Transition Process from Non-Precipitating Cumuli to Precipitating Convective Clouds over Mountains : Observation by Ka-band Doppler Radar and Stereo Photogrammetry

Convective storms are frequently initiated over mountains under weak synoptic forcing conditions. However, the initiation process of such convective storms is not well understood due to a lack of observations, especially the transition process from non-precipitating cumuli to precipitating convective clouds. To investigate the initiation process, we conducted observations around the mountains in the Kanto region, Japan on 18 August 2011 using a 35 GHz (Ka-band) Doppler radar and a pair of digital cameras. The evolution of convective clouds was classified into three stages: convective clouds visible but not detected by the Ka-band radar (stage 0), convective clouds detectable by the Ka-band radar with reflectivity below 15 dBZ (stage 1), and convective clouds accompanied by descending echoes corresponding to precipitation (stage 2). During the transition process from stage 1 to stage 2, weak radar echoes rose to the higher level and reflectivity rapidly increased. This phenomenon suggests that drizzle particles produced in a preexisting convective cloud were lifted by a newly developed updraft, and raindrops were formed rapidly by coalescence of the drizzle particles and cloud droplets. This hypothetical process explains the precipitation echo formation in the lower layer frequently observed in the mountainous area in the Kanto region.


Introduction
A convective cloud is formed when the surface air is lifted above the level of free convection (LFC) if the environmental temperature lapse rate is greater than the moist adiabatic lapse rate (e.g., Holton 1992).Therefore, when a convective storm is generated, some force is needed to lift the surface air above the LFC.When a large-scale forcing is absent, convective storms frequently form near colliding gust fronts, boundaries of cold-air pools, and over mountains in warm and humid environments (Lima and Wilson 2008).In this paper, the formation process of convective storms over mountains under weak large-scale forcing is explored.
The mechanism for convective storm initiation over mountains in warm and humid environments is explained as follows: when a mountain slope is heated by solar radiation, the air above the slope becomes buoyant due to the horizontal temperature gradient between the air on the slope and the ambient air.As a result, an upslope wind is generated (Orville 1964), which transports water vapor from the surrounding plains to the mountain (Kimura and Kuwagata 1995).Thus, the air above the mountain becomes more unstable than that over the surrounding plains, and convective storms are initiated during the daytime (Sato andKimura 2003, 2005).
The development of convective storms over mountains is not from a single deep convective storm but from successive shallow convective clouds.Zehnder et al. (2007) observed the formation of convective storms over the Santa Catalina Mountains in southern Arizona using stereo photogrammetry.They found that a deep cumulonimbus cloud with a cloud top height of 10 km formed after the successive formation of shallow convective clouds.Zehnder et al. (2009) conducted radiosonde observations in the same area and suggested that the successively formed shallow cumuli produced favorable conditions for the subsequent development of deep convection by weakening the stability in the middle layer or moistening midlevel dry air.This process is called "preconditioning".Because the humidity of midlevel air is very important for the vertical development of convective clouds (Takemi et al. 2004), moistening midlevel air is the key process in preconditioning for deep convection.Kirshbaum (2011) simulated the formation of convective storms over the Black Forest Mountains in central Europe using a two-dimensional cloud-resolving model.He showed that midlevel dry air, which inhibited the development of deep convection, was moistened by the detrainment of cloud water from shallow cumuli before the onset of deep convective storms.
From the aforementioned studies, it was concluded that favorable conditions for the onset of deep convective storms over mountains are produced through the transportation of water vapor from surrounding plains to the mountains via upslope winds and water vapor transportation from the lower layer to the middle level by shallow convective clouds.However, the abovementioned studies focused on how the favorable conditions for the onset of deep convective storms are produced over mountains, and it is still unknown how the structure of convective clouds evolves during the initiation process of precipitation.Observational studies of convective cloud structure, including dynamics and microphysics, are essential for understanding the initiation process of convective storms.However, such studies are lacking because of the difficulty of observing non-precipitating clouds due to their weak radar reflectivity.
Techniques for observing non-precipitating clouds include stereo photogrammetry (Knight et al. 1983;Zehnder et al. 2007Zehnder et al. , 2009;;Kobayashi et al. 2012;Nishiwaki et al. 2013), use of Bragg scatter of radars (Knight and Miller 1998), and observations by millimeter-wave radars (Kobayashi et al. 2011;Sakurai et al. 2012;Nishiwaki et al. 2013).Among these meth-ods, the combined use of stereo photogrammetry and millimeter-wave radars is effective for these types of studies because both the outline and internal structure of convective clouds can be observed simultaneously.Nishiwaki et al. (2013) attempted to observe the initiation process of convective storms over mountains in Nikko, Japan, by a combination of stereo photogrammetry and a Ka-band radar.However, their Kaband radar did not have enough sensitivity to detect the internal structure of non-precipitating convective clouds due to a hardware problem.Therefore, they mainly used X-band radars.
In this study, we observed convective clouds initiated over mountains in Kanto, Japan using stereo photogrammetry and Ka-band Doppler radar.The minimum detectable reflectivity of the Ka-band radar used in this study is about −20 dBZ, which is small enough to observe the early stage of precipitating convective clouds.The data were analyzed with a focus on the transition process from non-precipitating cumuli to a precipitating convective cloud.The observed case has already been reported with regards to the first radar echoes (Sakurai et al. 2012) and the precipitation cores detected by X-band radars (Shusse et al. 2015).In contrast to these studies, we use stereo photogrammetry data and discuss the development of convective clouds from an earlier stage to the beginning of the precipitation stage.Japan Standard Time (JST; UTC + 9 hours) and altitude above sea level are used throughout this paper.

