SEEDING SUMMERTIME CONVECTIVE CLOUDS TOI INCREASE BLACK HILLS RAINFALL

A. S. Dennis U.S. Bureau of Reclamation Denver CO 80225

Abstract. Data from two Reclamation-sponsored randomized cloud seeding experiments near the Black Hills of South Dakota are combined with data on size distributions of showers to estimate the potential for modifying summertime precipitation. The analysis suggests that increases amounting to roughly 12 percent of the natural precipitation are possible through a combination of microphysical and dynamic effects. Convective clouds around 6 km deep are judged to be the most promising targets for seeding to increase summer rainfall.

I. INTRODUCTION cumulus clouds, well away from other From 1963 through 1972 the South showers, with updrafts of at least Dakota School of Mines and Technology 2 m-s I below cloud base and cloud top (School of Mines) conducted a series temperatures less than -10°C were cloud seeding experiments in the western selected as test cases. The existence Dakotas to test effects of seeding upon of a radar echo did not prevent the rainfall from summertime convective selection of a cluster of growing clouds clouds. Almost all of the experiments as a test case." were conducted as a part of the Bureau of Reclamation’s Project Skywater. By definition, each test case lasted 1 hour, which is about twice the The most important of the School of lifetime of a typical convective cell in Mines projects near the Black Hills were a multicell thunderstorm. The principal the Rapid Project and Project Cloud response variable was the radar- Catcher. Both projects dealt with estimated rainfall (RER), which was summertime convective clouds near the determined from X-band radar data eastern edge of the Black Hills. The recorded by the on-line computer. No purpose of this paper is to combine corrections for attenuation were findings from the Rapid Project of 1966 attempted. For data logging purposes, to 1968 (Dennis and Koscielski, 1969) the area around the radar site was laid and Project Cloud Catcher I of 1969 and off in squares I0 nautical miles on a 1970 (Dennis et al., 1975a) with side. A test case normally occupied previously published information on size two squares at one time, and all echoes distributions of convective showers in in the designated squares were order to assess the potential impact of considered to be part of the test case. seeding convective clouds upon total summer rainfall over and near the Black In 1969 and 1970, a three-way Hills. randomization was used, with the choices being no-seed, salt seed, and silver 2. CLOUD CATCHER: AN EXPERIMENT ON iodide seed. All seeding was done from INDIVIDUAL CLOUDS AND CLOUD CLUSTERS an aircraft operating in updrafts below cloud base. Attention was concentrated 2.1 Experimental Desiqn on updrafts under new cloud towers Cloud Catcher was directed from a adjacent to existing showers. radar site about I0 km east of Rapid City. It was one of the first cloud- Salt seeding was accomplished by seeding experiments to use an on-line releasing up to 50 kg of powdered sodium computer for recording and preliminary chloride during the first 30 minutes of processing of radar data (Boardman and each test case. Seeding of a silver Smith, 1974).¯ It used a single, moving iodide case was accomplished by burning target area to test the effects of seed- flares, each containing 120 g of silver ing on isolated convective clouds or iodide, one at a time in place in wing- cloud clusters. Quoting from Dennis et mounted racks. A flare burned for about al. (1975a), "Distinct clusters 5 minutes, and up to 6 flares were used

IBased on an oral presentation at the Conference on the Science and Technology of Cloud Seeding in the Black Hills, Rapid City, South Dakota, October 16-17, 1989.

117 on a test case if the updrafts lasted (RER) ~/s = -4.02 + 1.43(CDP) (i) long enough. There was no attempt to overseed the clouds to produce dramatic where RER is in kilotons 2 and CDP is in dynamic responses. kilometers.

