Research Collection
Working Paper
Existing urban transportation in Greater Jakarta Results of agent-based modelling
Author(s): Ilahi, Anugrah; Balać, Miloš; Axhausen, Kay W.
Publication Date: 2019-12
Permanent Link: https://doi.org/10.3929/ethz-b-000394347
Rights / License: In Copyright - Non-Commercial Use Permitted
This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use.
ETH Library 1 Existing urban transportation in Greater Jakarta: Results of
2 agent-based modelling
a,1,∗ a a 3 Anugrah Ilahi , Milos Balac , Kay W. Axhausen
a 4 ETH Zurich, Institute for Transport Planning and Systems, Stefano-Franscini-Platz 5, 8093 Zurich, 5 Switzerland
6 Abstract
7 Agent-based models have become popular over the years as more traditional modelling
8 approaches were not suitable for studying emerging transportation modes and transport
9 policies. However, creating a detailed scenario, which is the basis of the agent-based
10 model that represents the supply and demand for a study area, is not trivial and requires
11 substantial effort. This is even more the case when the study area is large and contains
12 many people living and performing many different daily activities. This is exactly the
13 case for the region of Greater Jakarta, the subject of this research. Greater Jakarta has
2 14 an area of approximately 8000 km and is home to approximately 30 million people.
15 Here we present the synthesis of the Greater Jakarta commuting scenario to help future
16 researchers generate large-scale scenarios in similar regions where the data is Limited or
17 difficult to gather. First runs, using an agent-based model MATSim, integrated with a
18 mode-choice model are presented and can be used as a backbone for future improvements
19 of the Greater Jakarta scenario and investigations of different transport related questions.
20 Keywords: MATSim; Agent-Based Model; Mode-Choice Model; Greater Jakarta
21 1. Introduction
22 Travel activities are affected by many different factors, including the mode of trans-
23 portation, the trip purpose, and social interactions with other people (Kitamura, 1988;
24 Axhausen and Grling, 1992). Bradley and Vovsha (2005) studied how intra-household
0 25 interactions changed trip decisions. For example, a family s morning travel activities
26 could be influenced by the needs of a child in the household. If a parent needed to drop
27 off the child at school before he or she went to the office, that decision impacted the way
∗Corresponding author Email address: [email protected] (Anugrah Ilahi) 28 the schedule was planned. Borgers et al. (2001); Gliebe and Koppelman (2005); Simma
29 and Axhausen (2001) noted that different time allocations for different activities greatly
30 impacted trip planning. In addition, Axhausen et al. (2002) discovered that existing
31 routines, weather, and the purpose of each trip made mapping transportation efforts a
32 challenging undertaking. Arentze and Timmermans (2004) found several types of con-
33 straints on travel such as household constraints, spatial constraints, time constraints, and
34 spatial-temporal constraints. An example of time constraints was the opening and closing
35 times of different stores or shopping places. In addition, the distance is a constraint on
36 travel. The farther away sometime was, the longer it would take to travel there from a
37 home location. Trip activities can be influenced by bigger group interactions; however,
38 people are more greatly influenced by their personal needs, occupations, ethnicity, na-
39 tionality, or interests. One of the greatest influences on a persons trip decisions was their
40 family or other people with whom they spent most of their time. For this purpose, we
41 developed an agent-based model that was able to simulate complex interactions based on
42 previous studies. There are several agent-based models, such as ORIENT/RV (Axhausen,
43 1989), TRANSIMS (Smith et al., 1995), SimMobility (Adnan et al., 2016), SimTRAVEL
44 (Pendyala et al., 2012), Multi-Agent Transport Simulation MATSim (MATSim) (Balmer
45 et al., 2006; Horni et al., 2016), and GEMSim (Saprykin et al., 2019). However, we make
46 use of MATSim in this research which has been shown to be suitable to model large-scale
47 cities in Singapore (Erath et al., 2012), was able to include microbuses in simulation
48 (Neumann et al., 2015), and utilized joint activities between household and household
49 members as can be seen in Dubernet and Axhausen (2015). In addition, it could also
50 simulate the impact of emerging transportation options and policies, such as the impact
51 of car-sharing (Balac et al., 2019), Urban Air Mobility (Balac et al., 2018), bike-sharing,
52 congestion pricing, automated vehicles or equity effects, which are hard to investigate on
53 a suitable level using the more traditional modeling techniques (Horni et al., 2016).