Study area and instruments
Figure 1 shows the location of the study area and the arrangement of instruments used in this study.The study area is in the mountainous region in the western part of Kanto, Japan.This area is usually dominated by subtropical high pressure during summer and shows clear diurnal variations of convective activity between the mountainous region and the surrounding plains (Sato and Kimura 2005).The Ka-band radar used in this study is a 35.35GHz scanning Doppler radar manufactured by Mitsubishi Electric Corporation.It transmits beams with a width of 0.28° and a pulse of 0.5 μs.The maximum observation range of the radar is 30 km, and the data are sampled at 50 m intervals in the radial direction.The minimum reflectivity detected by this radar is approximately −20 dBZ at a distance of 15 km.According to Sauvageot and Omar (1987), the threshold of reflectivity for distinguishing between drizzling and non-drizzling cloud is −15 dBZ for continental cumuli.Therefore, the Ka-band radar can observe convective clouds from the stage just prior to drizzle formation.The Nyquist velocity of 8.5 m s −1 is obtained using the spaced pulse-pair method (Hamazu et al. 2003).Unfolding of aliased Doppler velocities was conducted after the observation.
The Ka-band Doppler radar was installed at 35.8712°N and 139.2918°E to observe convective clouds initiated over the mountains to the west.Before a radar echo emerged, the radar was operated with sector scans for plan position indicator (PPI) at 5.6°, 7.3°, and 10.5° elevation angles in 212° azimuthal widths between 183° and 35°.The antenna scan speed for the sector PPI scans was 6.2° s −1 .After radar echoes were located, a combination of a sector PPI scan at an elevation angle of 7.1° and two Range Height Indicator (RHI) scans crossing the center of the radar echoes was repeated with manual operation.The azimuthal angle for the sector PPI was determined by operators at each scan to efficiently observe radar echoes.In most cases, azimuth angles from 210° to 330° were selected.It took between 16 and 19 s for each sector PPI scan.The azimuth angle for RHI scans was selected to cross the center of the radar echoes.It took between 12 and 20 s for each RHI scan.The combination of a sector PPI and two RHI scans was repeated at about 70 s intervals.In this paper, the start times of scans are presented for the radar data panels.
For the stereo photogrammetry, two digital cameras (RICOH CX-1) were set up at the Ka-band radar site (HDK) and at the top of Mt.Tenran (TRZ; 35.8619°N and 139.2918°E at an altitude of 197 m).The distance between the cameras was 1809 m.The size of the camera's image sensor was 6.2 × 4.7 mm and the number of pixels was 2592 × 1944.Though the two cameras were identical, the view angles were slightly different when we tested them using a theodolite: 60.42° × 41.35° for the camera at HDK and 60.27° × 41.27° for the camera at TRZ.The focal lengths of the lenses were derived from the view angles and the size of the image sensor as 5.32 mm and 5.34 mm, respectively.Photographs were taken automatically at 1 min intervals at the beginning of every minute.In order to simplify the method of the photogrammetry, the camera lenses were directed in the horizontal direction.Under this condition, an object on line l passing through the center of the lens P is projected at Q (x, y) on the image sensor (Fig. 2a) when the origin of the coordinate is the center of the image sensor O. Therefore, the azimuth angle α and the elevation angle ϕ relative to the line OP satisfy the following relationships: where f is the focal length of the camera lens.We can estimate the azimuth and the elevation angle of any objects in a picture from their coordinates in the image sensor using (1) and ( 2).When the relationship between object C and the two cameras A and B is represented as in Fig. 2b, the horizontal distances of C from the two cameras L 1 and L 2 , and the altitude of the object Z a can be calculated as follows (Kikuchi et al. 1988): where DZ is the devia- tion of the altitude of camera B from A, and Z A is the altitude of A.
In order to evaluate the accuracy of the stereo photogrammetry, we estimated the distance of Mt.Takamizu (Fig. 1b) from HDK (10.419 km) and its altitude (759 m) using photographs taken from HDK and TRZ.The errors were 478 m in the horizontal distance and 41 m in the vertical.These errors are small enough for our analysis.However, the error may increase depending on the position of the target.Thus, we estimated the errors for an imaginary target at a height of 5000 m, distance of 10 to 20 km from the radar site, and azimuth angle from 280° to 290°.These ranges correspond approximately to those of convective clouds observed in the present study.For this estimation, the same errors as that of Mt.Takamizu were given for azimuth and elevation angles.As a result, the errors in the distance were between 300 and 850 m and those for altitude were between 150 and 340 m.These values are small enough to not significantly impact our results.
For the analysis of the surface winds, the 10 min data from Automated Meteorological Data Acquisition System (AMeDAS) operated by the Japan Meteorological Agency (JMA) at Chichibu (CCB), Hatoyama (HTY), and Ogouchi (OGC) were used.The 10 min interval data derived from the Global Positioning System (GPS) network maintained by the Geospatial Information Authority of Japan were used for precipitable water vapor (PWV).This PWV dataset is distributed by Hitachi Zosen Corporation that calculates PWV from the zenith total delay analyzed using the RTNet software (Sato et al. 2009).For the upper sounding data, we used the observations by Tateno Aerological Observatory of the JMA located approximately 70 km east of the study area (Fig. 1a).For the surface rainfall, we used the JMA 1-km mesh 5 min interval composite radar data based on the operational C-band radars of the JMA.