Eighty Cloud Catcher cases were Dennis et al. (1975a) found recorded in 1969 and 1970, with 33 of indication that rainfall from salt cases them being no-seed cases, 18 silver exceeded that from no-seed cases of the iodide cases, and 29 salt cases. The same depth, but the statistical relative lack of silver iodide cases can significance of this result was somewhat be attributed in part to the lack of weak, with the p-value for the test for sufficiently powerful blocking in the differences in adjusted means being randomization scheme. In 1970 the random 0.06. [The test for differences in decision for the first test case of each adjusted means allowed for the fact that day was applied to subsequent cases on the distributions of CDP varied for the that day, if any. This arrangement, three classes of test cases. CDP which was intended to reduce averaged 6.5 km for the no-seed cases, possibilities of contamination, had the 5.2 km for the silver iodide cases, and unfortunate effect of aggravating the 5.8 km for the salt cases.] imbalance among the three classes of test cases. RER for the silver iodide cases exceeded that from no-seed cases of the 2.2 Effects of Seedinq on Individual same depth. CIt should be noted that, Clouds on figure I, the values of CDP for the The radar data from Cloud Catcher silver iodide cases are those actually showed that cloud seeding with either observed, rather than the values that silver iodide or salt can stimulate would have been observed had the clouds precipitation in cumulus congestus been left unseeded.) The difference in clouds. "First echoes" from new adjusted means was statistically precipitation shafts in clouds seeded significant, with the p-value being 0.01 with either silver iodide or powdered under the assumption that all test cases salt appeared closer to cloud base than were independent. Although the first echoes in unseeded clouds (Dennis significance is weakened by the fact and Koscielski, 1972). The average that in 1970 all cases occurring on one height of first echoes above cloud base day received the same treatment, this is was 3.4, 2.1, and 1.6 km for the no- a surprisingly positive result to emerge seed, silver iodide, and salt cases, from only two seasons of respectively. This result is evidence experimentation. It is assumed, for the that seeding speeded the formation of purposes of the exercise presented in precipitation in the clouds studied, as this paper, that real differences exist is the fact that clouds greater than between the no-seed and silver iodide 2 km deep generally precipitated after cases. seeding with either silver iodide or salt, whereas unseeded clouds did not Dennis et al. (1975a) derived the precipitate until they reached about following regression equation for the 4 km depth. Cloud Catcher silver iodide cases:

Dennis et al. (1975a) presented (RER)’/3 = -2.62 + 1.44(CDP) (2) statistical analysis of the Cloud Catcher test cases of 1969 and 1970. In where the units are as in (i). The particular, they studied RER as a difference in the intercepts of CI) and function of cloud depth (CDP), which was (2) is 1.40, which is very close to the defined as the highest radar echo top change associated with a l-km difference recorded in the test case minus the in CDP. That is, the value of RER for a height of the cloud base. They obtained silver iodide case approximates that for straight-line relationships by plotting a no-seed case with CDP i km greater. the cube root of RER against CDP. Figure i, which is patterned after their Table 1 has been prepared to figure 4, has been replotted using data provide additional insight into the from table A.2 of Dennis et al. (1974). implications of (I) and (2), particularly when combined with Dennis et al. (1975a) found the information to be presented below on the equation of the regression line for the frequency distributions of shower sizes. no-seed Cloud Catcher cases to be: Application of (1) and (2) leads columns 3 and 5 of table l,

s 2A metric ton is the mass of 1 m of water. A3. kiloton is equivalent to 103 m

118 respectively. Comparison of columns 3 storms can reduce their precipitation and 5 indicates that rainfall from efficiency, and the lack of silver clouds of any given depth up to i0 km is iodide cases with CDP exceeding i0 km in increased by silver iodide seeding as the Cloud Catcher sample, we can not practiced in CloudCatcher. Expressed draw any conclusions about the effects as a percentage of the expected or no- of silver iodide seeding on such clouds. seed rainfall, the calculated effect It is for this reason that the entries decreases from small to large clouds; in column 5 of table 1 end at CDP equal expressed in terms of rain volume, it to i0 km. continues to increase. 2.3 Consideration of Dynamic Effects It is unlikely that the linear The Cloud Catcher radar data relationship between (RER) I/s and CDP for indicated that the additional rainfall silver iodide cases shown in figure 1 associated with clouds of a given depth extends to clouds with depths as great was not due to increases in peak as ii or 12 km. Uncertainties about rainfall intensity, but rather to inferences based on regression lines are additions in space or time, or both, of greatest when one deals with the additional precipitating areas similar extremes of the size range, rather than to the natural showers (Dennis et al., the mid-range, and extrapolation is 1975a). In other words, the showers considered undesirable. The situation responded to seeding by becoming more is made worse in this case by the fact numerous, increasing in horizontal that the largest value of CDP among the extent, lasting longer, or some silver iodide cases was just i0.i km. combination of all three responses. In view of indications from some cloud Such a result could be produced by modeling studies (Orville, 1984, p.54) microphysical effects alone, but it is that silver iodide se~ding of large judged more likely to represent some

Table i. - Relative abundance of clouds, rainfall per case, and total contribution to convective rainfall, all as functions of cloud depth, for no-seed and silver iodide cases

(i) (2) (3) (4) (5) (6) (7) (8)

