27SOLA, 2015, Vol. 11, 27−30, doi:10.2151/sola.2015-007
Water vapor variations associated with a meso-γ scale convec-
tion were investigated using GNSS (Global Navigation Satellite
System) derived PWV (precipitable water vapor) and high resolu-
tion numerical model data with a 250 m horizontal grid interval.
A rapid increase of GNSS-derived PWV that occurred prior to the
initiation of surface rainfall was well simulated by the numerical
In the model, PWV values began to increase 16 min before
the rainfall occurred at the surface. A local maximum of PWV
was formed because of the generation of shallow free convection
and surface water vapor ux convergence due to a lifting of an air
parcel at approximately 1 km elevation by a preceding surface
wind convergence. Due to the existence of a stable inversion layer
between 2.2 and 3.5 km elevation, the shallow free convection
took 11 min to rise above the inversion layer to form a deep con-
vection. These results suggest that observation of local distribu-
tions of GNSS-derived PWV is useful for monitoring the genera-
tion of deep moist convection.
(Citation: Oigawa, M., E. Realini, and T. Tsuda, 2015: Study
of water vapor variations associated with meso-γ scale convec-
tion: Comparison between GNSS and non-hydrostatic model data.
SOLA, 11, 27−30, doi:10.2151/sola.2015-007.)
Water vapor is a source of energy for moist convection and
has an inuence on its developments. Therefore, understanding
relationships between variations of water vapor and moist convec-
tive activities is important for the prediction of heavy rainfall. For
example, it is reported that initiations of localized heavy rainfall
are affected by inhomogeneous distributions of surface water
vapor in the meso-γ scale by reducing CIN (convective inhibition)
(Bodine et al. 2010).
In Japan, studies of moisture variations associated with moist
convection have been conducted by analyzing PWV (precipitable
water vapor) derived from a nationwide GNSS (Global Navigation
Satellite System) observation network known as GEONET (GNSS
Earth Observation Network System), with mean inter-station dis-
tances of approximately 20 km. GEONET studies have reported
that water vapor variations precede initiations of moist convec-
tions (Seko et al. 2007; Inoue and Inoue 2007). However, the
horizontal resolution of GEONET is not high enough to capture
the ne structure of a water vapor eld associated with a convec-
tion cell since the horizontal scale of isolated convection is on the
order of a couple of km.
A few observational studies have been conducted to inves-
tigate relationships between water vapor variations and meso-γ
scale convections by using regional GNSS networks with reso-
lutions lower than 10 km. Seko et al. (2004) employed 75 GPS
receivers within an area of 20 × 20 km2 around Tsukuba, Japan
and analyzed the thickening of the humid boundary layer before
the generation of a thunderstorm by using a tomography method
and GNSS derived slant path delay (SPD) data (Shoji et al. 2004).
Realini et al. (2012) and Sato et al. (2013) installed 17 GNSS
stations with a horizontal spacing of 1−2 km near Uji campus of
Kyoto University, which is located south of Kyoto prefecture,
Japan. They succeeded to improve the horizontal resolution of
the retrieved PWV maps by analyzing PWV from high elevation
SPDs. However, it is difcult to understand physical relationships
between water vapor variations and the occurrence of meso-γ
scale convections only from the information of observed PWV
and surface rainfall. Therefore, the objective of this study is to
clarify the cause for meso-γ scale water vapor uctuations prior
to moist convections by conducting a numerical downscaling
experiment of a mesoscale non-hydrostatic model and comparing
observed PWV data with output model data.
This manuscript is organized as follows: Section 2 describes
analytical results based on observational data, such as weather
radar, GNSS-derived PWV, and rain-gauge. Section 3 describes
the explanation about the analyzed model and its analysis result.
Relationships between simulated surface rainfall and simulated
PWV and the associated three-dimensional movement of water
vapor were the focus of this analysis. Discussion and summary are
presented in Section 4.
