MATSim Special session
Authors: Golan Ben-Dor1, Bella Dmitrieva2, Michał Maciejewski3,4, Joschka Bischoff4
Eran Ben-Elia2, Itzhak Benenson1
1Tel Aviv University; 2Ben-Gurion University of the Negev, 3Poznan University of
Technology, 4TU Berlin
Subject: MATSim simulations in the Tel Aviv Metropolitan Area: Direct competition
between public transport and cars on the same roadway
The Israeli road network is still car-oriented. In the Tel Aviv metropolitan area there are 147
km of dedicated bus lanes. Compared to other metropolitans around the world, we see that
Tel Aviv is miles behind (Bocher, 2014). Simulation of the Tel Aviv traffic demands
comprehensive processing of the car and public transport (PT) competition on the same
MATSim possesses the ability of simulating concurrent use of the road network by PT and
private cars, but this ability remains unexploited. However, vast majority of MATSim
simulations ignore competition between private cars and buses for the same network
(Horni, Nagel, & Axhausen, 2016). Our application focuses on calibrating MATSim for this
purpose. The goal of the this calibration is to apply MATSim for predicting the effects of
transportation network changes in relation to the Light Rail (LRT) construction in the Tel Aviv
metropolitan area (TLV below) that started in 2014 and will continue until 2021.
Transportation in the Tel Aviv metropolitan area
The investments in the PT network in TLV are far
below the necessary level (Figure 1) and bus trips
share is major among the modes of TLV PT (Table
1) (Nir et al., 2015).
Buses and private cars compete for the same
road space in TLV. This is possibly the reason that
the only two existing MATSim applications in
Israel have ignored PT and focused only on
private car traffic. Bekhor et al. (2010) created a
synthetic TLV population of car users using the
generator of the "Tel Aviv activity-based model"
(Cambridge Systematics, 2008) and applied it on EMME2-based road network for TLV that is
used for metropolitan transportation planning. Dobler & Horni (2014) enriched the
application by including a toll road #6 and enabling agents re-planning abilities to include
Table 1 – Mode share for daily travelers in TLV (Amir, 2010)
Inside the Tel Aviv Metropolitan
Outside of the Metropolitan Area
Figure 1. Investment in PT per person (Nir et al., 2015)
(1) To extend the TLV MATSim application by including all currently available and projected
modes, with an emphasis on bus, train and paratransit.
(2) To validate the TLV MATSim application with the LRT construction examples.
Current State of the TLV MATSim application
The data on the TLV road network are supplied by “Netivey Ayalon” LTD, the agency
responsible for the maintenance and updating of the metropolitan model.
The current road network contains 13,109 links and 5,152 junctions. The transit module
contains all transit lines operating inside the TLV area. A dozen of operators that are active
in the area exploit 1,151 transit lines (1,122 Buses and 29 Trains) that have 6,013 stops.
Figure 2 presents the view of entire TLV network at 06:00, while Figure 3 presents the state
of the system at 07:00, when many buses are running, and at 7:30, with a moving train.
Figure 3. (a) TLV transportation system at 07:00 (many buses running) and (b) at 7:30, with a
Figure 2. General view of the TLV transportation network, 06:00
Problems of public transport scalability
A typical MATSim scenario uses a 10% population sample, and consequently 10% of the
original private car fleet. To reflect the system for 100% of users, MATSim modifies several
basic parameters of the traffic flow queue model (Rieser et al., 2014), the most important
being flow capacity and storage capacity:
𝑓𝑙𝑜𝑤 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 = ( 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑙𝑖𝑛𝑘
𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝑝𝑒𝑟𝑖𝑜𝑑 𝑜𝑓 𝑛𝑒𝑡𝑤𝑜𝑟𝑘) × 𝑓𝑙𝑜𝑤 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝑓𝑎𝑐𝑡𝑜𝑟
𝑆𝑡𝑜𝑟𝑎𝑔𝑒 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 = (𝑙𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑙𝑖𝑛𝑘× 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑙𝑎𝑛𝑒𝑠 𝑜𝑓 𝑙𝑖𝑛𝑘
𝑒𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑒𝑙𝑙 𝑠𝑖𝑧𝑒 ) × 𝑠𝑡𝑜𝑟𝑎𝑔𝑒 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝑓𝑎𝑐𝑡𝑜𝑟
The above scheme works well either when only private cars are simulated, or when private
cars and PT vehicles are simulated on separate networks. However, mixing PT vehicles and
private cars in one traffic stream dominated by the latter ones turns out problematic.
Specific issues include:
- The number of the PT vehicles cannot be reduced proportionally to the percentage of
travelers participating in the scenario, as conducted with private cars: The use of 10%
instead of 100% of buses would reduce the frequency of buses 10-fold.