2 48.4 0.00 0 0.02

8 1.5 410 620 700 1050 930 1400 9 0.93 690 640 ii00 1020 1400 1300 i0 0.59 ii00 650 1600 940 - 940 ii 0.38 1600 610 - 610 - 610 12 0.24 2300 550 - 550 - 550

i00 4220 7190 9830

(I) Cloud depth (CDP) in kilometers. (2) Relative number of cases for ~ = 0.35 -I, no rmalized to i00 . (3) Rainfall (10S.m s) per no-seed case. (4) Contribution to total convective rainfall (arbitrary units) all clouds expected with CDP in range indicated. (5) Rainfall (10S.mS) ’ per silver iodide case according to Eq. (2), assuming no change in CDP due to seeding. (6) Contribution to total convective rainfall (arbitrary units) all clouds expected with CDP in range indicated, assuming all clouds seeded with silver iodide and no change in CDP due to seeding. (7) Rainfall (10S.m s) per silver iodide case according to Eq. (2), assuming an increase in CDP of 600 m due to seeding. In this case, the CDP shown in column 1 refers to the cloud left unseeded. The observed clouds would be 600 m taller. (8) Contribution to total convective rainfall (arbitrary units) all clouds expected with (unseeded) CDP in range indicated, assuming all clouds seeded with silver iodide and an increase of 600 m in CDP due to seeding for all values of CDP up to 9 km.

119 ~ O AgI -- Ag I TREND E-! NO SEED -- NO SEED TREND

16T

141

. i ¯ ~.... O ’-T ..

2T ./.~_ ..o...._-,.~.~ c 1 2 3 4 5 6 7 8 9 10 11 CLOUD DEPTH £igu£e Z, - Scatter diagram comparing cube root of radar- estimated rainfall (RER) to cloud depth (CDf) for no-seed and silver iodide cases o£ Project Cloud Catcher in k969-20. dynamic effects as well, perhaps an be observed rather than the no-seed intensification of secondary cloud values as shown. Comparison of columns towers around the major precipitation 5 and 7 shows that the additional impact cells, of a 600-m increase in cloud height upon rainfall would be comparable to that The analysis to this.point has revealed by the regression analysis. taken no account of possible increases in CDP, the depth of the tallest We repeat, for emphasis, that the precipitation echo in each test case. increases in CDP would not be uniform. An increase in CDP due to seeding would Cloud model runs using North Dakota shift a point on figure 1 corresponding Pilot Project data (Dennis et al., to a seeded cloud some distance to the 1975b) suggest that the increases in right. Analysis of the Cloud Catcher cloud height are greatest for clouds radar and aircraft data with the aid of with (unseeded) top temperatures in the cloud models showed no increases in CDP range from -I0 to -30°C, that is, to associated with salt seeding, but clouds with CDP from 2 to 6 or 7 km, and suggested increases in CDP due to silver nonexistent for most cases with CDP iodide seeding, with maximum cloud..top greater than 10 km (smith et al., 1986). in a test case rising, on average, by Because of the many uncertainties, there perhaps 600 m (Dennis et al., 1975a). are no entries in column 7 for clouds The model analysis showed no skill in whose depth is ~0 km or greater. assigning increases in CDP to particular days; however, the increases must vary 2.4 Potential Effects of Seedinq from day to day, or even from cloud to Assemblaqes of Convective CiQ.~.d~s cloud, depending on the ambient In order to determine the potential conditions. Any changes in rainfall impact upon Black Hills summer rainfall associated with changes in CDP would be of seeding all available convective in addition to those indicated in column clouds, it is necessary to develop some 5 of table i. Column 7 of. table 1 statistics regarding the size provides an estimate of the total distributions of natural clouds and the rainfall increases produced by the contributions of the different size combination of increases in CDP (a categories to the natural rainfall. For possible dynamic effect) and changes in this purpose, it is not sufficient to the relationship between RER and CDP look at the size distribution of the (probably a combination of dynamic and Cloud Catcher cases, because many microphysical effects). The numbers in factors influenced their selection, column 7 were obtained by adding 600 m including location with respect to the to CDP and applying (2); it would be the enhanced cloud tops that would actually

