Indonesian Journal of Geography Vol. 5151 No.

No.2,3,August 20192019
December (199-206)
(273 - 284)
DOI: http://dx.doi.org/10.22146/ijg.44914
http://dx.doi.org/ 10.22146/ijg.44924
RESEARCH ARTICLE

The Effect of
Improving Baseline Component
Numerical Correlation
Weather Prediction on theEvents
of Rainfall DesignUsing
of GNSS
Radar Data
Network Configuration
Assimilation for Sermo Reservoir Deformation Monitoring
Yulaikhah
Miranti Indri
1,3
, Subagyo Pramumijoyo
Hastuti1*), Jaka Anugrah2
, Nurrohmat Widjajanti
Ivanda Paski 2
), and Fatkhuroyan
3
Fatkhuroyan2)
1
Ph.D.
1) Student, Doctoral
Meteorological Agency of Study ProgramMedan,
Kualanamu of Geomatics
BMKG,Engineering,
Indonesia Department of Geodetic Engineering,
Faculty
2) of Engineering,
Research and DevelopmentUniversitas
Center,Gadjah
BMKG,Mada, Indonesia
Indonesia
2
Department of Geological Engineering, Faculty of Engineering, Universitas Gadjah Mada, Indonesia
3
Department of Geodetic Engineering, Faculty of Engineering, Universitas Gadjah Mada, Yogyakarta, Indonesia

Received: 2019-05-18 Abstract Data assimilation is a method to improve initial atmospheric conditions in numerical
Accepted: 2019-09-30
Received: 2019-05-18 Abstractprediction.
weather This of
The condition study
the aims to investigate
geological structurethe
in effect of assimilation
the surrounding of Doppler
Sermo reservoirweather
shows
Accepted: 2019-07-29 that there
radar dataisina fault crossing
Weather the reservoir.
Research Deformation
Forecasting (WRF) monitoring
numerical modelof that for
faultthe
hasprediction
been carried of
out byrain
heavy conducting
events inGNSS campaigns at
the Jabodetabek 15 with
area monitoring stations simultaneously.
dates representing four seasons However,
respectivelythoseon
Keywords: campaigns
WRFDA; 20 Februarywere
2017,not well designed.
3 April With2017,
2017, 13 June such and
a design, it took 2017.
9 November many The instruments and
reflectivity (Z)spent
and
DA – radar;
Keywords:
much velocity
radial money. For (V)the next
data GNSS
from Plancampaign,
Position itIndicator
should be designed
(PPI) so that
product and the optimal network
reflectivity (Z) data
weather
baselineradar;
correlation, configuration
from Constantis Altitude
obtainedPPIand (CAPPI)
the cost can be reduced.
product were In the designusing
assimilated of deformation
WRFDA (WRF monitoring
Data
heavy rains
sensitivity, network, sensitivity
Assimilation) criteria
numerical modelbecome
with very
3DVar important
(The Threefor Dimensional
detecting theVariational)
deformations. In GNSS
system. The
GNSS network,
deformation relativeofpositioning,
output radar datathe baseline components
assimilation and withoutareassimilation
correlated, but thisnumerical
of the correlationmodel
is often
ofignored.
WRF is
Correspondent Email: This research
verified examined
by spatial the effect
with GSMaP of and
data baseline component
by point correlations
with precipitation on the design
observation results
data. of the
In general,
[email protected]
GNSS configuration of the Sermo Fault network based on sensitivity criterion.
WRF radar assimilation provides a better simulation of spatial and point rain events compared In this case, the
western
to the WRFsidemodel
of the without
fault wasassimilation.
taken as a reference,
Moreover,while the other side
improvements in as anprediction
rain object moving would rela-
be
Corespondent Email: tivelyvisible
more againstinthe western
areas close side. Thissources,
to radar study found that the baseline
not echo-blocked fromcomponent correlation
fixed objects, and duringaffects
the
[email protected] the results of GNSS network configuration. Considering the correlation could result a sensitive
rainy season.
network configuration with a fewer baseline; therefore, the cost and time of field surveys can be
reduced. It can be said that the baseline component correlation needs to be taken into account in
the configuration design of deformation monitoring network.
© 2019 by the authors. This article is an open access article distributed under the terms and conditions of the Creative
Commons Attribution(CC BY NC) licensehttps://creativecommons.org/licenses/by-nc/4.0

