• Cooper, W. A., R. T. Bruintjes, and G. K. Mather, 1997: Calculations pertaining to hygroscopic seeding with flares. J. Appl. Meteor.,36, 1449–1469.

  • Dixon, M. J., and G. K. Mather, 1986: Programme for Atmospheric Water Supply—Phase 1, 1983–1986, Vol. 3. WRC Rep. 133/3/88, 55 pp. [Available from Water Research Commission, P.O. Box 824, Pretoria 0001, South Africa.].

  • ——, and G. Wiener, 1993: TITAN: Thunderstorm Identification, Tracking, Analysis, and Nowcasting—a radar-based methodology. J. Atmos. Oceanic Technol.,10, 785–797.

  • Mather G. K., M. J. Dixon, and J. M. de Jager, 1996: Assessing the potential for rain augmentation. The Nelspruit randomized convective cloud seeding experiment. J. Appl. Meteor.,35, 1465–1482.

  • ——, D. E. Terblanche, and F. E. Steffens, 1997a: National Precipitation Research Programme: Final report for the period 1993–1996. WRC Rep. 726/1/97, 147 pp. [Available from Water Research Commission, P.O. Box 824, Pretoria 0001, South Africa.].

  • ——, ——, ——, and L. Fletcher, 1997b: Results of the South African cloud-seeding experiments using hygroscopic flares. J. Appl. Meteor.,36, 1433–1447.

  • Terblanche, D. E., 1996: A simple digital signal processing method to simulate linear and quadratic responses from a radar’s logarithmic receiver. J. Atmos. Oceanic Technol.,13, 533–538.

  • Woodley, W. L., A. G. Barriston, J. A. Flueck, and R. Biondini, 1983:The Florida Area Cumulus Experiment’s second phase (FACE-2). Part II: Replicated and confirmatory analysis. J. Climate Appl. Meteor.,22, 1529–1540.

  • View in gallery

    A map of the Northern Province showing the main topographical features of the area, the location of the radar close to Tzaneen, and its 134.4-km data collection range.

  • View in gallery

    Quartile analysis, from time of origin, of rain mass at lowest scan for the seeded and control storms. Here C1 represents the 194 control storms, C2 represents the 60 largest control storms, and the bold line represents the 60 seeded storms. (a) First quartile; (b) second quartile (median); (c) third quartile.

  • View in gallery

    The first (Q1), second (median) (Q2), and third quartile (Q3) of the rain mass at lowest scan for the 60 seeded (solid lines) and matching 60 unseeded storms (dotted lines) in 10-min time intervals from time of origin.

  • View in gallery

    Accumulated mean rain mass of the 60 seeded storms (solid line) in comparison with the accumulated mean rain mass from the matching 60 unseeded storms (dotted line) in 10-min time windows from time of origin.

  • View in gallery

    A map of SAWB rainfall districts overlayed by the radar area as well as the seeded storm tracks and contours of normalized rainfall values for Mar 1995.

  • View in gallery

    Same as Fig. 5, but for Oct 1995–Mar 1996.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 3 3 0
PDF Downloads 0 0 0

Toward the Operational Application of Hygroscopic Flares for Rainfall Enhancement in South Africa

View More View Less
  • a Bethlehem Precipitation Research Project, South African Weather Bureau, Bethlehem, South Africa
  • b Centre for Applied Statistics, University of South Africa, Pretoria, South Africa
  • c Bethlehem Precipitation Research Project, South African Weather Bureau, Bethlehem, South Africa
  • d CloudQuest, Nelspruit, South Africa
© Get Permissions
Full access

Abstract

A major challenge of any operational cloud seeding project is the evaluation of the results. This paper describes the development of verification techniques based on data collected during the first South African operational rainfall enhancement project in which hygroscopic flares were used to seed the bases of convective storms. Radar storm properties as well as historical rainfall records were used in exploratory studies. The storm-scale analyses are viewed as extremely important, because individual storms are the units that are seeded. Their response to seeding has to be consistent with that of the seeded group in a randomized experiment using the same seeding technology before a positive effect on area rainfall can be expected. Sixty storms were selected for seeding, mostly early in their lifetimes. This permits a time-of-origin analysis in which the group of seeded storms can be compared to a “control” group of unseeded storms from the time they were first identified as 30-dBZ radar storm volumes. One such control group was obtained by selecting unseeded storms by using certain threshold criteria obtained from the seeded storms. Another control group was obtained by simply selecting the 60 largest storms from the set of unseeded storms meeting the threshold criteria. Yet another control group was obtained by matching the seeded storms, in the first 20 min of their lifetimes, before seeding effects can be expected, with a corresponding set of unseeded storms. Comparisons with the National Precipitation Research Programme’s randomized hygroscopic flare seeding experiment database show consistency in the way seeded storms reacted toward producing more rainfall. The analyses on historic rainfall suggest trends in the same direction, but it is shown that one has to be careful in interpreting these trends. The importance of quantitatively linking storm-scale seeding effects to apparent area effects is highlighted.

