Comprehensive Reply

by

Arthur L. Rangno and Peter V. Hobbs

Cloud and Aerosol Research Group,

Department of Atmospheric Sciences,

University of Washington, Seattle, Washington 98195-1640

to

"Comments on A New Look at the Israeli Cloud Seeding

Experiments’" by D. Rosenfeld*

*Rosenfeld’s Comments appeared in J. Appl. Meteor., 36, 260-271, 1997.

Rangno and Hobbs’ brief Reply was in the same issue (J. Appl. Meteor., 36, 272-276).

May 1997

This Extended Reply is also available on the Cloud and Aerosol Research Group’s home page:

http://cargsun2.atmos.washington.edu

NOTE ON THE BACKGROUND TO THIS COMPREHENSIVE REPLY

The Israeli cloud seeding experiments hold a unique place in meteorology, because they havebeen widely viewed as one of the very few (perhaps the only) demonstration that precipitation onthe ground can be significantly modified by artificial seeding.

In 1995 we published a critique of the Israeli cloud seeding experiments (Rangno and Hobbs1995). Subsequently, D. Rosenfeld, a former student of the late Abraham Gagin (a leader of theIsraeli experiments), who is continuing the work of his mentor/submitted for publication in theJournal of Applied Meteorology extensive Comments on our critique (Rosenfeld 1997). Weprepared a comprehensive Reply to Rosenfeld’s Comments. However, at the request and adviceof the Editor of the Journal (R. Koenig), we published a considerably abbreviated Reply (Rangnoand Hobbs 1997). In the first paragraph of that abbreviated Reply, we noted that we would makeavailable our detailed rebuttal to Rosenfeld’s Comments; the present report is that rebuttal.

Discussions of the pros and cons of the Israeli experiments, involving as they do both physicaland statistical arguments and assessments of the relative roles of natural meteorological processesand artificially-induced effects, are necessarily detailed, often lengthy, and not always conclusive.However, as in a good detective story, attention to detail has its rewards. In carrying out ourstudies of cloud seeding over many years, we have certainly learned the wisdom of RichardFeynman’s philosophy, as summarized by Gleick (1992): "He believed in the primacy of doubt;not as a blemish upon our ability to know but as the essence of knowing."

Peter V. HobbsArthur L. Rangno

May 1997

-ii-

CONTENTS

Page

1 Overview. 1

2 Responses to Rosenfeld’s specific comments concerning cloud

microstructures.......................................3

3 Responses to Rosenfeld’s specific comments concerning seeding

and seeding logistics in Israeli I 8

4 Response to Rosenfeld’s specific comments concerning the statistical

evaluations of the Israeli experiments...................................................... 12

5 Concluding remarks...............................................................................22

References ....................................................-..-- 22

-m-

1. Overview

Rosenfeld (1997) (hereafter R) agrees withus (Rangno and Hobbs 1995-hereafter RH95)that earlier reports by A. Gagin and hisassociates1 that the clouds of Israel contain fewnatural ice particles at cloud top temperaturesabove -21 C, do not contain droplets ^23 umdiameter in the riming-splintering temperaturezone (-2.5 to -8" C), and never form rain bythe collision-coalescence process are allincorrect. As anticipated by Rangno andHobbs (1988), it has now been observed thatconcentrations of ice particles of 20-300 perliter occur in clouds in Israel at cloud toptemperatures ^-14 C (Levin 1994; Levin et al.1996), even though it was reported for manyyears that these clouds were virtually free ofice until cloud top temperatures fell to -14 C orless, and averaged only 3 per liter at -20C(e.g., GN74, GN76, GN81, G75, G80, G81.G86, GG87). Also, precipitation does form inIsrael from both the ice process and thecollision-coalescence mechanism (e.g.,Rangno 1988).

Even though R now agrees with us onAese basic new facts about Israeli clouds, ourviews diverge sharply on two main points:(1) whether it has been demonstrated that theartificial seeding of clouds increased rainfall inIsrael during two randomized statisticalexperiments, and (2) whether naturally high iceparticle concentrations reduce seedingpotential.

Rosenfeld believes that cloud seeding inboth Israeli experiments caused widespreadincreases or decreases in rainfall, depending onthe absence or presence of dust/haze. Forexample, R contends that seeding had effectson precipitation in the various target areas ofthe Israeli experiments, in the buffer zone(which was designed not to be seeded), and inJordan and Lebanon. We, on the other hand,concluded that the HUJ investigatorsmisinterpreted natural patterns of rainfall inboth of the Israeli experiments as being due to

Gagin and Neumann 1973, 1974, 1976, 1981-hereafter GN73, GN74, GN76, GN81; Gagin 1975,1980, 1981, 1986-hereafter G75, G80, G81, G86; andGagin and Gabriel 1987-hereafter GG87. We willrefer to these workers collectively as the "HebrewUniversity ofJerusalem (HUJ) investigators."

seeding effects. We arrived at this conclusionbecause:

Similar (in some cases even larger)anomalies in rainfall to those that the HUJinvestigators attributed to cloud seedingoccurred in both of the Israeli cloud seedingexperiments in regions where any effects ofseeding should have been minimal or non-existent.

Although Rosenfeld and Farbstein(1992-hereafter RF92), Rosenfeld and Nirel(1996) and R agree with RH95 that warm rainand high ice particle concentrations arecommon in clouds in Israel with tops warmerthan -21 C, R believes that such findings donot compromise, nor cast any doubt, on thecontention that rainfall in Israel can beincreased by cloud seeding. This is because Rbelieves that the presence of warm rain and/orice multiplication does not affect the staticseeding potential of polar maritime clouds overIsrael, except when those clouds are alsoaffected by dust/haze from deserts to thesouthwest of Israel. Rosenfeld and Farbstein(1992) and Rosenfeld and Nirel (1996) believethat excessive ice formation caused bydust/haze makes the clouds of Israel unsuitablefor seeding on about half the days with rain innorthern Israel (north of Tel Aviv), and for themajority of the days with rain fromapproximately Tel Aviv southward (i.e., mostof Israel). Rosenfeld and Farbstein concludedthat seeding on dust/haze days may havedecreased rainfall by about 10-25%. Webelieve Ae latter analysis is flawed, and thatseeding had little effect on rainfall under anyconditions (see Section 4).

Rosenfeld believes that Ae potential forincreasing rainfall by cloud seeding is notcompromised even when Ae targeted cloudsnaturally produce high concentrations of iceparticles soon after cloud tops cool to below-10 C. We, on the other hand, as well asmany oAers (e.g., Braham 1964, 1979;Dennis 1980, 1989; Sax et al. 1975; G75;Cotton 1986; Silverman 1986), including R in1989, believe Aat high natural ice particleconcentrations in moderately supercooledclouds severely compromise their staticseeding potential.

Today we know that Ae clouds of Israel,even wiA tops warmer Aan about -14 C, have10s to 100s per liter of natural ice particles(Levin 1994; Levin et al. 1996); Aat Ae onset

of precipitation from clouds in Israel occurs at

cloud top temperatures >-10 C (e.g., Rangno1988; Rosenfeld and Gagin 1989); and, thatthe collision-coalescence process for rainformation is active in Israel (Rangno 1988).Therefore, by inference, the criteria for theproduction of high ice particle concentrationsby the Hallett-Mossop mechanism are met inIsrael. Consequently, the numerous claims bythe HUJ investigators that ice particleconcentrations are perpetually low in Israel,and therefore cloud seeding might increaserainfall in Israel when cloud temperatures arebetween -12 and -21 C, now lacks a physicalfoundation.

We (RH95) pointed out that, irrespectiveof the presence of dust/haze, cloud basetemperatures in Israel (generally >-5 C)virtually guarantee high natural ice particleconcentrations when cloud top temperaturesare in the range where the strongest seedingeffects were reported by the HUJ investigators(-12 to -21 C). Moreover, fetches of moist,unstable boundary layer air over theMediterranean Sea virtually guarantees theaddition of large salt and biogenic particles thatcan contribute to the broadening of the clouddroplet spectra. Thus, the days on which largecloud droplets, naturally high concentrations ofice particles, and warm rain can compromisecloud seeding to increase rainfall are likelymore numerous than those days in whichprecipitating clouds are affected by dust/haze.

Rosenfeld’s proposal that dust/haze causesimmediate glaciation in moderately supercooledclouds is an interesting, but unproven,hypothesis. On the other hand, the profoundeffects of large cloud droplets ($ 23 umdiameter) in producing high ice particleconcentrations in maturing and aging andmoderately supercooled clouds is anestablished fact (e.g., Koenig 1963; Braham1964; Mossop 1970, 1985; Ono 1972; Hobbset al. 1980; Hobbs and Rangno 1985, 1990;Rangno and Hobbs 1991, 1994).

In the first Israeli cloud seedingexperiment (hereafter Israeli I), seeding wascarried out during only about 25-35% of thetime that precipitation fell, or 65-70 h per entirerainy season per target area. Indeed, thenumber of hours of seeding per target is onlyslightly larger than the average number of dayswith rain (= 55)!

Contrary to the assertion of R, the resultsof the second Israeli cloud seeding experiment(hereafter Israeli II) have not been fullyreported. Both Gabriel (1967) and GN74emphasized that all days, including dry days,must be included in the evaluation of Israeli n(as they were in Israeli I). Such an evaluationfor Israeli II has not appeared. Second, RH95noted that rain fell in northern Israel on daysthat were not included in the statistical analyses(e.g., GN81). Third, the statistical crossoverresults that have been reported for Israeli n donot replicate those reported for Israeli I, eitherin toto or in the details of the target areas thatexperienced apparent seeding effects (e.g.,Gabriel and Rosenfeld 1990-hereafterGR90).

At first glance, it would appear that seedingin Israeli n decreased rainfall in the southtarget area (STA) but increased it in the northtarget area (NTA) by almost exactly the sameamount, resulting in a null crossover result(GR90). However, it has been establishedbeyond a doubt that heavy rainfall in the NTAon seeded days was also experienced on thosesame days in the south target area (e.g.,GR90) and in central and southern Lebanon(RH95). Thus, rainfall was not lighter thannormal in the STA on seeded days, rather itwas much heavier than normal on control days(which were the days on which the NTA wasseeded). These facts, first established byGR90, are a formidable refutation of RF92 andR’s hypothesis that seeding decreased rainfallin the STA of Israeli n. Consequently, weconcluded that a "false positive" (in the NTA)and a "false negative" (in the STA) occurred inIsraeli II, one of the several explanationsoffered by GR90.

Control and target stations were notspecified in advance of either Israeli I or II, acondition that is necessary for proper statisticalexperimentation (e.g.. Thorn 1957; Court1960; Dennis 1980). The control and targetstations in Israeli n were unevenly distributed,with a clustering of control stations in zoneswhere seed/no seed precipitation ratios wereanomalously low from a regional viewpoint(Fig. 1); target and control stations alsochanged from one analysis to another. Norwere the target stations evenly distributed,except in Gabriel’s (1967) analysis of Israeli I(Fig. la). More significantly, the target andcontrol stations used by GN81 in their analysis

(a) (b) (c)

Figure 1. a) Rainfall stations (dots) used by Gabriel (1967). See Fig. 13 ofRH95 for further information, b) Rainfall stations(triangles) used by GN81 in their target-control analysis of Israeli II. The shaded area represents the catchment area of LakeKinneret (the primary target area for Israeli n). The hatched area denotes the extreme northern coastal strip from which GN81chose nine control stations. The complete control area (from which a total of seventeen control stations were chosen byGN81) is marked by "C". The various target sub-areas analyzed by GN81 are denoted by "N". (c) The target (N) and control(C) rainfall stations used by GR90 (dots). A surface wind rose is shown in (a) for Bet Dagan, Israel, for those rawinsondelaunches when rain was falling at or within 60 km of Bet Dagan, and at or within about 90 min of the launch dme, for theperiod January-March 1978, November March, 1978-79 through 1984-85, January February 1986, and November throughMarch 1986-87. The number in the center of the wind rose represents those cases with a calm wind.

of Israeli n are not the same as those used byGabriel (1967) to evaluate Israeli I (Fig. Ib) inthose regions where the target and controlareas overlapped in the two experiments.Finally, we note that Gabriel and Rosenfeld(1990) were apparently not able to duplicatethe results of GN81 when they used data fromthe same stations; they therefore used manydifferent stations (compare Fig. Ib with Fig.Ic). Our concern in this matter is that therewere more than 500 raingauges to choose fromat the. time of Israeli I (Atlas of Israel 1970),and 436 Israeli rainfall stations had completerecords during the second Israeli experiment(personal communication, 1997, fromR. Ben-Sarah, Chief of Climatic VerificationSection, Israel Meteorological Service). TheHUJ investigators used only 182 raingaugestations in the analysis of Israeli II (GR90).