Observation case
Simultaneous observation by the Ka-band Doppler radar and the digital cameras was conducted as a part of the Lifecycle of Cumulonimbus Experiment (LCbEx; Iwanami et al. 2015) in August 2011 during the Tokyo Metropolitan Area Convection Study (TOMAX; Nakatani et al. 2015).The stereo camera observations were sometimes disturbed by low-level clouds near the cameras but they were successful on 18 August 2011.Figure 3 shows the surface pressure chart at 0900 JST.The Kanto region lay within a warm air mass to the south of the stationary front.There were two low-pressure systems along the stationary front but they were not strong enough to significantly affect the study area.
The vertical profile of the equivalent potential temperature θ e observed at Tateno at 0900 JST (Fig. 4) indicates a potentially unstable layer extending from the surface to the 5500 m level although several thin stable layers were found in this layer.Above the 5500 m level, the deviation of θ e from the equivalent potential temperature of a hypothetically saturated atmosphere θ e ¢ is relatively large, suggesting that the air was dry in this layer.The hodograph (Fig. 4b) indicates that there was a southerly wind of about 2 m s −1 near the surface and southwesterly of approximately 11 m s −1 at the 7717 m level.Therefore, the direction of the vertical wind shear between the surface and the 7717 m level was from the east-southeast to the west-northwest.According to the analysis by the University of Wyoming (http://weather.uwyo.edu/upperair/sounding.html), the convective available potential energy (CAPE) and convection inhibition (CIN) of this sounding were 2028 J kg −1 and − 67 J kg −1 , respectively, indicating a favorable conditions for the development of deep convective storms.The lifting condensation level (LCL), LFC, and bulk Richardson number were 759 m, 2422 m, and 363, respectively.
Figure 5a shows the distribution of the surface wind at 1100 JST.According to the climatological study in this region by Sato and Kimura (2005), a southeasterly wind from the plain to the mountains prevails around noon in August.The wind pattern in this case is a little different from that of Sato and Kimura (2005); i.e., a southwesterly wind prevails to the south of the  study area instead of a southeasterly.However, weak southeasterly wind toward the mountains was found near the study area.The PWV increased from 1000 to 1100 JST in the study area (Fig. 5b), suggesting that horizontal convergence of water vapor occurred during this period.
Figure 6a shows the time variation of horizontal convergence calculated from the surface wind at three AMeDAS points (HTY, CCB, and OGC in Fig. 1b).The graph indicates convergence from 0900 to 1200 JST and divergence after 1240 JST.The PWV at NGR and CCB increased in the morning and gradually decreased in the afternoon.This tendency is a typical diurnal variation of the PWV over mountains (Sato and Kimura 2005).However, the PWV at NGR rapidly increased around 1000 JST and decreased after 1100 JST.This suggests that a small-scale convergence of water vapor occurred around NGR. Unfortunately, we did not observe the vertical distribution of water vapor.Therefore, we cannot evaluate how cumulus convections modified the environmental conditions before the occurrence of intense precipitation.A peak in the precipitation intensity of 136.5 mm h −1 was observed at 1120 JST by the JMA C-band radar.Intense precipitation continued until 1640 JST (Fig. 6c), but the rainfall area was very narrow (figure not shown).