54 Our contributions in this paper were threefold: first, to the best of the authors knowl-
55 edge, this paper will be the first to make use of an agent-based model that incorporates all
56 of mode transport available including microbuses, called angkots, as a form of transporta-
2 57 tion and that simulates the daily behaviors of the people performing their daily activities
58 in Greater Jakarta. There are previous studies that had been conducted in Jakarta. Yagi
59 and Mohammadian (2010) simulated mode and destination choices based on discrete
60 choice modelling, and Dharmowijoyo et al. (2016) measured variability of travel patterns
61 in Greater Jakarta; however, those studies do not take in to account Greater Jakarta as
62 a whole object of study using agent-based modelling. Second, our model used a novel
63 approach that integrated the mode-choice model in MATsim simulation that allowed it
64 to filter unnecessary plans. As shown in H¨orlet al. (2018, 2019), this integration can
65 give a faster convergence speed of simulations than previous scoring based models which
66 estimated all the plan options available (Balmer et al., 2006; Horni et al., 2016). Third,
67 this paper will also add to the growing literature on modelling large-scale cities, especially
68 when the data is scarce or difficult to obtain. Finally, different from a conventional four
69 step model, each individual is simulated as an agent, with each of them having their own
70 attributes, such as sociodemographic features, activity locations, and modes of trans-
71 portation. The attributes of each agent are used as an input plan, and by using iterative
72 processing, we found the best plan that maximized utility.
73 The remainder of this paper is structured as follows: The following section describes
74 the case studies of Greater Jakarta. The third section explains the MATSim framework.
75 The fourth section presents mode-choice in MATSim. The fifth section presents the first
76 results obtained for the commuting population of the Greater Jakarta. Finally, the last
77 section presents conclusions, limitations, and further recommendations.
78 2. Case studies of Greater Jakarta
79 Greater Jakarta is comprised of nine cities and four regencies. The province of Jakarta
80 itself has five cities. The other four cities are Bogor, Depok, Tangerang, and Bekasi. The
81 four regencies are Tengerang, South Tangerang, Bogor, and Bekasi. Greater Jakarta,
82 known as Jabodetabek, has a population of approximately 30 million inhabitants. Greater
83 Jakarta has a significant role in the national economy in Indonesia producing more than
84 IDR 1,500 trillion (USD 113.51 Billion ). In 2012, Jakarta was the primary contributor
3 85 to the Jabodetabek GDP, with a share of 72.44%; therefore, the economic activities and
86 employment opportunities of Jabodetabek are concentrated in the city of Jakarta. At the
87 national level, Jabodetabek contributes 18.48% of Indonesias GDP. Around 3.6 million
88 people commute within, into, and out of Jakarta (BPS-Statistics, 2014). Therefore, traffic
89 congestion is generated by daily activities of commuters coming to Jakarta. The sample
90 of the data that we used is from a JICA study (JICA, 2009), as can be seen in Table 1.
91 The total number of respondents were 334,973. 69.91% of respondents were male, and
92 47.58% of respondents who worked had university degrees. Most of respondents, 71%,
93 were in the agglomeration area (the area outside of Jakarta), and 28.74% were living in
94 Jakarta.
4 Table 1: Sample summary statistics
Categorical variables N % Categorical variables N % Share Share Person is male ** 234,181 69.91 Car ownership University degree*** 96,542 47.58 No car 164’364 82.95 Employed** 202,924 60.58 1 12,703 7.10 Age categories ** 2 1,486 0.83 < 6years old 9,212 2.75 3 270 0.15 6 - 12 years old 62,086 18.53 > 3 130 0.07 12 - 18 years old 53,286 15.91 Motorcycle ownership 18 - 24 years old 34,627 10.34 No Motorcycle 53’009 29.62 24 -32 years old 51,435 15.35 1 92,668 51.78 32 - 42 years old 61,023 18.22 2 26,274 14.68 42 - 60 years old 57,565 17.18 3 5,635 3.15 > 60 years old 5,739 1.71 > 3 1,367 0.76 Income (in IDR per NMT ownership month)* No NMT 146,724 81.99 No answer 1,445 0.81 1 24,786 13.85 < 1 M 28,024 15.66 2 5,722 3.20 1 M - 3 M 116,461 65.08 3 1,192 0.67 3 M - 5 M 23,369 13.06 > 3 529 0.30 5 M - 8 M 6,746 3.77 Driving license** 8 M - 15 M 2,216 1.24 Motorcycle 98,854 29.56 > 15 M 692 0.39 Private car 7,410 2.22 Total