- As a result, the size of the PT vehicle should be scaled down proportionally to the
reduction of the flow/storage capacities. MATSim contains a solution to resolve this
issue – the PCE value (Passenger Car Equivalent) that proportionally reduces the size of
different vehicle types, but the influence of the PCE on the model output has not been
- Traffic simulation of a 10% sample exacerbates congestion as traffic flow is not so fine-
grained anymore. Because private cars can change their route to avoid congestion, while
PT vehicles must follow fixed routes, it is the PT vehicles that are mostly affected by
scenario down sampling.
Problems of public transport scalability in the TLV scenario
To investigate the MATSim abilities for the case of the concurrent use of the roads by private
cars and PT, we investigated a 10% population scenario and full PT fleet and schedule for
morning hours. To scale down the traffic flow model properly, we used the following
𝑓𝑙𝑜𝑤 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝑓𝑎𝑐𝑡𝑜𝑟 = 0.1,
𝑓𝑙𝑜𝑤 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦 𝑓𝑎𝑐𝑡𝑜𝑟 = 0.18
(Nagel, 2016a), 𝑃𝐶𝐸 𝐶𝐴𝑅 = 1
(Nagel, 2016b) (size remains the
same, but the amount of cars is
reduced 10 times), 𝑃𝐶𝐸 𝐵𝑈𝑆 = 0.3
(instead of 3.0, which is the
original bus PCE; the PT fleet size
and PT schedule remain the
same). The concurrent use of
roads by private and PT vehicles
results in essential and unrealistic bus delays in this scenario (Figure 4).
Figure 4. TLV 10% scenario, route-time diagram: The unrealistic 2-hour
delay for Dan line #63 in the model (blue) relative to the transit schedule
Vehicle Speed (Relative to Link Free Speed)
To explain the delay, Figure 5 presents a MATSim snapshot of link 2822 consisting of 51.6
vehicles/hour capacity that is simultaneously used by cars and several bus lines. As one can
see, when buses of several lines entered this link after the private car (shortly before 06:19),
the flow capacity of a link is overused and the buses cannot leave it for long.
Figure 5. Private car C (corresponding to 10 real cars) leaves the link at 16:19:41. Until then,
7 buses that entered the link between 06:16:32 - 16:19:41 are bunched and cannot leave it.
Possible solution of public transport scalability in the TLV scenario
To resolve the problem of unrealistic bus delays, we modified the queueing model of the
mobility simulator (QSim), by allowing the buses to ignore the link flow capacity restriction
Only private cars (in contrast to buses) are kept on the link until its flow capacity
accumulator recovers. As a result, when the private car leaves the link, the buses directly
behind it can immediately move over the intersection (provided that there is enough space
for them on the next link). Figure 6 illustrates this solution for the traffic situation on same
Figure 6. Traffic on the link 2822 for the modified
QSim, 10% scenario: (a) Private car C enters the
link at 06:13:52, bus B1 is about to enter the link
a moment after; (b) Bus B2, enters the link at
06:15:00, B1 caught up C (that is obscured by
B1); (c) private car C is about to leave the link at
06:15:15 and this will immediately release B1
and then, very soon, B2.
The corresponding changes to the MATSim’s QSim code can be viewed here:
We are currently investigating the consequences of this solution and its effectiveness. The
results will be presented in the paper.
1. Amir, R. (2010). Reorganization of Public Transport in Tel Aviv Metropolitan Area. Paper
presented at The Israeli Association of Transport Research.
2. Bekhor, S., Dobler, C., & Axhausen, K. W. (2010). Integration of Activity-Based with
Agent-Based Models: an Example from the Tel Aviv Model and MATSim. Washington,
D.C: 90th Annual Meeting of the Transportation Research Board.
3. Bocher, B. (2014). The National Program for Prioritizing Public Transport in Israel. 34th
Annual Meeting of the Israeli Association of Transport Research, Israel.
4. Dobler, C., & Horni, A. (2014). Update and Extension of the MATSim Model of Tel Aviv.
Arbeitsberichte Verkehrs- und Raumplanung. IVT, ETH Zürich. Zürich. Retrieved from
5. Horni, A., Nagel, K., Axhausen, K. W. (2016). The multi-agent transport simulation
MATSim (Vol. 9). London: Ubiquity.
6. Nagel, K. (2016a). Can I run MATSim with sample populations, e.g. with 10% of the true
population size? - MATSim community space - confluence. Retrieved from
7. Nagel, K. (2016b). Mixed traffic - MATSim community space - confluenc. Retrieved from
8. Nir, S., Ashak, R., Shiftan, Y., Balasha, D., Berechman, J., Magin, D., Szeinuk, M. (2015).
Development of PT. 35th Annual Meeting of the Israeli Association of Transport
9. Rieser, M., Dobler, C., Dubernet, T., Grether, D., Horni, A., Lammel, G., . . . Nagel, K.
(2014, 9.12.2014). MATSim User Guide Development Version. Retrieved from
10. Cambridge Systematics. (2008). Tel-Aviv Activity Schedule Travel Demand Model System:
a Tour-Based Approach. Report prepared for the Israel Ministry of Transport and Road