Evolution stages of convective clouds
Many small cumuli appeared and decayed in the After that, convective echoes E4 -E6 formed around preexisting echoes (1054:25 JST).From the tracks of the convection echoes (Fig. 7i), it can be seen that all the echoes except E7 and E8 appeared within 3 km of the location where E1 was first identified.The E7 Figure 8a shows the time variation of the maximum cloud top height analyzed by stereo photogrammetry, and the time-height cross-section of the maximum radar reflectivity derived from all sector PPIs and RHIs of the Ka-band radar.The cumulus ensemble was first identified at 1023 JST from a photograph; however, we could not conduct a stereo photogrammetric analysis until 1032 JST for the reasons outlined in Section 3.2.The maximum cloud top heights were  2013) also classified the evolution stages of convective clouds generated over mountains based on X-band radar observations.Their stage 1 (clouds were visible but not be detected by X-band radars) is subdivided into stages 0 and 1 in the present study, i.e., the radar reflectivity in stage 1 in the present case was very weak to be clearly detected by the X-band radar.Their stage 2 with X-band radar echoes limited to mountainous areas corresponds to stage 2 in the present study.Stage 3 in Nishiwaki et al. (2013), in which radar developed over the plain, did not occur in the present case.In this research, we concentrate on the transition process from stage 1 to stage 2.
Figure 8c shows the temporal variation of the echo top height based on RHIs.For the convective echoes in stage 1 (E1, E2, and E3), the maximum echo top height was lower than 6 km, whereas some of the convective echoes in stage 2 (E6, E7, and E9) developed at an altitude over 6.5 km.The echo top height of E8 was significantly lower than that of the other echoes.Insufficient data were obtained for the echo top height of other convective echoes to discuss their evolution.