expenditures (in Passenger vehicle 2,512 0.75 IDR per month)* Motorcycle and car 13,195 3.95 No answer 1,818 1.02 Motorcycle and 1’385 0.41 < 1 M 69,646 38.92 passenger vehicle 1 M - 3 M 96,849 54.12 No license 211,087 63.12 3 M - 5 M 8,204 4.58 Most frequently used 5 M - 8 M 1,825 1.02 mode of transport 8 M - 15 M 506 0.28 No answer 4,795 2.68 > 15 M 105 0.06 Commuter rail 7,554 4.22 Transport expendi- BRT 7,264 4.06 tures (in IDR per Feeder 46,090 25.76 month)* Taxi 171 0.10 No answer 16,798 9.39 Motorcycle taxi 6,524 3.65 < 1 M 150,430 84.06 Car 7,225 4.04 1 M - 3 M 11,229 6.27 Motorcycle 94,507 52.81 > 3 M 496 0.28 NMT 3,383 1.89 Spatial household lo- Others 1,440 0.80 cation * At time the survey was conducted, 7,470 IDR DKI Jakarta 53,084 28.72 was equivalent to 1 US Dollars Agglomeration 131,781 71.28 ** The calculation is based on the person who has either work or school activities *** The calculation is based on person who has work activities
5 95 3. MATSim framework
96 This study utilizes a Multi-Agent Transport Simulation (MATSim), which performs
97 a microscopic simulation of daily schedules of synthetic persons performing activities in
98 the study area. The persons in MATSim are called agents. Each agent having its own
99 plan that represents its daily schedule of activities connected by trips. The plans are
100 simulated using the mobility simulation (mobsim) for several iterations. Before the start
101 of each simulation, some of the agents can change a part of their plan in the re-planning
102 phase. This simulation cycle can be seen in Figure 1.
Figure 1: The MATSim loop
Source: MATSim book
103 In this work, we used a slightly different iterative approach proposed by H¨orlet al.
104 (2018, 2019) that integrated mode-choice models with the microsimulation in MATSim.
105 In this approach, agents can change their modes of travel based on an implementation
106 of a discrete-mode choice model. Since the mode-choice model in the re-planning phase
107 used estimates on the travel times, travel costs, waiting times etc. From the previous
108 iteration, the scoring phase was no longer needed. This approach was used to investigate
109 mode choices for the commuting population of the Greater Jakarta population.
110 3.1. Traffic flow model
111 Traffic simulation model in MATSim used a queue-based approach, which had two
112 attributes: storage and flow capacity. Storage capacity defined how many cars could be
113 stored at a time on a road link, and flow capacity defined the outflow capacity of a link.
6 114 3.2. Population synthesis
115 Commuting population was synthesized using the data gathered by the Japan Inter-
116 national Cooperation Agency (JICA) in 2009 (JICA, 2009), which included 3% of the
117 households in Greater Jakarta. To expand the sample and to synthesize the complete
118 commuting population, we utilized a Bayesian network approach and Generalized Raking
119 (GR) as shown in (Ilahi and Axhausen, 2019; M¨uller,2017; Sun and Erath, 2015). The
120 synthesized population contained approximately 20 million agents, which represented the
121 people who have either work or school activities based on a statistics bureau report at
122 the province level (Jabodetabek consists of 3 provinces: Jakarta, West Java and Banten).
123 There are two activity chains in this population: home work home, and home school
124 home. The locations of home, work, and school activities were based on the addresses
125 provided by the respondents and assigned a coordinate randomly drawn from an area
126 created by drawing a 1 km radius circle around the coordinates of each address. Unfor-
127 tunately, starting times of work activities as well as their duration was not reported in
128 the survey conducted by JICA; therefore, we distributed the starting times and activity
129 durations based on the behavior of people living in the Ile-de-France region, as the area
130 of the city is similar and the data was publicly available.
131 3.3. Network
132 The network of Greater Jakarta was used to create the scenario based on the (OSM)
133 OpenStreetMap (http://openstreetmap.com). The network of Greater Jakarta was ex-
134 tracted from OSM using Java Osmosis. The network consisted of 472,205 nodes and
135 1,214,769 links. The links were classified by their function and parameter, such as speed,
136 lane capacity, and road hierarchy. The road classification was reported in Table 2. The
137 network used coordinate system EPSG:5330 for Batavia/Jakarta.