© 2019 by the authors. This article is an open access article distributed under the terms and conditions of the Creative
Commons Attribution(CC BY NC) licensehttps://creativecommons.org/licenses/by-nc/4.0
1. Introduction In the last few years, That NWP has often used are
According to Nugroho (2002), most of the mesoscale models of Weather Research and Forecasting
1. Introduction (WRF).
flooding
Sermo thatreservoir
occurredisinlocated
Jabodetabek
in thewas causedpart
western by
the lastThe three years, deformation
advantages of WRF are monitoring
that they arehas been
efficient
heavy rain. The prediction of accurate
of Yogyakarta, Indonesia. It was built by damming heavy rain is
carried out because
and flexible by conducting
they canGNSS
be usedcampaigns. However,
on supercomputers
important in order to build a flood early warning
Ngrancah river and officially operated in 1997. It can system.
those campaigns
to laptops to studywere not well
various designed.
dynamic Observations
interactions on a
However, operational
hold 25 million cubic weather
metersprediction
of water capabilities
and servesofa were
scale ofcarried
metersout simultaneously
to thousands at 15 (Skamarock
of kilometers monitoring
Indonesian Agency for Meteorology, Climatology,
vital role as a water reservoir from which water is then and
stations distributed
et al., 2008). Besides,around
WRFthe fault. With
is included assuch a design,
open-source
Geophysics (BMKG) are only good at forecasts
distributed by the Water Utilities (PDAM) serving the of rain
itsotook
that many instrumentsofand
the configuration thespent
model much money.
can be For
adjusted
or not for
needs rainclean
dichotomies, while predictions
water, irrigation, and floodofprevention.
heavy and
the
andnext GNSSwith
adjusted campaign,
researchit needs.
shouldIn bethe designed so that
WRF model,
heftyThe
rainfall
condition of the geological structure inet al.,
have low predictive abilities (Gustari the the
thereoptimal
are threenetwork
ways toconfiguration is obtained
improve the accuracy and the
of weather
2012).
Sermo reservoir and surrounding have an interesting cost can be reduced.
predictions, namely improving initiation data (initial
Predicting
phenomenon. rain is not
Overlaying a simple
geological map matter.
and LandsatThe
In general,
conditions network optimization
and boundaries), numerical designtechniques,can
accurate predictions of rain require in-depth
imagery show that there are reverse and thrust faults knowledge
be
andclassified into several
parameterization. Dueorders, namely
to limited costs zero, first,
and time,
of the cross
which weather system of(Figure
the reservoir a region which
1). This includesis
condition
characteristics, physical processes, and dynamic the improvement
second, and third of initiation data is the main
orders (Halicioglu focus.
& Ozener,
confirmed by (Widagdo, Pramumijoyo, Harijoko, & One way
2008; to fix
Kuang, this Mehrabi
1996; is to assimilate
& Voosoghi, the data. Data
2014). A
weather
Setiawan, processes.
2016) in their Comprehensive
research aboutunderstanding
the geological
of assimilation
geodetic is a method
network needs oftoimproving
be designed initiation
to meet datathe
as
structure of rock distributionnecessary
the weather system is in the areatoofget accurate
Kulonprogo. an inputofmodel by calculating
predictions. Furthermore, understanding a complete criteria accuracy, reliability,observational
and low cost.data into a
However,
They found that the secondary structure which controls
weather system can inbeKulonprogo
done objectively agrid model system
deformation (Skamarock
monitoring et al., 2008).
network must meet one
the rock distribution mountainbyis using
in the
modeling that requires formulations that Southwest-
describing more WRF Data that
criterion, Assimilation (WRFDA)
is, sensitivity to theis occurring
a specific
form of Northwest-Southeast normal fault, WRF program for assimilation data. WRFDA has
the weatherreverse
Northeast system fault,
as a function of time.
and North-Northwest lateral deformation (Benzao & Shaorong, 1995; Even-Tzur,
several assimilation techniques,
2002). Several study has been done to design the including The
fault. The similar description is also found systems
One of the weather modeling in the main is
Three-Dimensional
Numerical Weather Prediction (NWP)
report of Sermo Reservoir Project Details Design which answers optimum geodetic Variational (3DVar), monitoring
and deformation The Four-
the challengesPekerjaan
related toUmum,
scientific methods needed to Dimensional Variational (4DVar),
network, wherein accuracy and reliability have and Ensemble
been
(Departemen 1985). Kalman Filter (EnKF). From various studies, 4DVar
predict the weather through physical
The fault, henceforth referred to as the calculations
Sermo fault, in the most used criteria. Mehrabi and Voosoghi (2014)
simulating sophisticated atmospheres (Sagita, 2017). and EnKF show good potential in assimilating data,
potentially affects the Sermo Dam deformation. In used the precision criteria with analytical methods
but this technique requires high computational costs.
IMPROVING NUMERICAL WEATHER PREDICTION Miranti Indri Hastuti, et.al

Meanwhile, 3DVar is considered to have the best also did a new experiment with the latest engineering
performance of all types of methods in analyzing research on the assimilation of radar 3DVar data in
hydrometeorological elements with excellent Jakarta using Z data of Constant Altitude Plan Position
computational efficiency (Liu et al., 2013). Indicator (CAPPI) products as the results provide
Assimilated data in WRFDA are surface air better spatial rain distribution simulations
observation data, upper-air observations, satellites, and The use of CAPPI Z data in radar assimilation is
radar. One of the most frequently conducted researches new in Indonesia. Therefore, it is interesting to research
is the assimilation of radar data. The advantage of the assimilating 3DVar technique radar data using Z
radar data compared to other data is that it has a high CAPPI data. The difference with previous research is
resolution (the resolution is higher than the resolution that there is a comparison test between the performance
of the mesoscale model) with full coverage. Based of the assimilation model of Z CAPPI, Z PPI, V PPI, and
on this, the assimilation of radar data is expected to the combination of Z and V PPI in terms of improving
improve the ability to predict rain significantly. initiation data on the WRF model for heavy rain events
Xiao et al. (2005) assimilated doppler radial in Jabodetabek. Improvements to the best models are
velocity from Korean Jindo radar, Li et al. (2010) expected to increase the accuracy of predictions of
assimilated doppler radial velocity in Space and Time heavy rainfall and temporal variations.
Analysis System (STMAS) project at ESRL (Earth
System Research Laboratory). The studies stated that 2. The Methods
the radial velocity (V) PPI (Position Plan Indicator)
In this study, some primary data are used.
data assimilation could improve quantitative rainfall
Global Forecast System (GFS) used as an initial data
prediction capabilities or short-term Quantitative
model, and C-Band Doppler Radar (CDR) data
precipitation forecasting (QPF). Furthermore, research
used as observation data. The used data for model
on the assimilation of 3DVar radar data uses two data,
verification in this research are data Global Satellite
namely product radial velocity data PPI (V PPI) and
Mapping of Precipitation (GSMaP) for spatial analysis
reflectivity (Z) of PPI products (Z PPI) (Xiao and Sun,
and observational data from the meteorological
2007; Sugimoto et al., 2009). Based on these studies, the
station owned BMKG around Jakarta region includes
assimilation of data from the combination of Z and V
Cengkareng, Ponbet, Kemayoran, Tanjung Periuk, and
PPI had a positive impact on the ability of QPF rather
Citeko for point analysis.
than using one of the two flats. In addition, assimilation
WRF model for research use three domains
of combination Z and V PPI data is more stable and
configuration. The third domain has a resolution of
does not cause significant errors.
three kilometers and covering the Jabodetabek area
In Indonesia, Satrya (2012) conducted a study on
(in Figure 1). Model parameterization configuration
the assimilation of data on the combination of Z and V
uses the best configuration conducting from Gustari
PPI 3DVar methods in the Bandung region. Based on
et al. (2014) in WRF 3.9.1. Weather assimilation model
the conclusion, the biggest sampling technique on Z PPI
obtained from the package of WRFDA with 3DVar
data as WRFDA input is the best in simulating heavy
technique using radar data as an observation. A brief
rain. In addition, Paski et al. (2017) also conducted
description of the weather assimilation simulations
the same study in Lampung region. Paski et al. (2019)
provided in Table 1.