Convective clouds in stage 0
The convective clouds developing into echo E1 were photographed at 1023 JST initially only from TRZ.The camera at HDK could not take pictures of these convective clouds until 1032 JST because the view was obstructed by low-level clouds.Therefore, we identified the cloud from its morphological similarity until 1032 JST.Figures 9a and 9b show the photographs of the clouds at 1032:00 JST.The red points in the pictures are the characteristic points commonly found in both pictures.The location and the height of the characteristic points were analyzed by stereo photogrammetry and plotted on the map in Fig. 9c.The clouds formed in a valley region between two mountains, which roughly coincides with the initial location of E1 in Fig. 7.The maximum cloud top height was 5250 m, which is a little lower than the dry layer above the 5500 m level observed by the upper sounding (Fig. 4).As mentioned previously, no detectable radar echo was observed until 1032:13 JST.

Convective clouds in stage 1
In stage 1, four convective echoes (E1 -E4) were developed and decayed (Fig. 8b).Convective echo E1 appeared at 1032:13 JST at a height of about 4 km in the sector PPI at an elevation angle of 10.3°, whereas convective echo E2 formed to the rear of E1 at 1041:40 JST in the levels between 2 and 4 km (figure not shown).Figure 10 shows the photographs of E1 and E2 with the characteristic points (Fig. 10a), the sector PPI of radar reflectivity at an elevation angle of 7.1° with the location of characteristic points and contour of altitudes (Fig. 10b), the RHI of radar reflectivity and the height of the characteristic points (Fig. 10c), and the RHI of Doppler velocity (Fig. 10d).In the photograph, it seems like the cloud on top of E1 was higher than E2 but this is an illusion due to the different distances of the clouds from the camera.Actually, the cloud top heights of E1 and E2 were not very different from each other; both were between 4700 and 5900 m.The characteristic points in the photograph roughly correspond to the periphery of the radar echoes (Fig. 10b), suggesting that the stereo photogrammetry shows reasonable results.According to the RHIs, both E1 and E2 comprise radar echoes weaker than 0 dBZ.The echo top height of E2 is about 4000 m, approximately 1000 m lower than the cloud top height in the same cross-section (Fig. 10c).The Doppler velocity generally indicates negative values (velocity component approaching radar), and the absolute value increases with height, corresponding to the vertical wind shear shown in Fig. 4b.In order to discuss the averaged structure of radar echoes E1 and E2, Fig. 11 shows the maximum radar reflectivity, the maximum cloud top height from the stereo photogrammetry, and the averaged Doppler velocity during 7 min from 1039 to 1046 JST.The peak values of reflectivity were 12.3 dBZ at the 2 km level in E1 and 8.3 dBZ at the 3 km level in E2.These values correspond to 0.28 mm h −1 and 0.18 mm h −1 of precipitation intensity, respectively, if we assume the empirical relationship for reflectivity greater than 0 dBZ proposed by Sauvageot and Omar (1987).Near the echo top, the radar reflectivity decreases with height, and there is no echo region between the cloud top and the echo top.The absolute values of the Doppler velocity increase with height, indicating the existence of vertical wind shear.
Convective echo E3 was first detected in the RHI at 1047:45 JST in the level between 4 and 5 km (figure not shown).Figure 12 shows the vertical cross-section of the maximum radar reflectivity and mean Doppler velocity around E3 for 6 min from 1049 to 1055 JST.Though other convective echoes (decaying E1 and E2) can be seen in the cross-sections, the maximum reflectivity was mainly from E3 in its mature stage.As for the cloud top height, we used the data at 1049 JST and from 1053 to 1055 JST because the data from 1050 to 1052 JST at TRZ were not obtained due to a battery issue.The maximum reflectivity is 3.9 dBZ at the 3.5 km level, and there is no echo region between the echo top and the cloud top for X < 14 km.The absolute values of the Doppler velocity exceed 14 m s −1 around the 4.5 km level, which is much higher than the wind speed at the same level in Tateno (Fig. 4b).An RHI scan for E4 was not conducted.
Common features of convective clouds in stage 1 are as follows: 1) the peak values of radar reflectivity are less than 15 dBZ (corresponding to 0.38 mm h −1 according to Sauvageot and Omar 1987) and they are found below the 3.5 km level and 2) there is an echofree region between the cloud top and the echo top.However, the maximum reflectivities of the three echoes were slightly different: 12.3 dBZ, 8.3 dBZ, and 3.9 dBZ in E1, E2, and E3, respectively.