138 3.4. Public transport network and counting stations
139 In Jakarta and in Indonesia in general, there are different modes of transportation
140 available, some of them formal, like a car, motorcycle, commuter rail, Bus Rapid Transit
141 (BRT), big buses, medium size buses, and some informal ones like microbuses (called
7 Table 2: Road network classification in the MATSim Greater Jakarta scenario
Hierarchy Highway-type Lanes Free Free speed Lane One way speed(m/s) Factor capacity 1 Motorway 2 33.33 1.0 2000 True 2 Trunk 1 22.22 1.0 2000 False 3 Primary 1 22.22 1.0 1500 False 4 Secondary 1 8.33 1.0 1000 False 5 Tertiary 1 6.94 1.0 600 False 6 Minor 1 11.11 1.0 600 False 6 Residential 1 4.16 1.0 600 False 6 Living street 1 2.77 1.0 300 False 7 Rail 1 44.44 1.0 9999 True
142 angkot, which is an informal service without a fixed schedule). Microbuses, which have
143 a small size, have an ability to become door to door services in Greater Jakarta. Simu-
144 lation of microbuses in an agent-based model has been previously used in the South-
145 African context (Neumann et al., 2015). To create a public transport network, we
146 need OSM and GTFS (General Transit Feed Specification) data; however, there is no
147 publicly available GTFS data for Greater Jakarta. Therefore, we have manually con-
148 structed the public transport schedules using the data from a company called Trafi
149 (https://www.trafi.com/id/jakarta). The data that we scraped from traffic websites are
150 formatted to GTFS structure data (Google, 2019). There are several important files that
151 must be available, which can be seen in Table 3. OSM network and manually constructed
152 GTFS schedules were converted to MATSim format using the pt2matsim extension (Po-
153 letti, 2016). BRT lines are categorized as dedicated lanes. In the end, the transit schedule
154 is mapped to the MATSim network using the same extension. Finally, we obtained public
155 transport lines within the network. There are 1,756 public transport lines. There are
156 325 BRT lines (including other bus companies that operate in BRT lines), 421 Bus lines,
157 22 commuter rail lines, and 988 microbus lines. There are also 20 counting stations that
158 count number of vehicles in 15min bins as can be seen in Figure 2.
8 Table 3: Overview of GTFS files
File Description agency.txt Public transport company that operates the public transport lines. In our case, these consist of BRT, Rail, Angkot, Buses. BRT are all public transports that operate in busway/dedicated line (Transjakarta), Buses are all big/medium buses that operate in mixed traffic. Rail consists of commuter lines that operate by public company railway (PT. KAI). Angkot are all microbus lines. stops.txt Station/shelter location, which consist coordinate of shelter station, name of shelter/station. routes.txt The name of public transport routes/lines. A route is a group of trips that are displayed to riders as a single service. trips.txt Trips for each route. A trip is a sequence of two or more stops that occurs at specific time. stop times.txt Times that a vehicle arrives at and departs from individual stops for each trip.
Figure 2: MATSim network, public transport line, counting station
159 3.5. Private and public transport vehicles
160 As the vehicles have different sizes and capacities in our simulation, we classified
161 private and public transport vehicles as in Table 4. Cars and motorcycles were classified as
9 162 private vehicles. BRT, buses, commuter, and microbuses are classified as public transport.
Table 4: Transit and private vehicles
Mode Name Symbol Length/Width Capacity Number of [m] Seats/Standing lines Private Car car 4.3/1.6 7/0 - vehicles Motorcycle mc 1.7/1.0 2/0 - Public BRT pt 2.5/50 50/30 325 transport Bus bus 2.5/35 35/15 421 Commuter rail 240/2.8 2.8/1000 22 Micro-bus angkot 4.2/1.6 1.6/8 988
163 4. Mode-choice in MATSim
164 The utility function for different modes were based on the values estimated for the
165 city of Zurich (H¨orlet al., 2018, 2019). However, the parameters had to be calibrated
166 for the model to fit the behavior of people in Greater Jakarta. As the people in Jakarta
167 have lower income levels and are more cost sensitive than people in Zurich, we modified
168 the cost parameter accordingly. The formulation of the utility functions for modes used
169 in the scenario are presented here:
UpublicT ransport =βnumberOfT ransfers,P T × XnumberOfT ransfers,P T + βinV ehicleT ime,P T
× XinV ehicleT ime,P T + βtransferT ime,P T × XtransferT ime,P T + (1)
βaccessEgrestime,P T × XaccessEgrestime,P T + βCost × Xcost,P T
UCar =ASCCar + βtravelT ime,Car × XtravelT ime,Car + βtravelT ime,Car
× θparkingSearchT ime,Car + βtravelT ime,Car × θaccessEgressT ime,Car + βCost (2)
× XdistanceCost,Car + βCost × Xcost,Car
UMotorcycle =ASCMC + βtravelT ime,MC × XtravelT ime,MC + βtravelT ime,MC
× θaccessEgressT ime,MC + βCost × XdistanceCost,MC (3)
+ βCost × Xcost,MC
UW alk =ASCNMT + βtravelT ime,W alk × XtravelT ime,W alk (4)
10 170 Table 5 shows the parameters obtained after calibration, and the result can be seen
171 in Fig 3. We also included the value of per km cost for cars and motorcycles, which were
172 0.05 USD and 0.015 USD respectively, and were based on fuel prices in Jakarta.