* Deceased.

Corresponding author address: Deon E. Terblanche, SA Weather Bureau, Bethlehem Weather Office, Private Bag X15, 9700 Bethlehem, South Africa.

deon@metsys.weathersa.co.za

Abstract

A major challenge of any operational cloud seeding project is the evaluation of the results. This paper describes the development of verification techniques based on data collected during the first South African operational rainfall enhancement project in which hygroscopic flares were used to seed the bases of convective storms. Radar storm properties as well as historical rainfall records were used in exploratory studies. The storm-scale analyses are viewed as extremely important, because individual storms are the units that are seeded. Their response to seeding has to be consistent with that of the seeded group in a randomized experiment using the same seeding technology before a positive effect on area rainfall can be expected. Sixty storms were selected for seeding, mostly early in their lifetimes. This permits a time-of-origin analysis in which the group of seeded storms can be compared to a “control” group of unseeded storms from the time they were first identified as 30-dBZ radar storm volumes. One such control group was obtained by selecting unseeded storms by using certain threshold criteria obtained from the seeded storms. Another control group was obtained by simply selecting the 60 largest storms from the set of unseeded storms meeting the threshold criteria. Yet another control group was obtained by matching the seeded storms, in the first 20 min of their lifetimes, before seeding effects can be expected, with a corresponding set of unseeded storms. Comparisons with the National Precipitation Research Programme’s randomized hygroscopic flare seeding experiment database show consistency in the way seeded storms reacted toward producing more rainfall. The analyses on historic rainfall suggest trends in the same direction, but it is shown that one has to be careful in interpreting these trends. The importance of quantitatively linking storm-scale seeding effects to apparent area effects is highlighted.

* Deceased.

Corresponding author address: Deon E. Terblanche, SA Weather Bureau, Bethlehem Weather Office, Private Bag X15, 9700 Bethlehem, South Africa.

deon@metsys.weathersa.co.za

Introduction

As scientists working in the field of rainfall enhancement research, we should never lose sight of the ultimate goal of the research, which is to develop viable techniques that could be used operationally to alleviate water scarcity. During an international workshop on water resource management held in South Africa in 1997, it was stated that, worldwide, the freshwater consumption has tripled in the past 50 yr. By 2025, for example, over a billion Africans will live in water-stressed countries.

A major challenge of any operational cloud seeding project, when randomized experiments are excluded, is the evaluation of the results. The transition from a scientific research experiment to an area-wide operational application is a necessary but highly complex step with many pitfalls. In this transition, the experimental unit and the response variables are often changed to an entity that is much less understood by the scientists and that includes even more variables. Moving upscale from convective storms in a research environment already has its challenges. An example of this challenge is using a rain day as unit in the Florida Area Cumulus Experiment, phase 2 and how that affected the outcome of the experiment (Woodley et al. 1983). Furthermore, within the scientific community, and especially among the statisticians, there is often the perception that this transition to operations goes hand in hand with a loss of scientific credibility. In this paper, the approach followed in South Africa to address some of these issues is presented. It is hoped that this approach will provide guidance on some of the issues central to a successful operational convective cloud seeding program for rainfall enhancement.

Early in 1995, the South African National Precipitation Research Programme (NPRP)’s randomized cloud seeding experiment using hygroscopic flares (Mather et al. 1997b) was nearing completion. It was at this time that the Northern Province government approached South Africa’s Water Research Commission (WRC), the South African Weather Bureau (SAWB), and the Department of Water Affairs and Forestry to employ cloud seeding at very short notice as an emergency response to drought in the province. This request presented the South African research team and its funding organizations with a dilemma. After careful consideration, it was decided to react positively to the request, the reason being that this request presented an opportunity that could facilitate a smooth transition to the ultimate operational application of rainfall enhancement methods as the randomized seeding activities were concluded. There was also the need to determine how the personnel and infrastructure would cope within an operational environment. Before the inception of the Northern Province project, it was decided that radar data should be collected and analyzed in a way similar to that applied in the randomized experiment, to ensure the possibility of developing scientific verification techniques. The analyses were primarily to be based on radar storm properties to facilitate comparisons with the randomized experiment database. Before the experiment was initiated, it was decided that radar-estimated rain mass at the lowest scan was once again the property to be focused on, using quartile analyses. Therefore, the local research team continued to study their familiar experimental unit, namely, the convective storm. Because further details on the storm-based analysis, including the manner in which control storms were to be identified, were not defined before the experiment started, all the analyses reported on in this paper should be viewed as exploratory in nature. In addition, SAWB historical rainfall records were also to be used in an exploratory manner.