Figure 1. Domain configuration for WRF models

274
 Indonesian Journal of Geography, Vol. 51 No. 3, December 2019 :273 - 284

Table 1. Brief description of the scenario assimilations carried out.

Code Experiment Name Z CAPPI Z PPI V PPI
S1 CTL no no no
S2 ZCAPPI yes no no
S3 ZPPI no yes no
S4 VPPI no no yes
S5 ZVPPI no yes yes

Radar volumetric (.vol) volumetric data is In CSV format data, separate data based on
converted into two types of formats, the polar elevation or sweep. In each sweep, the data is also
coordinates netCDF and CSV using EDGE applications. separated based on reflectivity (Z), radial (V), spectral
Then, both types of radar data formats will each go width (W) and filter unfilled (U) reflectivity. This
through different processes. The polar coordinates research uses Z PPI and V PPI data as assimilation data.
netCDF radar data will be processed with the wradlib- CSV radar data formats will be processed in numerical
python application to convert the format into netCDF data analysis to combine Z PPI and V PPI data in one
cartesian coordinates, as well as process the data into a .txt format file. Then, the combined data Z PPI and
netCDF format CAPPI product. The CAPPI calculation V PPI (.txt) will be processed with the biggest value
specifications that have been designed following sampling technique on a 3 km x 3 km grid (according
research Permana et al. (2016). Furthermore, CAPPI to the domain of previous *.ctl) using numerical data
netCDF format data is processed from the maximum analysis and computation applications (Paski et al.,
sampling value in one grid model of 3 km x 3 km 2017). The output data of the R application will result in
(according to the domain of previous *.ctl) using the R the output of radar data in ASCII format or according
Studio application. The output data of the R application to the WRFDA application input format. See Figure 2
will result in the output of radar data in ASCII format for details.
or according to the WRFDA application input format.

Figure 2. Flowchart for making radar data input on WRF models

275
IMPROVING NUMERICAL WEATHER PREDICTION Miranti Indri Hastuti, et.al

The analysis carried out in this study begins warm-rain schemes using Z data (Wang et al., 2012).
by knowing the impact of radar data assimilation on Temperature parameters play a role in the convection
changes in weather parameters on the input data. process where the higher the temperature, the greater
Furthermore, spatial and temporal verification of the the convection process that is likely to the occurrence
model prediction results is done using comparative of heavy rain (Jo Han et al., 2010).
data from GSMap and surface observation results. Figure 3 shows the difference of WRF initial
The verification was carried out to determine the data for surface temperature parameters between 5
performance of the WRF model without assimilation simulations on February 20th, 2017. In the southeast-to-
and with assimilation using prediction skills. POD, west study area, the temperature of ZCAPPI simulation
FAR by using a contingency table on the dichotomy of is warmer than CTL assimilation, ranging from 22-240
rainfall events or not and looking for the percentage of C to 26-27° C, moreover at Kemayoran, Tanjung Periuk,
hits, underestimates, and overestimations based on the Ponbet, and Cengkareng, area of ​​warmer temperatures
category of light rain, moderate rain, heavy rain, and are widespread. In Z PPI simulation also changes the
hefty rain. temperature to more temperate in the southeast to the
west of the study area. However, temperature of study
3. Result and Discussion areas such as Kemayoran, Tanjung Periuk, Ponbet,
and Cengkareng is getting colder. Meanwhile, there
The Effect of Radar Data Assimilation on WRF Initial
is no difference between ZVPPI simulation and ZPPI
Parameter Data
simulation. Besides, VPPI assimilation does not change
Based on table 2, which discussed the effects of at all the temperature parameters.
each assimilation scenario, showed that the significant
impact of radar data assimilation on the initial WRF
Humidity Parameter
data occurred in the ZCAPPI and ZVPPI simulations
by modifying four initial data parameters (temperature, The humidity parameter is the concentration of
humidity, water vapor mixing ratio, and wind). water vapor in the atmosphere. In this study, we examine
Meanwhile, the ZPPI simulation only modified three the 850 mb humidity because if the value is high in
parameters (temperature, humidity, and mixing ratio WRF initial data, then it indicates the moisture content
of water vapor) while the least is the VPPI simulation is large enough for convective cloud growth process that
which only modified one parameter (wind). has the potential to cause moderate to heavy rain (Liu et
In essence, Z CAPPI and Z PPI data have the same al., 2018). Figure 4 shows the initial data difference on
type of data, namely Z data, only in Z CAPPI data is Z the 850 mb humidity parameters between 5 simulations
data at certain fixed altitudes from data every scanning on February 20, 2017. Compared to CTL simulation,
elevation, while Z PPI data is Z data from one scanning the humidity of ZCAPPI and ZPPI simulations don’t
elevation. In this study, ZCAPPI simulations turned change significantly. However, generally, the air gets
out to have a higher value than ZPPI simulations (Fig. damper so it can be ascertained that rainfall will be
3). So, it was able to change wind parameters through high as well. Location of value changes occurring in
a negative temperature change scheme, while ZPPI areas near radar such as Kemayoran, Tanjung Periuk,
simulations were unable to change wind parameters Ponbet, and Cengkareng by 84% to 86%. Similarly,
(Wang et al., 2016). areas far from radar such as Citeko, the humidity value
is getting higher, for example 96% to 98%. Meanwhile,
humidity increments in ZVPPI simulation are similar
Temperature Parameter
to ZPPI simulation, while VPPI simulation shows no
In the 3DVAR WRF assimilation, initial change at all or equal to CTL simulation.
temperature data increments are calculated from