In the remainder of this Reply, we respondin more detail to Rosenfeld’s specificcomments on our paper. For the convenienceof the reader, our responses are grouped underthe following headings: cloud microstructures,seeding and seeding logistics in Israeli I, andstatistical evaluations of the Israeliexperiments. In the final section, we make afew concluding remarks.

2. Responses to Rosenfeld’s specificcomments concerning cloudmicrostructures

In his Abstract, R states: "the existence ofcoalescence and ice multiplication in some

of the Israeli clouds in no way precludesenhancement of precipitation (from cloudseeding) even from those clouds."

First, high ice particle concentrations arelikely to occur in most (not "some") Israeliclouds. This is because warm (>5 C) cloudbases predominate, and the air masses thatbring rain-bearing clouds to Israel generallyhave an appreciable fetch across theMediterranean Sea. Both of these factors giverise to large cloud droplets, which areconducive to ice multiplication. In addition to

these factors, R has postulated that dust/haze,which he asserts occurs on more than half ofthe days with rain in Israel, causes or amplifiesice multiplication (RF92). In fact, RF92

concluded that because of dust/haze, cloudseeding in Israel can actually decrease rainfall.

In none of the many analyses of Israeli II(GN76, GN81, G81, G86) was it found thatthe seeding of clouds that contained naturallyhigh ice particle concentrations (previouslybelieved to be confined to those clouds withecho tops <-21 C) produced any effect onrainfall. Therefore, in asserting that icemultiplication in Israeli clouds does not affecttheir cloud seeding potential, R contradictswhat the HUJ investigators stated for manyyears and in numerous publications, namely,that the presence of high ice particleconcentrations has a drastic effect on thepotential for increasing rain in Israel byglaciogenic static seeding. Further, R (1989)himself claimed there was no static seedingpotential in these cases.

In his Abstract, R states that we concludedthat there is no physical basis forglaciogenic seeding in Israel.

Because of natural ice multiplication, wedoubt that there is seeding potential for cloudswith top temperatures in the range for whichthe greatest seeding effects in Israel werereported by the HUJ investigators (viz., -12 to-21 C). In Rangno and Hobbs (1988) and inRH95, we suggested that the onset ofsignificant concentrations of ice particles inIsraeli clouds occurs near -10 C. If this iscorrect, a static seeding potential might bepresent in some clouds with tops warmer thanabout -10 C. However, the HUJ investigatorsconcluded that the rainfall produced by seedingsuch clouds is trivial (e.g., G86) because theseeding agent is ineffective at thesetemperatures. Moreover, in winter in Israel,clouds with top temperatures around -10 Care relatively thin (<3 km), it was argued, andthe growth of precipitation embryos is limited(e.g., G81, G86). However, we noted that theresults of seeding clouds in northwest flow at850 hPa appeared promising, as reported byGN74, since such seeding might sometimesinvolve clouds with low cloud basetemperatures and perhaps less likelihood ofnaturally high ice particle concentrations.

In summary, based on new physicalinsights and observations of themicrostructures of clouds in Israel, we (RH95)

suggested that a small seeding "window"might exist in Israel for clouds with toptemperatures between about -5 and -10 C, assuggested by statistical analyses of Israeli I andn (e.g., GN74, GN81, G86, GG87).

In his Introduction, R acknowledges that (asfirst pointed out by RH88 and Rangno 1988)cloud structures and the processes leading to

rain formation in Israel are more com-plicated than reported/or many years by theHUJ investigators. However, R believesthat seeding can still increase rainfall fromwintertime clouds in Israel, even though icemultiplication and droplet growth bycollision-coalescence occur in these clouds.

Rainfall in Israel is dominated by clouds inpolar air masses in which ice multiplicationand/or droplet growth by collision-coalescenceare active. We believe that rainfall from suchclouds is unlikely to be enhanced by cloudseeding because: 1) the clouds wouldnormally contain high ice particleconcentrations at relatively modestsupercoolings; 2) these clouds have littlecapability ofresponding "dynamically"because of their relatively low liquid watercontents; and, 3) the clouds are often cappedby strong stable layers that restrict their verticalgrowth and therefore dynamic seeding effects(e.g., Druyan and Sant 1978; Rangno 1988).

In his Section 2a, R expresses his convictionthat the potential for increasing rainfallthrough the seeding of clouds with naturallyhigh ice particle concentrations (due to icemultiplication) is no less than for low ice-producing clouds that were previouslythought to exist in Israel.

It was thought previously that clouds inIsrael had low ice particle concentrations andwere therefore ripe for seeding at cloud toptemperatures ^-21 C. Today, the evidencestrongly suggests that clouds over Israelfrequently contain high ice particleconcentrations, except perhaps at temperatures>-10 C, and that they can develop rain solelyby the collision-coalescence process.Rosenfeld’s view that these new findingshaving little impact on cloud seeding potential

in Israel is consistent with his refutation of theevidence for "lucky draws" in Israeli I and II.This is because it is not possible, on the onehand, to accept that the clouds in the primaryseeding "window" defined by the HUJinvestigators (i.e., cloud tops from -12 to -21C) have little seeding potential (because of thehigh natural ice particle concentrations nowknown to be present in these clouds) and, atthe same time, accept the hypothesis thatartificial seeding increased rainfall in the Israeliexperiments.

In his Section 2b, R questions the cloud toptemperature range assigned by us to theclouds shown in Fig. 9a and 9b of RH95. Heasserts that we deduced that haze waspresent from these photographs alone.

The cloud top temperature of "about -14C" that we assigned to this cloud at 1600 LTwas obtained as follows. From the height ofthe lifting condensation level (LCL),determined from the temperature and dew pointmeasurements of IMS coastal stations at 12and 15 UTC, the distance from the surface tothe base of the cloud shown in Fig. 9a and 9bof RH95 was estimated to be 0.5 to 0.6 kmASL. To be conservative in our estimate ofcloud top height, which we derived from thecloud base height estimate (e.g., Malkus andScorer 1955), we increased the cloud baseheight to 0.7 km ASL. The cloud thickness inFig. 9b of RH95 is 5.6 units of cloud baseheight, or 3.9 km thick. Adding the distancefrom the surface (0.7 km), results in a cloudtop height of 4.6 km ASL. The IMS soundingfrom Bet Dagan, Israel, at 12 UTC (actuallylaunched at about 1230 LT, or 3.5 h prior tothe photographs in question), shows thetemperature at 4.6 km to be -14 C (Fig. 11 inRH95).

Rosenfeld claims, erroneously, that hisradar data is for the same times as thephotographs shown in Fig. 9 of RH95. Ourphotographs were taken at 1556 and 1600 localtime, which were 8 and 4 mins earlier than R’sradar data, which was at 1604 LT (Fig. 1 inR).

Rosenfeld claims that the turrets associatedwith this cloud complex continued to grow (by800 m in 5 mins) after 1604 local time, from5200 to 6000 m ASL. At this rate of rise of the

4

cloud top, it would have been 640 m lower, orat 4560 m, at 1600 LT. This is 60 m lowerthan the height we estimated in RH, but thiswould have a negligible effect on thetemperature we estimated. Note that the top ofthis cloud was thus several degrees warmer at

1556 LST (Fig. 9a of RH95) when iceparticles were undoubtedly already developing.

For the purpose of perspective on thebehavior of this cloud relative to a radarclimatology in Israel compiled by Rosenfeldand Gagin (1989), we note that the thicknessof the cloud (^3.9 km by either R’s estimate orours) is far greater, and the cloud toptemperature is well below, the values forice/precipitation formation in Israel (e.g.,RH88; Rangno 1988; Gagin and Rosenfeld1989; Levin 1994; RH95; Levin et al. 1996).Figure 2 shows our estimate of the height ofthis cloud above the freezing level relative to

radar echo data in Israel from Rosenfeld andGagin (1989). It can be seen that it would havebeen extraordinary if this cloud had not shownsigns of converting to ice by the time it hadreached-14 C.

Our statement that haze was present on thisday was not based on our visual observationsalone. Restricted visibility (10 km), whichsatisfies RF92’s criterion for a "dust/haze"day2, was reported on 15 January 1986 by thecoastal stations of the Israeli MeteorologicalService (IMS) at 1200 and 1500 UTC (1400and 1700 local time)-also see Fig. 3.

In his Section 2c, R agrees that shallowclouds (with marginally supercooled tops)can produce precipitation in Israel.However, he asserts that these clouds areexceptions, because they have long lifetimesand tend to be more stratiform in character,and produce precipitation efficiently. Heasserts that artificial seeding of cumuliformclouds with shorter lifetimes might result inenhanced rainfall by decreasing the timerequired for the formation of rain.

This is an old (and still unproven) cloudseeding hypothesis, which R now applies adhoc to the Israeli experiments. On the onehand, there is lime doubt that small, isolated

2 Although the air carrying these clouds did not

pass over deserts-see Figs lOa and lOb in RH95.

clouds, which momentarily shoot turrets up to

heights where the temperature is -5 C orlower, and which often do not produceprecipitation naturally, can be seeded toproduce spectacular results-aloft (e.g.,Cooper and Lawson 1984). However, it isvery unlikely that such clouds could producethe rainfall needed to explain the statisticalresults of the Israeli cloud seedingexperiments, and, in particular, could not haveproduced the increase in runoff that Ben-Zvi(1988) has attributed to cloud seeding.

First, such clouds would rarely have beentreated at the key moment in their life cycle(before the liquid water disappeared) by theaircraft line-seeding method used in Israeli I.Second, clouds with lifetimes too short toproduce precipitation naturally are unlikely toreach temperatures as low as -12 to -21 C,which is the temperature range where strongseeding effects on precipitation were reportedby GN76, GN81, G81, G86, and GG87.

In RH95 we showed that the developmentof precipitation in the clouds of Israel can beextremely rapid (-10 min). Therefore, anyseeding "window" (assuming that it existsbefore high ice particle concentrations developin ascending turrets that reach temperaturesbetween -12 and -21 C) will be present onlybriefly. Is it likely that the aircraft line-seedingmethod used in the Israeli experiments, inwhich no attempt is made to target specificclouds, could have effectively treated cloudsthat naturally glaciate so quickly? We thinknot.

We agree with R that more airbornemeasurements, made in a variety of weathersituations over a substantial period of time,will be needed to define the degree of"seedability" of Israeli clouds, to answer thequestions we have raised concerning theefficiency of the aircraft line-seeding methodused in Israeli I, and to establish the viabilityof the "corkscrew" seeding trajectoriespostulated by R to explain possible inadvertentseeding of the BZ in Israeli I (see Section 3below).

In his Section 2d, R questions therepresentativeness of Levin’s (1994) andLevin et al.’s (1996) measurements onIsraeli clouds.

Temperature (C)

-5.5 -13 -20.5

Figure 2. Volume of rain derived from radar measurements in Israel during the 1982-83 rainfallseason as a function of maximum radar echo height above the freezing level (afterRosenfeld and Gagin 1989). The temperatures (shown at the top of diagram) werederived using the saturated-adiabadc lapse rate and a cloud base temperature of 8 C at

0.7 km ASL.

1 2 3 4 5 6 7 8 10 15Maximum Radar Echo Height Above Freezing Level (km)

Figure 3. Photographs taken on 15 January 1986 in windy conditions a) at1458 LT from Tel Aviv beach looking to the south and showinga surface haze layer, and b) at 1600 LT from Tel Aviv beachlooking to the west and showing crepuscular rays caused byaerosols below the bases of approaching stratocumulus clouds.

Apparently, Rosenfeld does not see theimport of the recent airborne measurements ofLevin (1994) and Levin et al. (1996). While Rhighlights how few measurements Levinobtained, he does not acknowledge howremote the chances would have been for Levinto obtain the measurements he did if the earlierreports by the HUJ investigators on themicrostructures of clouds in Israel werecorrect. Had Levin sampled thousands ofclouds, found only a few cases that differedfrom the earlier reports, and published thosefew cases, one would have reason to beconcerned about their representativeness. Butthis is not what happened. In only a fewmeasurements, Levin (1994) and Levin et al.(1996) found several key differences from theearlier reports. More importantly. Levin’s fewmeasurements are compatible with a globaldata set on "continental" cumuliform clouds(e.g., Rangno and Hobbs, 1988, 1995), andwith the descriptions of Israeli clouds given byRangno (1988).

In his Section 2e, R asserts that theprofound difference between Gagin’s cloudmeasurements and those of Rangno (1988)and Levin (1994) and (Levin et al. (1996)can be attributed to the fact that Gaginfocused his airborne sampling on youngclouds, and apparently had little regardforice particle concentrations that may havedeveloped later in the life cycle of theclouds.