120 "REVIEWED" radar, visibility below cloud base, and the other uncertainties in the analysis. a need to develop a statistical base by All results should be interpreted as sampling clouds rather uniformly indicating only general trends, rather distributed along the size range from than as exact calculations of the cumulus congestus to moderate possible rainfall increases. thundershowers. Column 2 of table I, which shows Numerous studies have shown that the relative frequency of clouds of small convective clouds greatly different sizes, was obtained by outnumber large ones. For example, applying (4) with the parameter ~ set I Dennis and Fernald (1963) found that the 0.35 km and normalizing so that the radii of radar echoes from showers in total number of clouds considered came. many parts of the world follow an out to I00. This is a truncated exponential distribution. Miller et al. distribution. Ignoring clouds with CDP (1975) fitted exponential distributions exceeding 12.5 km is of little to the diameters of echoes in the North consequence as far as number of clouds Dakota Pilot Project radar data from is concerned, although such clouds may 1972 to determine the parameter ~ in be significant in terms of total rainfall. On the other hand, there are N = No exp(-~D) (3) many clouds with CDP less than 2 km. Dennis and Fernald (1963) quoted work where N is the number of shower echoes earlier authors showing that the with diameters between D and (D+dD), O exponential distribution they observed is a constant reflecting the number of for radar echoes from isolated showers showers in the space-time domain extends down-scale to the visual studied, and the parameter ~ dimensions of small cumulus clouds. characterizes the size distribution. However, we can ignore such clouds in They found ~ for the entire data set to the present case because they do not be about 0.35 km "I , in good agreement produce any precipitation whether seeded with Dennis and Fernald (1963). As the or unseeded. height of a convective cloud tends to approximate its diameter (e.g., Miller The numbers in column 4 of table 1 et al., 1975), we assume that the are the products of the corresponding exponential relationship with a similar numbers in columns 2 and 3. They show value for ~ holds for CDP.. the relative contributions of clouds of different sizes to the total natural Equation (3) was derived from convective rainfall over the Black hourly snapshot views. As there is a Hills, and peak for CDP around i0 km. positive correlation between echo size As the bases of summertime convective and echo lifetime, a large shower has a clouds .near the Black Hills are, on greater chance of showing up in such a average, very close to 3 km above sea census than does a smaller one. Miller level, the i0 km value corresponds to et al. (1975) studied a supplementary large storms with cloud tops about 13 km sample of 479 echoes, from which echoes above sea level. This number agrees that were obviously multi-cellular were with the common perception that much excluded; they found a roughly linear Black Hills summer rain falls from large relationship between echo diameter and thunderstorms. echo lifetime (r = 0.65). Therefore, order to make table 1 reflect more The entries in column 6 of table 1 closely the size distribution of all for all values of CDP up to i0 km were echoes that form, exist, and dissipate, obtained by multiplying the as opposed to the size distribution corresponding numbers in columns 2 and observed at a given moment, we have 5, and represent the expected rainfall modified (3) as follows: from silver iodide seeded showers according to (2) and (4). On -I N’ = No D exp(-~D) (4) assumption that seeding would not be done on clouds more than I0 km deep or, There is a further complication in that if it were done, would have no effect, the snapshot views must catch some the entries of column 4 for such clouds undefined fraction of the echoes in are repeated in column 6. their growing or dissipating stages, thereby leading to underestimates of A comparison of columns 4 and 6 their maximum size. No satisfactory ignores the possibility of increases in method of adjustment for this second- CDP but suggests, nevertheless, that order complication has been devised. silver iodide seeding can increase the Therefore no allowance is made for it in total rainfall from isolated summertime the calculations that follow. It is convective clouds by a factor of roughly considered a minor factor compared to 1.7. It is instructive to consider