Table 2. Effect of radar data assimilation on WRF initial parameter data

Weather Parameter in WRF
WRF Assimilation
Temperature Humidity Mixing Ratio Wind
ZCAPPI √ √ √ √
ZPPI √ √ √ -
VPPI - - √ -
ZVPPI √ √ √ √

276
 Indonesian Journal of Geography, Vol. 51 No. 3, December 2019 :273 - 284

Figure 3. Initial data of surface air temperature Figure 4. Initial data of humidity parameter
parameter

Water Vapor Mixing Ratio Parameters (Ma et al, 2018). In WRF 3DVAR assimilation, the
The parameter of the water vapor mixing ratio is initial wind data increment comes from ZCAPPI and
the ratio of moisture mass to the dry air mass present radial velocity data. ZCAPPI assimilation through
in one particular volume (Wang et al, 2013). The negative temperature change (Wang et al, 2016), while
mixing ratio also represents the moisture content of the assimilation of radial velocity data is directly from wind
moisture present in the air so that with the higher the parameter element u, v, w (Routray et al., 2010)
mixing ratio, the greater the chance of rain (Jones et al, Figure 6 shows the initial data difference on
2014). 850 mb layer wind parameters between 5 scenarios
Figure 5 shows the initial data difference for water of assimilations on February 20, 2017. Spatially,
vapor mixing ratio parameter between 5 simulations ZCAPPI, VPPI, and ZVPPI simulations significantly
on February 20, 2017. Compared to CTL simulation, alter the wind parameters, shows that CALM wind
there is a significant change in water vapor mixing ratio speed is widespread in most Southwest Jabodetabek.
of ZCAPPI, ZPPI, and ZVPPI simulations in western This indicates that convective cloud growth is not
region study. In areas close to radar sites such as hindered by the wind. Wind speed of Kemayoran,
Kemayoran, Tanjung Periuk, Ponbet, and Cengkareng, Tanjung Periuk, Ponbet, and Cengkareng area (near
there was an increase in mixing ratio from 0.0132 to radar location) also weakened, namely Cengkareng
0.0136 in ZPPI and ZVPPI simulations, whereas 0,014 in and Ponbet from 10 knots to 8 knots and Kemayoran
ZCAPPI experiment. This increase indicates the higher and Tanjung Periuk areas from 14 knots to 10 knots.
moisture content of the water or wetter conditions so Meanwhile, Citeko which far away from radar location
that there is more potential for heavy rains. In contrast, did not change significantly, for example, wind speed
areas far from radar such as Citeko there was a decrease CALM on CTL and assimilated simulations. Meanwhile,
in the value of mixing ratios, from 0.0152 to 0.014 in wind direction parameters did not change significantly
ZCAPPI, ZPPI, and ZVPPI simulations. Furthermore, in each assimilation experiments. Furthermore, ZPPI
VPPI simulation did not change at all in mixing ratio. simulation does not occur at all wind changes in both
direction and speed.
Wind Parameter
The wind parameter acts as a mass vapor
supply from the outside or as a humidity transport

277
IMPROVING NUMERICAL WEATHER PREDICTION Miranti Indri Hastuti, et.al

Figure 5. Initial data of water vapor mixing ratio Figure 6. Initial data of 850 mb layer wind
parameter parameter

Lapse Rate lapse rate at atmospheric altitudes of 1000 - 10000 km.