The Northern Province project received approval by all parties on 9 February 1995, and the radar was moved to the new site and was running by 27 February. The first operational seeding occurred on 3 March 1995.

This paper covers the NPRP Northern Province operations for the period 1995–97, a phase of the experiment in which our colleague Graeme Mather played a major and final role in the longstanding South African rainfall enhancement effort. His contributions to the Northern Province project ranged from assistance with setting up of the project infrastructure, its procedures, and radar control to flying one of the seeding aircraft and performing data analyses. In the first section of the paper, the experimental location, infrastructure, and procedures are described. This section is followed by the exploratory analyses on storm-track properties (with specific reference to the randomized experiment data) and the historical rainfall data. In the conclusions, the valuable lessons learned from this experiment, in the context of consequent developments regarding rainfall enhancement in South Africa, are given.

Location, infrastructure, and procedures

The Northern Province represents the northeasternmost region of South Africa. It borders on Mozambique to the east, Zimbabwe to the north, and Botswana to the northwest. A large percentage of the annual rainfall of the province is from summer convective storms that are strongly affected by the local topography. The main topographical features of the province, some of the major towns, provincial and national borders, and the boundaries of SAWB’s rainfall districts are shown in Fig. 1.

The project’s C-band radar situated close to the Tzaneen airport (23.86°S, 30.31°E, 503 m above mean sea level), about 20 km east of the town, was operated in computer-controlled volume-scan mode, collecting data to a maximum range of 134.4 km. The radar data were collected on tape to allow storm property analyses using the system developed by Dixon and Mather (1986). Detail on the software and how storms are identified, tracked, and their properties computed are given by Mather et al. (1996). This same software was used for the randomized experiment analysis (Mather et al. 1997b), and, as before, 30-dBZ storm volumes were the experimental units, or cases, that were tracked and studied. A Z–R relationship of Z = 200R1.6 was used to estimate rain rate R from base-scan reflectivity Z. Extensive effort has gone into improving radar signal processing and calibrations (Terblanche 1996), and comparing radar-estimated rainfall to that measured by rain gauges in South Africa (Mather et al. 1997a). Especially under convective storm conditions, the superior sampling capabilities of radar makes the use of gauge measurements as ground truth somewhat less than ideal.

The Thunderstorm Identification, Tracking, Analysis and Nowcasting (TITAN; Dixon and Wiener 1993) real-time system was installed at the radar, primarily to assist during operations. This application was the first use of TITAN in a weather modification environment and, although not used for the analysis presented in this paper, nevertheless proved to be a valuable operational tool for the radar operator. A radio-telemetry system was used to provide the radar operator with the aircraft global positioning system positions in real time and this information was also displayed on the TITAN system. Two seeding aircraft were used, an SAWB Aerocommander 690A carrying 36 flares and a WRC Commander 500S carrying 20 flares. The flares produce a spectrum of small (average diameter of 0.5 μm) hygroscopic particles (mostly potassium chloride and sodium chloride), and seeding was conducted in the updraft areas of the storms, just below cloud base. The main updraft area in the South African storms is normally on the upwind side of the storms. More detail on the physical and chemical properties of the flares, the criteria for selecting storms for seeding, and the manner in which seeding was done is given by Mather et al. (1997b). In contrast to the randomized experiment and because of the operational nature of the Tzaneen project, the number of flares per case was limited only by the maximum number of flares carried per aircraft. However, once airborne, the seeding pilot could select a new case (i.e., a different storm) only if he had 10 or more flares left after completing a previous case. A concerted effort was made to select storms for seeding early in their lifetime. By doing this, bias caused by intentionally selecting the best storms was limited because the aircraft was normally on its way to the storm before it reached 30 dBZ for the first time. During March 1995 and the period October 1995 to March 1996 a total of 60 storms were seeded.

Analysis of the radar data

The analysis of the radar data is an attempt to evaluate this operational cloud seeding program and to quantify the effects of the seeding material on the radar-estimated rain mass of the storms. This analysis relies on comparisons of seeded storms to control storms that are extracted from the Tzaneen radar data records. The results of this analysis, especially the seed–no seed differences, have been compared with the results of the randomized seeding experiment conducted in the Carolina and Bethlehem areas (Mather et al. 1997b). [The randomized experiment’s areas are a few hundred kilometers to the south of the Tzaneen area, in the Mpumalanga and Free State Provinces (see Fig. 1).]

A major problem that faces researchers in analyzing results from operational cloud seeding programs is the selecting of a group of unseeded (control) storms from the radar records that are comparable to the seeded storms. If it was possible to find a common time base, such as the time at which the decision as to whether to seed was made, then meaningful comparisons could be made. A common time base could be found if there was some way of assigning a decision time to selected control storms, that is, the time at which the storms would have been seeded if they were selected as targets. This process is bound to be somewhat subjective and therefore open to debate. A second, more objective method, proposed by Dr. Mather, is to compare the seeded and selected control storms from the time at which the radar storm tracking software started identifying the storms as 30-dBZ volumes. This kind of analysis is feasible in Tzaneen, where every effort was made to seed the storms early in their lifetimes, resulting in about half the seeding events commencing in the first 15 min after time of origin of the 30-dBZ storm. We call this a time-of-origin analysis.

Obtaining a general group of control storms

The first task was the selection of some group of control storms from the Tzaneen radar data. Control should be read in the context of the operational nature of the project. Based upon the observed characteristics of the seeded storms for the 1995/96 season, the following threshold criteria were applied by Dr. Mather to the unseeded storms of that season to select a control group that is at least comparable to the seeded group:

  • minimum storm duration (decimal hours), 0.55
  • time of origin (SAST), 1100–1900
  • mean range from radar (km), 10–95
  • minimum storm echo top (m), 4600
  • minimum storm volume (km3), 55
  • minimum storm area (km2), 20
  • minimum storm rain flux (m3 s−1), 34
By applying these criteria to the 1995/96-season Tzaneen radar records, 194 storms were obtained. When the average storm volume, rain mass, duration, and so on of these 194 control storms were compared with those of the 60 seeded storms, it became apparent that the control group’s storms are substantially smaller. Hence the 60 largest storms from the set of 194 storms were selected as an additional set of control storms. This subset of 60 storms are, on average, larger than the 60 seeded storms. The 60 largest storms can be considered as an “equivalent group of the largest comparable unseeded storms from the area.”

Despite the fact that range limits were set when identifying the control storms, it was found that the seeded storms’ average range from the radar was just over 45 km as compared with about 62 km for the control groups (including the group discussed in the next section). Although this difference was found to be statistically significant, it was also found that the correlation between mean range and rain mass for storms for all groups (seeded, control, and all storms) was small and not statistically significant. It is important to investigate this aspect because the C-band radar used is affected by attenuation, and, if the control storms are, on average, located further out from the radar than the seeded groups, they could artificially appear weaker. Although of concern here, it appears not to have been a major factor in the observed differences between the groups.

Quartiles were computed for the radar-estimated rain mass, in 5-min time intervals from storm origin, produced by the 60 seeded storms, the 194 control storms (C1), and the 60 largest control storms (C2). These quartiles divide the rain mass of each sample into four equal groups. The first quartile represents the smaller storms, the second is the median, and the third quartile indicates how the larger storms are behaving.

If we can assume that a randomly chosen group of control storms would follow the seeded group for at least the first few time intervals after time of origin (before seeding effects are expected on rain mass), then Figs. 2a–c, which represent the three quartiles, suggest that such a group would lie somewhere between the 194 and the 60 largest control storms extracted from the Tzaneen radar data. Because each quartile exhibits a different initial growth rate in rain mass (about 20 metric ktons per 10-min time interval for the first quartile, 40 for the second, and 70 for the third), the idea of selecting controls by matching the seeded storms with unseeded storms that have the closest initial average size and growth rate in rain mass seemed feasible. Clearly, using the 60 largest unseeded storms will bias the analysis against a positive seeding result. Nonetheless, the seeded storms’ rain mass in the second and third quartiles (Figs. 2b and 2c) peaks higher and later, and lasts longer, than the rain mass from the 60 largest control storms.

Obtaining a more specific group of “control” storms by pairing

There is a marked difference in the initial growth rate (in rain mass) among storms in the three quartiles, as seen in the previous section. The results from the NPRP randomized experiment indicated that it is unlikely that rain mass would be affected by seeding within the first 20 min after seeding started. An additional control group can therefore be obtained by matching each of the 60 seeded storms with an unseeded storm in the total dataset (both seasons) that best resembles it with respect to average rain mass at the lowest scan during the first 20 min from the time of origin. A least squares criterion was used for this purpose; the matching unseeded storms were selected in such a way as to minimize the sum of the squares of the differences between rain masses of the seeded and unseeded storms during the first 20-min intervals after being identified as a 30-dBZ storm.

Using rain mass for the matching also resulted in no initial bias between rain mass of the seeded and control groups. The quartiles of the distributions of rain mass for the seeded and unseeded storms are displayed in Fig. 3.

Inspection of Figs. 2 and 3 indicates the same trends of higher and later peaks in the second and third quartiles for the seeded storms’ rain mass as compared with the control storms, with even the first quartile showing this trend in Fig. 3. An accumulated mean rain mass analysis of the seeded group and the control group obtained in this manner indicates a possible 66% average increase in accumulated rain mass over a 100-min interval since time of origin in the seeded group, as shown in Fig. 4.

Although this experiment is not randomized, permutation tests were used to test for a difference between the quartiles of the seeded and the matching unseeded storms’ rain mass at lowest scan. In this case one has to be careful with the interpretation of p values, because seeding may possibly not be the only difference between the two groups of storms—preselection of suitable storms to be seeded may be another or a partial explanation for the difference. The results obtained should be interpreted only as the strength of the differences between the two groups. The p values were obtained from the permutation test for the null hypothesis that the quartiles are the same, against the one-sided alternative that the quartile of the seeded group is larger than the corresponding quartile of the unseeded group. The differences between the quartiles of the 60 seeded and the 60 matched control storms as well as the results from the permutation tests are given in Table 1. The positions of the quartiles q were calculated using the formula for discrete values [i.e., the position of qi is i(n + 1)/4, where i = 1, 2, and 3, and n is the number of observations]. Note that zero is not included in the intervals for the differences between the medians from the 40–50 min time intervals onward, and for the third quartiles from the 30–40 min time intervals onward. This result indicates that there are indeed differences between the medians and between the quartiles of the seeded and control storms for these time intervals.

Comparisons with the NPRP randomized experiment

The manner in which the operational experiment was conducted and the data were collected and analyzed allows one to compare the differences between the Tzaneen seeded and control groups to the differences observed in the randomized hygroscopic flare seeding experiment (Mather et al. 1997b). The motivation for doing this comparison is to verify whether common seeding signatures can be isolated. This comparison will also provide additional confidence in the validity of the control group. Although the storm characteristics as such might differ between the project areas because they are separated by a few hundred kilometers, one can expect similarities in the hygroscopic seeding signatures as observed in the radar seed–control differences. As for the permutation test, in this section the 60 storms obtained by matching are used as the control group for the operational dataset throughout.

A comparison between the quartiles of the seeded storms in the two experiments (Mather et al. 1997b) shows that the seeded storms from both groups peak at about the same time, but the Tzaneen storm rain mass in all three quartiles exceeds those from the randomized experiment in later time windows, despite the fact that the Tzaneen storms were initially smaller. This result could be related to the longer seeding periods employed in the operational program. Most of the storms in the randomized experiment were seeded with a maximum of 10 flares (approximately 20-min seeding time).

We decided to use p values for the seed–control differences in variables, extracted by the software developed by Dixon and Mather (1986), of the randomized experiment and the Tzaneen data as a way to identify the strength of common trends in the time evolution of seed–control differences. These differences were compared in 10-min time intervals from decision time in the case of the randomized experiment and from time of origin in the case of the Tzaneen data, because most of the Tzaneen storms were seeded early in their lifetimes. As before, keep in mind that these p values are not used to measure statistical significance but rather are used in a purely exploratory manner to indicate the strength of the differences. The variables that scored p values of 0.05 or less between seed and control in the same time interval for both datasets show a trend that is commensurate with the hygroscopic seeding hypothesis, which is based on enhancing coalescence in continental convective storms and thereby improving the rainfall production efficiency of these storms, as set out in Mather et al. (1997b).

For both datasets the maximum rate of increase of the height of peak reflectivity (in the 0–10-min interval) exhibited the greatest seed–control difference. This result is followed by the height of the vertical centroid (in the 10–20-min interval) and then the height of the peak dBZ (in the 20–30-min interval). Differences between the seeded and control storms result from stronger radar returns (greater reflectivity) in the upper storm regions caused by the enhancement or acceleration of the coalescence process (Cooper et al. 1997). In the 30–40-min interval, the maximum rate of increase of rain mass at base scan is identified, followed by base-scan area and base-scan rain mass in the time intervals up to 60 min. Also of interest is the observation that mass as a function of height, percentage area and volume as functions of reflectivity, the vertical centroid, and the reflectivity-weighted centroid also score strongly in the intervals after 40 min. This result is, as has been pointed out by Mather et al. (1997b), indicative of a top-heavy, dense storm in which precipitation formation is probably still active.

Consider for a moment the natural evolution of a storm. As the storm builds, the volume (area) occupied by greater reflectivity increases, reaching a peak at around 20 min after the time of storm origin. At around this time, the storm is in its most compact state. The first rainfall at cloud base is often the most intense. This intense phase is followed by a gradual spreading out of the storm into cloud with a weaker echo profile (production of an anvil, etc.) and by less intense rainfall, which spreads over a larger area at cloud base. The findings discussed in this section suggest that the hygroscopic seeding is preserving the storm’s denseness by keeping the storm volume and lowest scan area filled with greater reflectivity. If this suggestion is indeed what this observation means, then it is a very important observation, because it clearly distinguishes seeded storm behavior from that of natural storms.

Analysis of the rainfall data

The objective of the rainfall analyses is to develop a methodology whereby a seeding response, assuming one exists, can be detected from routine rainfall measurements made within the target area of an operational seeding program. District rainfall records spanning a period in excess of 70 yr were obtained from SAWB for the districts shown in Fig. 1. The analysis was split into two parts: 1) March 1995, which was the first month of operational seeding, and 2) the rainy season from 1 October 1995 to 31 March 1996. The rainfall characteristics of the districts analyzed are vastly different, with topography and the proximity of the escarpment playing major roles. For comparison purposes, the mean and standard deviations for the 6-month period were taken into account by calculating normalized values z = (xm)/s, where x is the observed value, m is the mean, and s is the standard deviation. In this manner, the differences in natural variability in the districts’ rainfall, prior to the seeding operations, are taken to account.

Figure 5 depicts the contour plot of z values for March 1995 with the seeded storm tracks superimposed. For the period, the upper-level steering winds were predominantly from the southeast, resulting in storms generally moving from the southeast to the northwest. A maximum in the z values is approximately centered on the target district.

The seeded storm tracks and the contour analysis of the z values for the 1995/96 season are shown in Fig. 6. District 49 was targeted most frequently during the season, and a local maximum in the z values can be discerned. The 1995/96 rainy season was one of the most unusual of the historical record to date, and many maximum rainfall records were broken.

Although the historical data are more than sufficient, several summer seasons with operational seeding in the area are required to establish the existence of a seeding response in rainfall records. Of more importance should be attempts to relate the apparent storm-scale seeding effect on rainfall discussed in section 3 to the observed district rainfall features. With the information available, it is possible to do simple calculations that highlight the issues that should probably receive the bulk of attention by those involved in rainfall enhancement but are often ignored.

Consider district 49 for the 1995/96 season. From Fig. 6 it can be seen that a conservative estimate of the z-value deviation in this district would be one standard deviation—238 mm. The surface area of this district is approximately 12 800 km2, implying that 3 × 106 metric kilotons of “additional” rainfall was needed for the observed z-value pattern. From Fig. 4, on the other hand, it can be seen that seeded storms, on average, probably produce 500 kilotons more rain than their natural counterparts. Therefore, for the observed z-value pattern to be linked realistically to cloud seeding activities, on the order of 6000 storms needed to be seeded. It is therefore most unlikely that the observed pattern is related in any significant way to the cloud seeding activity. We call this the “two orders of magnitude challenge.” To address this challenge, much better use will have to be made of modern technology, including mesoscale models and sensitive multiparameter radar, which would, for example, be able to forecast and identify boundary layer convergence fields as the precursors of convective development. Detailed radar-based storm climate datasets also need to be compiled to identify areas with a high frequency of seedable storms, and so on.

The above exercise highlights the importance of

  • determining a storm-scale cloud seeding effect,
  • establishing whether the storm-based seeding effect on rainfall quantitatively matches apparent area effects, and
  • recognizing the logistical challenges of seeding the number of storms that would produce significant area rainfall effects.