Table 5: Calibrated parameters of mode choice model
Mode Parameters Estimation Public Trans- βnumberOfTransers, PT. -0.170 port βinVehicleTime, PT [min-1] -0.120 βtransferTime, PT [min-1] -0.048 βaccessEgressTime, PT [min-1] -0.080 Car αCar 1.227 βtravelTime, Car [min-1] -0.066 Motorcycle αMC 1.227 βtravelTime, MC [min-1] -0.100 Walk αwalk 1.430 βtravelTime,walk [min-1] -0.141 Other βCost -0.030 Calibration θparkingSearchTime, Car [min] 4.000 θaccessEgressTime, Car [min] 4.000 θaccessEgressTime,MC [min] 2.000
173 5. First result
174 5.1. Mode shares
175 The mode-share results for a simulated sample of 5 (%) are shown in Figure 3. We
176 can observe some differences in the mode shares, especially for a car mode. However,
177 looking at motorcycles and cars together, the statistics match very well. One of the
178 possible reasons for underestimating the amount of car trips is the unreliability of car
179 and motorcycle ownership data, which is not always available in the JICA dataset.
180 5.2. Traffic counts
181 There are 20 counting locations in our network. The traffic volume results for one
182 of the most important roads in central Jakarta (Thamrin Street) can be seen Figure 4,
183 which presents the comparison between traffic volume in MATSim and traffic count data,
184 from 1 am to 4 pm. While having lower counts for cars makes sense (as we only simulated
185 commuters and also had a lower share of cars in the simulation), having a larger number of
186 motorcycle numbers is a concern. There are several possible reasons for this discrepancy.
11 Figure 3: Mode shares comparison between JICA study and MATSim
187 The counting data might not be complete, the MATSim might overestimate the number
188 of motorcycles taking this route, or the traffic count data might not be complete. Further
189 investigations should find suitable explanations and make the appropriate adjustments
190 to the data.
Figure 4: Count comparison for motorcycle (left) and car (right) on Thamrin Street
191 The results, in Table 6, show that 37.67% of the trips consisted of a distance between
192 1 to 5 km, 22.04% had a distance of 5 to 10 km, and 7.72% were trips with a distance
193 that was less than 1 km. If we add the trips higher than 10 Km together, we get 33%, or
194 Greater than 54 % were trips further than 5 km. It shows that there were a significant
195 share of people that were living a far distance from their activity locations. It is because
196 the people in Greater Jakarta find it difficult to find homes near their activity locations.
12 197 We cannot compare this result with commuter trip surveys from Statistic Indonesia in
198 Jakarta (BPS-Statistics, 2014), because our model is a mixture of commuter and non-
199 commuter trips. Additionally, we only simulated a person who has trip activities.
Table 6: Trip Distance Band
Distance Trips %Trips Less than 1Km 105,372 7.72 1 5 km 514,413 37.67 5 10 km 300,938 22.04 10 20 km 248,785 18.22 20 30 km 95,119 6.97 More than 30 km 100,813 7.38
200 5.3. Accessibility of public transport infrastructures
201 We measured the accessibility of public transport infrastructures in Greater Jakarta.
202 Accessibility can be defined as the availability of means of urban transport to support
203 the travel of individuals from their homes to the destination location (Dalvi and Martin,
204 1976). It can also be measured based on several assumptions about travel behaviors and
205 transport supply (Pirie, 1979). The method to measure accessibility needs more improve-
206 ments, especially to capture activity patterns using activity based models. As Geurs and
207 van Wee (2004) explained, there are four common methodologies to measure accessibility,
208 such as infrastructure-based, location-based, person-based, and utility-based, although
209 those methodologies could not capture intense travel behaviors.
210 There is a rich literature of studies that try to measures accessibility. Paez et al.
211 (2012) reviewed several measures of accessibility with the focus on normative (i.e. pre-
212 scriptive), and positive aspects (i.e. descriptive) in the city of Montreal, Canada. They
213 identified the gap between desired (as normatively defined) and actual levels (as revealed).