Rosenfeld repeats the suggestion that wemade in RH88, and again in RH95, to explainwhy Gagin and his colleagues did not detectice multiplication in clouds in Israel.However, Gagin stated on numerousoccasions that ice multiplication did not occurin Israeli clouds without any qualificationsabout cloud life history (e.g., GN74; GN76;GN81; G75; G81; G86). For example, G75,in discussing the results of the few agingclouds he did sample, wrote: "The probableabsence of any significant time-dependent (ice)enhancement mechanisms.. .is furtheremphasized by the absence of any systematicor significant change of (ice) concentrationwith time."

In his Section 2f, R appears to contend thatonly dust/haze can produce icemultiplication andwarm rain in Israel, andthat dust/haze is limited to certain winddirections.

Many factors contribute to the large clouddroplets required for ice multiplication andwarm rain, including high cloud basetemperatures, low cloud condensation nucleusconcentrations, large cloud condensation nuclei(which may or may not include dust/hazeparticles), and cloud depth. Once clouddroplets attain diameters in excess of about 23urn, and reach slight to moderatesupercoolings (>-20 C), ice particleproduction can be prolific (e.g., Mossop 1970;Mossop et al. 1967, 1968, 1972; Ono 1972;Hobbs and Rangno 1985. 1990; Biyth andLatham 1993), particularly when precipitation-sized drops are present (e.g., Chisnell andLatham 1976; Lamb et al. 1981).

With cloud bases averaging 5-10 C inIsrael, there is no reason to believe that largedrops, and therefore high ice particleconcentrations, require the presence ofdust/haze and certain wind directions. Forexample, we have not found in the literature acloud droplet spectrum for a continentalcumulus cloud with base temperature ^5 Cthat did not have a droplet spectrum conduciveto the production of high ice particleconcentrations at 2-3 km above cloud base(which is typical of where the riming-splintering zone is located in Israeli clouds).

Some of the most spectacular cases of highice particle concentrations that have beendocumented occurred in very clean air (e.g.,Mossop etal. 1968; Hobbs and Rangno 1985,1990). Further, Rangno (1988) described acase in Israel when rain was produced by thecollision-coalescence process in clouds on anearly calm day and ice formed in clouds withtops at -9 C embedded in a northwest flow.Levin (1994) reported ice particleconcentrations of 10 per liter in clouds withtops at -10 C in northwest flow. In none ofthese cases was it likely that dust/haze from thedeserts of southwest Israel affected the clouds.

In his Section 7a-c, R restates his belief inthe dust haze hypothesis of RF92 and itsrole in decreasing rainfall in the STA. He

6

cites several papers to support his assertionthat dust/haze has a profound influence oncloud microstructures in Israel.

If dust has a profound effect on iceformation in clouds (in essence, providingnearly instant glaciation and leaving no seeding"window"), why was not such a dramaticeffect on clouds detected long ago, particularlyin Israel (where measurements of clouds weremade over many years by Gagin and hiscolleagues) or in the numerous measurementsof ice nucleus concentrations and cloudstructures that have been made throughout theworld?

Are the apparent decreases in rainfall inJordan, and throughout Israel and Lebanon, onSTA seeded days in Israeli n (Fig. 4) reallydue to seeding on dusty days in the STA, as Rwould have us believe, or are they part of awidespread natural rainfall pattern that pro-duced an illusion of decreases in rainfall due toseeding on STA days (and increases on NTAdays)? We believe the answer is obvious.

In his Section 7a-c, R challenges ourstatement that there is no evidence in RF92that dust was carried into the clouds inIsraeli II.

In RH95 we questioned whether dust/hazeobserved at Elat in the extreme south of Israelmeant that clouds that precipitated onJerusalem, or on Lake Kinneret 12 h later,were affected by this dust. (In R’s view, ifany one of 13 IMS observing stations reportdust/haze at any time of the day, that dayqualifies as one in which precipitating cloudsare affected by dust.) Neither in RF92, nor insubsequent papers, is there any evidence thatthe clouds on any particular day of Israeli 11were affected by dust/haze.

On the other hand, we do not doubt thatthere are some days in Israel when clouds docontain dust particles. But do these particleshave a significant effect on cloud micro-structures and rainfall? Do they occur in rain-bearing clouds with long fetches across theMediterranean Sea, which typifies mostshowery periods in Israel (e.g.. Levin et al.1996)? Are restrictions to visibility alwayscaused by dust/haze? Are dropletconcentrations in clouds on dust/haze days

comparable to days without dust/haze? Does afew hours of seeding really cause significantdecreases in rainfall from such clouds? Untilanswers are available to such questions, thedust/haze hypothesis will remain just that: anunproven, if interesting, hypothesis.

Also, consider the following implicitaspects of R’s postulates about the clouds ofIsrael. While rejecting the pre-1988 reports oncloud microstructures in Israel by Gagin andhis colleagues, R postulates (implicitly) thatthere are two processes that cause high iceparticle concentrations in Israeli clouds. In thefirst type, which might be thought of a "good"ice multiplication (because, according to R, itleaves a "window of opportunity" for a rainfallenhancement caused by a dynamic-likeresponse, albeit in clouds seeded via the staticmethod!), there are high ice particle concentra-tions at cloud top temperatures <-12 C, but fora few minutes as these clouds ascend above the-10 C level, they have sufficiently low iceparticle concentrations for silver iodide to beeffective in accelerating and enhancing rain.For example, R believes that in Israeli I theseeding agent reached sufficient clouds duringthis narrow seeding "window" to have causedstatistically significant increases in rainfall,including regions that were not targeted forseeding, and in which line seeding occurred onan average of only 4 h per day. We are notconvinced.

The second type of ice multiplicationpostulated by R might be thought of as "bad"ice multiplication (because when it occurs Rbelieves that seeding with silver iodide reducesprecipitation, even in clouds with moderatelysupercooled tops). According to R, this type ofice multiplication occurs when precipitatingclouds are present in northern and central Israeland dust/haze is reported in (usually) the dryportions of southern Israel. However, rainfalland dust/haze are usually not co-located eitherin time or space.

Rosenfeld envisions that dust/haze fromthe southern deserts of Israel affects clouds tothe north in an extreme way: ice particles formwith such rapidity and in such high concentra-tions, even in moderately supercooled clouds(tops ^-21 C), that glaciation is virtuallyinstantaneous. Therefore, there is no seeding"window" even in the building stages of theseclouds. In fact, because so many more of thesesituations occur in central Israel, R believes

that rainfall was decreased on seeded days inthe CTA and the STA of Israeli I and H,respectively, due to "overseeding".

Rosenfeld and Farbstein (1992) did notreconcile their conclusions regardingoverseeding of clouds with the fact that cloudtops in the STA average only -16 C (GN74),3C higher than in the NTA. Nor did theyaddress why G75 asserted that the clouds ofIsrael were impervious to overseeding,regardless of cloud top temperature.

In his Section 7e, R states that severalfactors prevented the HUJ investigatorsfrom acknowledging or learning that theclouds of Israel frequently exhibit high iceparticle concentrations and, on occasions,rain formation via the collision-coalescenceprocess. He asserts that "the time available

for glaciation", among other things, mayhave been why these researchers did notdetect high ice particle concentrations andwarm rain occurrences.

Rosenfeld states that because high iceparticle concentrations tend to occur during themiddle and later stages in the life cycle of acloud, this may have prevented the HUJinvestigators from observing that clouds inIsrael precipitated at much shallower depths,and at significantly warmer cloud toptemperatures, than they reported. Apparently,these researchers failed to notice the dramaticeffects of high ice particle concentrations, eventhough they sampled more than 100 cloudswith an instrumented aircraft (e.g., GN74;G71; G75), used an aircraft to verifyvertically-pointed X-band wavelength radarmeasurements of the tops ofprecipitatingclouds over Jerusalem (e.g., G80; Rosenfeld1980), and used a powerful C-band radarbeginning in the 1970s to assess cloud topheights (see Section 2.2 of R’s Comment).Rangno (1988), on the other hand, detectedthis phenomenon visually from the ground andthrough the use of rawinsondes. Also, afterjust a few flights, it was readily apparent toLevin (1994) that high concentrations of iceparticles were common in clouds over Israel.Further, Fig. 2 of this Extended Reply (whichis adapted from Rosenfeld and Gagin 1989)reveals most of these phenomena.

In his Section 8, R discusses the radar-inferred cloud top temperatures in Israeli IIand cloud seeding effects. He asserts thatthe consistency between those studies is agood argument for believing thatstratifications of seeding effects in Israel bycloud top temperatures determined fromradar are valid.

In RH95 we questioned the reliability ofcloud top temperature data for Israeli n derivedfrom radar by the HUJ investigators. Gaginand Neumann (1981), for example, stated thatthey measured the top of every cell in the NTAof Israeli n. In fact, they could not have doneso because the radar was too far from the NTAto measure all of the cloud tops in that region(personal communication, 1994, from KarlRosner, Chief Meteorologist for Israeli n).Also, as noted by GN76, a "modal" radar toptemperature3 of-15 C assigned to an entireday, can be misleading if, for example, all ofthe rainfall fell from a band of clouds with atop temperature of -24 C. This problem hasnever been satisfactorily resolved, nor is itnow by R.

We also note that the radar used by theHUJ investigators was installed at Ben GurionAirport in the late 1970s (e.g., G80; Rosenfeld1980). Based in part on observations madewith this radar, G80, G81, G86, and GG87concluded that rain did not develop either bycollision-coalescence or by the ice process inclouds in Israel with tops warmer than about-14 C, a conclusion now known to beincorrect. Perhaps this radar was improperlycalibrated.

3. Responses to Rosenfeld’s specificcomments concerning seeding andseeding logistics in Israeli I

In his Section 3b, R postulates that thepercentage of suitable clouds seeded inIsraeli I was about 60% (rather than 25-35%estimated by RH95) and that this wasenough to produce statistically significantmodifications in rainfall. He bases his

3Modal cloud top temperatures are the radar-derivedcloud top temperatures used by the HUJ investigatorsto stratify seeding effects in the NTA of Israeli II.

estimate on two claims: 1) that the hours ofrainfall per season from clouds in Israel withtops warmer than -5 C (which areunresponsive to seeding) is significant, andthe inclusion of these clouds in ourestimates of hours of "showery weather"inflated our estimates of the clouds perseason that were suitable for seeding; and,2) that the HUJ investigators knew that rain

fell from clouds with temperatures >-5 Cand they did not seed these clouds.

Rosenfeld’s claims raise several interestingissues. They are apparently based onGabriel’s (1967) statement: "The aircraft takesoff whenever cloud conditions appearfavorable, but seeding is carried out only afterthe cloud seeding officer has ascertained thatcloud tops reach or exceed the -5 C level." Asdiscussed by RH95, this statement could nothave been true for many of the actual cloudseeding events in Israeli I. Consider, forexample, the case ofnighttime seeding.According to the Chief Meteorologist forIsraeli I (personal communication, 1994, fromKari Rosner, Chief Meteorologist for Israeli I),the seeding aircraft was sent up and seedingtook place at cloud base whenever rain wasreported during nighttime hours. It waspresumed that the precipitating clouds wereappropriate for seeding. Clearly, it was not

possible for a cloud seeding officer aboard theaircraft at night to ascertain whether or notsome cloud tops rose to the -5 C level. Also,in convecdve situations, cloud top heights varygreatly, and undoubtedly some "unsuitable"clouds were seeded as the aircraft flew backand forth at cloud base along the seeding track.Further, the data of Rosenfeld and Gagin(1989) for the 1982-83 rainfall seasonsuggests that rain from such clouds comprisesa small fraction of convective rain events inIsrael (see Fig. 2 of this Extended Reply).

In view of the many requirements for"suitable" clouds for seeding, we believe thatour estimate that 25-35% of the seedingmaterial was released in suitable cloudconditions is, if anything, generous, and thatseeding, as carried out in Israeli I, wasunlikely to have produced a statisticallysignificant effect on rainfall in the target areas.Moreover, we believe that it is virtuallyimpossible that on the few occasions when the

Buffer Zone (BZ) may have been inadvertentlyseeded that this could have caused the highrainfall that occurred in the BZ, which showedthe most significant statistical increases inrainfall on days when the CTA was seeded(e.g., Wurtele 1971).

In his Section 3c, R states that we ignoredthe paper by Gagin andArroyo (1985)concerning the dispersion of seedingmaterial in Israeli II. He claims that thispaper shows that the seeding coverage inIsraeli I was considerably greater than weestimated.

In the RH95 discussion of seedingefficiency, to which R refers, we confinedourselves to Israeli I for which sufficient datahave been published to make an assessment.Gagin and Arroyo’s (1985) paper (hereafterGA85) is concerned with Israeli n. Theairborne line-seeding path used for the NTA ofIsraeli II was considerably shorter in lengththan the path used in Israeli I (54 km versus 75km). Further, GA85’s estimates of thedispersion of the seeding agent are based onseveral questionable assumptions. Forexample, they assume that the aircraft flewunder a continuous updraft for the entire 54km, and that this updrafts transported theseeding agent straight up to the -15 C levelwhere it dispersed downwind across the targetin an enormously wide plume relative toGaussian model plume predictions. It isapparent from R’s own radar studies ofconvection that it is unlikely that the seedingmaterial was carried up in a continuous updraftalong an entire seeding track (see Fig. 1 of R’sComment and Fig. 3 of the present Reply).

Even with these optimized assumptionsconcerning the dispersion of the seeding agentin Israeli n, GA85 found that line seeding inthe NTA could have affected only half of thevolume of air -15 C. Thus, even if thegenerous assumptions concerning dispersionmade by GA85 are applied to Israeli I, thefraction of the air affected by the seeding agentwhen the aircraft was flying at -15 C isreduced to 25% simply due to the longerseeding track (by 50%) used in Israeli I.We note also that, in apparent recognition

of the poor efficiency of seeding in Israeli I, asecond aircraft and a network of ground

seeding generators was added in Israeli n(GN81) and, in today’s operational cloudseeding program in Israel, two aircraft flysimultaneously in opposite directions.

Rosenfeld has also misunderstood what wedid. There are two seeding scenarios in RH95:1) when the aircraft is line seeding betweencumulus cloud updrafts and the seedingmaterial is not transported dtrecdy upward(i.e., the sky overhead is clear, or contains astratiform cloud), and 2) when the aircraft isbeneath the base of a cumulus cloud andreleases silver iodide into the updraft feedingthe cloud. Recall that the aircraft almostalways flew in a straight, prescribed path,rather than performing "sorties" in search ofcumulus updrafts.

Our first calculation was for the firstscenario (no updraft). Therefore, we used aGaussian plume dispersion model. Toexaggerate the vertical dispersion, we assumeda slightly superadiabatic lapse rate.Rosenfeld has confused this calculation withour discussion of dispersion when a seedingagent is released in the updraft at the base of aconvective cloud. We concluded that since theseeding aircraft flew under a cloud just once,not more than one or two turrets (or evenportions of turrets) could have been seeded.

At night the seeding aircraft was directedtoward regions of precipitation detected byradar (personal communication, 1995, fromKarl Rosner, Chief Meteorologist for Israeli I).Yet, as R acknowledges, if a cloud is alreadyprecipitating it is probably too late for seedingto have any affect on rainfall.

In view of these many factors, we standbyour conclusion that it is doubtful that enough"suitable" clouds were seeded in Israeli I tosignificantly affect rainfall.

In his Section 2c, R asks us to consider thecase of line seeding from an aircraft/lyingupwind of orographic clouds in Israel.

Line seeding at cloud base is potentially aviable method for intermittently seedingstanding orographic clouds, providing theaircraft track is short. However, even thisscenario has problems in Israel. First, most ofthe hilly terrain is in the eastern portion of thecountry. Therefore, even if the stationaryclouds that form over these hills were deep

enough for ice formation to occur (tops >3-4km ASL on most rain days), there would notbe sufficient time for ice crystals nucleated at<-10C to grow and fallout in Israel. Atemperature of -10 C is referred to herebecause, according to G81 and GN81, cloudswith tops warmer than -10 C, and certainlythose clouds with tops warmer than -5 C,cannot be seeded effectively with silver iodide.

Standing orographic stratocumulus cloudsdo, in fact, form frequently over the JudeanHills and in the hilly regions in the north ofIsrael. However, since these clouds areformed by uplift over hilly regions that are <1km in elevation, they are not very deep, andoften topped by strong stable layers (Rangno1988). Such clouds rarely reach -10 C.

Also, much of the rain that falls in the hillyregions in northern and eastern Israel derivesfrom cumulonimbus complexes, whichoriginate upwind over the Mediterranean Sea.When these clouds pass over Israel they aregenerally in their mature and decaying stages,and contain high concentrations of ice particlesat relatively small supercoolings. These cloudsmay also overrun shallower orographic clouds,where accretion and riming can increaserainfall. While temporarily over the hillyregions, such dissipating clouds could bemistaken for simple orographic clouds.

In his Section Id, R states that our estimate

of the amount of rain that would have had tohave been produced by seeding in Israeli Iis too large by a factor of two because wedid not consider control days. He claimsthat for Israeli I we calculated that about1,000,000 m~^ of water would have had tohave been produced by each gram of silveriodide.

Nowhere does the number 1,000,000 m-3appear in RH95. We estimated that 500,000m’3 of water would have had to be producedby seeding in Israeli I to explain the amount ofrainfall attributed to seeding by GN74 in Israeland by Brier et al. (1973) in Syria and Jordan.Our number was derived by assuming that 1%of the seeding material reached "suitable"clouds, namely, those clouds with toptemperatures from -12 to -21C but not in theirmature or dissipating stages. However, if onewere to assume that RF92’s report of

10

decreases in rainfall of 10-30% due to seedingon dust/haze days actually occurred, even ourestimate of 500,000 m-3 per gram of silveriodide would be conservative because thesilver iodide that entered "suitable" clouds ondays without dust/haze would have had tomake up for the decrease in rain caused byseeding on dust/haze days!

In his Section 4b, R questions ourcharacterization of the seeding efforts inIsraeli I as "wasted".

The word "wasted" does not appear inRH95. It has to be recognized however that inany seeding operation forecasts can go awry:conditions that might be suitable for seedingare not forecast correctly, and forecasts ofsuitable conditions are not fulfilled; also,aircraft downtimes occur. These realities oflife reduce the efficiency of any seedingproject. Rosenfeld implies that the seedingoperations in the Israeli experiments wereperfectly executed, with the seeding materialrarely released in unsuitable conditions.

In his Section 4e, R recounts someobservations concerning cloud base heightsover the Mediterranean Sea, and a shallow,offshore-flowing wind regime from theinterior of Israel that can develop over thecoastal plain. He posits that a zone ofenhanced convection, which can occur

offshore of Israel in some of thesesituations, could have been seeded. Hebelieves this explains how the BZ wasinadvertently seeded on a routine basis inIsraeli I and that it accounts for thestatistically significant greater rainfall inthe BZ on CTA seeded days in Israeli I.

Because of the troublesome bias in rainfallon CTA seeded days in the BZ and along thecoast of Israel in Israeli I, R has proposed acomplicated scenario to explain why the BZmay have been seeded. By contrast, weattribute the bias in rainfall in the BZ on CTAseeded days to an uneven draw in naturalrainfall that led to the misperception of aseeding effect.

Complicated scenarios, such as the oneproposed by R, have a number of steps that

have to be fulfilled to be valid. For example,each of the following nine steps would have tobe realized before the scenario postulated by Rcould produce rain due to seeding in the BZand other coastal locations in Israel:

1) Silver iodide is released at cloud base(about 0.8 km ASL) in westerly,onshore flow.

2) Turbulence at the release level causesthe silver iodide plume to deepenrapidly downward (as well as upward).

3) The silver iodide plume reaches downto levels at or very near the surfacebefore reaching the Judean Hills, about40 km inland (otherwise it would becarried over those hills and not returnto the west).

4) An ESE wind would have to exist atand just above the surface on thecoastline, and offshore, for anappreciable distance; also, for anappreciable distance inland.

(In the afternoon, and in strongpressure gradient situations inpostfrontal periods when rain oftenoccurs in Israel, there is rarely a windwith an easterly component at the coast(Neumann 1951). Also, of the 4 h perday that seeding was carried out inIsraeli I, fewer hours occurred at night(Gabriel and Neumann 1978; RH95)when an ESE wind is more likely to bepresent. Wind rose data for theIMS rawinsonde site at Bet Dagan,which is located about 7 km from thecoastline, is shown in Fig. la. It canbe seen that when rain falls at this site,or within about 60 km as determinedfrom IMS observing stations, there is awind from the southeast quadrant lessthan 15% of the time!)

5) Rosenfeld assumes that most of theseeding agent affects cloudsdownwind. However, he alsosuggests that, to the east of the seedingline, a portion of the silver iodideplume filters downward into theshallow easterly flow at the surface,and then "corkscrews" to the west andnorthwest while rising back to cloudbase. Silver iodide must not only becarried back to the coast by thisshallow flow, it must be carried far

11

enough offshore to the west to enhancerainfall in the BZ.

(How far upwind of the BZ mustthe silver iodide be carried? Assuminga 5 m s"1 updraft, 15 min forprecipitation formation after the silveriodide enters cloud base, and 15 minfor the precipitation to reach theground, silver iodide entering a cloudbase at 0.8 km ASL would requireabout 40 min to rise, formprecipitation, and fall to the ground.For a westerly wind of 15 m s’1between cloud base and the 600 hPalevel (where temperatures typically firstgo below about -10 C on showerydays in Israel), the silver iodide wouldhave to enter clouds that are movingtoward the BZ at a distance of no lessthan 25-30 km west of the Israelicoastline. However, east winds at thesurface, which might carry the silveriodide westward, are rarely greater than5 m s"1. At 5 m s’1, it would take thesilver iodide in this layer about 1 ^hours to reach a location far enoughupwind to have the possibility ofcreating ice particles aloft that couldfallout as rain in the BZ.)

6) During its 1 ^ hour drift to the west,the shallow easterly flow must not bedestroyed by convection currentsarising from the wanner MediterraneanSea.

(Rosenfeld does not address whythe "corkscrew" seeding phenomenonhe postulates for Israeli I did not affectIsraeli II, where ground generatorswere more likely to inject seedingmaterial into the easterly surface flow.)

7) If the silver iodide arrives at aappropriate point upwind of the BZ,clouds suitable for seeding must bepresent; these clouds must be largeenough to precipitate, but not so largethat sufficient concentrations of naturalice particles ’are present for rainfallproduction. Cloud top temperaturesupwind of the BZ must, according tothe HUJ investigators, be between -12and -21 C, a span ofjust 1 km incloud top height!

8) The seeding plume that is carried intothe upwind clouds must not only be

ingested by clouds with appropriatecloud top temperatures but also,according to R, the clouds must be atan appropriate stage in their life cycle(i.e., the young, building stage), andbe seeded with the optimumconcentrations of silver iodideparticles, to enhance rainfall.

9) Finally, according to R’s dust/hazehypothesis, this inadvertent seedingmust not have occurred on the majorityof days in the CTA in which rain wasdecreased because of the presence of asuper ice-nucleating dust/haze aerosol.

If we assume (generously) that there is an80% chance of independently fulfilling each ofthe above requirements, the probability that allof the requirements would simultaneously bemet on any one occasion (which R’s scenariorequires) is about 13%. Hence, we do notbelieve R’s scenario is viable.

4. Response to Rosenfeld "s specificcomments concerning the statisticalevaluations of the Israeli experiments

In his Abstract, R asserts that "the results(of the Israeli seeding experiments) showsignificant positive effects in northernIsrael."

This statement is misleading. First, torecap the published results of the Israeliexperiments to which R alludes: in Israeli I,increases in rainfall due to seeding weresuggested in both targets, but with muchgreater increases in the CTA (e.g., Gabriel1967; Wurtele 1971; GN74). In Israeli H, onecan accept an unambiguous statisticallysignificant increase in rainfall due to seeding inthe NTA only if one ignores the well-established fact that naturally heavier rain fellregionally in Israel on NTA seeded days (asfirst reported by GR90), and that these heavierrains extended into Lebanon and Jordan(RH95). If there is a glimmer of hope thatthere was a seeding effect in Israeli n, it restson the attempt by RF92 to use a numericalmodel, based on rawinsonde profiles, toaccount for the naturally heavier rainfall onNTA seeded days.

12

On the other hand, if the rainfall in the non-seeded target is used as the control rainfall inIsraeli I (as GN74 recommended for Israeli II),the results for Israeli I (from Gabriel and Baras1970 andWurtele 1971) are strikingly differentfrom previously published results, and they dosupport R’s contention concerning Israeli Ibecause the double ratio for the NTA is 1.28(and clearly significant), while the result forthe CTA is 1.03 and insignificant. However,if one accepts the large double ratio in the NTAas due to seeding, one also has to accept thatseeding effects were produced at coastlinelocations that were situated virtually under thepath of the line seeding, and were produced byvery few hours of seeding. We believe thatthese statistical results are furthermanifestations of a Type I error (i.e., luckydraw) for the NTA of Israeli I.

In his Abstract, R states that in his viewIsraeli I confirmed Israeli II.

Rosenfeld does not state which result wasconfirmed: positive, negative, null, or allthree, all of which have been suggested by theHUJ investigators! For example, R states thathe now believes that seeding decreasedrainfall in the center target area (CTA) of IsraeliI due to dust/haze that affected the clouds onmost CTA-seeded days. This is a new claim,which we find particularly interesting in viewof the earlier analyses of Israeli I (e.g.,Wurtele 1971; GN74) which suggested that themain effect of seeding was to increase rainfallin the CTA more than in the NTA.

In his Abstract, R states that a target/controlanalyses of the operational seeding programin northern Israel (which followed Israeli IIand is still continuing) shows significant(-6%) increases in rainfall due to seeding.

While the results of a non-randomizedcloud seeding experiment can be encouraging,they are rarely scientifically convincing. Littleis known about the operational cloud seedingprogram in Israel. We do not know, forexample, how much seeding material is usedor for how long. Are one, two or threeseeding aircraft used? Do they flysimultaneously? How many ground-based’silver iodide generators are there? On what

days has seeding occurred? What radar is usedto infer cloud properties? What are the effectsof storm types on the analysis? What are thecriteria for a seeding operation? Why are theapparent effects of seeding on rainfall so smallcompared to those reported by GN81?However, perhaps what is most needed is acomprehensive and independent evaluation ofthe results of the operational cloud seedingprogram in Israel, along with comprehensivecloud measurements.

Other evaluations of the effects of thecurrent operational cloud seeding program inIsrael have produced differing results.Benjamin! and Harpaz (1986) found nostatistically significant evidence for increasedrunoff in the target area around Lake Kinneret.However, Ben-Zvi (1988) did. The findingsofBen-Zvi (1988), and those ofNirel andRosenfeld (1995) who reported a 6% increasein rainfall due to seeding, are dependent, tovarying degrees, on the use of control datafrom the northern coastal strip and from theplains regions of Israel, rather than on rainfall(or runoff) from nearby areas in similar terrain,such as the central hill region adjacent to theoperational target area, or from southernLebanon, both of which experienced heavierthan normal rainfall on NTA seeded days andwhose rainfall is highly correlated with theoperational target.

Rosenfeld states in his Abstract that there ismounting evidence that desert dust isresponsible for the apparent differences inthe effects of seeding on rainfall in northand south Israel.

Rosenfeld describes the results of cloudseeding in Israel as though they are uniformlythe same (viz, increases in rainfall in the NTAand decreases in the STA due to dust/haze).Rosenfeld believes that the statistical results ofIsraeli I were misanalyzed by Gabriel andBaras (1970), Wurtele (1971) and GN74 (i.e.,heavier rain in the CTA and BZ wereincorrectly attributed to seeding), and that hisown interpretation is the correct one.Rosenfeld first raised this interestinghypothesis in 1989; we recommend that hesubmit his reanalysis of Israeli I for formalpublication.

13

In his Introduction, R states that two "luckydraws" (for Israeli I and II) could not haveoccurred.

First, lucky or unlucky draws seem to turn

up rather often in cloud seeding experiments(e.g.. Brier and Enger 1952; Lovasich et al.1971; Gelhaus et al. 1974; Hobbs and Rangno1978; Rangno 1979; Mieike 1979; Grant et al.1979; Nickerson 1979), particularly whentarget and control station are not specified inadvance.

Second, had the original statistical designsfor Israeli I and n been strictly followed, withall of Ae target and control stations listed inadvance and adhered to, two "lucky" draws ina row would have, indeed, been unlikely,although not impossible. However, it isarguable whether two "lucky" draws in a rowactually occurred in the Israeli experiments.From the standpoint of the CTA and STA ofthese experiments, R could just as well arguethat there was a "lucky" draw in Israeli I(apparent increases in rainfall due to seeding)followed by an "unlucky" draw in Israeli II(apparent decreases in rainfall due to seeding).Such ambiguity in possible interpretations ofthese experiments is evidence of problems inthe statistical draw.

Rosenfeld is not disturbed by the fact thatin Israeli I Ae statistical analyses indicate thatAe greatest effect of seeding occurred in thebuffer zone (BZ), even though the BZ wasprobably seeded for only a few hours eachrainy season (e.g., Wurtele 1971; RH95).Further, Ae greatest statistical significance forIsraeli I found by GN74 (to which R refers)was enhanced by Ae inclusion of the BZ,which was not only designed not to be seededbut has been stated on many occasions by AeHUJ investigators not to have been seeded(Gabriel 1967; GN73; GN81; GG87). Yet,Ae addition of rainfall data from Ae BZimprovedAe statistical results Aat suggestedseeding increased rainfall in Israeli I. Webelieve Aat Ais result, when combined wiAAe oAer evidence presented by RH95, showsAat Israeli I was compromised by naturalrainfall patterns Aat led to Ae misperception ofseeding effects. This supports R’s (1989) ownstatement: "It is not likely Aat Ae buffer area(of Israeli I), only due to inadvertentcontamination, will be positively affected twice

as strong (1.32 and significant) as the actualcenter target area (1.16 and insignificant)."

A statistically significant result indicatingincreases in precipitation due to artificialseeding was achieved in Israeli n only when alarge portion of Ae data was omitted (Aat forAe STA) in Ae many analyses prior to GR90,and Ae published analyses were confined toAe NTA using a target/control raAer Aan acrossover evaluation (e.g., GN76; GN81;G81; G86; GG87). The seeding effect in AeNTA disappears when the STA is used as acontrol on NTA seeded days, which was anoriginal requirement of Ae statistical design inorder to avoid meteorological biases (e.g.,GN74).

There is no question Aat Ae target/controlanalyses for Israeli n shows that rainfall on AeNTA seeded days (in which a seeding effecthas been claimed by Ae HUJ investigators)was unusually heavy over a wide region (e.g.,GR90; RH95). How could it have beenconcluded Aat seeding increased rainfall inIsraeli II in Ae face of such clear evidence AatAe experiment was compromised by a luckydraw? As a starting point, exactly Ae samecontrol and target stations should have beenused in Israeli II as were used in Israeli I. Ouranalysis shows Aat Ae supposed effect ofseeding on rainfall in Ae NTA (reported byGN81) may have been dependent, in part atleast, on the post-factum choice of controlstations in a small coastal strip that hadanomalously low seed/no seed ratios (see Fig.Ib of Ais Reply and Fig. 17 of RH95).

In his Introduction, R emphasizes his beliefthat the statistical results of the Israeliexperiments provide sufficient plausibilitythat seeding increased rainfall since theywere "black, box" experiments.

We have already pointed out Aat Aevarious statistical analyses ofAe Israeliexperiments have not yielded consistentresults, and Aat Aey are subject to variousinterpretations. For example, in our view, Aestatistical results for the BZ of Israeli I alonemake that experiment ambiguous. If onebelieves Aat the BZ, which exhibited Aestrongest statistical result in Israeli I, waslargely unseeded when Ae CTA was seeded,Aen a lucky draw (false positive) did occur for

14

the CTA. On the other hand, if one believesthat the BZ was routinely inadvertently seeded,then seeding may have been responsible for theheavier rain in the CTA on seeded days.However, in doing so, one would have to:

1) accept that a very poor job was done oftargeting seeding effects in Israeli I; 2) rejectthe assessment of the Chief Meteorologist forIsraeli I that seeding could have affected theBZ only "5-10% of the dme" (Wurtele 1971;RH95), together with a similar conclusion byRH95 based on a more detailed analysis of tenseasons of data; 4) reject the statements of theHUJ investigators that the BZ was not seeded(e.g., Gabriel 1967; GN81); 5) ignore theanalysis of GN74, who found that the BZ hadeven more rain on CTA seeded days whenseeding was not carried out; and, 6) explainwhy there was not a similar effect when theSTA was seeded in Israeli n.

In his Section, 4c, R contends that thepossibility of a lucky draw in Israeli I is asremote as the tests of statisticalsignificance suggest. He further asserts thatsince we have accepted the null results ofthe crossover analysis of Israeli II, weshould accept the crossover analysis ofIsraeli I, which suggested a 15% increase in

rainfall on seeded days overall.

Were it not for the strong statisticallysignificant increases in rainfall in the BZ(which was supposed not to have beenseeded), a natural bias indicating heaviernatural rainfall on seeded days in Israeli coastalzones too close to the line of seeding to havebeen affected by seeding, a similar naturalrainfall bias that favored rainfall on seededdays in extreme southwest Jordan (Brier et al.1973), and the lack of a sound physical basisfor expecting that seeding might increaserainfall in Israel, there would indeed be noreason to doubt the crossover results of theIsraeli I experiment.

Are we inconsistent in accepting the nullresults of Israeli n crossover evaluation whilededucing that the crossover results of Israeli Iwere impacted by a storm bias? The problem,as we see it, is that the results of these twoexperiments are inconsistent In Israeli I,seeding appeared to have increased rainfall themost in the southernmost target; in Israeli II,

nearly statistically significant decreases inrainfall due to seeding appeared to haveoccurred in the southernmost target (GR90;RF92). Thus, even if one accepts prima faciethe outcome of Israeli I (and ignores thewarning signs presented by the BZ), onewould conclude that the crossover results ofIsraeli I were not confirmed by the crossoverresults of Israeli n.

The types of storms that affected these twoexperiments were also different In Israeli I,the rainfall patterns on seeded days were morelocalized and contained within small regions(Brier et al. 1973; Rosenfeld 1989). Thesepatterns favored the perception of a strongpositive seeding effect in the CTA of Israeli I,although heavy storms in the CTA had littleeffect in the NTA (Rosenfeld 1989).

Israeli n, on the other hand, wasdominated by storm conditions on seeded daysthat affected larger regions. For example,when heavy rain fell on NTA seeded days, therain was also often heavy on those same daysin the STA (GR90). We extended this findingto as far north as Beirut, Lebanon, andsouthern Jordan (RH95). Thus, any increasein rainfall that might have been produced by.seeding of the STA was masked by the heavyrain that fell on the control days for the STA(which were the seeded days for the NTA).

On the other hand, in Israeli II, R(mistakenly in our opinion) interprets vagariesin the random draw as increases in rainfall (inthe NTA) and decreases (in the STA) due toseeding.

In his Section 4c, R claims that Wurtele’sfindings concerning the BZ, are "irrelevant".

We disagree. Wurtele’s enigmatic findingsare as important today as they were when firstpublished. The results she found for the BZcan now be seen as a "red flag," which shouldhave raised further questions about the HUJinvestigators’ statistical analyses of Israeli I.

In his Section 4d, R posits a scenario thatappears to resolve the vastly disparateresults of Israeli I and II. He believes thatin both experiments seeding decreasedrainfall in the CTA and STA (contrary to

several previously published analyses). Healso claims that the increase in rainfall in

15

the NTA due to seeding in Israeli I wascloser to 40% (instead of the 8% increaseclaimed by GN74) and that there was a 7%decrease in rainfall in the CTA (instead ofthe 22% increase claimed by GN74).

Rosenfeld’s speculations are interesting.He believes that the previously publishedresults for the CTA of Israeli I are incorrect.Whereas, Gabriel (1967), Wurtele (1971),Brier et al. (1973), and GN74 all concludedthat most of the apparent increases in rainfalldue to seeding occurred in the CTA, R thinksthat seeding produced an overall decrease inrainfall in the CTA.

However, R’s speculations are based on amajor flaw. As pointed out above, and byRosenfeld (1989), the storms that affectedIsraeli I and n were quite different both in theirtype and regional extent. For example, inIsraeli I the storms on CTA seeded days didnot affect most of Israel, Jordan and Lebanon,as they did on NTA seeded days in Israeli n(Brier et al. 1973; Rosenfeld 1989).

It is worth repeating R’s own conclusionsin this regard (Rosenfeld 1989): "Theoccurrence of heavy rain days was muchgreater in Israeli 1 as compared to Israeli 2.Israeli 1 had 19 days with more than 40 mm inone or both of the target areas, as opposed toonly 3 such days in Israeli 2. The nature of thedaily rainfall is remarkably different betweenthe two experiments." In his Table 10,Rosenfeld (1989) notes that for Israeli I theelimination of the heavy rain days causes littlechange in the apparent increases in rainfall dueto seeding in the NTA, but it causes theapparent increase in rainfall due to seeding inthe CTA to diminish by 23%. Thissubstantiates our point that the heavier rains inIsraeli I were not evenly distributed betweenthe two target areas.

Hence, inferences about the rainfall in theNTA of Israeli I based on the rainfall in theCTA are misleading. For example, using acluster of stations in extreme southernLebanon, Brier et al. (1973) found no evidencethat the rain was heavier in southern Lebanonon CTA seeded days in Israeli I. Hence, thestronger storms on CTA seeded days did notextend across the NTA to southern Lebanon,as would be required by R’s hypothesis of

translating results from the CTA to the entireNTA.

Were the upper-level troughs, or "cutofflows", farther south on CTA seeded days inIsraeli I, so that a (natural) rainfall maximumoccurred in the south and decreasednorthward? Were NTA seeded (and thus CTAcontrol) days affected by storms in Israel I thatproduced heavy rain in southern Lebanon?Were "coastal fronts", of the type described byKhain et al. (1993) and Rosenfeld and Nirel(1996), more numerous on CTA seeded daysin Israel I? We do not know the answers tothese and many other questions. However,several natural meteorological scenarios suchas those mentioned above, which could haveconcentrated rainfall in the CTA, could alsoexplain the increases in rainfall that wereattributed to seeding in Israeli I. In this contextit is illuminating to quote Wurtele (1971):"...according to M. G. Wurtele and A. Gagin,the meteorological characteristics of the daysthat are associated with moderate or strong(seeding) effects are descriptive of situationsfor which greater than average natural rainfallmay be expected."

In his Section 4f, R asserts that we arepremature to dismiss possible seedingeffects when the airflow is from thenorthwest at 850 hPa.

We agree with R that one of the mostpromising statistical results in Israeli I was thatindicating increases in rainfall due to seeding innorthwest flow at 850 hPa (GN74; RH95).However, we questioned the practicality ofincreasing rainfall significantly by randomlyseeding this flow regime, since in Israeli I thetotal number of experimental days withnorthwest flow was only about eighty in sixand a half seasons (GN74), and many of thosedays had sporadic, light rainfall. Thus, thesmall sample size that would accrue for thisregime in a random statistical experiment(about 6 seeded and control days each season),would make it difficult to arrive at a firmevaluation unless the experiment was carriedout over many years.

In RH95 we pointed out that rain falls fromshallow clouds wiA tops ^-10 C in northwestflow. This conflicts with R’s dust/hazehypothesis, since it shows that the presence of

16

dust/haze from deserts to the southwest is notthe major factor required for ice multiplicationand/or collision-coalescence of rain in Israel,and it raises questions about how muchpotential there is for seeding effects even inthese clouds.

In his Section 5, R objects to ourcharacterization of Israeli II as being"widely viewed" as a confirmatoryexperiment. He describes the target/controlevaluations /or Israeli II as exploratoryanalyses.

We are puzzled by R’s refutation of Israelin being a confirmatory experiment. It wasdescribed as such by Tukey et al. (1978),Simpson (1979), Kenr (1982), Silverman(1986), Cotton (1986), Dennis (1989), andCotton and Pieike (1992, 1995), without anyobjections from the HUJ investigators!

In his Section 5, R states Israeli II wasdesigned first as a single target!controlrandomized experiment, and that thecrossover design was subsidiary.

We appreciate R providing a summaryfrom a 1969 meeting of the Israeli RainCommittee (consisting of J. Neumann, A.Gagin, A. Kali, M. Bitaron, and K. R.Gabriel). However, this document (originallyin Hebrew) does not substantiate R’s claim thatIsraeli II -was, primarily a target/controlexperiment. While the target/control design ofthe experiment was discussed by theCommittee, and was clearly a component ofIsraeli n (which we never doubted), thisdocument shows that the Committee voted to

approve the full crossover design of Israeli nby a vote of three to two.

Second, based on information provided bythe HUJ investigators, the National Academyof Science’s Panel on Weather and ClimateModification (1973) described Israeli II as a"two-target, crossover, randomized; one area(N) has an adjacent control area."

Finally, in the physical descriptions of theIsraeli cloud seeding experiments given byGN74, the description of the Israeli nexperiment begins as follows: "As in the firstexperiment, there are two experiment areas,North and South. These areas are larger than

those of the first experiment. Again, we canuse the randomized crossover scheme."Concerning the early results of Israeli n,GN74 begin their discussion of the crossoverdesign by stating: "At the time of writing thischapter (April, 1972), we are close to the endof the third season of seeding in the frameworkof the second experiment...the RDR (rootdouble ratio) North versus South is about1.10; the single area ratio for the North alone isover 1.2 while for the South it is less than 1.However, none of those results is significantas yet."

We refer R to the discussion of GN74 onthe purpose and value of the crossover design,which we described in similar terms in RH95.While praising the attributes of the crossoverdesign that they were later to drop, GN74wrote: "The great merit of the crossoverdesign and, especially that of the test statistic isthat it eliminates to some extent, thetroublesome and misleading effects of thenatural (emphasis by GN74) fluctuations inrainfall."

No one can read GN74, and the emphasisthat they place on the RDR results, and theirlauding of the crossover design overtarget/control designs, and not clearly andunambiguously understand that the primaryconcern of GN74 in April 1972 (when GN74was being written) was the result of the Israelin crossover experiment. The target/controlaspect of Israeli n was just as clearlysubsidiary (not the other way around as R nowclaims). We urge the reader to study GN74,and the 1973 National Academy of Science’sreport, to confirm the emphasis that was placedon the crossover design.

Rosenfeld is correct on one point: thecrossover results of Israeli n were downplayedwhen the results did not confirm Israeli I. Infact, by 1981, the results of the Israeli IIcrossover experiment not only becamesubsidiary to the target/control results, theywere not mentioned at all (GN81)!

In his Section 5f, R states that the"experiment days" in Israeli II werepredetermined as those days withmeasurable rain at three stations in theBZ.

17

A rainfall criterion for the crossover designof Israeli n is not mentioned in the Israeli RainCommittee meeting of 1969. However, such acriterion for determining, retrospectively,experimental days is not unreasonable for theIsraeli n crossover experiment, because theoccurrence of rain in the BZ is a reasonable testfor the presence of suitable clouds in the targetareas on either side of the BZ. However, forthe target/control evaluations for northernIsrael and its hilly regions, the BZ rainfallcriterion is not a particularly good choice. Forexample, we found many days (about 50)when rain occurred in extreme northern Israelin or around the NTA, but these days were notincluded in the GN81 analyses because therewas not rain in the BZ. Hence, the criterion ofrain in the BZ to ensure the presence ofsuitable clouds in the NTA failed on numerousoccasions in Israeli II.

In Section 6b of his Comments, Rcontests our conclusions that theanalyses of Israeli II are, thus far,incomplete because numerous days^withrain in the NTA have not been analyzedand only a fraction of the available IMSraingauge data has been used.

With regard to the completeness of thereporting of Israeli n, we quote Gabriel(1967): "It would be advantageous to restrictthe evaluation of the experiment to those dayson which seeding is feasible and to exclude thelarge number of days without rain cloudswhich cannot add to the sensitivity of theexperiment. However, this is notpermissible."

Also we quote GN74: "Yet statisticalsignificance testing of the results of seeding isonly meaningful if the data of all (emphasis byGN74) days randomly allocated to seeding,whether actually seeded or not, are consideredtogether."

To date, there is no published analysis ofIsraeli n that meets the requirements specifiedin the above quotations from Gabriel (1967)and GN74. For example, in Israeli 11 arandom selection was made to seed one of thetargets every day between November 1 andApril 30 for the years 1969-70 through 1974-75 (for a total of 1087 days). So far, with theexception of results mentioned by R in his

present Comments, formal results have beenpresented for only 388 (or fewer) days (e.g.,GN81; GR90). Hence, we repeat: the resultsof Israeli n have not been fully reported.

We appreciate the additional resultsregarding wider analyses of Israeli II nowreported by R. While these results alleviatesome of our concerns, we recommend thatthey be incorporated into a comprehensivedescription of Israeli n and that this besubmitted for formal publication.

In his Section 6b, R states that the selectionof the control gauges in Israeli II was basedsolely on "historical continuity andreliability".

First, R’s statement is inaccurate. InRH95, we used the rainfall data published bythe IMS in their Annual Climatological Datafor Israel. To our astonishment, many of thesemain IMS rain gauges were not used by GR90in their analyses, despite having continuousrecords throughout the Israeli n experiment,including, for example, the gauge atJerusalem! Some of the gauges not used byGR90 are noted in Fig. 4. Hence, manyquestions remain about the choices by bothGN81 and GR90 of target and controlraingauges in Israeli n (see below).

On days when the NTA was seeded weexpected to find an isolated anomaly in theseed/no seed ratios in the NTA, beginningsome distance downwind of the line ofseeding. Elsewhere we expected to find nearunity values for the seed/no seed ratios. Toour surprise, we found high seed/no seedratios on NTA seeded days from centralLebanon to southern Jordan (RH95). In fact,the only anomaly in rainfall (from a regionalviewpoint) on the NTA seeded days of Israelin were relatively low (near unity), seed/noseed ratios in a narrow coastal strip north ofHaifa (see Fig. 17 in RH95).

We then observed that the number ofcontrol stations that GN81 selected ( sometimeduring or after the conclusion of Israeli n)from this narrow coastal strip wasdisproportionate relative to the density ofcontrol stations that they selected in other areaswhere the seed/no seed ratios were not so low.GN81 used nine control stations from thenarrow coastal strip with low seed/no seed

18

34N

33N -\

32N -I

Location andnames of stations (a) (b)

Southern Lebanon

* Beirut* Marjayoun

IsraeliNorthern Coast*NahariyyaHaifa/Qiraty AttaNir Ezyon

IsraeliNorthern HillsMishmar Ha-EmeqNazarethAfula

North TargetMt. Kena’an*Dafna*DeganiaAlef

0.85 1.02

0.78 1.11

0.87 1.05

1.00 1.16(SAR)

Buffer ZoneEn Ha Horesh

*Hefzi-Bah

0.77 1.09

Israeli Central 0.64 1.03and South Coast

Tel AvivYavne/Kefar MordecaiNegba

*Sa’ad*Lod

Israeli Southern Hillsand Jordan Valley*TiratZevi*JerusalemBeit JimalLahav

0.65 0.97

35E

Figure 4. Figure 17 from Rangno and Hobbs (1995) with the names of the stations used by themfor the calculation of seed/no-seed ratios and double ratios. See the caption to Fig. 17 ofRangno and Hobbs (1995) for further information. Stations listed on the right andpreceded by a star are IMS meteorological stations not used by GR90.

ratios (shaded area in Fig. Ib of the presentReply), but only seventeen control stationsfrom the rest of the control areas that togetherwere about ten times larger in area than thenarrow coastal strip. Also, the gauges used byGN81 were not the same as those used forIsraeli I by Gabriel (1967) who employed anearly uniform grid of gauges across Israel(compare Fig. la with Fig. Ib).

Rosenfeld states the control gauges werechosen by GN81 on the basis of "historicalcontinuity and reliability". If this is true, thenthe operators of rain gauges in the extremenorthern coastal strip of Israel exhibited aremarkable propensity for maintaining"historical continuity and reliability" comparedto their colleagues who operated rain gauges inthe remainder of the control area in Fig. 4-since GN81 used almost every available gaugefrom the extreme coastal strip as a control intheir analysis of Israeli II, but not so in theother regions (compare Figs. 4 and Fig. Ib ofthe present Reply)! Also, if this is true, whydid GR90 not use data from the sameraingauges as GN81 (compare Fig. Ib withFig. Ic).

We suggest that it would have been moreappropriate for the IMS to have determinedwhich gauges should have been used in theanalysis of the Israeli experiments, both in thetarget and the control areas. Further, in viewof the importance of these experiments, weurge that the IMS daily rainfall data,rawinsondes, surface observations, and therandom decisions for each day of Israeli I andn, be made freely available to any interestedparty.

In Section 6c of his Comments, R asserts

that the orographic rainfall bias in Israeli II,which is apparent on NTA seeded days(Gabriel and Rosenfeld 1990; RH95), hasbeen sufficiently accounted for by using

rainfall predictions from a numerical modelbased on rawinsondes launched twice-a-dayby the IMS during Israeli II (RF92). Hestates that a) adding data from the Beirut,Lebanon, soundings is not necessary; b)sensitivity tests in which the number ofrawinsonde profiles was doubled resulted innegligible improvements in the modelpredictions of rainfall; and, c) RH95

exaggerated the difficulty of predictingconvective rainfall amounts.

Rosenfeld’s statement that higher spatialand temporal data (i.e., using fourrawinsondes per day instead of two) does notimprove his model’s predictions of rainfall, issurprising. We believe it reflects the crudity ofR’s model. Also, R does not indicate how heaccounted for the large errors in relativehumidity during daylight hours in the raw-insonde humidity elements that were used inthe Israeli II era (Hill 1980). Airborne studies,in which humidity profiles measured withrawinsondes were compared with actual cloudfields, often showed little con-elation betweenthese profiles and the nearby presence ofclouds (United States Air Force 1969).

We note also that in rejecting thesuggestion in RH95 to use rawinsonde datafrom Beirut, Lebanon, in his numerical model,R also rejects the design for Israeli nrecommended by the Israeli Rain Committee in1969. That committee recommended that:"Temperature and upper level data (upper levelwind, humidity, etc.). Interpolation betweenBelt Dagan and Beirut, twice a day" be used inevaluating Israeli n.