2. Observational results
2.1 Synoptic conditions and the rainfall system
This study focuses on a heavy rainfall event on August 13−14,
2012, around Uji, Kyoto. Ishihara et al. (2013) reported that MCSs
(meso-scale convective systems) produced 400 mm of accumu-
lated rainfall in 13 h due to unstable atmospheric conditions, i.e.
the CAPE (convective available potential energy) was 2340 J kg−1
at 21 LST on 13 August, 2012 at Shionomisaki, which is located
170 km south of Uji. Moist convections could occur anywhere
around Kyoto due to the humid air supply near the surface in
proximity to the stationary front, south of which the MCSs were
generated. GEONET-derived PWV values at 21 LST on 13
August, 2012 at Shionomisaki and Uji were more than 60 mm and
this fact suggests that CAPE value was also high in Uji. Figure
1 shows movement of a precipitation cell, which was part of a
back-building type MCS, south of Kyoto. A new precipitation cell,
which passed over the Uji campus within 30 min, was observed
south-west of the Uji campus at 2000 LST on 13 August, 2012.
2.2 GNSS-based observations of PWV
Temporal variations of GNSS-derived PWV and surface rain-
fall amounts derived from a rain gauge associated with the pas-
sage of the precipitation cell were observed at the Uji campus (Fig.
2). PWV data used here was derived from the GNSS receiver at
the Uji campus, which is one of the receivers of the hyper-dense
GNSS network used by the authors in earlier studies (Realini et al.
2012; Sato et al. 2013). Precipitation intensity at the GNSS station
rapidly increased from 2010 LST on 13 August, 2012, reached
a maximum of more than 60 mm h−1 within a few minutes, and
diminished by 2020 LST. The GNSS-derived PWV started to in-
crease approximately 10 min before the intensication of surface
precipitation. As a result, PWV value increased by approximately
4 mm in 10 min.
Study of Water Vapor Variations Associated with Meso-γ Scale Convection:
Comparison between GNSS and Non-Hydrostatic Model Data
Masanori Oigawa1, Eugenio Realini1, 2, and Toshitaka Tsuda1
1Research Institute for Sustainable Humanosphere (RISH), Kyoto University, Gokasho, Uji, Kyoto, Japan
2Geomatics Research & Development (GReD) srl, Via Cavour 2, Lomazzo, Como, Italy
Corresponding author: Masanori Oigawa, Research Institute for Sustain-
able Humanosphere (RISH), Kyoto University, Gokasho, Uji, Kyoto 611-
0011, Japan. E-mail: firstname.lastname@example.org. ©2015, the
Meteorological Society of Japan.
28 Oigawa et al., Study of Water Vapor Variations Associated with Meso-γ Scale Convection
increased from 40 m to 886 m as their height increased. The prog-
nostic variables were wind components, temperature, pressure and
all water-related quantities such as vapor, rain, clouds, snow, grau-
pel, and ice clouds (all three-dimensional).
Although experiments had been conducted by setting initial
times of 2 km-NHM and 250 m-NHM as 1800 LST and 1900 LST
on 13 August, 2012 respectively, the observed precipitation
cell was not reproduced. In contrast, if we set initial times as
0000 LST and 0100 LST on 14 August, 2012 for 2 km-NHM and
250 m-NHM, respectively, a back-building type MCS was sim-
ulated by 2 km-NHM and a precipitation cell which was part of
the MCS and initiated around Uji was simulated by 250 m-NHM.
Figure 3 shows the precipitation intensity maps simulated by
the 250 m-NHM. Surface precipitation around Uji started at
0118 LST. Subsequently, the cell continued to grow and reached
an intensity of more than 10 mm h−1 at 0128 LST.