214 Chen et al. (2013) developed a reliable space-time prism model to analyze service areas
215 with travel time uncertainty and continued their research to evaluate accessibility un-
216 der travel time uncertainty for large-scale urban areas. This research was inspired by a
217 placed-based framework (Chen et al., 2017). Hallgrimsdottir et al. (2016) examined ac-
218 cessibility policy for older people and wheelchair users in Sweden between 2004 and 2014
219 for the outdoor environment. Pyrialakou et al. (2016) measured, identified, evaluated,
13 220 and quantified transport disadvantages in the U.S with measure accessibility, mobility,
221 and travel behaviors. Maroto and Zofo (2016) measured accessibility infrastructure in dif-
222 ferent regions using the non-parametric frontier approach (DEA) with a dynamic scope
223 (Malmquis indices). L¨attman et al. (2016) developed a Perceived Accessibility Scale
224 (PAC) that measured peoples perceptions when they traveled with a specific mode of
225 transport.
226 Dong et al. (2006) used ABA (activity-based accessibility) to measure all individual
227 activities, to integrate scheduling and travel characteristics, and to examine all trips and
228 activities throughout the day. Dubernet and Axhausen (2016) measured accessibility
229 using a joint destination-mode choice model and MATSim. There are three basic models
230 that should be calculated to measure accessibility, i.e. destination choice, mode choice,
231 and route choice.
232 The measurements of accessibility in this study were MATSim accessibility extensions
233 seen as potential accessibility (Ziemke, 2016) and also from the LIMA project (H¨orl,
234 2019). Hansen (1959) defined the potential accessibility and calculated it for the whole
235 scope of activity facilities (e.g. shopping, leisure, etc.). In our case, the accessibility
236 was calculated based on a primary tour of home-to-work-to-home and home-to-school-to-
237 home. There are three activity facilities that we measured, such as home facilities, work
238 facilities, and school facilities. The mathematical form, which was based on the LIMA
239 project, is as follows:
X U(i) = log( oj × exp(−β × tij)) (5) j
240 When oj is the number of activity facilities in zone j, tij is the travel time from zone i
241 to j and β is a configurable factor. The results of accessibility in Greater Jakarta can
242 be seen in Figure 5. It shows that accessibility for all Greater Jakarta is not well served
243 and central Jakarta has the highest level of accessibility. It means that more options of
244 public transport lines and facilities are available in central Jakarta, such as the BRT line,
st 245 commuter rail line, and the 1 phase of the Mass rapid Transit (MRT) line.
14 246 The relatively high accessibility can also be seen in city of Tangerang, Bekasi, Depok,
247 and Bogor, where there are settlement areas. In addition, commuters from those cities
248 do their activities in Jakarta. There are well known universities such as University of
249 Indonesia and Bogor Institute of Agriculture, located in Depok and Bogor. Lower acces-
250 sibility can be seen in the regency area. The reason is that public transportation is quite
251 limited in the regencies and the number of inhabitants is more concentrated in the city.
252 Moreover, the local budget allocations in the regencies are smaller than they are in the
253 city, so, there are less funds available for public transport infrastructures.
15 Figure 5: Accessibility of public transport lines in several location
254 6. Discussion and conclusions
255 In this paper, we use an agent-based modelling framework to simulate the commuting
256 population of Greater Jakarta. The methodology presented here also utilizes a novel
16 257 approach that integrates mode choice with a micro-simulation in MATSim. The results
258 show that differences between the MATSim and JICA mode shares are very low. Further
259 research is needed in order to represent complete travel behavior of people living in
260 Greater Jakarta. Nevertheless, this research provides the backbone on which further
261 research can be built. The accessibility of public transportation in Greater Jakarta needs
262 to be improved to accommodate all the demand; however, it will take a considerable
263 amount of money and time for construction, due to the size of the city and its population.
264 We also see that the regency areas have lower accessibility than the city; therefore, the
265 government also need to focus on development in the regency areas and manage the
266 settlement areas to create a more compact area and minimize the sprawl. Furthermore,
267 there are significant shares of people living far from activity locations due to the high
268 cost of housing near activity locations. Additionally, as these residents are typically
269 Indonesian, they tend to live in landed houses that make it more difficult and expensive
270 to stay near activity locations.
271 In this research, we only synthesized the population that performed mandatory ac-
272 tivities. The next step is naturally to expand this to include secondary activities. This
273 is currently not possible, as the data on travel-behaviors in Jakarta does not exist. The
274 parameters of the mode choice model are based on the Zurich model which is a limiting
275 factor. In future work, we hope to obtain the necessary stated-preference data to be able
276 to estimate a mode-choice model that is more fitting to the population of Jakarta. We
277 use 5% of the population in our simulations, as it is time intensive to simulate a popu-
278 lation of 20 million people. Further research should also focus on using larger samples
279 to make sure the results stay stable and measure the impact of other emerging modes of
280 transportation.
