Integration of Different Mobility Behaviors and Intermodal Trips in MATSim

Round 1

Reviewer 1 Report

This paper presents an extension to the MATSIM simulator in order to take into account intermodal trips and mode choice decision-making. It is applied on Wien city.

Even if the state of the art of existing approaches is presented and the choices of algorithms are well discussed with relation to the literature, the paper is missing a clear lack in existing works. In particular, there is no reference to previous works that provide bad or not as good as possible results due to a lack that the proposed paper will tackle. 
More justification should thus be provided about the need of what is proposed in this article. 
In addition, this is only at the end of the article that possible (and dedicated?) applications are presented. The justifications could be stronger if these applications would have been presented earlier in the paper. 

The paper does not take into account shared bikes (or cars). A discussion on that point should be introduced, as it removes so hypotheses that are made in the paper (in particular that a trip during the day cannot be done by car if the agent did not take its car in the morning).

The authors have conducted experiments to evaluate, calibrate, and validate their model. Nevertheless, the results are only provided for the article model. It would be interesting to compare it with other existing approaches to identify whether it improves the quality of results or to identify where it improves the results (or decreases their quality).

Finally, the results could include data about execution time, in order to evaluate the increase introduced by this additional component.

Author Response

Please find our answers to your comments in the document attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear Authors,

Thank you for the paper. First, I want to give you some of the major comments:

• You present a description of MATSim model set-up for the Vienna region. This is interesting in itself, since my assumption is that an agent-based model for that region does not currently exist.
• Beyond this you state that you are introducing a new intermodal router called Ariadne. It is not clear if this router was already presented in a previous publication or not. If it was not, then its description in this paper is shallow and it is not clear why this approach is any better than the current SwissRailRaptor router used in MATSim. Later in the paper you provide a reference to some previous work, but it is still not clear why developing this new router brings anything to the MATSim or an agent-based community, considering that the framework is closed source. You claim that Ariadne can combine multiple modes with public transport, even beyond first and last leg, but this is nowhere used in this paper, so that feature is irrelevant. Furthermore, B+R or P+R are nowhere mentioned in the results.
• Potential use cases presented at the end of the article are without any specific implementation useless to a reader, as we are all aware by now what kind of studies one can do with an agent-based model. You should either remove that section or do a complete case study that would fit the scope of a journal.
• The description of the complete model is not detailed enough, we miss a lot of information on how the synthetic population was generated, based on which data and methods. This makes it unclear for which studies the final model can be used.
• When you calibrate a transport simulation you should be aware that calibration should involve two of the speed, volume and density. This means that calibrating only on volume does not mean that your system is actually calibrated properly. You need two of these dimensions to calibrate a transport model.

Other comments:

• The main tool used in this study – MATSim is incorrectly referenced
• SUMO also has a proper publication; please do not use a website
• Page 1, line 19, ‘today’s’ and ‘nowadays’ is redundant, remove one.
• Page 1. Line 20: what does “more individual” mean? It is either individual or it is not
• Page 1, line 22, you say that “tours instead of trips allowed a specification of mode choices”. Why is that the case? One can also specify mode-choices on a trip level.
• Page 1, line 23, mentioning activity-based models is out of place and not connected to the story. You also say “these:, which are those models?
• Page 1, you mention one difference between MATSim and TRANSMIS. Is that the only difference? Why compare it to only TRANSIMS, there are other agent-based models as well – i.e., SimMobility, POLARIS
• Reference 9 is informal; I am sure there are proper publications to cite different MATSim models.
• Reference 15 on Page 2 has a proper publication as far as I know. It should be this one:

Hörl, S., & Balac, M. (2021). Synthetic population and travel demand for Paris and Île-de-France based on open and publicly available data. Transportation Research Part C: Emerging Technologies, 130, 103291.

• Page 4, lines 111 and 112, the sentence misses some words
• Page 4, you say you use 100 iterations. Is this enough for convergence? What is your measure of convergence?
• Page 4, lines 124-126, are you saying that in MATSim agents have to use the same mode throughout the day? I am not sure that is the case
• If you cache all routing in Ariadne, how do you deal with changing travel or waiting times due to evolution patterns in MATSim?
• Page 6, lines 213-217, it is not clear to me what is happening here
• Page 6, how do you adjust activity times in those specific cases? And how often this happens?
• Page 7, line 253, did the survey cover whole Austria or part?
• Line 258, the definition of VoL is not clear to me
• Page 8, you report a fixed cost for car. How is this used in the utility functions of MATSim?
• I assume that Ariadne does mode-choice together with routing. Can you explain to me how does this work together with MATSim later scoring? Are the same parameters used in routing as in the scoring?
• You use in some places MATSim wordings and parameter names, which are for general reader confusing.
• Page 9, lines 312-314, why is that parameter defined as the sum of the other two?
• Page 9, line 328, what do you mean by ours 24-30? Is that the simulation of the following day? Can you not just add those to the traffic of the morning?
• I am not sure that I agree that your model fits well with the observed data, as sometimes the differences are substantially large (i.e., mean trip distance for bike is 30% higher)
• You also mention that your model only considers individuals above 17 and does not consider external traffic. Why don’t you then only compare the data for those individuals. Currently I have a feeling like you are comparing apples and oranges.
• It would be good to show mode share in distance bins, to see if the model is able to react properly based on distance.
• Figure 7 shows also differences probably due to the fact that you are not ensuring that speeds on the roads are correct
• It would be nice to tell us how does the national travel survey looks, what data is collected, how many observations, etc.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

You paper is a really good written and easy to read.

