brief description of the composition of these three sets of independent data are given. 3. Basic Data a. Climax I data sample (251 cases) During the peri od 1960- 65 there were 283 experimental days defined for the Climax experiment. Preliminary results of this entire sample were previously discussed Orant and Mielke, 1967). Chappell (1967) in a rurther analysis of the 1960-65 Climax data found tln t several of the experi- mental events had wind directions that could not bring the seeding material toward the primary target area. If only those experimental events that have 500 mb wind directions between 210 degrees and 360 degrees inclusive are considered (as originally defined for the experiment), the sample reduces to 251'cases. b. Climax II data sample (127 cases) During the period 1965-68 there were 231 experimental days defined for the Climax experi- ment. Five of these events were also found to have 500 mb wind flows outside the prescribed direction interval, reducing the originaf sample to 226 cases. Further investigation indicated that on 99 other experi- mental days seeding from other activities close upstream (100 miles or less) could have affected either the primary target area or the control area. Therefore, there are only 127 "clean" experimental days available during this period for analysis with controls. The overall variations 'Of seeding effects with 500 mb temperatures for the. original samples composed of 231 experimental days (Grant, et al. , 1969) is essentially duplicated by the "clean sample" comprised of 127 days. c. Wolf Creek I data sample (362 cases) There were 362 days that met the prescribed criteria for an experimental event at Wolf Creek Pass during the four win!; er seasons from 1964-68. It should be mentioned that in this total sample there are 59 designated seeded cases where actually no seeding was conducted. These 59 cases were heavily biased toward low precipitation amounts. They have been retained in the seeded group since there is no acceptable way to remove their counter- part from the non-seeded events. It is likely that seeding effects are being diluted by the retention of these 59 cases in the total sample. The results shown for the samples at Climax are for two sensors located within a few feet of one another at the High Altitude Observatory of the University of Colorado, near the summit of Fremont Pass. Consideration of elevation and spatial variation of precipitation over a network of 65 stations does not substantially alter the results presented below (Mielke, et al. ,1969). The results shown for the Wolf Creek I sample are based on precipitation amounts recorded near the summit of Wolf Creek Pass. This sensor is an 8 inch shielded recordir:g t:;:pe g2.ge 2.S is one of the sensors in the 14