Although the initiation place of the simulated cell was shift-
ed of about 5 km southward with respect to the observation, the
difference of the initiation place is small and its moving direction
agreed well with the observation result. In addition, the fact that
the observed and simulated precipitation cells were both part of
the same type of MCSs indicates that these cells were generated
under ambient conditions very similar to those of the observation
result. Figure 4 shows the temporal variations of PWV and the
precipitation intensity as reproduced by the 250 m-NHM simu-
lation. These values were calculated at a location approximately
3. Simulation results
3.1 Design of the simulation and validation of the model data
Analyzed numerical model data were derived from the Japan
Meteorological Agency Non-hydrostatic Model (JMA-NHM)
(Saito et al. 2007). First, numerical simulation using a horizontal
grid interval of 2 km (2 km-NHM) were performed to express
unstable background condition and the MCS south of Kyoto. The
2 km-NHM has 451 × 451 horizontal grid points and a horizontal
domain centered at 35°N, 135°E on a Lambert conformal projec-
tion. The initial and boundary conditions were set on the basis of
JMA mesoscale analysis data (NPD/JMA 2013). It is commonly
known that phenomena expressible by numerical models have
horizontal scales at least more than 5 times that of the models’
grid intervals. Since we focus on water vapor variations within
few kilo meters, we conducted downscaling to a horizontal grid
interval of 250 m (250 m-NHM). The 250 m-NHM has 161 × 161
horizontal grid points and a horizontal domain centered at
34.88°N, 135.77°E on a Lambert conformal projection. Three-ice
bulk microphysics was used in both experiments (Lin et al. 1983;
Murakami et al. 1990), with no cumulus parameterization. A
hybrid terrain following coordinate is adopted as the vertical coor-
dinate and the vertical resolution was 50 levels from the surface to
the model top (21801 m) in both models. The lowest atmospheric
level was 20 m above the surface, and the depth of the layers
Fig. 1. Maps of precipitation intensity observed by weather radar in 10
min intervals from 1950 LST to 2020 LST on August 13. The location of
the Uji campus is indicated by the “▲” symbol.
Fig. 2. Temporal variations of GNSS-PWV (red) and precipitation inten-
sity (green) observed at the Uji campus from 1950 LST to 2020 LST on
Fig. 3. Time variation of precipitation intensity (mm h−1) simulated by the
250 m grid interval simulation. The location of the Uji campus is indicated
by the black “▲” symbol. Simulated PWV variation at the point indicated
by the red “×” symbol was compared with observations.
Fig. 4. Time variations of PWV (red) and precipitation intensity (green)
as reproduced by the 250 m-NHM simulation at the point indicated by the
red “×” symbol in Fig. 3.
29SOLA, 2015, Vol. 11, 27−30, doi:10.2151/sola.2015-007
5.5 km south-southeastward of the Uji campus, indicated in Fig.
3. A rapid PWV increase prior to the intensication of surface
precipitation was successfully simulated and this result suggests
the validity of the numerical modeling, particularly regarding the
relationship between water vapor variation and rainfall intensity.
For the above reasons, the data of 250 m-NHM is worth using for
analysis at least to investigate the physics of the observed precipi-
3.2 Simulated meso-γ scale PWV uctuations
Figure 5 shows maps of PWV and precipitation intensity as-
sociated with the precipitation cell and its evolution over time in
6 min intervals from the 250 m-NHM. The indicated horizontal
domain has an area of 12.8 km × 8.9 km. The PWV values in
the eastern side have relatively lower values, because mountains
cover this region. Surface precipitation started at 0118 LST. We
focused on the local PWV maximum indicated by a “▲” in the
gure at 0118 LST. This maximum formed before the start of sur-
face precipitation and can be recognized by its shape at 0106 LST,
12 min before the start of rainfall. The horizontal velocity of the
local PWV maximum was estimated by assuming constant move-
ment from 0106 LST to 0130 LST and the locations of the local
PWV maxima for each time are indicated by “▲” symbols in Fig.
5. Vertical proles of water vapor and wind at these points were
analyzed to investigate the mechanism for the PWV increase asso-
ciated with the cumulonimbus cloud.
Figure 6a shows temporal variations of the PWV anomaly and
VIL (vertically integrated liquid water) at the local PWV maxi-
mum. The PWV started to increase at 0102 LST, 16 min before
the start of the surface rainfall. Figure 6b shows the vertical distri-
bution of the vertical wind velocities and its temporal evolution.
Before 0118 LST, a downward wind region can be seen at an
elevation of approximately 2 km, with upward wind motion oc-
curring below this region. After 0118 LST, upward wind velocities
became strong and deep convection was generated. The down-
ward wind region prior to 0118 LST was likely an inversion layer.
Before 0118 LST, upward wind velocities were more than 4 m s−1
and we can deduce that a shallow free convection had already gen-
erated under the downward wind region. Figure 6b shows a strong
upward wind region at a height of 1 km that can be traced back to
0107 LST. In addition, the rate of the PWV increase became large
at 0107 LST. Therefore, the initiation of the free convection oc-
curred at approximately 0107 LST. Before 0107 LST, horizontal
surface wind convergence can be seen in Fig. 6c and this conver-
gence caused the lifting of an air parcel around 1 km elevation.