The Lapse rate is defined as a decrease in The highest lapse rate occurs in ZCAPPI simulation.
temperature against the height of the atmosphere or Then, the second-highest lapse rate occurred in
a negative vertical temperature gradient (Tyasjono, ZVPPI simulation, then followed by ZPPI, VPPI,
2007). The value of the lapse rate is related to the and CTL simulations. Meanwhile, VPPI simulation
cooling of the ambient temperature quickly. This has not changed. This is due to the VPPI simulation
significant and positive lapse rate initial data effects cannot improve temperature. Besides, the lapse rate
on the WRF calculation and prediction process at atmospheric altitude 1000 - 10000 km in all WRF
(Janiskova, 2015). In general, from table 3, it can be models is positive, so there is no inversion in the
seen that CTL and assimilated simulations on February atmosphere layer.
20, 2017, has a significant difference in temperature

Table 3. Lapse Rate

Layer (km) Without Z CAPPI Z PPI V PPI Z & V PPI
1000-2000 4.307 6.674 7.304 4.307 8.304
2000-3000 5.84 5.499 4.286 5.84 4.286
3000-4000 6.448 5.219 4.882 6.448 4.882
4000-5000 6.512 5.699 5.42 6.512 5.42
5000-6000 5.996 4.903 6.192 5.996 6.192
6000-7000 5.083 6.091 5.362 5.083 5.362
7000-8000 5.368 5.96 6.071 5.368 6.071
8000-9000 6.901 8.038 6.763 6.901 6.763
9000-10000 8.512 8.013 8.765 8.512 8.765
10000-11000 8.794 8.998 9.274 8.794 8.274
Lapse Rate 6.1074 6.2328 6.1161 6.1074 6.227222

278
 Indonesian Journal of Geography, Vol. 51 No. 3, December 2019 :273 - 284

Figure 7. Spatial rain distribution on February 20, 2017, between (a) GSMaP, (b) Without Assimilation, (c) Z
CAPPI Assimilation, (d) Z PPI Assimilation, (e) V PPI Assimilation, (f) Z & V PPI Combination Assimilation

Analysis and Verification of Spatial Rain rain area is seen in the southeastern part of Jabodetabek,
Distribution Bogor and surrounding areas. Meanwhile, the light
Based on Figure 7, it shows the rainfall simulation rain area in VPPI simulation extends to the south
results and rainfall accumulation between GSMaP of Jabodetabek, for example in most areas of Bogor.
observation data and the simulation results using CTL, However, the area of ​​heavy rain on VPPI simulation
ZCAPPI, ZPPI, VPPI, and ZVPPI on 20 February 2017. is almost absent in Jabodetabek area. In general, the
In GSMAP data, shows rainfall of Jabodetabek area simulated GSMAP rain-simulation results similar to
dominated by moderate rain (20 - 50 mm). However, assimilated simulations differ only in rain intensity
the southern part of Bogor area happened light rain with the most superior models are ZCAPPI and ZVPPI
(<20 mm). Moreover, the western and eastern parts simulations because it is more susceptible to heavy
of Jabodetabek are heavy rains (50 to 100 mm), such rainfall than other assimilated simulations.
as southwestern Tangerang, northern Bogor, eastern
Depok, eastern Bekasi, southwestern Bekasi and most Verification of time series predictions for rainfall
of Java Sea. accumulation for three hours
Compared to GSMaP observation data, the Figure. 8a shows the time series of accumulated
moderate rainfall area of CTL simulation is narrow. As rainfall for three hours in CTL or assimilated simulations
a result, the area of ​​light rain increases to northwest and at the Soekarno Hatta Cengkareng Meteorological
northeast of Jabodetabek, in Tangerang, West Jakarta, station on February 20th, 2017 tends not to be similar to
South Jakarta, western Depok, and the southeast and observational data, namely rain starting at 12.00 UTC,
northeast of Bekasi. Likewise, the area of ​​heavy rain also whereas according to observation data occurs around
changed places and slightly narrowed in the southern 18.00 UTC. In addition, the rainfall accumulation
region of Greater Jakarta, which is in Bogor. in the observation data is underestimated, especially
Compared to the CTL experiment, spatial rainfall CTL simulation, and then there is an improvement in
of assimilated simulations does not change significantly, accumulation of rain which is closest to the observation
but generally, the broad-range rainfall area in assimilated data, those are Z CAPPI and ZVPPI simulations.
simulations is wider than CTL, as well as with higher Figure. 8b shows the time series of accumulated
rainfall. It’s just that the area of ​​moderate-heavy rain rainfall for three hours in CTL or assimilated simulations
remains narrower than GSMaP data. The highest at Ponbet climatological station tends not to be similar
rainfall is in ZCAPPI and ZVPPI simulations occur in to observation data. Based on observation data it
the southeastern part of Jabodetabek, the northeastern appears no more rain after at 09.00 UTC, but on CTL or
Bogor and southeastern Bekasi. In addition, the light assimilated simulations, the rain returned where there
rain area’s ZCAPPI and ZVPPI simulations in northwest was an increasingly rising line pattern. In addition,
Jabodetabek also appear to be narrowing. the rain accumulation of the WRF model is entirely
In ZPPI simulation, the light rain area decreases overestimate compared to observation data, especially
in the northwest part of Jabodetabek while the heavy CTL simulation, and then there is an improvement

279
IMPROVING NUMERICAL WEATHER PREDICTION Miranti Indri Hastuti, et.al

in rainfall accumulation which is the closest to the Rain Dichotomy Verification (yes / no)
observation data, that is ZPPI simulation. To compare the numerical experiments, three
Figure. 8c shows the time series of accumulated statistical indicators PC, POD, and FAR (Jakubiak,
rainfall for three hours in CTL or assimilated 2014) are evaluated using rain dichotomy (yes / no)
simulations at Tanjung Periuk tends to be similar to (Wiegand, 2015; Fatkhuroyan et al, 2019). Radar
observational data with the same start of rain starting observation distance will have a significant influence on
at 00.00 UTC, then there was heavy rain at 18.00 UTC. the accuracy of the radar parameters used to assimilate
However, the accumulation of rain that occurred in the data. The evaluation methodology based on the
the WRF simulations was entirely underestimating, point-to-point comparison between model-generated
especially CTL simulation and then there was an variables and observations. This is a two-dimensional
improvement in rainfall accumulation which is closest matrix where each element counts the number of
to the observation data with the lowest underestimation occurrences in which the gauge measurements and the
occurring in ZCAPPI and ZVPPI simulations. model forecasts exceeded or failed to reach a certain
Figure. 8d shows a time series of accumulated threshold for a given forecast period. The table elements
rainfall for three hours in WRF without assimilation or are defined as: A-model forecast and gauge measurement
WRF assimilation of radar data at Citeko meteorological exceeded the threshold; B-model forecast exceeded the
station tends not to be similar to observation data. It threshold but measurement not; C-model forecast did
can be seen from the graph that rainfall distribution not reach the threshold but measurement exceeded it;
at observational data occurs continuously at 00.00 and D-model forecast and measurement did not reach
- 06.00 UTC, then returns to rain starting at 18.00 the threshold, see Table 4. Based on the above elements,
UTC, whereas in WRF without assimilation and radar the Proportion Correct is defined as PC = (A+D)/N,
assimilation rain occurs at 03.00 UTC continuously with N holding the total number of observations being
with different intensities. Moreover, the accumulation verified (N=A+B+C+D). The Probability of Detection
of daily rainfall occurs in each WRF is overestimate, is defined as POD = A/(A+C), the False Alarm Ratio is
with the highest being is in WRF V PPI assimilation. defined as FAR = B/(A+B).

a) b)

c) d)

Figure 8. Time series of accumulated rainfall for three hours based on

4 rainfall observed data’s meteorological stations

280
 Indonesian Journal of Geography, Vol. 51 No. 3, December 2019 :273 - 284

Table 4. 2x2 Contingency Tabel

Observation: Observation :
Yes No
Forecast: Yes A B
Forecast: No C D

Figure 9. Histogram of Skill forecast (a) PC, (b) POD, (c) FAR

Based on figure 9a, it can be seen that all WRF Based on figure 9b, it can be seen that assimilated
model scenarios are assimilating radar data improve PC radar simulations improve POD skills. The highest POD
skills. The highest PC value occurred at the Cengkareng value occurred at Kemayoran observation point with
area observation point of 0.95, which occurred ZCAPPI an absolute value of 1.00, which occurred in ZCAPPI,
and ZVPPI simulations. Furthermore, at Ponbet ZPPI, and ZVPPI simulations. Furthermore, at Ponbet
observation point, the highest PC value which occurred observation point, POD the highest value of 0.95
in ZCAPPI simulation is 0.91. At the Kemayoran occurs in ZCAPPI and ZVPPI simulations. Likewise,
observation, the highest PC value of 0.83 occurred at Tanjung Periuk observation, the point highest POD
in ZVPPI simulation. At Tanjung Periuk observation value of 0.95 also occurs in ZCAPPI simulation. At
point, the highest PC value of 0.75 occurred in ZCAPPI Cengkareng observation point, the highest POD value
simulation. Finally, at Citeko observation point, the of 0.94 occurs in ZVPPI simulation. And finally, at
highest PC value of 0.75 occurred in ZCAPPI, ZPPI, Citeko observation point, the highest PC value of 0.83
and ZVCAPPI, but this value is lowest compare to occurs in ZVPPI simulation, which is this value is the
PC value at other observation points. Meanwhile, lowest than the highest POD value at other observation
the lowest PC value always occurs in CTL simulation points. Meanwhile, the lowest PC value occurred in
at all observation points followed by VPPI or ZPPI CTL simulation at all observation points followed by
simulation. ZPPI or VPPI simulation.

281
IMPROVING NUMERICAL WEATHER PREDICTION Miranti Indri Hastuti, et.al

Based on figure 9c, it is clearly seen that the heavy rain category with the most superior model
assimilated radar simulations improve FAR skills. The is ZCAPPI simulation, while the category of very heavy
lowest FAR value occurred at the Ponbet observation rain - the average underestimate percentage reached
point, which arises in ZCAPPI simulation by 0.25 and 99.48% with the most superior model is ZCAPPI
ZV PPI simulation by 0.28. Furthermore, at the Tanjung simulation.
Periuk observation point also had a low FAR value of
0.28, which occurred in ZCAPPI simulation and as Discussion
much as 0.31 in ZVPPI simulation. At the Kemayoran
In this study, the most significant modification
observation point, the lowest FAR value occurred
of the parameters of temperature, humidity, and water
in ZCAPPI and ZVPPI simulations by 0.3. At the
vapor mixing ratio occur in ZCAPPI simulation, then
Cengkareng observation point, the lowest FAR value
followed by ZVPPI simulation, and ZPPI simulation.
of 0.32 occurred in ZCAPPI simulation. Finally, at the
The most significant modification of wind parameter
Citeko region observation point, the lowest FAR value
values occurs in VPPI simulation, then followed
of 0.44 occurred in ZCAPPI simulation, where this
by ZVPPI and ZCAPPI simulations. This result is
value is the highest than the lowest FAR value at other
consistent with the research of Sun and Wang (2013),
observation points. Meanwhile, the highest FAR value
which states that assimilation of V data has a significant
occurred in WRF without assimilation at all observation
impact on wind analysis, while analysis of temperature,
points, followed by ZPPI or VPPI simulations.
humidity, and water vapor mixing ratio are secondary.
Conversely, assimilation of Z data has a direct effect on
Prediction Verification based on Rain Category the analysis of temperature, humidity, and water vapor
Moreover, rain category verification is investigated mixing ratio while wind analysis is secondary.
to see improvement of simulations (Ricciardelli et Based on the summary ranking of skill prediction
al., 2018). Generally, from table 5, the percentage (see table 6), it can be clearly stated that ZCAPPI and
of assimilated simulations performance is better ZVPPI simulations are the best rain prediction both
than CTL simulation. This can be seen from the best spatially and periodically. It was clearly seen from the 1st
sequence always on assimilated simulations. Simulation best sequence only in the two assimilated simulations.
performance is best in the light rain category, because This is consistent with the effects on the initial WRF
the average percentage of hits is better than the other data, where those both assimilated simulations can
rain categories, with the most superior model is ZCAPPI improve WRF performance. This is consistent with
simulation. For the moderate rain category, the rate of the research of Sun and Wang (2013), which states
hits also looks dominant with the best model is ZVPPI that when physical parameters (air temperature, air
simulation. The performance of assimilated model is humidity, water vapor mixing ratio) and dynamic
quite good for the moderate rain category, while in the parameters (wind) are calculated in the 3DVar WRF
category of heavy rain and very heavy, underestimate assimilation system, errors can be minimized in order
conditions are still very dominant. Furthermore, the to predict more accurate rainfall.
average underestimates percentage reached 76.4% in