121 where the predicted increases come from; from clouds or cloud clusters with this can be seen in figure 2, which values of CDP around 6 km. shows the rainfall contributions as a function of CDP for both no-seed and In summary, the Cloud Catcher silver iodide cases, as well as in results, combined with some modeling table i. Small clouds are numerous and studies and data on size distributions respond well to seeding (percentage- of shower echoes, indicate that silver wise), but they produce very little rain iodide seeding could increase total whether seeded or not, so the impact of rainfall from isolated convective clouds seeding them is not great. On the other or cloud clusters near the Black Hills hand, large clouds are scarce and the by a factor somewhere between 1.5 and effect of seeding (percentage-wise) 2.5, with much of the increase coming drops off as one moves to large clouds. from moderate shower clouds or small The contribution to the additional thunderstorms with tops from 7 to I0 km rainfall peaks for CDP between 6 and above sea level. 7 km. Clouds in this size range are significantly smaller than the ones that In order to draw conclusions about produce the bulk of the natural the impact of seeding isolated rainfall. convective clouds upon total summer rainfall near the Black Hil~s, one ’ ~ SEED 2 ~ SEED 1 ii~!~ NO SEED requires information on the contribution that isolated convective systems make to 15OO that rainfall. It is also necessary to consider the possibility that seeding 1250 one convective cloud might suppress rainfall from other clouds in the vicinity through dynamic effects (e.g., A 1000i Simpson and Dennis, 1974). Some I information about both of these points N 750’, can be drawn from the results of the F Rapid Project although, as we shall see, A L 5oo some questions remained unanswered L following its completion. 250 3. THE RAPID PROJECT: AN ~XPEREMENT ON AREA RAINFALL 0 2 3 4 5 6 7 8 9 10 11 12 3.1Experimental Desiqn CLOUD DEPTH (km) The Rapid Project of 1966-68 has Figure 2. - Contributions to total been described by Dennis and Koscielski shower rainfall from clouds (1969). It used a randomized crossover of different depths under design, with two pairs of target areas assumptions of no seeding, laid out east of the Black Hills, one silver iodide seeding pair for days with southwest flow and without changes in cloud one pair for days with northwest flow. depth (Seed i), and silver This arrangement minimized cross-target iodide seeding with 600-m contamination. Each target area increases in cloud depth for encompassed2. about 2500 km clouds with depth less than i0 km (Seed 2). Days were classified in advance as southwest-flow or northwest-flow days. The entries in column 8 of table 1 They were also classified according to for all values of CDP up to 9 km are the the synoptic situation into four types, products of the corresponding numbers in of which the most important were shower columns 2 and 7. They account for the days and storm days. Shower days and possibility of increases in CDP of 600 m storm days had the same requirements as a dynamic response to silver iodide regarding precipitable water and upper seeding. On the assumption that there winds as determined from the Rapid City would be no increases in maximum height rawinsonde. Storm days were for clouds with CDP of i0 km or greater characterized by forcing by large-scale (Smith et al., 1986), the entries weather systems, as evidenced by column 6 are repeated in column 8 for positive vorticity advection at the these large clouds. The differences 500-mb level; they normally produced between columns 4 and 8 of table 1 (see substantial rainfall, often from squall also fig. 2) suggest a 2.3-fold increase lines. Shower days were characterized in total shower rainfall through a by absence of positive vorticity combination of microphysical effects, advection at the 500-mb level; they with the maximum contribution coming normally produced only isolated showers,

122 which were often limited to the Black suppression effects in the unseeded Hills. target, the Rapid Project results foreshadowed those obtained later on The static seeding approach was Cloud Catcher. That is, the most followed. Ice crystal formation was promising clouds to be seeded for induced in supercooled cumulus clouds potential rainfall increases are cumulus for microphysical effects, with the congestus and towering cumulus, and the principal effect sought being an rainfall increases obtainable on days increase in precipitation efficiency. when such clouds predominate are very The experimenters were aware that large, when expressed as a percentage of dynamic effects might be produced too, the natural rainfall on those days. but dynamic effects were not Adding up the numbers in table 2 for deliberately sought. The seeding rate shower days pertaining to seed and no- was kept low to reduce the possibility seed targets yields totals of 4.87 and of overseeding. Several seeding agents 1.73, respectively. The ratio of those were used. By far the most common two numbers is 2.8, which agrees treatment was the release of silver surprisingly well with the estimate of iodide from an acetone generator on an 2.3 obtained from the Cloud Catcher aircraft flying in updrafts below cloud data. One can hypothesize that the base. Sodium iodide was used as the Cloud Catcher cases correspond to the solubilizing agent. Only one generator shower days of the Rapid Project, and was used at a time, with the release that the storms on the Rapid Project rateI. being approximately 300 g.h storm days are samples of a different type of phenomenon, which is completely ¯ Seeding directions were radioed to outside of the shower size distributions the pilot from the radar station east of expressed by (3) and (4). While further Rapid City. The objective was to treat study might enable one to improve on all promising cumulus congestus and this assumption, it is not possible to towering cumulus clouds that were over be more precise at this time. or approaching the seed target area. Seeding was sometimes suspended when the Table 2*. Average rainfall (in mm) per only target clouds available were large, gauge per operational day heavily glaciated cumulonimbus clouds. and seed/no-seed ratios in target areas for all days of 3.2 Results given type on Rapid Project The Rapid Project included 98 (Number of days by category follows operational days, of which 27 were north-area entries) shower days and 50 were storm days. A more extensive breakdown is given in table 2. SW-flow NW-flow