Summary and conclusions

The radar analyses

A new approach to analyzing results from an operational rain augmentation program is presented, which is based upon the selection of comparable control storms and compares seeded and control storm properties from the time of origin of the storms. Key elements that make this analysis possible are early seeding in the lifetimes of the storms and objective radar storm analysis software such as that of Dixon and Mather (1986) as used here or the improved TITAN system of Dixon and Wiener (1993). Similar radar-measured storm properties probably are responding in both the seeded Tzaneen and the randomly selected seeded storms of NPRP, when compared with their control storms. This analysis also provides an idea of what the storm-scale rainfall increase from seeding might be. As more experience is gained from additional seasons of operational seeding in the Tzaneen area, it may be possible to construct a stand-alone analysis that will not depend so heavily on the experience gained from the randomized experiment.

Significant differences in seeded and control storm-track properties that are measured by radar in both analyses in order of appearance from time of origin or decision time are

  • maximum rates of increase in the height of peak dBZ,
  • maximum rates of increase in lowest scan radar reflectivity,
  • increases in storm volumes and areas occupied by greater radar reflectivity, and
  • increases in radar-estimated rain mass.
Issues that should be kept in mind in this type of analysis include
  • verifying what effects differences in range from the radar could have on the results, and
  • the need to compile a detailed radar-based storm climate dataset to investigate how topographic and other effects are modifying storm behavior within an area and between adjacent areas.

The rainfall analyses

Many analyses have been carried out on the rainfall recorded in March 1995 and for the 1995/96 season, none of which at this stage is statistically significant. Two anomalies in the rainfall patterns close to and in the operational target area around Tzaneen have been recorded. If these trends continue in succeeding seasons, rigorous statistical conclusions may be drawn. However, the evaluation of an operational cloud seeding effort should not depend solely on measured rainfall. It has to be supported by the analysis of the radar data in such a way that the total storm-based seeding effect can be quantitatively linked to any observed rainfall anomalies. Based on simple calculations done in this paper, it becomes clear that the observed district rainfall anomalies in the period studied are probably not related to the cloud seeding activities.

Developments since 1997 and the way forward

The NPRP was terminated in April 1997 when the WRC was informed by the Ministry of Water Affairs and Forestry that they should focus their financial support toward more critical water needs in South Africa. Fortunately, a new project proposal for the South African Rainfall Enhancement Programme (SAREP), in which there was a stronger emphasis on operational application and capacity building, received a favorable response by the departments of Agriculture, Water Affairs and Forestry, and Environment Affairs and Tourism. Toward the end of November 1997 the required funds were transferred to WRC to contract with the University of South Africa and CloudQuest for their contributions to the new program. Seeding operations again started in the Tzaneen area in December 1998. The experience gained from the NPRP randomized experiment and the previous work in the Northern Province described in this paper were valuable in deciding on the approach to be followed. A significant step forward was the decision that the TITAN software (Dixon and Wiener 1993), apart from its established use during operations, will form the basis of all storm-based analyses in SAREP.

The time-of-origin analysis could turn out to be a good way of studying natural storm behavior. If normal storm behavior can be well understood, then the matching procedure used to obtain a “control” group could be refined, and differences brought about by seeding should be more readily revealed.

It may be possible to refine the selection of control storms chosen from the Tzaneen radar data. As long as the seeded storms are outperforming an equivalent number of the largest unseeded storms in the area, however, it will be difficult to refute the claim that the seeding is indeed producing more rainfall on storm scale in the area.

Operating the radar on a 24-h basis so as to collect data from all storms (and other rainfall-producing systems in the experimental area) would facilitate these analyses and give a handle on possible area effects. Final evidence of such an increase will have to appear in the analyses of the district rainfall and streamflow records. This analysis may take 3–5 yr to reach acceptable levels of statistical significance, but only if the logistical challenges of seeding the necessary number of storms that will allow linking total storm-scale effect to area effects are addressed. The agricultural and hydrological communities in South Africa, who are now funding the operations, should also in the meantime employ models to understand the possible benefit–cost ratio of the operations.

Acknowledgments

This paper represents some of Dr. Graeme Mather’s final unpublished ideas and work in South Africa. Graeme passed away on 15 August 1997. On 13 January 1999, Mr. Rob Parsons, our coauthor and Graeme’s successor at CloudQuest, also passed away. We sadly miss them and their contributions, especially in the light of the challenges faced by the scientific community regarding weather modification in the new millennium.

