An aircraft cloud-seeding program was conducted in northern Jiangxi Province in eastern China during 2008–14. Supercooled clouds were seeded at their base and/or top with AgI. Seeding at the cloud-top level was normally initiated in the temperature range (−7 to −12°C). Base seeding with flares was carried out when the clouds were mature and precipitating, whereas top seeding was usually performed before the clouds began to precipitate. Each individual seeding operation released 80–120 g of AgI.
The seeding period was selected as the winter months (November–February) during 2008–14. Little or no seeding was conducted during the winter months from 1978 to 1997 and therefore it was selected as the control period. A total of 71 operational aircraft cloud-seeding campaigns were carried out during the seeding period. The downwind effect of seeding is inextricably associated with the motion of upper-level winds and cloud systems during the seeding period. Mean wind fields for the seeding period were therefore determined by obtaining the 700-, 600-, and 500-hPa pressure level data from the ECMWF reanalysis dataset (Fig. 1). The wind flow during the seeding period was mainly from the west and the winds had an average speed of more than 15 m s−1. The weather stations used in this study are all National Automatic Stations (Table 1).
Figure 1. Locations of stations used to assess the extra-area effect of operational cloud seeding in Jiangxi Province in winter months (November–February) during 2008–14. The daily mean wind field at 600 hPa for a total of 71 cloud-seeded days is also shown. The blue, yellow, and green dots donate the weather stations in the seeded (target) region, weather stations in the downwind area, and control stations, respectively. The areas outlined by black and blue dash-dotted lines are the range of operational seeding and identified target region, respectively. The radar is located at station Nanchang (28.59°N, 115.9°E). Heights are given in meters above the mean sea level. All stations are described in Table 1.
Station ID Latitude (°N) Longitude (°E) Control Pingxiang (PG) 57786 27.63 113.85 Lianhua (LH) 57789 27.13 113.95 Anfu (AF) 57798 27.40 114.60 Yongxin (YOX) 57891 26.93 114.25 Jinggangshan (JGS) 57894 26.58 114.17 Target Yongxiu (YX) 58509 29.05 115.82 Duchang (DC) 58517 29.27 116.20 Gaoan (GA) 58605 28.42 115.38 Nanchang (NC) 58607 28.55 115.95 Zhangshu (ZS) 58608 28.07 115.55 Jiujiang (JJ) 58502 29.73 116.00 Jinxian (JX) 58614 28.38 116.27 Chongren (CR) 58710 27.77 116.05 Xinyu (XY) 57796 27.80 114.93 Downwind Wannian (WN) 58615 28.68 117.08 Yujiang (YJ) 58616 28.20 116.82 Leping (LP) 58620 28.97 117.13 Dongxiang (DX) 58622 28.95 117.58 Shangrao (SR) 58623 28.47 117.92 Yiyang (YY) 58624 28.40 117.43 Hengfeng (HF) 58625 28.42 117.60 Guixi (GX) 58626 28.30 117.23 Qianshan(QX) 58629 28.32 117.70 Jinxi (JIX) 58712 27.92 116.78 Zixi (ZX) 58713 27.72 117.07 Wuyuan (WY) 58529 29.27 117.85
Table 1. Information of the control, target, and downwind National Automatic Stations used in this study
In addition to the weather stations in the target region, a number of stations downwind of the target area were selected to explore the changes in precipitation in the downwind area caused by cloud seeding (see Fig. 1 and Table 1 for locations of these stations). The stations are representative and were not contaminated by other cloud-seeding activities during the selected time periods.
To quantify the impact of aircraft seeding on seasonal precipitation in Jiangxi Province, we adapted the commonly used historical regression method for evaluating the effect of cloud seeding (Dennis, 1980). This regression equation is based on the historical relationship between variables in the designated target area and those in the selected control area. Records of variables to be tested are acquired for a historical period of over 20 years (Griffith et al., 2005). Following the well-established practices (e.g., Solak et al., 2003), the seasonal accumulated precipitation was taken into account as an important input variable. Five stations in northwestern Jiangxi and upwind of the target area were selected as the control group (Table 1). Among all the stations, the seasonal average precipitation at these five stations had the best correlation with the downwind group (Table 2). Linear regression equations were then established between the control group and each downwind station (Table 3) for the seeding period.