Transition from stage 1 to stage 2
In stage 2, convective echoes E5 -E10 developed.Among them, E7 developed most strongly and reached 26.9 dBZ in the sector PPI scan (Fig. 8b).Here we focus on the development of E7. Figure 13 shows the vertical cross-sections of the radar reflectivity of E5 and E7 (right panels) with sector PPI at the nearest time (left panels).At 1059:54 JST, E5 and E7 exhibited similar features to convective clouds in stage 1; the peaks in the reflectivity were 2 dBZ at the 2 km level in E5 and 6.0 dBZ at the 3 km level in E7, and the reflectivity decreased with height near the echo top in both convective echoes.The echo top height of E7 was 5.6 km.Unfortunately, we could not obtain the stereo photogrammetry data for this time due to the obstruction of the camera view by other clouds at both sites.At 1101:09 JST, the echo top of E7 increased to 6.1 km, whereas the peak reflectivity remained at 5.9 dBZ.The peak reflectivity and the maximum echo top height rapidly increased after this time; they reached 13.4 dBZ and 6.3 km, respectively, at 1102:23 JST, and 25.7 dBZ and 6.7 km at 1104:55 JST.According to the upper sounding in Tateno, the temperature at the echo top at 1104:55 JST was −8 °C.
The Doppler velocity in the same cross-section suggests an existence of strong inflow from the west into the cloud at 1059:54 JST (the black arrow in Fig. 14).This inflow still existed at 1101:09 and 1102:23 JST but became obscure at 1104:55 JST.It is possible that the inflow may have been a part of the circulation associated with an updraft within the cloud but the estimation of wind vectors is difficult using a single Doppler radar.