Table 5. Ranking and Percentage Average of Skill Prediction Based on Rain Category
The Best Order Percentage Average
Category*
1 2 3 4 5 Underestimate Hits Overestimate
Light S1 S4 S3 S2 S0 0% 89,6 % 10,4 %
Moderate S4 S1 S2 S3 S0 2,8 % 86,6% 8,6 %
Heavy S1 S4 S2 S3 S0 75,8 % 23,4 % 0,8 %
Very Heavy S1 S4 S3 S2 S0 99,8 % 0,2 % 0%
* The rain category based on Perka BMKG NO: KEP. 009 (2010)

Table 6. Summary Ranking of Skill Prediction by Spatial and Point

Point
WRF Models Spatial
Time Series Rain Dichotomy Rain Category
CTL 5 5 5 5
ZCAPPI 1 2 1 1
ZPPI 3 3 4 4
VPPI 4 4 3 3
ZVPPI 2 1 2 2

282
 Indonesian Journal of Geography, Vol. 51 No. 3, December 2019 :273 - 284

4. Conclusion Sugimoto, S., N.A Crook., Sun J., & Xiao Q. (2009). An
Examination of WRF 3DVar Radar Data Assimilation
Generally, the WRF model of radar data on Its Capabilityin Retrieving Unobserved Variables and
assimilation provides a better simulation of spatial Forecasting Precipitation through Observing System
and point rainfall events than the WRF model without Simulation Experiments, American Meteorological
assimilation. The improvement of rainfall predictions Society, 137, 4011-4029.
will be more visible in areas close to the radar source, Satrya, L. I. (2012). Asimilasi Data Radar dalam Penerapan
that is not blocked by echo from fixed objects such as Prediksi Cuaca Numerik di Indonesia (Studi Kasus di
mountains or hills, and during the rainy season. From Jawa Barat), Skripsi, Meteorologi, ITB, Bandung.
the four assimilation model scenarios carried out in Paski, J.A.I. & Gustari, I. (2017), Pengaruh Asimilasi Data
this study, it can be concluded that the best sequence Radar C-Band Dalam Prediksi Cuaca Numerik (Studi
of WRF scenarios is ZCAPPI, ZVPPI, VPPI, and ZPPI Kasu s di Lampung, Jurnal Meteorologi dan Geofisika,
18 (2), 55-64.
simulations, respectively.
Paski, J.A.I., Permana, D.S., Hastuti, M.I. and Sudewi, R.S.S.
(2019). Dampak Asimilasi Data Radar Produk Cappi
Acknowledgment pada Prediksi Kejadian Hujan Lebat di Jabodetabek
The authors would like to acknowledge BMKG Menggunakan Model WRF-3DVAR. Jurnal Meteorologi
Indonesia for providing weather radar and synoptic dan Geofisika, 20(1), pp.47-54.
data. Permana, D. S., Hutapea, T.D.F, Praja, A.S., & Fatkhuroyan,
Muzayanah, L.F. (2006). Pengolahan Multi Data Format
Radar Cuaca Menggunakan Wradlib Berbasis Python.
References Jurnal Meteorologi dan Geofisika, 17(3), 157-164.
Nugroho, S.P. (2002). Evaluasi dan Analisis Curah Hujan Wang, H., Bruyère, C., Duda, M., Dudhia, J., Gill, D., Lin, H.
Sebagai Penyebab Bencana Banjir Jakarta. Jurnal Sains C., Michalakes, J., Rizvi, S., & Zhang, X. (2016). WRF-
dan Teknologi Modifikasi Cuaca, 3(2), 91-97. ARW Version 3 Modeling System User’s Guide, National
Gustari, I. ,Hadi T. W. ,Hadi S. , & Renggono F. (2012). Center for Atmospheric Research, Amerika Serikat.
Akurasi Prediksi Curah Hujan Harian Operasional di Wang, H., Sun, J., Zhang, X., Huang, X. Y., & Auligne, T.
Jabodetabek : Perbandingan Dengan Model WRF. Jurnal (2012). Radar Data Assimilation with WRF 4D-Var.
Meteorologi dan Geofisika, 13, 119-130 Part I: System Development and Preliminary Testing.
Gustari, I. (2014). Perbaikan Prediksi Cuaca Numerik Monthly Weather Review, 141, 2224-2244.
Kejadian Hujan Sangat Lebat Terkait dengan Sistem Wang, H., Sun. J., Fan. S., & Huang X. Y. (2013). Indirect
Awan di Jabodetabek Menggunakan Asimilasi Data Assimilation of Radar Reflectivity with WRF 3D-Var
Radar C-Band, Disertation, Sains Kebumian, ITB, and Its Impact on Prediction of Four Summertime
Bandung. Convective Events. Journal of Applied Meteorology
Sagita, N. (2017). Asimilasi Model Weather Research And and Climatology, 52, 889-902. https://doi.org/10.1175/