Station Correlation p-value PG 0.89 6.91E–08 LH 0.88 1.88E–07 AF 0.91 1.88E–08 YOX 0.84 2.27E–06 JGS 0.83 2.99E–06
Table 2. Correlations of precipitation among the five control sites and the group of downwind sites. The one-tailed p-values were obtained by using the Student’s t-test assuming unequal variances. The same method was also applied to test the correlations among the five control sites and group of target sites, and similar results were obtained (not listed)
Station Regression equation WY y = 1.1012x – 16.1339 LP y = 1.1403x – 30.4658 DX y = 1.116x – 14.130 WN y = 1.134x – 19.785 YY y = 1.1551x – 22.4279 HF y = 1.1201x – 15.8786 SR y = 1.1176x – 7.2871 YJ y = 1.1616x – 23.54239 GX y = 1.1370x – 20.6839 QX y = 1.0396x + 5.3449 JIX y = 1.0029x + 44.0250 ZX y = 0.96337x + 40.06729
Table 3. Linear regression equation for winter rainfall (mm) between the control group (represented by x) and each downwind station (represented by y)
The goodness of fit of the regression model was evaluated by using scatterplots between the predicted (calculated by using the relationship in Table 3) and observed precipitation in the downwind region during 1978–97. The determination coefficient R2 of each regression equation in Table 3 is shown in Fig. 2. The line 1:1 indicates good agreement between the predicted and observed rainfall, and the determination coefficients show that this model performs well.
Figure 2. Scatterplots of the rainfall fitted by linear regression vs. the observed rainfall at the 12 downwind stations during 1978–97. Blue lines represent the 1:1 line. R2 is the determination coefficient.
Equations were also developed between the control group and each of the nine stations in the target region by using the same method. These regression equations were used to predict the natural precipitation at the target and downwind stations. The observed rainfall and its prediction in the seeded seasons were compared and the effects of seeding were tested. The statistical method remains uncertain because only the precipitation falling on the ground is evaluated. In addition to the amount of precipitation, other factors such as the cloud height, cloud liquid water content, cloud base temperature, and boundary layer depth, may also contribute to the observed differences in precipitation processes due to cloud seeding. It is therefore necessary to establish the relationship between the statistical results and physical variations in clouds.
Radar-based evaluation methods have become more generally used in the evaluation of cloud-seeding experiments during the past 20 years by either tracking a reflectivity maximum or advecting a circle around the seeding target (Maier et al., 2009). The physical testing in this study was based on a seeding case study using radar data from a station at Nanchang (28.59°N, 115.9°E) in Jiangxi Province, which is part of the New Generation Weather Radar System (CINRAD) of China. The temporal resolution was 6 minutes and a monitoring radius of 200 km was achieved by this radar system. A fuzzy logic algorithm was used to remove anomalous propagation echoes (Kessinger et al., 2003; Liu et al., 2007).
A sophisticated algorithm similar to the Thunderstorm Identification, Tracking, Analysis, and Nowcasting (TITAN) algorithm (Dixon and Wiener, 1993; Han et al., 2009) was applied to track the seeded cloud unit (target unit) under the investigation and to search for alternative control units. The most objectively comparable control unit was then defined and tracked by applying the following steps.
(1) The target area was determined by using the seeding trajectory information combined with the dispersion of the seeding agent. The target unit was defined when an echo first reached a threshold of 30 dBZ within a volume > 10 km3 in this area. Other units outside the target area that satisfied the same tracking criteria were also tracked as reserved units.
(2) All units were tracked before and after the initiation of seeding. Although their complete life history could be tracked, the pre-seeding history was defined as about 30 min prior to the initiation of seeding, and tracking was terminated when echoes larger than the threshold of 30 dBZ faded away.
(3) The meteorological and topographical similarity is required while selecting control units to match the target units. Potential control units should not be contaminated by the cloud-seeding activities.
(4) The echo-top height (km), echo volume (km3), maximum reflectivity (dBZ), vertically integrated liquid water content (VILWC; kg m−2), and precipitation flux (m3 s−1) calculated from the maximum radar reflectivity factor, were recorded (Marshall and Palmer, 1948). Multiple potential units with a history similar to the pre-seeding history of the target unit were defined automatically. The most probable unit was determined as the control unit by using our assessment system. Evolution of the control unit after the certain history stage was regarded as natural development of the target unit.
Variations in these parameters between the target and control units were analyzed over time. The seeding operation is only considered to be effective if these parameters increase after cloud seeding.