Discussion
The development of convective clouds that initiated over the mountains in Kanto was analyzed using observational data obtained from a Ka-band Doppler radar and stereo photogrammetry.The evolution of convective clouds was classified into three stages.In this section, we will discuss the microphysical and dynamical structure of convective clouds in each stage focusing on the transition process from stage 1 to stage 2.
After the first convective clouds appeared, no detectable reflectivity was observed by the Ka-band radar from 1023 to 1032 JST (stage 0).The minimum detectable reflectivity of the Ka-band radar was −16 dBZ at a range of 19 km from where the first clouds appeared.According to the stereo photogrammetry, the cloud top reached up to 5 km (Fig. 8) at the end of stage 0.Under the environmental conditions of the upper sounding at Tateno (Fig. 4), liquid water content at 5 km would be 7.9 g m −3 if a saturated air parcel at 1 km (897.7 hPa, 24.2°C) rose to this altitude along the moist adiabatic curve.When we assume Rayleigh scattering of monodisperse cloud droplets, the relationships among the number concentration (N c ) and mean diameter (D) of cloud droplets, liquid water content (LWC), and radar reflectivity (Z) are written as follows:  where ρ w is the density of water.N c and D at the top of the clouds are estimated as 1100 cm −3 and 24 μm, respectively, under the assumption that LWC was 7.9 g m −3 and the radar reflectivity was −16 dBZ.Murakami et al. (2015) observed convective clouds in summertime in Kochi, Japan, using airborne instruments.They reported that the number concentration of cloud droplets in convective clouds was from 400 to 1500 cm −3 .Therefore, the estimated number concentration here is a reasonable value.Because the convective clouds in stage 0 probably consisted of high concentration and low dispersed cloud droplets (Fig. 15a), they could not be detected by the Ka-band Doppler radar due to weak scattering.
In stage 1, radar reflectivity weaker than 15 dBZ was observed in the middle level of the clouds.Since radar reflectivity is represented by an integral of the sixth power of droplet diameter under the assumption of Rayleigh scattering, radar reflectivity is very sensitive to the droplet diameter.According to the study by Sauvageot and Omar (1987), the reflectivity threshold between drizzling and non-drizzling continental cumuli is −15 dBZ, and a radar reflectivity of 15 dBZ corresponds to a rainfall intensity of 0.38 mm h −1 .Moreover, no descending echo was observed during stage 1.These facts suggest that the convective clouds in stage 1 included many drizzle particles.It is reasonable to consider that the drizzle particles were produced by the collision coalescence of cloud droplets because it takes a long time for drizzle formation through the condensation growth (e.g., Rogers and Yau 1988) (Fig. 15b).The echo-free region near the cloud top is considered to be due to cloud droplets with reflectivity weaker than the minimum detectable value by the Ka-band radar.It is likely that more sensitive millimeter-wave radar could detect the structure near the cloud top.
In the transition process from stage 1 to stage 2, the radar echo rose to a higher level and the reflectivity increased rapidly (Fig. 13).The rise of the radar echo suggests the existence of updrafts transporting drizzle particles to the upper layer.An inflow found in the Doppler velocity (Fig. 14) can be interpreted as a compensating flow for the updraft (Fig. 15c).A possible reason for the updraft development is a moistening of the midlevel air due to detrainment of preexisting clouds, i.e., the dry layer above 5.5 km, which inhibited the development of deep convection in stages 0 and 1 was moistened by successive formation of convective clouds, and a favorable condition for the updraft to develop to the upper layer was created.Similar preconditioning processes were observed or simulated in other mountainous areas (Zehnder et al. 2009;Kirshbaum 2011).
The lifted echo in stage 2 rapidly increased the radar reflectivity.The maximum reflectivity increased from 6.0 to 25.7 dBZ in 5 min from 1059:54 to 1104:55 JST (Fig. 13).This rapid increase of radar reflectivity can be explained by coalescence of drizzle particles and cloud droplets; the cloud condensation nuclei would be activated in the rising air parcel and new cloud droplets formed.Since the size of new cloud droplets would be much smaller than the drizzle particles, they were accreted by drizzle particles due to the difference in their terminal fall velocity.As a result, drizzle particles could rapidly grow to raindrops and the radar reflectivity increased.Because the temperature of the echo top in stage 2 was estimated to be −8 °C, the possibility of raindrop formation through an ice phase process cannot be excluded.However, accretion of cloud droplets would still be an important process even if the drizzle particles were frozen.
It is notable that strong radar echoes corresponding to raindrops formed in the layer warmer than −10 °C.From a statistical analysis using X-band radar by Kobayashi and Inatomi (2003), the first radar echoes (the radar echo observed by their X-band radar (> 13 dBZ) for the first time inside a convective cloud which subsequently attained a reflectivity of more than 28 dBZ) frequently appeared in the layer from 2 to 3 km over the mountains in Kanto in summer.The altitudes of the first radar echoes observed by an X-band radar can be interpreted as the altitude wherein precipitation particles are formed.Their results suggest that precipitating particles form in the lower layer in the mountainous region in Kanto.The drizzle recirculation discussed in the present study could explain the raindrop formation at such low altitudes.This process would be most likely to occur over mountains wherein an ensemble of shallow cumuli evolves into taller clouds.However, further studies are needed to confirm this hypothetical process.

Summary
The development of convective clouds initiated over mountains in the eastern part of Kanto, Japan, in summer was observed by Ka-band Doppler radar and stereo photogrammetry.The results can be summarized as follows: 1) The evolution of convective clouds was classified into three stages: convective clouds visible but not detected by the Ka-band radar (stage 0), convective clouds detectable by the Ka-band radar with reflectivity below 15 dBZ (stage 1), and convective clouds accompanied by descending echoes corresponding to precipitation (stage 2).2) It is considered that the convective clouds in stage 0 consisted of a high concentration of small droplets, and the convective clouds in stage 1 included drizzle particles formed by the collision coalescence of cloud droplets.3) In the transition process from stage 1 to stage 2, a weak radar echo rose to the upper level and quickly increased its reflectivity.This phenomenon was interpreted as drizzle particles produced in preexisting convective clouds being lifted by a newly developed updraft and raindrops being produced rapidly by coalescence of the drizzle and cloud droplets.In the future, we need to verify whether the suggestion of raindrop formation by drizzle recirculation in this paper is correct by observing the droplet size distribution and by numerical simulations with a cloud-resolving model, including detailed microphysics.
the University of Tsukuba for his support in the analysis.Appreciation is extended to Dr. Kazuo Saito of the Meteorological Research Institute and three anonymous reviewers who provided helpful comments.