The Rapid Project was evaluated on Shower days the basis of observations at a network of almost i00 rain gauges. Under the North area, seed N 0.99 / 6 0.91 / 7 randomized crossover design, there was a seed S 0.69 / 8 0.64 / 6 seed target area and a no-seed target Seed/no-seed ratio 1.4 1.4 area each operational day. The rain South area, seed S 1.83 1.14 seed N 0.30 0.i0 gauge data (table 2) show that the seed Seed/no-seed ratio 6.0 II target received more rain than the no- seed target on shower days, but less Storm days rain on storm days. Statistical analyses by Chang (1976) showed that the North area, seed N 2.18 /15 0.89 / 7 differences on shower days were likely seed S 2.82 / 14 4.32 / 14 real, with the p-values being 0.01 and Seed/no-seed ratio 0.77 0.21 0.09 for southwesterly flow and South area, seed S 2.36 1.37 northwesterly flow days respectively. seed N 2.97 3.18 However, the p-values for the storm Seed/no-seed ratio 0.79 0.43 days, 0.35 and 0.61 for southwesterly * after Dennis and Koscielski, 1969 flow and northwesterly flow days respectively, were too large to support We turn next to the possible effects any conclusions about effects of seeding of seeding isolated convective clouds or on those days. cloud clusters upon total seasonal rainfall. The indicated effect is If one assumes that the differences modest, in large part because of the in rainfall patterns on north-seed and scarcity of’shower days. The Rapid south-seed days represent rainfall Project, which was in the field 6 days a increases in the seed target rather than week from roughly June 1 to August15

123 for 3 years, logged only 27 shower days. randomized crossover experiments when For an operational project, one should strong dynamic effects are present or plan on about I0 days per season suspected. providing good seeding opportunities over a small, fixed target area. 4. SUGGESTIONS FOR FUTURE RESEARCK Cloud Catcher I left some unanswered The data in table 2 on numbers of questions about the effects of silver shower and storm days and on average iodide seeding on convective clouds, rainfall per operational day in the particularly when the convective cells unseeded targets indicate that shower are clustered together. It provided days contribute about 7 percent and evidence of significant rainfall storm days about 93 percent of the increases due to silver iodide seeding natural summer rainfall on the east side of clouds of moderate size, but it did of the Black Hills. (Contributions from not make clear how the individual cells other types of days, which include reacted to produce the apparent frontal overrunning situations, are increases. These unanswered questions negligible after early June.) Combining were to be addressed in Cloud the data in table 2 for seeded targets Catcher II, which was in the field in on shower days and unseeded targets on 1971 and 1972. It used an S-band radar storm days, we find that the shower days to avoid attenuation by precipitation, a now contribute some 16 percent of the floating target design, and quite total rainfall, which is increased sophisticated programming in the on-line thereby by about 12 percent. computer which controlled the radar scans, logged and processed the radar The preceding calculations rest on data, and provided real-time graphic the assumption that results of Cloud displays to the project meteorologist Catcher and the Rapid Project can be (Boardman and Smith, 1974). extrapolated to larger areas.. In fact, Unfortunately, seeding suspensions interactions among the seeded clouds following the Black Hills flood of 1972 might complicate matters considerably. and program cuts in 1973 led to A seeding strategy for increasing termination of Cloud Catcher II before rainfall can not be selected its potential could be realized. intelligently without having the target Lingering questions on the part of the area clearly in mind. Consider, for general public about whether or not example, the Rapid Project shower days cloud seeding contributed to the 1972 with southwesterly flow (table 2). flood make it doubtful whether seeding There is no doubt that, in both the of Black Hills summer clouds will be north and south targets, rainfall was attempted again during this century. heaviest on days when the particular Nevertheless, this section considers target was seeded. [We do not know what what a new round of experiments miqht fraction of this favorable result was entail. due to suppression effects in the no- seed target.] Therefore, if one wants In assessing the status of seeding more rain in the north target area, he of convective clouds, it is of some should seed it; if one wants more rain value to compare the South Dakota in the south target area, he should seed results with those obtained subsequently it. However, if the objective is to in other places. Statistical analyses maximize total rainfall over the two of the second Florida Area Cumulus target areas, the recommended treatment Experiment (FACE-2) failed to prove the would be to seed the south target existence of rainfall increases for always. Seeding the south target either the total target area or for the yielded an average rainfall across the floating targets encompassing the cells two target areas of approximately 1.3 selected for treatment (Woodley et al., mm, compared to an average of only 0.65 1983). Subsequently, Gagin et al. mm when the north target was seeded. (1986) used computerized cell-tracking techniques in a detailed post-hoc study Of course, the Rapid Project did not of convective cells within seeded and test the more obvious treatment for unseeded cloud systems on FACE-2. They maximizing total rainfall, namely, found evidence suggestive of rainfall seeding clouds in both target areas on increases for clouds that were treated the same day, whenever and wherever they early in their lifetimes with more than appeared, nor, for that matter, did it 8 silver iodide flares. Their results test the option of not seeding at all. for such cases were in good agreement This exercise points out the need for with the estimates based on table i, clearly defined objectives in selecting namely, that rainfall from a moderate a seeding strategy, as well as the shower cloud can be multiplied by a ambiguities that arise in the factor of 2 or 3 by silver iodide interpretation of results from seeding. However, unlike Dennis et al.