The maximum horizontal convergence formed at an elevation of
1.7 km at 0118 LST, following the breakthrough of the inversion
layer (Fig. 6c). The humid boundary layer thickened with time due
to shallow convection and the following horizontal convergence
of moisture near the surface from 0107 LST to 0118 LST (Fig.
6d), resulting in the increase of PWV.
Figure 7 (left) shows the vertical distributions of equivalent
potential temperature and the saturated equivalent potential tem-
perature at 0105 LST. An inversion layer can be seen between 2.2
and 3.5 km elevation and a strong convective instability is evident
between 1.00 and 1.28 km elevation. Air at elevations between
0.73 and 1.08 km was saturated with water vapor at 0105 LST and
it was conrmed that an air parcel in this layer was located at the
LFC (level of free convection) (Fig. 7 (left)). The height of the
convectively unstable layer increased with time after the shallow
convection was initiated at 0107 LST and the inversion layer can
no longer be seen at 0120 LST after its breakthrough at 0118 LST.
4. Summary and discussion
Variations of water vapor associated with a meso-γ scale
convection were investigated using a non-hydrostatic mesoscale
Fig. 5. Maps of PWV (shade) and precipitation intensity (contour: 1 mm
h−1) at time steps of 6 min reproduced by the 250 m grid interval sim-
ulation from 0100 LST to 0130 LST on August 14. Vertical proles of
physical values were analyzed by tracing the convection cell at the points
depicted by “▲” symbol. The “▲” symbol at 0124 LST corresponds to
the red “×” symbol in Fig. 3.
Fig. 6. (a) Time variations of PWV increase from the initial time of the
250 m grid interval simulation (line) and the Vertically Integrated Liquid
water (VIL) (boxes) within the convection. (b) Vertical distributions of
vertical wind velocity and its time evolution within the convection. (c)
Vertical distributions of horizontal wind divergence and its time evolution
within the convection. (d) Vertical distributions of water vapor mixing
ratio and its time evolution within the convection. Location of proles of
each times are indicated by the “▲” symbol in Fig. 5.
30 Oigawa et al., Study of Water Vapor Variations Associated with Meso-γ Scale Convection
model and observational data. Temporal variations of PWV and
surface precipitation intensity associated with the passage of a
convection cell that was part of a back-building type MCS, which
brought heavy rainfall to Uji on 13−14 August, 2012, were ob-
served using a GNSS receiver and rain gauge. Based on observa-
tions at the Uji campus, PWV values started to increase rapidly
approximately10 min before the intensication of surface precipi-
A downscaling simulation was performed to simulate water
vapor variations within a cumulonimbus cloud that passed over
the Uji campus using the JMA-NHM by setting a horizontal grid
interval to 250 m. Although there is a small gap in the formation
time between the simulated and observed rain bands, back-build-
ing type MCSs were simulated by a 2 km-NHM. Furthermore,
the validity of the downscaling results was conrmed because
the 250 m grid interval simulation successfully simulated the ob-
served rapid PWV increase prior to the surface rainfall.
In the model, PWV associated with moist convection began
to increase at 0102 LST, 16 min before the start of surface rainfall
(at 0118 LST) and local maximums of PWV were formed prior to
the rainfall. Analysis of output data revealed that free convection
was initiated at 0107 LST, 11 min before the surface rainfall, due
to the lifting of an air parcel at approximately 1 km elevation.
The air parcel became buoyant because the air column around
1 km elevation was highly convectively unstable, allowing the air
parcel to reach the LFC, and the horizontal wind had previously
converged near the surface before the initiation of the free con-
vection. It was also revealed that it took 11 min for the free con-
vection to rise above a stable inversion layer that existed between
2.2 and 3.5 km elevation to generate deep convection. Therefore,
the PWV increase from 0102 to 0107 LST was caused by surface
wind convergence and the PWV increase from 0107 to 0118 LST
was caused by shallow free convection under the inversion layer.
Then, rainfall at the surface started at 0118 LST.
During this heavy rainfall event, a large area was affected by
high PWV values of more than 60 mm. Under such atmospheric
conditions, the distribution of surface wind convergence may be
more important than local convergence of moisture to determine
where the convection is initiated. However, even under such
conditions, observation of the local distribution of PWV is useful
for monitoring convective activity because local convergence of
moisture is detectable from PWV maps and air parcels require
time to form deep convections with the presence of an inversion
The numerical model used in this study was provided by
the JMA and the initial and boundary data were provided by the
“Meteorological Research Consortium.” This work is partly sup-
ported by a Grant-in-Aid for the Japan Society for the Promotion
of Science (JSPS) Fellows. The authors appreciate the cooperation
offered by Dr. Kazutoshi Sato in setting up GNSS and rain gauge
observation instruments. The authors would like to express their
gratitude to Dr. Takuya Kawabata, Dr. Hiromu Seko, and Dr.
Yoshinori Shoji of Meteorological Research Institute for providing
useful comments and suggestions. We extend our thanks to Dr.
Atsuki Shinbori for his support to use IDL software and the other
members of the Tsuda Laboratory of RISH, Kyoto University, for
providing useful comments.
Bodine, D., P. L. Heinselman, B. L. Cheong, R. D. Palmer, and
D. Michaud, 2010: A case study on the impact of moisture
variability on convective initiation using radar refractivity
retrievals. J. Appl. Meteorol. Clim., 49, 1766−1788.
Inoue, H. Y., and T. Inoue, 2007: Characteristics of the water-vapor
eld over the Kanto district associated with summer thunder-
storm activities. SOLA, 3, 101−104.
Ishihara, M., and K. Takara, 2013: Organization of mesoscale
convective systems causing heavy rainfall and ash ood
around Uji, Kyoto on 14 August 2012. Disaster Prev. Res.
Inst. Annuals, 56B, 79−86.
Japan Meteological Agency, 2013: Outline of the operational
numerical weather prediction at the Japan Meteorological
Agency. WMO Technical Progress Report on the Global
Data Processing and Forecasting System (GDPFS) and
Numerical Weather Prediction (NWP) Research, 28−36.
Lin, Y.-L., R. D. Farley, and H. D. Orville, 1983: Bulk parametar-
ization of the snow elds in a cloud model. J. Climate Appl.
Meteor., 22, 1065−1092.
Murakami, M., 1990: Numerical modeling of dynamical and
micro physical evolution of an isolated convective cloud. –
the 19 July 1981 CCOPE cloud. J. Meteor. Soc. Japan, 68,
Realini, E., T. Tsuda, K. Sato, M. Oigawa, and Y. Iwaki, 2012:
Analysis of the temporal and spatial variability of the wet
troposphere at a local scale by high-rate PPP using a dense
GNSS network. ION GNSS 2012, At Nashville, Tennessee,
Saito, K., J. Ishida, K. Aranami, T. Hara, T. Segawa, M. Narita,
and Y. Honda, 2007: Nonhydrostatic atmospheric models
and operational development at JMA. J. Meteor. Soc. Japan,
Sato, K., E. Realini, T. Tsuda, M. Oigawa, Y. Iwaki, Y. Shoji, and
H. Seko, 2013: A high-resolution, precipitable water vapor
monitoring system using a dense network of GNSS receiv-
ers. J. Disaster Research, 8, 37−47.
Seko, H., N. Nakamura, Y. Shoji, and T. Iwabuchi, 2004: The
meso-γ scale water vapor distribution associated with a thun-
derstorm calculated from a dense network of GPS receivers.
J. Meter. Soc. Japan, 82, 569−586.
Seko, H., Y. Shoji, and F. Fujibe, 2007: Evolution and air ow
structure of a Kanto thunderstorm on 21 July, 1999 (the
Nerima Heavy Rainfall Event). J. Meter. Soc. Japan, 85,
Shoji, Y., H. Nakamura, T. Iwabuchi, K. Aonashi, H. Seko, K.
Mishima, A. Itagaki, R. Ichikawa, and R. Ohtani, 2004:
Tsukuba GPS dense net campaign observation: Improvement
in GPS analysis ofslant path delay by stacking one-way post-
t phase residuals. J. Meteor. Soc. Japan, 82, 301−314.
Manuscript received 8 December 2014, accepted 17 February 2015
Fig. 7. (Left) Vertical distributions of equivalent potential temperature
(black) and saturated equivalent potential temperature (blue broken line)
at 0105 LST. (Right) Vertical proles of equivalent potential temperature
within the convection cell at 0107 LST (black), 0114 LST (blue broken
line) and 0120 LST (red). Locations of each prole are indicated by the
“▲” symbol in Fig. 5.