Fig. 1 .
Fig. 1.(a) Observation area (blue frame) and the location of Tateno Aerological Observatory.Topographic contours are drawn at every 500 m.(b) Arrangement of instruments for observation.Symbols represent Ka-band Doppler radar (+), digital cameras (•), GPS receivers of the Geospatial Information Authority of Japan (), and AMeDAS of the JMA (□).The white solid curve indicates the observation range of the Ka-band radar, and the yellow lines demarcate the camera views.The origin of the X-and Y-axes is at the radar site.

Fig. 2 .
Fig. 2. (a) An object on line l running through the center of the camera lens P is projected at point Q on the image sensor.Here α and ϕ in- dicate the azimuth and elevation angles, respectively, of the object relative line PO.The length of PO corresponds to the focal length of the camera lens.(b) Geometric relationship between two cameras (A and B) and a target (C ).

Fig. 4 .
Fig. 4. Upper sounding at Tateno at 0900 JST on 18 August 2011.(a) Vertical profiles of the potential temperature θ, equivalent potential temperature θ e , and equivalent potential tem- perature θ e ¢ of a hypothetically saturated atmo- sphere with the same temperature profile.(b) Hodograph, wherein the numerals near the plots indicate altitude (m).

Fig. 5 .
Fig. 5. (a) Distribution of the surface wind at 1100 JST and (b) deviation of PWV at 1100 JST from 1000 JST on 18 August 2011.Topographic contours are drawn at every 500 m.The frames with solid lines indicate the observation area shown in Fig. 1.

Fig. 7 .
Fig. 7. (a) -(h) Distribution of radar reflectivity obtained by the sector PPI scans at an elevation angle of 7.1°.The origin of the coordinates is the Ka-band radar site.(i) Tracks of convective echoes.

Fig. 8 .
Fig. 8. (a) Time-height section of the maximum radar reflectivity in the area of X = 5 to 30 km and Y = −10 to 10 km obtained by the sector PPI and RHI scans.The black triangles are the maximum cloud top heights derived from the stereo photogrammetry (the data before 1032 JST and after 1102 JST were not obtained).(b) Time variations of maximum radar reflectivity of convective echoes in the sector PPIs at an elevation angle of 7.1° and (c) echo top height of convective echoes based on RHIs.

Fig. 9 .
Fig. 9. Photographs of clouds from (a) HDK and (b) TRZ at 1032:00 JST on 18 August 2011.Red spots indicate the characteristic points common to both pictures.(c) Locations and heights of the red spots on topographic contours.

Fig. 10
Fig. 10.(a) Photograph of clouds from HDK at 1043:00 JST on 18 August 2011 with characteristic points (red spots).The yellow line indicates azimuth angle of 281°.(b) Sector PPI of radar reflectivity at an elevation angle of 7.1° at 1043:30 JST (color) with the characteristic points (black spots) and contours of altitude (dashed lines; unit is km).The red line indicates an azimuth angle of 281°.(c) RHI of reflectivity and (d) Doppler velocity at 1043:05 JST at an azimuth angle of 281°.Black spots are the cloud top within ± 5° from the azimuth angle.Because the noise threshold differs between reflectivity and Doppler velocity, the numbers of data points differ.

Fig. 11 .
Fig. 11.(a) Maximum reflectivity (color) and cloud top (black dots) and (b) averaged Doppler velocity in the period from 1039 to 1046 JST.Eleven RHIs in azimuth angles from 280° to 283° were used.The plots of cloud top are the maximum values in 0.5 km horizontal intervals.The topography is represented by black shading.Radar echoes near the ground were not obtained due to radar beam blockage by the terrain.

Fig. 12 .
Fig. 12. Same as Fig. 11, except that nine RHIs in azimuth angles from 287° to 290° were used, and the periods from 1049 to 1055 JST for the radar data and 1049 JST and from 1053 to 1055 JST for the cloud top.

Fig. 14 .
Fig. 14.Same as right panels in Fig. 13 but for Doppler velocity.The black arrow indicates the inflow discussed in the text.