Although R’s model results reflect some ofthe natural bias in rainfall on seeded days thatfavored the NTA of Israeli n, it is difficult tobelieve that with the use of twice-a-dayrawinsonde data alone (and without usingrawinsonde data from Beirut, which is closerto the target than Belt Dagan) that his modelcan account for differences in natural rainfallbetween seeded and control days, which wereless than 10 percent in Israeli II, and couldnot be improved with more frequent soundingprofile data.

Finally, only RH95 have used rainfall datafrom southern Lebanon as a control for theNTA, as was suggested by the Israeli RainCommittee in 1969.

Therefore, just as it is prudent to declaretarget and control stations prior to a cloudseeding experiment, it is also good practice todeclare prior to a statistical experiment whichmeteorological variables will be used forevaluation (e.g., Flueck 1986).

In his Section 6d, R gives several reasonswhy he believes that the conclusions we

19

reached regarding the occurrence of a Type Istatistical error ("lucky draw") in Israeli IIon NTA seeded days are invalid. Hecontends that since our analysis wasconducted post facto it is invalid comparedto what he asserts is the a priori designevaluation of GN81. He points out the needfor a high correlation between target areasin crossover experiments. Finally, he states

that Fig. 17 of RH95 (Fig. 4 in the presentReply) includes seeded days in some regionsas part of the regional analysis.

Some statements by R must be correctedbefore we can respond to these comments.There is no evidence that the proper elementsfor an a priori design and evaluation were inplace prior to Israeli II. For example, thecontrol stations were not specified in advance,and an analysis using the same gauges as thoseused in Israeli I has still not been presented.Also, analysis of Israeli II is incomplete. With436 raingauges available for analysis of IsraeliII, virtually any result could be produced (e.g.,Thorn 1957).

Rosenfeld’s discussion of our wider arealanalysis for Israeli n, which included Lebanonand Jordan (RH95), highlights a seriousdeficiency in the design and previousevaluations of both Israeli I and n. Thedesigns either did not include analyses of thewider regional distribution of precipitation/runoff on seeded and control days, whichwould have alerted the HUJ investigators to

potential problems, or, when the design didspecify the use of regional data, such as fromLebanon (see meeting of the Israeli RainCommittee in 1969), these data were not used,

Rosenfeld believes that postfacto regionalanalyses should be disregarded if they are not

specified in the original design of a cloudseeding experiment. In fact, post factoanalyses have a long and valuable tradition incloud seeding experiments. For example,numerous post facto analyses were carried outfor Project Whitetop that shed important lighton that experiment (e.g., Braham 1979). Arethese analyses invalid?

Rosenfeld is correct in stating that we usedrelatively few rainfall stations in our reanalysisof Israeli n. We wished we could have used

more, particularly for Israel4 and LebanonHowever, we used all of the data we couldobtain, all of the Israeli stations we used arecontained in the IMS publication. AnnualClimatological Summary for Israel, and weexpanded the analyses of GR90 and RF92 byadding as many stations as we could fromLebanon and Jordan. Apart from somestations in the Israeli deserts that had zeromedian rainfall on experimental days, nostation for which we had data was omitted inour analysis.

We agree with R that it is best to choosetarget and control regions that are highlycorrelated. However, when the bias in stormsis as strong as it was on NTA seeded days inIsraeli n, a consistent pattern of seed/no seedratios among both highly correlated andmoderately correlated stations with the NTArainfall will be evident, as indeed it was.

By showing the regional values of seed/noseed ratios for the seeded days of the NTA andSTA of Israeli II (see Fig. 17 in RH95 andFig. 4 in present Reply), we demonstrated thatIsraeli n was dwarfed by natural events thatproduced a regionally consistent pattern oflarge seed/no seed ratios on NTA seeded days,and that this pattern was inadvertentlyinterpreted as "increases" and "decreases" inrainfall due to cloud seeding by RF92. Forexample, low seed/no seed ratios (1) arepresent throughout the region of our analysison STA seeded days, which one coulderroneously interpret as seeding-induceddecreases in rainfall over this entire region, not

just in the STA and downwind in Jordan asdoes R.

Rosenfeld has also overlooked the findingof Gabriel and Rosenfeld (1990) who foundthat it was the heavy rainfall (25-45% higherthan the climatological normal daily rainfall!)on STA control days (the NTA seeded days),

Geophysical data are now seen as a valuableresource by some countries and some charge large sumsfor such data. In 1987, the cost of obtaining the fulldata set for the Israeli experiments was estimated by theIMS to be S50,000 (personal communication, 1987,from A. Vardi, Deputy Director, IMS)!

With the exception of data for Beirut andMariayoun, the rainfall data for the dense Lebanonnetwork had numerous missing months during Israeli IandH.

20

and not exceptionally light rain on seeded daysin the STA, that produced the appearance ofdecreased rainfall due to seeding in the STA.It is worth quoting GR90: "Otherwise, onewould need to explain why there was somuch more rain in the south when the northwas being seeded (our italics for emphasis);as of now, no explanation is available,especially as the prevailing wind direction isfrom the southwest." In view of this statement(which R co-authored), we find it difficult to

comprehend R’s sole reliance on hishypothesis that rain in the STA of Israeli nwas decreased by seeding. How does Rexplain why rain was so much heavier in theSTA on NTA seeded days?

To provide just an even draw, the naturalrain on the STA seeded days would have hadto have been 25-45% heavier than normal!Since the seeded and control days comprisedall of the rain days in Israel, R is implying thatduring Israeli n these were six straight rainseasons with total rainfall 25-45% abovenormal! However, this did not occur either inthe seeded or in the unseeded target areas ofIsraeli n (Atlas of Israel, 1985). This is whywe disagree with RF92 and R’s conclusionthat seeding decreased rainfall in the STA (andincreased it in the NTA) in Israeli n.

A test of the claim by RF92 that seedingincreased rainfall in the northern interior ofIsrael, while decreasing rainfall in the centralpans of Israel, during the thirteen and a halfseasons of the two experiments can be checkedin a rudimentary way by using historicalrainfall data for a much longer period than thatused by GR90. According to R, seeding in thenorth part of Israel should increase the gradientin rainfall from the northern coast (north ofHaifa) to the interior hill region beginning nearMar Meron and Mt. Kana’an in extremenorthern Israel and on eastward downwind ofthe seeding line. Also, according to RF92 andR, seeding should have especially increasedrainfall gradients in the north-south direction inIsrael, since it increased rainfall in the northinterior and decreased rainfall in the southinterior (e.g., near Jerusalem).

We have tested these hypotheses using thedetailed isoheyts of rainfall for each rainfallseason from 1932-33 through 1975-76,compiled and analyzed by the IMS, that appearin the Atlas of Israel (1985). Figure 5 showsthe results of this survey of rainfall gradients

using the departure from the medians of eachgradient (computed for the full 44 seasons) asa measure of any seeding-induced effects inrainfall gradients. While the data are notsufficiently precise to be tested statistically, itis obvious that there is not even rudimentarysupport for RF92’s claim that seedingincreased rainfall in the NTA while decreasingit in the STA during Israeli n. Figure 5indicates that the rainy seasons of Israeli n, onwhich RF92 based their "increases-in-the-north-and-decreases-in-the-south" hypothesis,run exactly counter to that hypothesis in boththe east-west and north-south directions!

In conclusion, we believe that the limitedareal scope of the rainfall analyses by RF92(and by R in his present Comment) led them toconfuse widespread natural rainfall events withseeding effects.

In his Section 7c, R claims once again thatthere is not a contradiction between theresults of Israeli I and II and the dust/haze

hypothesis of RF92.

We disagree. No one who is familiar withthe several previous analyses of Israeli I (i. e.,Gabriel 1967; Gabriel and Baras 1970; Wurtele1971; GN74), all of whom concluded that thebest indications of a seeding effect in Israeli Iwere in the CTA and in the BZ on days whenthe wind was from the southwest quadrant,could not be astonished at R’s claim that therainfall in Israeli I in the CTA was actuallydecreased by seeding.

As we have pointed out already, R’sreasoning is flawed because it is based on theassumption that rainfall patterns in Israeli I andn were similar. As noted by RH95, and againin the present Reply, in Israeli I the heavierrainfall was more localized on CTA seededdays than it was on the seeded days of theNTA in Israeli H (e.g.. Brier et al. 1973;Rosenfeld 1989; RH95).

Rosenfeld’s view that in Israeli I rainfall inthe CTA was decreased by seeding, but that inthe NTA it was substantially increased (>30%)by seeding, is an interesting new idea. Wereiterate our suggestion that R submit a paperfor publication to substantiate this claim.

Rangno and Hobbs (1995) noted thatregional values of the seed/no seed ratios

21

^̂ 1.40a<u

I 120

0

^^D2S3

5̂e

^

1.00

.80

.60

.40

.20

0

-.20

-.40 [--.60’-

^ ^\ la. ^d.rIsraeli I Israeli II

&Q 1935 1940 1945 1950 1955 1960 1961 1965 1967 1968/69 1970 1975 1976

Rainy Season Year Ending

Figure 5. Departures from rainfall gradients in Israel for each rainfall season (mid-October through April) from 1932-33 through 1975-76. The open bars are the differences in average rainfall between the coast north of Haifa and the maximum in the northernhills around Mt. Kana’an. The black bars are the differences in rainfall between the rainfall maxima in the northern hillsaround Mt. Kana’an and those in the southern hills near Jerusalem. An upward/downward bar indicates that the difference inrainfall between the two regions is greater/less than the 44-year median. Therefore, for the periods 1961-1967 and 1969-1975(which are the years of Israeli I and II), an upward/downward bar supports/refutes RF92’s hypotheses that seeding 1)increased rainfall in the northern hill region compared with the northern coast, and 2) increased rainfall in the north target areabut decreased rainfall in the southern hill region.

for Israeli II were not displayed by RF92

for the dust/haze and non-dust/haze days.We suggested that the display of such valueswould have helped readers understandRF92’s evaluations of Israeli II. However,in his Section 7d R asserts that regionalvalues of the these ratios are superfluous inthe evaluation of cloud seeding experiments.

Seed/no seed ratios provide a first step inthe evaluation of cloud seeding experiments,and this statistic has been a standardcomponent of such evaluations for manydecades. These ratios are vital in order to seeif the expectation of an anomaly in them,produced by cloud seeding, is reasonablylocalized to regions downwind from whereseeding took place. In our view, the failure to

give proper weight to regional seed/no seedratios accounts for the many of the erroneousconclusions that have been drawn for theIsraeli cloud seeding experiments.

5. Concluding remarks

Many unanswered questions remainconcerning the Israeli cloud seedingexperiments, the clouds of Israel, and theintriguing new hypotheses about seedingeffects and cloud seeding potential that havebeen postulated by R and Reisin et al. (1996).It is clear that the questions outnumber theanswers, and that the results of the Israelicloud seeding experiments are, at best,ambiguous.

We join with R in urging a comprehensivestudy of clouds in the Middle East with respectto their seeding potential, the efficacy of lineseeding, and many other questions that wehave raised. We recommend that such a studybe carried out by a team of experts under theauspices of an international organization, suchas the World Meteorological Organization.

REFERENCES

Atlas of Israel, 1970: Published by theSurvey of Israel, Ministry of Labour,Jerusalem, and Elsevier Publishing Co.,Amsterdam.

1985: Published by the Survey ofIsrael and Macmillan Pub. Co.

Ben-Zvi, A,, 1988: Enhancement of runofffrom a small watershed by cloud seeding.J. Hydrology, 101, 291-303.

Benjamini, Y., and Y. Harpaz, 1986:Observational rainfall-runoff analysis forestimating effects of cloud seeding onwater resources in northern Israel. J.Hydrology, 83, 299-306.

Biyth, A. M., and J. Latham. 1993: Iceparticles in New Mexican summertimecumulus clouds. Quart. J. Roy. Meteor.Soc., 119, 91-120.

Braham, R. R., Jr., 1964: What is the role ofice in summer rain-showers? J. Atmos.Sci., 21, 640-646.

1979: Field experimentation inweather modification. J. Amer. Statist.Assoc., 74, 57-68.

Brier, G. W., and I. Enger, 1952: An analysisof the results of the 1951 cloud seedingoperations in central Arizona. Bull. Amer.Meteor. Soc. 23, 208-210.

L. 0. Grant, and P. W. Mieike, Jr.,1973: An evaluation of extended areaeffects from attempts to modify localclouds and cloud systems, Proc.MWO/IAMAP Scien. Conf. on WeatherModification. Switzerland. WorldMeteor. Org., Geneva, 439-447.