124 (1975a), Gagin et al. (1986) found Project storm days with southwesterly difference in the relationships between flow, there were indications of rainfall cell height and other variables, decreases at the edge of the Black including rainfall volume, for unseeded Hills, but of substantial increases some and silver iodide seeded cells. 30 to 40 km northeastward, still within Therefore, they interpreted all of the the target areas (Dennis et al., 1976). apparent FACE seeding effect in terms of It appears that, for that particular changes in cell height. situation, additional organization or clustering induced by seeding was still On the other hand, Rosenfeld and beneficial over the open prairie, but Woodley (1989) found for a seeding could not be produced (or was not experiment in Texas a "large indicated beneficial) over the Black Hills effect of treatment on cell rain volume, themselves. despite the small indicated effect on maximum cell heights." It is These inferences are only tentative, interesting to note that the Texas but suggest a possible strategy to be result, which is only tentative due to followed in any.future field the small number of cases analyzed to experiments. On shower days, only date, is in line with the Cloud Catcher clouds supported by low level seeding hypothesis, which stressed a convergence induced by the Black Hills combination of microphysical and dynamic . themselves have any chance of producing effects, rather than the hypothesis in significant showers. Seeding would be the Texas project design, which was directed at the most promising growing patterned after FACE and considered clouds, wherever they occurred. On increases in cell height as the storm days, as evidenced by positive principal, if not the sole, source of vorticity advection (synoptic-scale additional rainfall (Rosenfeld and forcing), systems over the Black Hills Woodley, 1989). would organize with no help, and seeding there likely would have no effect. Any future experiments near the However, clouds forming in the less Black Hills should be designed to shed disturbed regions might be of a size light on these questions, as well as (CDP around 4 to 6 km) that could interactions between the clouds and the stimulated to further growth and Black Hills themselves. Seeding clouds organization. Evaluation should be on over the Black Hills to maximize both a cell or cluster basis and an precipitation requires consideration of area-wide basis to study the larger wind, temperature, and moisture fields scale effects. over the entire area. Kuo and orville (1973) showed how the interactions Continued work with powerful between the Black Hills and the air numerical models, such as the new three- streams passing over them set up dimensional models developed by Clark preferred areas for cloud and shower and others (e.g., Smolarkiewicz et al., formation. Because seeding influences 1988), is required to sort out the cloud dynamics, one should not assume various possible effects. In time, it that seeding over one part of the Black should be possible to integrate detailed Hills does not influence subsequent parameterizations of microphysical cloud developments elsewhere. processes in individual clouds into three-dimensional mesoscale models A unifying concept, which seems to capable of simulating air motions over bring together various pieces of and around the Black Hills. The output evidence, is that rainfall increases are of such models would suggest additional produced when seeding helps convective seeding hypotheses, and might even clouds merge into clusters, rather than provide statistical controls for seeding continuing as single-cell showers experiments. This would be an extremely (Dennis et al., 1976). Such important development, as it would organization promotes the precipitation provide to single-area experiments a efficiency of clouds. It also appears degree of sensitivity now obtainable to increase the total amount of water only with target-control or randomized vapor processed, which accounts for the crossover experiments. As we have seen, great attention paid to cloud mergers in the interpretation of results from cloud seeding experiments in Florida and target-control or randomized crossover Texas (e.g., simpson and Dennis, 1974; experiments is ambiguous when strong Rosenfeld and Woodley, 1989). dynamic effects are present or suspected. The apparent success of seeding on Rapid Project shower days has already been noted, as well as the lack of overall success on storm days. On Rapid 4. ACKNOWLEDGMENTS Gagin, A., D. Rosenfeld, W. L. Woodley The author thanks Paul Smith, Harold and R. E. Lopez, 1986: Results of Orville, and Jim Miller of the School of seeding for dynamic effects on rain-cell Mines and Jon Medina of the U.S. Bureau properties in FACE-2. ~. Appl. Meteor., of Reclamation for useful discussions 2_~5, 3-13. during the preparation of this paper. Preparation of the paper was supported Kuo, J-T., and H. D. orville, 1973: A by the U.S. Bureau of Reclamation. radar climatology of summertime convective clouds in the Black Hi]Is. 5. REFERENCES J. Appl. Meteor,..~ 1~2, 359-368. Boardman, J. H., and P. L. Smith, Jr., 1974: A computer-generated four- Miller, J. R., Jr., A. S. Dennis, J. K. dimensional graphic display for weather Hirsch, and D. E. cain, i975: Statistics radar data. Bull. Amer. Meteor. Soc.~. of shower echoes in western North 55, 16-19. Dakota. Preprints 16th Radar Meteorology Conf., ~mer. Meteor. Soc., Boston, Chang, L. P., 1976: Reevaluation of the Mass., 391-394. Rapid Project data. M. S. Thesis, South Dakota School of Mines and Technology, Orville, H. D., 1984: A review of Rapid City, South Dakota. dynamic-mode seeding of summer cumuli. In Precipitation Enhancement - A Dennis, A. S., and F. G. Fernald, 1963: scientific Challenqe [R. R. Braham, Frequency distribution of shower sizes. Jr., Ed.), Meteor. Mono~.. 21, No. 43, J. Appl. Meteor.. 2, 767-769. Amer. Meteor. Soc., Boston, Mass., 43- 62. and A. Koscielski, 1969: Results of a randomized cloud seeding Rosenfeld, D., and W. L. Woodley, 1989: experiment in South Dakota. J. Appl.. Effects of cloud seeding in west Texas. Meteor., 8, 556-565. J. Appl. Meteor., 28, 1050-1080.