The Northern Province Tzaneen experiment was funded by the South African Weather Bureau, (Department of Environment Affairs and Tourism), the Water Research Commission, the Department of Water Affairs and Forestry, and the Northern Province Government. The authors are grateful for the dedicated contributions by the scientific and technical personnel and the pilots of the project.

REFERENCES

  • Cooper, W. A., R. T. Bruintjes, and G. K. Mather, 1997: Calculations pertaining to hygroscopic seeding with flares. J. Appl. Meteor.,36, 1449–1469.

  • Dixon, M. J., and G. K. Mather, 1986: Programme for Atmospheric Water Supply—Phase 1, 1983–1986, Vol. 3. WRC Rep. 133/3/88, 55 pp. [Available from Water Research Commission, P.O. Box 824, Pretoria 0001, South Africa.].

  • ——, and G. Wiener, 1993: TITAN: Thunderstorm Identification, Tracking, Analysis, and Nowcasting—a radar-based methodology. J. Atmos. Oceanic Technol.,10, 785–797.

  • Mather G. K., M. J. Dixon, and J. M. de Jager, 1996: Assessing the potential for rain augmentation. The Nelspruit randomized convective cloud seeding experiment. J. Appl. Meteor.,35, 1465–1482.

  • ——, D. E. Terblanche, and F. E. Steffens, 1997a: National Precipitation Research Programme: Final report for the period 1993–1996. WRC Rep. 726/1/97, 147 pp. [Available from Water Research Commission, P.O. Box 824, Pretoria 0001, South Africa.].

  • ——, ——, ——, and L. Fletcher, 1997b: Results of the South African cloud-seeding experiments using hygroscopic flares. J. Appl. Meteor.,36, 1433–1447.

  • Terblanche, D. E., 1996: A simple digital signal processing method to simulate linear and quadratic responses from a radar’s logarithmic receiver. J. Atmos. Oceanic Technol.,13, 533–538.

  • Woodley, W. L., A. G. Barriston, J. A. Flueck, and R. Biondini, 1983:The Florida Area Cumulus Experiment’s second phase (FACE-2). Part II: Replicated and confirmatory analysis. J. Climate Appl. Meteor.,22, 1529–1540.

Fig. 1.
Fig. 1.

A map of the Northern Province showing the main topographical features of the area, the location of the radar close to Tzaneen, and its 134.4-km data collection range.

Citation: Journal of Applied Meteorology 39, 11; 10.1175/1520-0450(2001)039<1811:TTOAOH>2.0.CO;2

Fig. 2.
Fig. 2.

Quartile analysis, from time of origin, of rain mass at lowest scan for the seeded and control storms. Here C1 represents the 194 control storms, C2 represents the 60 largest control storms, and the bold line represents the 60 seeded storms. (a) First quartile; (b) second quartile (median); (c) third quartile.

Citation: Journal of Applied Meteorology 39, 11; 10.1175/1520-0450(2001)039<1811:TTOAOH>2.0.CO;2

Fig. 3.
Fig. 3.

The first (Q1), second (median) (Q2), and third quartile (Q3) of the rain mass at lowest scan for the 60 seeded (solid lines) and matching 60 unseeded storms (dotted lines) in 10-min time intervals from time of origin.

Citation: Journal of Applied Meteorology 39, 11; 10.1175/1520-0450(2001)039<1811:TTOAOH>2.0.CO;2

Fig. 4.
Fig. 4.

Accumulated mean rain mass of the 60 seeded storms (solid line) in comparison with the accumulated mean rain mass from the matching 60 unseeded storms (dotted line) in 10-min time windows from time of origin.

Citation: Journal of Applied Meteorology 39, 11; 10.1175/1520-0450(2001)039<1811:TTOAOH>2.0.CO;2

Fig. 5.
Fig. 5.

A map of SAWB rainfall districts overlayed by the radar area as well as the seeded storm tracks and contours of normalized rainfall values for Mar 1995.

Citation: Journal of Applied Meteorology 39, 11; 10.1175/1520-0450(2001)039<1811:TTOAOH>2.0.CO;2

Fig. 6.
Fig. 6.

Same as Fig. 5, but for Oct 1995–Mar 1996.

Citation: Journal of Applied Meteorology 39, 11; 10.1175/1520-0450(2001)039<1811:TTOAOH>2.0.CO;2

Table 1.

Differences between the quartiles of the 60 seeded and the 60 matched control storms, the p values, and the corresponding 90% confidence intervals obtained from the permutation tests for rain mass at the lowest scan.

Table 1.
Save