Table 4 lists the average winter rainfall enhancement ratio [(observed rainfall – predicted rainfall)/predicted rainfall] for the 9-station target and 12-station downwind groups during 2008–14, calculated by using the regression analysis described in Section 2.2.
Region Enhancement ratio p-value Target 17.30% 0.25 Downwind 21.67% 0.0013
Table 4. Statistics for the enhancement ratio over the target and downwind regions. The one-tailed p-values were obtained by using the Student’s t-test assuming unequal variances
Figure 3 shows that the average rainfall enhancement ratio for the 9-station target group is 0.17, suggesting an increase of about 17.3% in the amount of rainfall during the 7-yr cloud-seeding operations. Most of the 9 stations had a positive ratio during 2008–14, except for 2011. Values mostly varied between 0 and 0.4. Previous studies (e.g., Jin et al., 2013) have shown that the observed winter rainfall in 2011 was clearly less than that in other years, which might, in part, explain the negative value. The increase in rainfall has a relatively large p-value (0.25), which is not significant.
Figure 3. Interannual variations in the average rainfall enhancement ratio [(observed – predicted)/predicted] for the (a) 9 stations in the target area and (b) 12 stations in the downwind area during winters of 2008–14.
Similarly, the average rainfall enhancement ratio of the 12-station downwind group is 0.21 (p = 0.0013), suggesting a clear increase in rainfall of about 21.67% after the 71 cloud-seeding operations. Table 5 lists the statistical effects with distance. The results show that winter rainfall in the downwind area has a fair chance of increasing by about 21.67% relative to the predicted rainfall after cloud-seeding operations. Values varied in the range 0–0.4. The enhancement ratio of the downwind stations was larger than that of target stations and did not decrease with distance from the target as expected. This may be because (1) dose of the catalyst released by the seeding aircraft as well as the seeding time were insufficient during this period since the aircraft seeding trajectory was not exactly scientific under some conditions, or (2) the wind speed was so high (> 15 m s−1; Fig. 1) that activation time of the catalyst was not sufficiently stable in the target area and the catalyst was probably transported to the downstream domain before the full activation occurred. Other environmental factors, including the effects of Poyang Lake, wind shear, humidity, vertical velocity, and moisture flux convergence, could not be excluded.
Station Distance from the target (km) Seeding effect (mm) Ratio (%) p-value 2008 2009 2010 2011 2012 2013 2014 YJ 20 34.63 60.89 64.46 10.01 29.10 70.13 –17.51 15.05 0.013 JIX 25 –12.98 49.79 72.50 96.35 119.86 74.23 –12.05 27.32 0.014 WN 35 58.26 97.90 116.55 31.60 53.95 16.82 42.47 26.09 0.002 GX 40 46.95 72.13 94.50 35.40 33.61 74.79 35.73 21.78 0.000 LP 44 88.97 86.48 58.24 –11.08 47.23 –9.64 246.41 29.02 0.035 ZX 73 –28.01 14.98 112.48 59.56 66.16 87.14 –13.25 30.48 0.038 YY 77 61.02 94.23 140.71 28.89 84.67 114.52 38.06 23.20 0.001 DX 80 99.87 137.87 117.60 –10.03 61.54 20.75 74.40 12.97 0.006 HF 90 72.92 76.20 177.45 39.23 90.89 108.17 20.25 20.51 0.002 WY 104 45.18 86.58 88.30 –37.33 –0.03 –2.64 103.42 21.45 0.049 QX 108 14.37 65.47 133.34 28.35 60.56 67.24 43.07 18.23 0.003 SR 126 21.25 60.67 160.15 61.06 58.94 72.67 23.57 14.75 0.005
Table 5. Winter seeding effects at each of the 12 downwind stations and p-values calculated by the Student’s t-test after cloud seeding during 2008–14. Distance between the target stations and each individual downwind station is also given
The seeded day (29 November 2014) was selected as a case for a-posteriori research. The cloud in this case was cumulus embedded stratus. The seeding was initiated at 1408 Beijing Time (BT) and terminated at 1505 BT. Figure 4 shows the target unit on the seeding trajectory, which is labeled as #8 in the tracking system. A total of 6 units were tracked as potential control units and are labeled as #12, #19, #29, #45, #49, and #51 in the tracking system.