, and , 1972: Height Simpson, J., and A. S. Dennis, 1974: and temperature of first echoes in Cumulus clouds and their modification. unseeded and seeded convective clouds in In Weather and Climate Modification South Dakota. J. Appl. Meteor., ii, (W.N. Hess, Ed.), John Wiley and Sons, 994-1000o New York, 229-281.

, P. L. smith, Jr., B. L. Smith, P. L., J. R. Miller, Jr., J. Davis, H. D. Orville, R. A. Schleusener, Hirsch, and H. D. orville, 1986: G. N. Johnson, J. H. Hirsch, D. E. Cain, Dynamic vs. microphysical effects of and A. Koscielski, 1974: Cloud seeding seeding: Some cloud model and radar to enhance summer rainfall in the observations. Preprints Tenth Conference Northern Plains. Report 74-10, on Weather Modification, Amer. Meteor. Institute of Atmospheric Sciences, South Soc., Boston, Mass., 175-178. Dakota School of Mines and Technology, Rapid City, South Dakota, 161 pp. Smolarkiewicz, ~. K., R. M. Rasmussen, and T. Clark, 1988: On the dynamics of ., A. Koscielski, D. E. Cain, J. Hawaiian cloud bands: Island forcing. H. Hirsch, and P. L. smith, Jr., 1975a: J. Atmos. Sci., 4~, 1872-1905. Analysis of radar observations of a randomized cloud seeding experiment. J_~. Woodley, W. L., A. Barnston, J. A. Appl. Meteor., 14, 897-908. Flueck, and R. 9iondini, 1983: The Florida Area Cumulus Experiment’s second , J. R. Miller, Jr., D. E. phase (FACE-2). Part II: Replicated and Cain, and R. L. Schwaller, 1975b: confirmatory analyses. J. ADDI~ Evaluation by Monte Carlo tests of Meteor., 22, 1529-1540. effects of cloud seeding on growing season rainfall in North Dakota. Appl. Meteor., 14, 959-969.

, J. H. Hirsch, and L. P. Chang, 1976: The role of low-level convergence in controlling convective rainfall and its possible modification by seeding. Papers Presented at Second WMO Scientific Conf. on Weather Modif., Boulder, Colo., World Meteor. Org., Geneva, 49-54.

126