Numerical prediction experiment over the United Arab Emirates by using

JMA-NHM
Akihiro Hashimoto1, Narihiro Orikasa1, Takuya Tajiri1 and Masataka Murakami2,1
1
Meteorological Research Institute, Japan Meteorological Agency, Tsukuba, Japan
2
Institute for Space-Earth Environmental Research, Nagoya University, Nagoya, Japan

1. Introduction 50˚ 400 grid cells 60˚

In 2015, the United Arab Emirates (UAE) Research Program 40794
40812
40783
OIFS
40798
40800
OIFM

OIYY
40821 5km-NHM
(UAEREP) for Rain Enhancement Science was launched in order
40827
OIAW
40811 OIZB
40829
KQXJ
40818

30˚ OIAG
40833
OIAM
40836
OISY

30˚
OIAA
40831
40835 30N

to promote scientific advancement and development of new

40550 OIKK
40841
40552
KQWM 40568
KQGV 40588 OIZH
40856
40570 40581
KWP40
40580 OISS
40848
OKBK 40851
40585 40853
OIBB
40858 40854 41710
40857

technology. Three projects among many proposed projects were

40596 40859
40373
OEPA 40592
40594

800 grid cells

OEKK
40377 KQAY

OIBJ

awarded the UAEREP prize. The authors are involved in one of

OISL 40878
OIZI
40879
OIKB
40875

OIKQ OIZJ
40417
OEDF 40882 40883
OIBK OIBL

400 grid cells

OEDR
40416
OBBI
41150

these three projects, “Advanced Study on Precipitation

800 grid cells

OIBS
40889OIBA
40890
OMRK
41184 40893 41759

2000 km
OEAH
40420 40898
OERK
40437 41170
OTBD OMSJ
41196
OMDB
41194 41756
40435 OERY
40438 KQIR 41198
OMFJ

800 km
Enhancement in Arid and Semi-Arid Regions,” which has four OEKJ OMAD
OMAA
41217
KQGX OMAL
41218
41244
Ha
41246

jar

subprojects; (1) data analysis for identifying areas suitable for

41258
OOMS
41256
M 4125341257
ou 41254
nta 41263
rt
41255
in
ese
41265 41267
s 41264 41268

cloud seeding, (2) laboratory experiments for characterization of lK

hal
iD
'A 1km-NHM
seeding material, (3) ground-based observation for evaluating the Rub
OEWD
41061

20˚ 41288

20N 20˚
800 km
occurrence frequency of seedable clouds and an airborne seeding
experiment for investigating seeding effects in clouds, and (4) OESH
41136
41314
41304

numerical modeling for realistic simulation of cloud seeding and

41315
OOSA
41316
41312

0 1000 2000 3000 4000

the seasonal evaluation of seeding effect. As a collaborative work 50E 60E

50˚ 2000 km
among subprojects, daily numerical weather prediction over the 60˚

UAE began in late 2016 to support safe and effective Fig. 1. Computational domain for the numerical prediction.
experiments of airborne cloud seeding and to evaluate model
performance. This article presents a description of the prediction
system and its preliminary results.
2. Numerical prediction system
The numerical prediction system was established based on the
Japan Meteorological Agency’s Non-Hydrostatic Model
(JMA-NHM, Saito et al., 2006) with several modifications, Weather charts at the surface, 850-, 700-, and 500-hPa planes
Distributions of IWP, LWP, Soil Moist. G2CB, Rain, TS, CB2CT,

mainly land surface configuration, as described in Hashimoto et

RH, and Tsoil
Time series of Precip., DSW, DLW, CLL, CLM, CLH,
P, Ts, and RH at AWS sites
Profiles of IWC and LWC at AWS sites

al. (2017).
The numerical prediction is conducted twice per day. In each
prediction cycle, the simulation with a 5-km horizontal resolution Distributions of IWP, LWP, Soil Moist. G2CB, Rain, TS, CB2CT, RH, and Tsoil

(5km-NHM) is performed first, followed by that with a 1-km

horizontal resolution (1km-NHM). Figure 1 shows the
computational domain. The 5km-NHM widely, but not
completely, covers the Middle East including arid and semi-arid
regions in the UAE. The 1km-NHM is embedded in the domain
of 5km-NHM. The standard latitude and longitude are at 20.00
ºN and 55.00 ºE, respectively, in the Lambert conformal conic
projection. The center of the domain is located at (24.00 ºN,
54.00 ºE) for the 5km-NHM, and at (24.25 ºN, 54.15 ºE) for the
1km-NHM. In the vertical direction, the two simulations are
configured identically as follows. The top height of the model
domain is 22 km. The vertical grid spacing stretches from 40 m
at the surface to 886 m at the top of the domain. Fifty vertical
layers are employed in a terrain-following coordinate system.
The initial time of the 5km-NHM forecast is 0400 or 1600
Gulf Standard Time (GST=UTC+4), which corresponds to the
forecast time FT = 6 h of the JMA’s Global Spectral Model
(GSM) forecast with initial times of 2200 or 1000 GST,
respectively. The initial time of the 1km-NHM forecast is shifted
12 h later from that of the 5km-NHM forecast. Boundary
conditions are provided by the JMA’s GSM forecast every 6 h.
The 1km-NHM is driven by the hourly meteorological field of
Corresponding author: Akihiro Hashimoto, Meteorological
Research Institute, 1-1 Nagamine, Tsukuba, 305-0052, Japan.
E-mail: ahashimo@mri-jma.go.jp
Fig. 2. Schedule for the weather prediction with the initial time
of 0400 GST. Fine black arrows indicate the data flow.
the 5km-NHM forecast in the one-way nesting manner. Figure 2
shows the schedule and data flow in the numerical prediction
system with the initial time of 0400 GST, for instance. The
integration time and time step are 48 h and 12 s for the
5km-NHM, and 27 h and 8 sec for the 1km-NHM, respectively.
Hourly output is obtained from each simulation.
3. Computational cost
Computations are conducted on the FUJITSU Supercomputer
PRIMEHPC FX100 at the Meteorological Research Institute
(MRI) of the JMA. Each job runs with 128 Multi Processor
Interface (MPI, 8 MPI x 16 nodes) for the 5km-NHM simulation
twice per day, which finishes in about one and a half hours. This
costs 17,520 node-hours for year-round simulations (16 nodes x
1.5 h x 2 runs x 365 days). Maximum memory usage is about 10
GB/node. For the 1km-NHM simulation, each job runs with 288
MPI (8 MPI x 36 nodes) twice per day, which also finishes in
about one and a half hours. This costs 39,420 node-hours for
year-round simulations (36 nodes x 1.5 h x 2 runs x 365 days).
Maximum memory usage is about 25 GB/node.
A single run of 5km-NHM requires 20.3 GB of free disk space
 0

(hPa)
0 820

(hPa)
1 820

(mm)
1 840

(mm)

PRESS.
2 840

PRESS.
2 860

PRECIP.
3 860

PRECIP.
3 880
4 880

LOCAL
4 900

LOCAL
5 900
5 920
6 920
6 82 83 84 82 83 84
16 82 83 84 1.0 82 83 84
International Airport observed by multi-wavelength microwave

COVERAGE
16 1.0
(a)

COVERAGE
0.8
OBS 5km-NHM 1km-NHM
10m 12
12 radiometer
0.8
0.6
(MP-3000, Radiometrics) and predicted by the
1km-NHM, respectively. Air temperature below a height of 2-km
10m

8 0.6
WS

8 0.4
WS

began0.4 to decrease several hours earlier (Fig. 4a) than the surface

CLOUD
4

CLOUD
0.2
4
0
air 0.2temperature drop (Fig. 3b). The predicted temperature
0.0
0 82 83 84 decrease
0.0 82 at the lower layer 83
preceding the surface84air temperature
82 83 84 1000 82 83 84
40
(b) drop
1000(Fig. 4b) was much weaker than the observed decrease (Fig.

m)−2)
36
40
32
36 4a).500
500 K-band Doppler radar (MRR-2, METEK), which was

m−2
28
32

DLW(W
(oC)

24
28
deployed at the same site, detected precipitation during the

DLW(W
(oC)

20
24 0
T2m

16
20 0
preceding decrease in low-level air temperature (Fig. 4c), while,
T2m

12
16

DSW,
8
12 −500
in −500
the prediction, the precipitation does not appear until just

DSW,
4
8 OBS 5km-NHM 1km-NHM
0
4
0
82 06 18 83 06 18 GST
84 before
−1000 82 the temperature drop
−1000
83 (Fig. 4d). This indicates
84 that the
100
82
23 83
24 84
cooling 82
of low-level air 83
due to evaporation 84 of precipitation
100
80
March 2017 particles was not represented very well in the prediction. To
Fig. 3 Observed and predicted results for (a) surface wind speed improve the prediction accuracy, it is necessary to consider the
(%)

80
(%)

60
predictability of synoptic scale disturbances as well as
RHRH

and (b) surface air temperature at Al Ain International Airport.

60
40
Black dot indicates the observation. Red and blue lines indicate
40
20
thermo-dynamical formulations of the prediction system.
the 5km- and
82
1km-NHM predictions,
20
83
respectively. 84 5. Summary
82 83 84
Julian Day of 2017 (GST)
Julian Day of 2017 (GST)
A numerical prediction system was established based on the
JMA-NHM to support airborne cloud seeding experiments and to
for input/output operation. We are archiving two-dimensional
evaluate model performance. The procedure of the numerical
data, vertical profiles above observation sites, and plotted graphs
prediction was described including the computational resources
for each simulation. The size of the archived data is 1.5 GB/run.
required for performing numerical simulations and for archiving
This needs the storage space of 949 GB/year. For the 1km-NHM,
the simulation results. Preliminary results of a comparison
84.4 GB/run and 1,678 GB/year are required for running a job
between ground-based observation and numerical prediction
and archiving the data, respectively.
were presented with respect to the event of surface air
4. Result temperature drop on 24 March 2017 at Al Ain International
Figure 3 shows the observed and predicted results for surface Airport. More validation and improvement of the model will be
wind speed and surface air temperature on 23 and 24 March 2017 the subject of future study.
at Al Ain International Airport, which will be the base of our Acknowledgement
airborne cloud seeding experiments planned in the summer of
This study was conducted as a part of the project “Advanced
2017. The 5km- and 1km-NHM predicted well the variation of
Study on Precipitation Enhancement in Arid and Semi-Arid
surface wind speed and air temperature on 23 March. In the
Regions” that is supported by the UAEREP for Rain
morning on 24 March, when a synoptic scale disturbance
Enhancement Science, an initiative of the Ministry of
accompanied with precipitation passed over the observation site,
Presidential Affairs, under the management of the National
a temperature drop of 6 ºC was observed. This temperature drop
Center of Meteorology and Seismology.
was predicted within a delay of a few hours by the 1km-NHM
simulation with an initial time of 1200 UTC on 23 March as References
compared to the actual event. For the 5km-NHM, we have three Hashimoto, A., M. Murakami and S. Haginoya, 2017: First
forecast results available for this temperature drop event with application of JMA-NHM to meteorological simulation
different initial times; 0000, 1200 and 2400 UTC on 23 March. over the United Arab Emirates. SOLA, submitted.
10 10
The best result was obtained from the forecast with an initial time Saito, K., T. Fujita, Y. Yamada, J. Ishida, Y. 4030Kumagai,
40
30
K. −30 −30
8 8
of 0000 UTC on 23 March, although the forecast shows a larger Aranami, S. Ohmori, R. Nagasawa, S. Kumagai,
20 20 C. Muroi, −20
Height (km)

−20
Height (km)

6 6 10 10 −10
delay time and a smaller magnitude of temperature drop than the T. Kato, H. Eito, Y. Yamazaki, 2006: The operational JMA
−10
0 0
0
4 4 0

observation and the 1km-NHM forecast. Figures 4a and 4b show nonhydrostatic mesoscale model. Mon. Wea. −20 −20Rev., 134,
−10 −10
10 10
2 2
time-height cross-sections of air temperature at Al Ain 0
1266–1298. −30 −30
−40
20 20

0 −40
82 82 83 83 84 84
10 10 100 100
10 40 10 10 40 40
8 (a) −30
Temperature (oC), MP-3000
30
8
8 8
8 (b) Temperature (oC), 1km-NHM
−30 −30 3010 10
30
Height (km)
(km)

20 1 −20 1 −20 20 20
Height (km)

Height (km)

Height(km)

−20
6 6 1 1
6 −10
10 6 6 −10 −10 10 10
0 0 0.10
Height

0 4 4 0 0
0.1
4 −10 4 4 −10 −10
10
−20 2 2 10 10 0.010.01
−20 −20
2 2 2
20 −30 20 20 −30 −30
0 0 0.001
0.001
0 −40 0 820 83 83 84 84 −40 −40
82
10 82 83 84 45 82 10 82
10 83 83 84 84 1001 1
10 100 10 10 100 100
8
8
(c)0.1 Radar reflectivity factor (dBZ), MRR-2
40
35 10 8
8 8
8
(d) Mixing ratio of liquid water (g kg-1), 1km-NHM
10 10
(dBZ)(km)
(km)

100.1
-1
(km)

0.1
Height(km)

Height (km)

Height(km)

6 30 6 6 1 1
1
6 25 1 6 6 1 1
0.
Height

0.
1

0.
1

0.
Height

Height

4 4 00
4 0
4 200.1 4 4 0.01 0.01
1 01
10
0.1-2
0.01 0.1
0.01
15 2 2
2
2 0.01
10 2 2 0.001 0.001 0.01 0.01
0 10 5 0 0 10-30.001
0.001
0
82 06 18 83 06 18 GST
84
0.001 0 820
82 06 18 83 83 06 18 84 GST
84
0.001 0.001
10 82 83 84 102 82 10 82
10 83 83 84 84 100 100
10 23 24 1 10 10 23 24 1 90 901
0.001

8 8 80 80
8
8 March 2017 101 8 8 March 2017
(mm h(km)

80
70 70
(km)

(km)

80
Height(km)

−1)
Height (km)

Height(km)

6 0.1 6 6 60 0.1 60 0.1

60
Fig. 4 Time-height cross sections of temperature (a) observed by multi-wavelength microwave radiometer and (b) predicted by the
60
6 6 6
100

0 50 50
100
Height

10 40 40
Height

0.

0.
Height

4 0.01 4 460 60 40 80 40 40
1

60
1km-NHM, (c) reflectivity factor observed by micro-rain radar and (d) mixing ratio of liquid water predicted by the 1km-NHM.
60 40 80 0. 0.
4 4 4 00
1
001 30 30
0.01
10 −1 0.01 0.01 0.01 0.01
2 2 2 20 20
2 2 2 80 80 10 10
0.1

0.001 0.001
0.001 0.001 0.001
0 0.001 10−2 0 0 0 0
0 0.001 0 820 83 83 84 84 0.001 0.001
82 83 84 82
10 82 83 84 100 82 82 83 83 84 84
10 100 10 10 Day Day 100 100
8 90 90 90
8 80 8 8 80 80
(km)

40
Height(km)

Height (km)

Height (km)

70 80 80
70 70
(g m−3)

6 60
6 60 60−1 6
10 6 60 60 60 60
80
100

100

50 50 50
Height

80 40 40
4 100
4 40 4 604 60 60 60 40 40 80 80 40 40
80
2
60 30 30 30
2 20 2 2 20 20
80 80
0
40 10
10 −2 10 10
0 0 0 0 0 0
82 83 84
10 82 83 84 10 82 82 83 83 84 84
Day Day Day
8 8
Height (km)

(m s−1)

6 6

4 4