MATSim-NYC: a virtual test bed for policies and transportation technologies in NYC, a reference guide

Two journal articles have been published related to the MATSim-NYC project so far, so I want to take this opportunity to provide a short overview of this work. This effort is a two-year project at C2SMART where we sought to develop a simulation model to evaluate travel demand impacts of transportation policies and technologies. This is the first of a "Network of Living Labs" using a uniform simulation platform across different cities so that new research in algorithms and operating policies can be consistently evaluated and compared under different market environments.

The tool is built from open-source software: MATSim, and designed as a supplement to tools already available to planning agencies, e.g. NYBPM used by NYMTC to evaluate long range transportation plans. Unlike those primary transportation planning tools, MATSim-NYC and the rest of the Network of Living Labs is developed with the intent to be a quick-response benchmarking tool for different emerging technologies or policies. If a public agency wanted to test certain transit alignments, different congestion pricing schemes, service regions for different mobility options, this tool can provide some consistent assessment of its impact on the city. The key strengths of this tool is the combination of dynamic traffic and heterogeneous activity scheduling behavior of the population through a multi-agent simulation. This means that the tool can capture the trade-offs between traffic congestion by time of day with the route and departure time decisions of its varied travelers at a citywide level.

The fabric of the city is captured by a synthetic population, which the details of its development can be found in the first paper in Transportation Research Part A:

https://www.sciencedirect.com/science/article/abs/pii/S096585642030745X

A synthetic population models each individual in the city as a unique agent with simulated characteristics. As a result, we can quantify heterogeneous effects of built environment policies on different segments of the population. For example, we use the synthetic population to show that if Amazon had succeeded in building the headquarters in Long Island City, it would have increased the peak morning trips from 5000 up to at least 8000 and be able to predict time of day distributions of trips by different modes like ride-hail ("for hire vehicles").

We can also use the synthetic population to show how Citi Bike's expansion plan could increase ridership by 92% (assuming level of service is kept up in those new regions -- which can be notably harder due to the lower population density).

These scenarios are conducted even without using the MATSim component. When traffic simulation and day-to-day behavioral responses are taken into account, we have the second paper in Transport Policy:

https://www.sciencedirect.com/science/article/abs/pii/S0967070X20309483

We describe the calibration and validation process for MATSim-NYC and show how we can use it to analyze scenarios that benefit from having the combination of heterogeneous activity scheduling behavior combined with dynamic traffic simulation. The primary example that we study in the paper is the congestion pricing plan for NYC. We compared a baseline scenario against a Schema 1 ($9.18 in peak hours, based off RPA's plan) and a Schema 2 ($14 in peak hours). The advantages of using MATSim-NYC become clearer: we can distinguish the benefits for each individual in the population, as shown by the cumulative distribution of change in daily consumer surplus ($) for each person among the segment who has a trip start or end in Manhattan (we call "Charging-related") versus others.

Not only this, we can also quantify the changes in number of auto trips by time of day for those two population segments. Our analysis suggests there is a peak toll price over which the benefits to Manhattanites are outweighed by the costs to other travelers in the city.

There are even more applications. The project report details an application looking at the Brooklyn-Queens Connector (BQX), which we predict to have 112K daily ridership with 18K travelers shifting from auto mode:

https://c2smart.engineering.nyu.edu/wp-content/uploads/69A3551747124-Development-Tech-Transfer-Multi-Agent-Virtual-Testbed-.pdf

We have also used the tool to evaluate the Brooklyn bus network redesign from the Marron Institute and used the MATSim-NYC to design an alternative service frequency design which can increase the farebox recovery ratio from an existing percentage of ~ 0.22 up to 0.35 and which 74% of new trips would be drawn from auto mode:

https://c2smart.engineering.nyu.edu/optimizing-a-redesigned-bus-network-using-open-source-simulation

Most recently, we have applied the tool to evaluate the impacts of COVID-19 and reopening strategies, in which our analysis suggests micromobility would play a bigger role:

https://www.dhirubhai.net/pulse/using-multi-agent-simulation-understand-impacts-covid-19-joseph-chow/

The tool is now being used in collaboration with researchers at CTECH/Cornell to study the trade-offs between COVID exposure and environmental impacts under different transit reopening strategies.

The synthetic population and MATSim-NYC files are open to public agencies and researchers via Zenodo, our open data repository (with registration/request):

https://zenodo.org/record/7430184 (note the link was updated 12/12/22)

The Network of Living Labs will be integrated into our center's urban data observatory so that different scenario results will be compiled on there over time. We are now proceeding with further improvements to performance, better incorporating urban deliveries, and consideration of underserved populations.

Other researchers on this project include Professor Kaan Ozbay, lead researcher Brian Yueshuai He, Jinkai Zhou, Ziyi Ma, Ding Wang, Di Sha, Mina Lee, and many others (both graduate and undergraduate students) who helped along the way. Much thanks to Joon Park from NYCDOT who helped share data for this project, to Milos Balac for his help with learning MATSim, and to Yubin Shen for his help with the IT side.