Forecasting (WRF) Dengan Data Observasi Untuk JAMC-D-12-0120.1
Prediksi Curah Hujan di Wilayah Jawa. Thesis, IPB. Lee J.-H., Lee H.-Ha., Choi Y., Kim H.-W., & Lee D.-K. (2010).
Bogor. Radar Data Assimilation for the Simulation of Mesoscale
Skamarock, William C., Klemp Joseph B., Dudhia J., Gill D O., Convective Systems. Advances in Atmospheric Sciences,
Barker D.M., Duda M G., Huang X. Y., Wang W., Powers 27(5), 1025–1042.
J G. (2008). A Description of the Advanced Research Jones, T.A., Otkin J.A., Stensurd D.J., & Knopfmeier, K. (2014).
WRF Version 3 NCAR Technical Note. Mesoscale and Forecast Evaluation of an Observing System Simulation
Microscale Meteorology Division. National Center for Experiment Assimilating Both Radar and Satellite Data.
Atmospheric Research, Boulder, Colorado, USA. Mon. Wea. Rev., 139, 755–773.
Liu, J., M. Bray., dan D. Han. (2013). A Study on WRF Radar Ma Y., Lu, M., Chen, H., Pan, M., Hong, Y. (2018).
Data Assimilation for Hydrological Rainfall Prediction. Atmospheric Moisture Transport Versus Precipitation
Hydrological Earth System, 17, 3095-3110. Across the Tibetan Plateau: A Mini-Review and Current
Liu, J., Tian J., Yan D., Li C., Yu F., & Shen F. (2018). Evaluation Challenges. Atmospheric Research. 209. 50-58.
of Doppler radar and GTS data assimilation for NWP Routray, A., Mohanty, U.C., Rizvi, S. R. H., Niyogi D., Osuri
rainfall prediction of an extreme summer storm in K. K., & Pradhan D. (2010). Impact of Doppler weather
northern China: from the hydrological perspective. radar data on numerical forecast of Indian monsoon
Hydrology and Earth System Sciences, 22, 4329-4348. depressions. Journal of the Royal Meteorology Society,
https://doi.org/10.5194/hess-22-4329-2018. 136, 1836-1850.
Li W., Xie Y., Deng S. M., & Wang Qi. (2010). Application of Tjasyono, B.H.K. (2009). Meteorologi Indonesia Volume 1,
the Multigrid Method to the Two-Dimensional Doppler Karakteristik dan sirkulasi Atmosfer, Badan Meteorologi
Radar Radial Velocity Data Assimilation. Journal of dan Geofisika, Jakarta.
Atmospheric and Oceanic Technology, 27, 319-322. Janiskova, M. (2015). Assimilation of cloud information from
https://doi.org/10.1175/2009JTECHA1271.1 space-borne radar and lidar: experimental study using a
Xiao, Q., Kuo, N. Y. H., Sun, J., Lee, W. C., Lim, E., Guo, Y. R., 1D+4D-Var technique. Journal of the Royal Meteorology
& Barker, D. M. (2005). Assimilation of Doppler Radar Society, 141, 2708-2725.
Observations with a Regional 3DVar System: Impact of Wiegand, B, 2015, Introduction to Numerical Weather
Doppler Velocities on Forecasts of a Heavy Rainfall Case. Prediction. htw saar – Hochschule für Technic und
Journal of Applied Meteorology, 44, 768- 788. Wirtschaft des Saarlandes (University of Applied

283
IMPROVING NUMERICAL WEATHER PREDICTION Miranti Indri Hastuti, et.al

Sciences). Fakultät für Ingenieurwissenschaften (School Ricciardelli, E., Paola F. D., Gentile, S., Cersosimo, A., Cimini
of Engineering). D., Galluci D., Geraldi, E., Larosa, S., Nilo, S. T., Ripepi,
Fatkhuroyan, A.S Praja, T. Wati. (2019). A preliminary E., Romano, F., & Voggiano, M. (2018). Analysis of
results of assessment of BMKG-WRF numerical model Livorno Heavy Rainfall Event: Examples of Satellite-
daily rainfall forecasts performance using categorical Based Observation Techniques in Support of Numerical
verification. IOP Conf. Ser.: Earth Environ. Sci. 284 Weather Prediction. Remote Sens. 2018, 10, 1549.
012013. doi:10.1088/1755-1315/284/1/012013 doi:10.3390/rs10101549
Sun, J. & Wang, H. (2013). Radar Data Assimilation with
WRF 4D-Var. Part II: Comparison with 3D-Var for a
Squall Line over the U.S. Great Plains. Mon. Wea. Rev.,
141, 2245–2264.

284