281 References
282 Adnan, M., F. Pereira, C. Lima Azevedo, K. Basak, M. Lovric, S. Raveau, Y. Zhu,
283 J. Ferreira, C. Zegras and M. Ben-Akiva (2016) Simmobility: A multi-scale integrated
284 agent-based simulation platform.
17 285 Arentze, T. A. and H. J. Timmermans (2004) A learning-based transportation oriented
286 simulation system, Transportation Research Part B: Methodological, 38 (7) 613 – 633.
287 Axhausen, K. (1989) Eine ereignirsorientierte simulation von aktivittenketten zur park-
288 standswahl.
289 Axhausen, K. W. and T. Grling (1992) Activity-based approaches to travel analysis:
290 conceptual frameworks, models, and research problems, Transport Reviews, 12 (4)
291 323–341.
292 Axhausen, K. W., A. Zimmermann, S. Sch¨onfelder,G. Rindsf¨userand T. Haupt (2002)
293 Observing the rhythms of daily life: A six-week travel diary, Transportation, 29 (2)
294 95–124.
295 Balac, M., H. Becker, F. Ciari and K. W. Axhausen (2019) Modeling competing free-
296 floating carsharing operators a case study for Zurich, Switzerland, Transportation
297 Research Part C: Emerging Technologies, 98, 101 – 117.
298 Balac, M., A. R. Vetrella, R. Rothfeld and B. Schmid (2018) Demand estimation for
299 aerial vehicles in urban settings, IEEE Intelligent Transportation Systems Magazine.
300 Balmer, M., K. W. Axhausen and K. Nagel (2006) Agent-based demand-modeling frame-
301 work for large-scale microsimulations, Transportation Research Record, (1985) 125–134.
302 Borgers, A., F. Hofman and H. Timmermans (2001) Conditional choice modelling of
303 time allocation among spouses in transport settings, European Journal of Transport
304 and Infrastructure Research, 2 (1).
305 BPS-Statistics (2014) Statistik Komuter Jabodetabek, Hasil Survei Komuter Jabodetabek
306 2014, Jakarta.
307 Bradley, M. and P. Vovsha (2005) A model for joint choice of daily activity pattern
308 typesof household members, Transportation, 32 (5) 545–571.
18 309 Chen, B. Y., Q. Li, D. Wang, S.-L. Shaw, W. H. K. Lam, H. Yuan and Z. Fang (2013)
310 Reliable spacetime prisms under travel time uncertainty, Annals of the Association of
311 American Geographers, 103 (6) 1502–1521.
312 Chen, B. Y., H. Yuan, Q. Li, D. Wang, S.-L. Shaw, H.-P. Chen and W. H. K. Lam
313 (2017) Measuring place-based accessibility under travel time uncertainty, International
314 Journal of Geographical Information Science, 31 (4) 783–804.
315 Dalvi, M. Q. and K. M. Martin (1976) The measurement of accessibility: Some prelimi-
316 nary results, Transportation, 5 (1) 17–42.
317 Dharmowijoyo, D. B. E., Y. O. Susilo and A. Karlstr¨om(2016) Day-to-day variability
318 in travellers’ activity-travel patterns in the Jakarta metropolitan area, Transportation,
319 43 (4) 601–621.
320 Dong, X., M. E. Ben-Akiva, J. L. Bowman and J. L. Walker (2006) Moving from trip-
321 based to activity-based measures of accessibility, Transportation Research Part A: pol-
322 icy and practice, 40 (2) 163–180.
323 Dubernet, T. and K. W. Axhausen (2015) Implementing a household joint activity-travel
324 multi- agent simulation tool: first results, Transportation, 42 (5) 753–769.
325 Dubernet, T. and K. W. Axhausen (2016) Using a joint destination-mode choice model
326 for developping accessibility measures, in 16th Swiss Transport Research Conference
327 (STRC 2016), Ascona.
328 Erath, A., P. J. Fourie, M. A. van Eggermond, S. A. Ordonez Medina, A. Chakirov and
329 K. W. Axhausen (2012) Large-scale agent-based transport demand model for Singa-
330 pore, Arbeitsberichte Verkehrs-und Raumplanung, 790.
331 Geurs, K. T. and B. van Wee (2004) Accessibility evaluation of land-use and transport
332 strategies: review and research directions, Journal of Transport Geography, 12 (2)
333 127–140.