Figure 4. Spatial distribution of the radar composite reflectivity (dBZ) for the target unit over northern Jiangxi Province at 1436 BT 29 November 2014 when the unit was seeded by an aircraft. Radar echoes were derived from volume scanned radar data deployed at Nanchang (28.59°N, 115.9°E). Units with echoes larger than the threshold of 30 dBZ were tracked and displayed. Control units were selected around the seeding period (1408–1505 BT). Six units were tracked as potential control units with labels of #12, #19, #29, #45, #49, and #51 (not marked here).
Weather conditions before the seeding were examined to verify the rationality of the seeding treatment. Figure 5 shows the 500-hPa geopotential height fields and 850-hPa wind fields at 0800 BT on the seeded day (Fig. 5a) and the water vapor image at 1400 BT 29 November 2014 (Fig. 5c). Most of southern China (including Jiangxi Province) was controlled by a southwesterly trough on this day (Fig. 5a). Figure 5a also shows the low-level jet at 850 hPa, which favors the transport of water vapor to the target area from southern oceans. Figure 5b shows that the low TBB clouds covered northern Jiangxi, indicating a higher water vapor content over the target area. This combination of prevailing southwesterly wind plus sufficient water vapor means that water vapor could be easily transported to northern Jiangxi at the time of seeding, favoring the cloud-seeding operation.
Figure 5. (a) 500-hPa geopotential height field (dagpm) and 850-hPa geopotential wind field (m s−1) at 0800 BT 29 November 2014 and (b) water vapor TBB (K) derived from FY-2 satellite data at 1400 BT 29 November 2014. Dark lines are the 500-hPa contour lines and arrows represent the wind direction and speed at 850 hPa.
Figure 6 shows the composite reflectivity of the target and control units after the ignition of seeding, allowing the evolution of clouds to be tracked over time. The echo of the target unit strengthened over time. There was a clear change in the size of the echo area and intensity from one hour after the initiation of seeding in the target unit, whereas the control unit disappeared in the Plan Position Indicator (PPI) display. It is shown that the control unit was decaying and eventually missed the recognition criteria of the tracking system. Lifetime of the target unit was therefore extended relative to that of the control unit as a result of seeding. This increasing trend persisted for almost two hours, indicating that the development stage of clouds (with echoes > 30 dBZ) was sufficiently extended to produce much more rainfall in the downwind region.
By using the 30 dBZ threshold, the target unit (#8) was tracked and 6 alternative control units (#12, #19, #29, #45, #49, and #51) were defined.
Figure 7 shows the variations in five physical parameters of the target unit and six control units derived from radar data before and after cloud seeding. Among the control units, #19 was found to be the most suitable control unit with which to observe the evolution of clouds before starting to seed the target unit. A clear difference in the lifetime—that is, the duration of echoes > 30 dBZ—of the target and control units was observed. Lifetime of the control units after seeding was 30 min, whereas for the target unit it was almost 2 h. The results indicated that the powerful development stage (with echoes > 30 dBZ) of the cloud was sufficiently extended for much more rainfall to occur. These parameters did not change that much in the control unit after cloud seeding. By contrast, these parameters did not increase so much in the first 30 min after seeding, but clearly increased 40–50 min after seeding, and this trend persisted for almost 2 h. The most sensitive radar parameters during seeding were the echo volume (Fig. 7b) and precipitation flux (Fig. 7e), suggesting that a larger precipitation potential was induced by cloud seeding.
Figure 7. Temporal evolutions of (a) echo-top height, (b) echo volume, (c) maximum reflectivity, (d) vertically integrated liquid water content (VILWC), and (e) precipitation flux derived from the S-band volume-scan radar data deployed at Nanchang (28.59°N, 115.9°E). The seeding period (1430–1436 BT) is shaded in pink during which the aircraft was passing through and seeding the target unit. The blue (#8), red (#19), and gray (#12, #29, #45, #49, and #51) lines denote the radar variables of the target unit, most suitable control unit, and alternative units selected during the matching process, respectively.
Values of these parameters were larger in the target unit than those in the control unit after seeding, resulting in a positive effect. Most AgI agents are activated 40–50 min after the initiation of seeding. The results showed that the effect of seeding persisted for two hours, leading to a longer period of positive effects. The seeded cloud might move out from the fixed target area, given that the cloud units moved at a speed of over 15 m s−1. Part of the increased precipitation induced by cloud seeding would fall into the extra area. This case study may therefore provide proof of a positive extra-area effect, assuming that the same technique is used on other seeding days with favorable synoptic conditions.
|Station||ID||Latitude (°N)||Longitude (°E)|