Chisnell, R. F., and J. Latham, 1976: Iceparticle multiplication in cumulus clouds.Quart. J. Roy. Meteor. Soc., 102, 133-156.

Cooper, W. A., and R. P. Lawson, 1984:Physical interpretation of results from theHIPLEX-1 experiment. /. Climate Appl.Meteor., 23, 523-540.

Cotton, W. R., 1986: Testing,implementation, and evolution of seedingconcepts-a review. In RainfallEnhancement-A Scientific Challenge,Meteor. Monogr., 21, No. 43, 139-149.

and R. A. Pieike, 1992: HumanImpacts on Weather and Climate, ASteRPress, Boulder, Colorado.

and 1995: HumanImpacts on Weather and Climate,Cambridge University Press, 40 West 20thStreet, New York, NY 10011-4211, USA,p. 17.

Court, A., 1960: Evaluation of cloud seedingtrials. J. Irrig. and Drainage Div., Proc.Am. Soc. Civ. Eng., 86, No. IR 1, 121-126.

22

Druyan, L. M., and Y. Sant, 1978: Objective12 h rainfall forecasts using a singleradiosonde. Bull. Amer. Meteor. Soc.,59, 1438-1441.

Dennis, A. S., 1980: Weather Modificationby Cloud Seeding. Academic Press,181 pp.

1989: Editorial, J. Appl. Meteor.,28, 1013.

Flueck, J., 1986: Principles and prescriptionsfor improved experimentation inprecipitation augmentation research. InPrecipitation Enhancement-AScientific Challenge, R. R. Braham, Jr.,Ed., Meteor. Monographs., 43, Amer.Meteor. Soc., Boston, 02108, pp. 155-171.

Gabriel, K. R., 1967: The Israeli artificialrainfall stimulation experiment: statisticalevaluation for the period 1961-1965. InProc. Fifth Berkeley Symposium onMathematical Statistics and Probability,Vol. 5, L. M. Le Cam and J. Neyman,eds.. University of California Press, 91-113.

and J. Baras, 1970: The Israelirainmaking experiment, 1961-1967: finalstatistical tables and evaluation. Tech.Rep., Dept. Meteor., Hebrew Universityof Jerusalem, Jerusalem, 47 pp.(Available from the Hebrew University ofJerusalem, Jerusalem, Israel).

and J. Neumann, 1978: A note ofexplanation on the 1961-67 Israeli rainfallstimulation experiment. /. Appl. Meteor.,17, 552-556.

and D. Rosenfeld, 1990: Thesecond Israeli rainfall stimulationexperiment: analysis of rainfall on bothtarget area. J. Appl. Meteor., 29, 1055-1067.

Gagin, A.. 1975: The ice phase in wintercontinental cumulus clouds. J. Atmos.Sci., 32, 1604-1614.

1980: The relationship between thedepth of cumuliform clouds and theirraindrop characteristics. J. Rech. Atmos.,14, 409-422.

1981: The Israeli rainfallenhancement experiments. A physicaloverview. J. Wea. Modif., 13, 1-13.

1986: Evaluation of "static" and"dynamic" seeding concepts throughanalyses of the Israeli n experiment and

FACE-2 experiments. In RainfallEnhancement-A Scientific Challenge,Meteor. Monogr., 21, No. 43, 63-70.

and M. Arroyo, 1985: Quantitativediffusion estimates of cloud seeding nucleireleased from airborne generators. /. Wea.Mod., 17, 59-70.

and K. R. Gabriel, 1987: Analysisof recording raingauge data for the Israelin experiment. Part I: Effects of cloudseeding on the components of dailyrainfall. /. Climate Appl. Meteor., 26,913-926.

and J. Neumann, 1973:Modification of subtropical winter cumulusclouds-cloud seeding and cloud physics inIsrael. Proc. Intern. TropicalMeteorology Meeting, Nairobi, Kenya,Amer. Meteor. Soc., 203-215.

and 1974: Rainstimulation and cloud physics in Israel. InClimate and Weather Modification, W.N. Hess, Ed. Wiley and Sons, New York454-494.

and 1976: The secondIsraeli cloud seeding experiment-the effectof seeding on varying cloud populations.Proc. 2ndWMO Scientific Conf. onWeather Modification, Boulder, 195-204. (Available from the World Meteor.Organization, 41 Ave. Giuseppe Moffs,Geneva 2, Switzerland.)

and____, 1981: The secondIsraeli randomized cloud seedingexperiment: evaluation of results. J. Appl.Meteor., 20, 1301-1311.

Gelhaus, J. W., A. S. Dennis, and M. R.Schock, 1974: Possibility of a Type Istatistical error in analysis of a randomizedcloud seeding project in South Dakota. J.Appl. Meteor., 13, 383-386.

Gleick, J., 1992: Genius: The Life andScience of Richard Feynman, PanAeon,532 pp.

Grant, L. 0., J. 0. Rhea, G. T. Meltesen, G.J. Mulvey, and P. W. Mieike, Jr., 1979:Continuing analysis of the Climax weathermodification experiments. Seventh Conf.On Planned and Inadvertent WeatherModification, Banff, Alberta. The Amer.Meteor. Soc., Boston, MA, 02108, J43-J45.

Hill, G. E., 1980: Re-examination of cloud-top temperatures used as criteria for

23

stratification of cloud seeding effects onwinter orographic clouds. /. Appl.Meteor. 19, 1167-1175.

Hobbs, P. V., M. K. Politovich, and L. F.Radke, 1980: The structures of summerconvective clouds in eastern Montana. I:natural clouds. /. Appl. Meteor., 19,645-663.

and A. L. Rangno, 1978: Areanalysis of the Skagit cloud seedingproject. J. Appl. Meteor., 17, 1661-1666.

and____, 1985: Ice particleconcentrations in clouds. J. Atmos. Sci.,42, 2523-2549.

and 1990: Rapiddevelopment of high ice particleconcentrations in small polar maritimeclouds. J. Atmos. Sci., 47, 2710-2722.

Ken-, R. A., 1982: Cloud seeding: onesuccess in 35 years. Science, 217, 519-522.

Khain, A. P., D. Rosenfeld, and I. L. Sednev,1993: Coastal effects in the easternMediterranean as seen from experimentsusing a cloud ensemble model with adetailed description of warm and icemicrophysical processes. Atmos. Res.,30, 295-319.

Koenig, L. R., 1963: The glaciating behaviorof small cumulonimbus clouds. J. Atmos.Sci., 20, 29-47.

Lamb, D. L., J. Hallett, and R. I. Sax, 1981:Mechanistic limitations to the release oflatent heat during the natural and artificialglaciation of deep convective clouds.Quart. J. Roy. Meteor. Soc., 107, 935-954.

Levin, Z., 1994: Effects of aerosolcomposition on the development of rain inthe eastern Mediterranean-potential effectsof global change. WMO Workshop onCloud Microphysics and Applications to

Global Change. Toronto, Ontario,Canada. World Meteor. Org., 115-120.

E. Ganor, and V. Gladstein, 1996:The effects of desert particles coated withsulfate on rain formation in the easternMediterranean. J. Appl. Meteor., 35,1511-1523.

Lovasich, J. L., J. Neyman, E. L. Scott, andM. A. Wells, 1971: Further studies of theeffects of cloud seeding in the Whitetop

experiment. Proc. of the National Acad.Sci., 6S, 147-151.

Malkus. J. S., and R. S. Scorer, 1995: Theerosion of cumulus towers. J. Meteor.,12, 43-57.

Mieike, P. W., Jr., 1979: Comment on fieldexperimentation in weather modification.J. Amer. Statist. Assoc., 74, 87-88.

Mossop, S. C., 1970: Concentrations of icecrystals in clouds. Bull. Amer. Meteor.Soc., 51, 474-479.

1985: Concentrations of ice crystalsin clouds. Bull. Amer. Meteor. Soc., 66,264-273.

Ono, A., and K. J. Heffeman,1967: Studies of ice crystals in naturalclouds. J. Atmos. Res.^. 1, 44-64.

R. E. Ruskin, and J. K. Hefferman,1968: Glaciation of a cumulus at -4 C. J.Atmos. Sci., 25, 889-899.

Cottis, R. E., and B. M. Bartlett,1972: Ice crystal concentrations incumulus and stratocumulus clouds. Quart.J. Roy. Meteor. Soc., 98. 105-123.

National Academy of Sciences-NationalResearch Council, Review Panel onWeather and Climate Modification, 1973:Weather Modification: Progress andProblems, 258 p. (Available from theSuperintendent of Documents,Government Printing Office, Washington,D.C. 20402.)

Neumann, J., 1951: Land breezes andnocturnal thunderstorms. J. Meteor., 8,60-67.

Nickerson, E. C., 1979: FACE rainfallresults: Seeding effect or naturalvariability? J. Appl. Meteor., 18, 1097-1105.

Nirel, R., and D. Rosenfeld, 1995: Estimationof the effect of operational seeding on rainamounts in Israel. J. Appl. Meteor., 34,2220-2229.

Ono, A., 1972: Evidence on the nature of icecrystal multiplication processes in naturalcloud. J. Res. Atmos., 6, 399-408.

Rangno, A. L., 1979: A reanalysis of theWolf Creek Pass cloud seedingexperiment. J. Appl. Meteor., 18, 579-605.

1988: Rain from clouds with topswarmer than -10 C in Israel. Quart. J.Roy. Meteor. Soc., 114, 495-513.

24

and P. V. Hobbs, 1988: Criteria forthe development of significantconcentrations of ice particles in cumulusclouds. Atmos. Res., 21, 1-13.

and 1991: Ice particleconcentrations in small, maritime polarcumuliform clouds. Quart. J. Roy.Meteor. Soc., 118, 105-126.

and 1994: Ice particleconcentrations and precipitationdevelopment in small continentalcumuliform clouds. Quart. J. Roy.Meteor. Soc., 120, 573-601.

and 1995: A new look atthe Israeli cloud seeding experiments. J.Appl. Meteor., 34, 1169-1193.

and 1997: Reply. J.Appl. Meteor., 36, 272-276.

Reisin, T., S. Tzivion, and Z. Levin, 1996:Seeding convective clouds with ice nucleior hygroscopic particles: a numerical studyusing a model with detailed microphysics.J. Appl. Meteor., 35, 1416-1433.

Rosenfeld, D., 1980: Characteristics of raincloud systems in Israel derived from radarand satellite images. M. S. Thesis, TheHebrew University of Jerusalem, 129 pp.(Available from the. Department ofMeteorology, Hebrew University ofJerusalem, Jerusalem, Israel).

1989: The divergent effects of cloudseeding under different physical conditionsin Israeli 1 and 2 experiments. Dept.Atmos. Sci., The Hebrew University ofJerusalem, Jerusalem, Israel, 42 pp.

1997: Comment on "A new look at

the Israeli cloud seeding experiments" byRangno and Hobbs. J. Appl. Meteor.,36, 260-271.

andH. Farbstein, 1992: Possibleinfluence of desert dust on seedability ofclouds in Israel. J. Appl. Meteor., 31,722-731.

and A. Gagin, 1989: Factorsgoverning the total rainfall yield fromcontinental convective clouds. J. Appl.Meteor., 28, 1015-1030.

and R. Nirel, 1996: Seedingeffectiveness-the interaction of desert dustand the southern margins of rain cloudsystems in Israel. /. Appl. Meteor., 35,1502-1510.

Sax, R. I., S. A. Changnon, L. 0. Grant, W.F. Hitchfield, P. V. Hobbs, A. M. Kahan,and J. Simpson, 1975: Weathermodification: where are we now andwhere are we going? J. Appl. Meteor.,14, 652-672.

Silverman, B. A., 1986: Static mode seedingof summer cumuli-a review. In RainfallEnhancement-A Scientific Challenge,Meteor. Monogr., 21, No. 43, 6-24.

Simpson, J., 1979: Comment on "Fieldexperimentation in weather modification".J. Amer. Statist., 74, 95-97.

Thorn, H. C. S., 1957: An evaluation of aseries of orographic cloud seedingoperations. Final Report of the AdvisoryCommittee on Weather Control., VolumeII, 25-50.

Tukey, J. W., D. R. Brillinger, and L. V.Jones, 1978: Report of the StatisticalTask Force to the Weather ModificationBoard, Vol. I. U.S. Government PrintingOffice, 229 pp.

United States Air Force, 1969: Use of theskew-T, log p diagram in analysis andforecasting. Air Weather Service Manual105-124, Department of the Air Force,Headquarters, Air Weather Service(MAC), Scott Air Force Base, Illinois62225.

Wurtele, Z., 1971: Analysis of the Israelicloud seeding experiment by means ofconcomitant meteorological variables. /.Appl. Meteor., 10, 1185-1192.

25