19 334 Gliebe, J. P. and F. S. Koppelman (2005) Modeling household activity–travel interactions
335 as parallel constrained choices, Transportation, 32 (5) 449–471.
336 Google (2019) General transit feed specification reference,
337 https://developers.google.com/transit/gtfs/reference/. Accessed: 2019-
338 01-13.
339 Hallgrimsdottir, B., H. Wennberg, H. Svensson and A. Sthl (2016) Implementation of ac-
340 cessibility policy in municipal transport planning progression and regression in Sweden
341 between 2004 and 2014, Transport Policy, 49, 196–205.
342 Hansen, W. G. (1959) How accessibility shapes land use, Journal of the American Institute
343 of Planners, 25 (2) 73–76.
344 H¨orl,S. (2019) Lima poc 3. Accessed: 2019-03-10.
345 H¨orl,S., M. Balac and K. W. Axhausen (2018) A first look at bridging discrete choice
346 modeling and agent-basedmicrosimulation in matsim, paper presented at the The 7th
347 International Workshop on Agent-based Mobility, Traffic and Transportation Models,
348 Methodologies and Applications (ABMTrans).
349 H¨orl,S., M. Balac and K. W. Axhausen (2019) Pairing discrete mode choice models and
350 agent-based transport simulation with matsim, paper presented at the 98th Annual
351 Meeting of the Transportation Research Board (TRB ).
352 Horni, A., K. Nagel and K. W. Axhausen (2016) The multi-agent transport simulation
353 MATSim, Ubiquity Press London.
354 Ilahi, A. and K. Axhausen (2019) Integrating bayesian network and generalized raking
355 for population synthesis in greater jakarta, Regional Studies, Regional Science, 6 (1)
356 623–636.
357 JICA (2009) Traffic data collected under the jabodetabek urban transport policy inte-
358 gration.
20 359 Kitamura, R. (1988) An evaluation of activity-based travel analysis, Transportation,
360 15 (1) 9–34.
361 L¨attman,K., L. E. Olsson and M. Friman (2016) Development and test of the perceived
362 accessibility scale (pac) in public transport, Journal of Transport Geography, 54, 257–
363 263.
364 Maroto, A. and J. L. Zofo (2016) Accessibility gains and road transport infrastructure in
365 spain: A productivity approach based on the malmquist index, Journal of Transport
366 Geography, 52, 143–152.
367 M¨uller,K. (2017) A generalized approach to population synthesis, Ph.D. Thesis, ETH
368 Zurich.
369 Neumann, A., D. Rder and J. W. Joubert (2015) Towards a simulation of minibuses in
370 south africa, Journal of Transport and Land Use, 8 (1) 137.
371 Paez, A., D. M. Scott and C. Morency (2012) Measuring accessibility: positive and
372 normative implementations of various accessibility indicators, Journal of Transport
373 Geography, 25, 141–153.
374 Pendyala, R., K. Konduri, Y.-C. Chiu, M. Hickman, H. Noh, P. Waddell, L. Wang,
375 D. You and B. Gardner (2012) An integrated land use–transport model system with
376 dynamic time-dependent activity-travel microsimulation, 91st Annual Meeting of the
377 Transportation Research Board, Washington, DC, 2303.
378 Pirie, G. H. (1979) Measuring accessibility: A review and proposal, Environment and
379 Planning A: Economy and Space, 11 (3) 299–312.
380 Poletti, F. (2016) Public transit mapping on multi-modal networks in matsim, Thesis.
381 Pyrialakou, V. D., K. Gkritza and J. D. Fricker (2016) Accessibility, mobility, and realized
382 travel behavior: Assessing transport disadvantage from a policy perspective, Journal
383 of Transport Geography, 51, 252–269.
21 384 Saprykin, A., N. Chokani and R. S. Abhari (2019) Gemsim: A gpu-accelerated multi-
385 modal mobility simulator for large-scale scenarios, Simulation Modelling Practice and
386 Theory, 94, 199 – 214.
387 Simma, A. and K. W. Axhausen (2001) Within-household allocation of travel: Case of
388 upper austria, Transportation Research Record, 1752 (1) 69–75.
389 Smith, L., R. Beckman and K. Baggerly (1995) Transims: Transportation analysis and
390 simulation system.
391 Sun, L. and A. Erath (2015) A bayesian network approach for population synthesis,
392 Transportation Research Part C, 61, 49–62.
393 Yagi, S. and A. K. Mohammadian (2010) An activity-based microsimulation model of
394 travel demand in the Jakarta metropolitan area, Journal of Choice Modelling, 3 (1) 32
395 – 57.
22