In the Introduction, try to found the better match to connect the travel modeling and micro-simulation tools. Now it looks like you are jumping from describing travel modeling to micro-simulation. Do the same with TRANSIMS and MATSim.

You mentioned several papers doing the modeling using MATSim. What are the differences between models in these papers, and the one you are proposing in your paper? What is the novelty of your paper in accordance with present research in this field?

Please describe the Ariadne software more briefly and its influence on intermodal trips.

In discussion it would be great if you can comprese your ideas how to improve intermodal trips in and outside of Vienna. The same can be applied in other similar cities.

Author Response

The response to your comments is written in the document attached.

Author Response File: Author Response.pdf

Reviewer 4 Report

Thank you for the opportunity to review this interesting paper. The improvement on intermodal trips in MATSim provided by this paper is beneficial and novel for the reproduction of urban traffic and the evaluation of transportation policies.

If the following revise is made, it will be easier for the reader to understand.

line 166:

- What kind of replanning strategy is SelectExpBeta in MATSim?

- The selection probabilities of AriadneReRoute, ReRoute and SelectExpBeta are set to 0.1, 0.1 and 0.8, respectively. What references or assumptions were these based on?

Author Response

Please find my response in the file attached.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Dear Authors,

Thank you for answering my previous comments. In some parts, the paper is improved by correcting some of the inconsistencies and providing additional information. The introduction reads much better, and the description of the generation of the synthetic population is clearer.

There are, however, still some fundamental problems with the approach and the results:

• The most important is your comment that you also have the densities by having access to hourly flows and lanes and capacities information from OSM. First, OSM does not provide any capacity information, only the number of lanes and length of the roads. Second, your understanding of traffic flow theory is questionable. Let me give you an example that will help you understand why speed or density together with the flow is necessary. Imagine a situation where you measure a flow of 300 vehicles/hour off-peak on a road segment with a saturation flow of 2000 vehicles/hour. Now imagine measuring the exact flow during peak period (i.e., the road is heavily congested) of 300 vehicles/hour. It is evident that densities and speeds at these two situations are different. This is a fundamental problem of your calibration approach, as you are only calibrating against the counts (i.e., flows) and neither of the speeds or densities. See for instance this work, which used both flows and speeds:

He, B. Y., Zhou, J., Ma, Z., Wang, D., Sha, D., Lee, M., ... & Ozbay, K. (2021). A validated multi-agent simulation testbed to evaluate congestion pricing policies on population segments by time-of-day in New York City. Transport Policy, 101, 145-161.

 

• The second issue is the usability of the model for forecasting that comes from the way it was calibrated and validated. It is not clear how the calibration was performed. First, you say that you calibrate the model using count data, but you do not mention which parameters were calibrated in this case. Second, on line 371, you mention that you calibrated the constants to match the modal split. This is inconsistent or lacks further explanation. Furthermore, you are using a disaggregated agent-based model, and then the only value you use to calibrate your model is the aggregated modal share. This disregards every possible advantage an agent-based model has. The modal share in distance bins figure shows us that your model is inaccurate and requires substantially more work to allow its academic or practical use.

Some minor points:

• I would highly suggest that you give the paper to a native English speaker for proofreading. There are both grammatical and structural errors, especially in the lack of use of commas.
• It would be helpful to show the original and after the calibration values of the constants.
• In the introduction, line 52, you mention that Horl and Balac “attempt” I am sure that the authors would disagree with the term used. My understanding is that they succeeded in their efforts. Moreover, you state that two datasets they used are not publicly accessible, which is not valid, as according to the paper, all datasets used are available as open- and publicly accessible data (as also stated in the title)
• I would advise comparing like with like when plotting the comparison between Simulation and Report. If you say that some trips in the Report are not represented in the Simulation, then remove those when making comparisons, and state what was filtered out and reason.
• Please also show average car speeds for different distance bins.
• The MATSim reference is still not correctly formatted.
• The title is misleading as intermodality is not present in the paper (even though the Ariadne framework can construct intermodal trips). The paper is more about setting up an agent-based model of Vienna than intermodality. In addition, integration of different mobility behaviors is a feature of MATSim, and you are only using this feature. I would advise finding a title better fitting the contents of the paper.

 

My advice to you is to improve your work with the above comments. I also advise you to be more critical of your work and point out clear limitations that will inform potential users of the model whether it is suitable for serious transport planning. The above comments can guide you in rewriting your limitation paragraph appropriately. 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf