What is the complexity involved in terms of data needs and model structure?
Incorporating intra-household interactions affects the model's ability to analyze behavior and policy in some cases where intra-household interactions are particularly important (pricing, shared ride, vehicle availability). Incorporating this capability is not a data issue (an activity-based travel diary is sufficient), but the model structure becomes more complicated.
How are intra-household interactions dealt with?
Intra-household interactions are now typical practice in ABMs although the actual implementation varies from model to model. There are several techniques and the peers recommend that NYMTC review models that are already implemented, such as methods the recently developed for advanced ABMs in San Diego, Phoenix, Chicago, and Los Angeles.
How reliable are the model outputs?
Daily travel generation can be validated against the (expanded) household survey, Census/American Community Survey data on journeys to work, traffic counts, and transit ridership data from MTA and NJT.
When developing a model, consider new trends related to activity patterns (e.g. telecommuting, flex-time).
NYMTC should borrow/adapt methods for population synthesis that have person-level and household-level control variables from the advanced ABMs developed in practice. See for example recent work on the PopGen program developed by Ram Pendyala at Arizona State (and implemented in Los Angeles, among other regions).
How much detail is necessary?
There are control and non-control variables in a population synthesis. The control variables are critical inputs, and the reliability of the synthesis depends on the quality and resolution of the control data. The non-control variables provide additional detail about individuals or households that is useful in segmenting models.
The panel advises NYMTC to do a good job describing the population (and activity centers) in the base year. The New York region is fairly stable overall, so the panel felt that NYMTC should not simply base data detail decisions on the need to forecast land-use growth. The design of the population synthesis relates to analysis needs, and it is important to know what travel market (or place) generates substantial travel and to be able to distinguish these markets/places in the NYBPM (which requires detailed segmentation).
What information from an establishment survey would benefit the development of the model?
An establishment survey should be designed so that NYMTC can construct better zone size measures (attractions) for use in destination choice. Well-done establishment surveys would allow NYMTC to perform some basic level of validation, and would collect some useful data on visitors and trucks. The panel feels that NYMTC should design a survey to identify characteristics that make certain places much more attractive, and to better understand and model associated generation, trip length distribution, activity duration and time-of-day considerations. NYMTC should keep in mind that agglomeration and uniqueness affect what types of people visit certain locations (see for example Fotheringham's work on Spatial Interaction models).
How to define and sample the establishments given the size, diversity and complexity of the region?
The panel recommends focusing on as few major types of establishments as possible (not more than 30 or so). Specifying this list must be done based on careful consideration of the travel characteristics associated with each type of establishment (activity generation, trip lengths to visit the establishment, portion of visitor travel, freight demand, and time-of-day patterns). The goal is to identify types of establishments that represent to a sufficient degree the breadth of activity centers in the region so that the findings from this survey can be generalized. Care should be taken to work this out iteratively with the consultant team responsible for that survey. Effective pre-testing will be important, since these surveys are new and the technology of data collection has not been yet established.
Is it important to consider which aspects of establishments can actually be forecast?
There needs to be recognition of the need for balance on both sides of this particular issue. Here again, the panel feels that the New York region is fairly mature, NYMTC should not simply base data detail decisions on the need to forecast land-use growth. In general, there should be a forecastable size variable developed for each establishment type.
Should the establishment survey be recast as a visitor survey?
Another approach would be to focus on hotels and get a daily pattern for visitors. This approach only covers people who stay in hotels, and the visitor diary data are very difficult to collect. The establishment focus is biased towards the particular types of establishments, but it is probably best to not think of the establishment survey as a visitor survey since a key recommendation is to focus on destination choice validation, which is distinct from representing visitor travel better. The establishment survey and the visitor survey are complementary and not duplicative.
There are many types of pricing, such as:
There are many additional complexities, as well:
The panel urges NYMTC not to confuse what a regional model can do with investment grade forecasting. Investment grade studies are highly focused efforts on a single corridor that represent a special issue that includes many additional aspects and specific modeling, data collection and post-modeling procedures. This is not to suggest that regional models are not useful in investment grade studies, but they are only one piece of a significant and focused data collection, forecasting, validation and post-processing effort.
How to incorporate complex pricing structures to reflect different payment methods, and to represent managed lanes?
Most of the things NYMTC would need to do in terms of representing pricing structures in the model have been done in other models. Parking and facility pricing, for example, have been handled a variety of different ways in models representing a spectrum of collection policies, with the possible exception of modeling non-trip cordon pricing policies. The panel recommends NYMTC consider a route type choice model to segment toll users from non-toll users and HOT/HOV lane users from general purpose lane users. This route choice model can be done as part of the assignment model, or within the mode choice model.
What mechanism can be implemented that allows users to easily update/modify tolls, fares, and other prices for policy and scenario analysis?
It is important to keep in mind the trade-off between what you CAN do and what is reasonable to do (need to carefully consider the need to explicitly represent all the variation in fare/toll schemes). It is possible to build a model to estimate which residents have toll passes, for example, but then it would be necessary to handle that detail in the assignment step. Rigorous validation of pricing models should be performed using the available data and traffic counts on the existing toll facilities.
What is the optimal choice set for the model given the users' needs? What strategies can be implemented to improve the transit model?
Is it feasible to have multiple calibrated choice sets in this model?
The panel feels that this is not a realistic option.
What procedure can be used to reflect bus operating delays on the highway network?
Link the transit network to the roadway network. Develop functions for bus level-of-service attributes related to the highway network congestion.
Are there examples in other places where New Starts forecasts come from ABM's, or examples like in the New York City region where there are four different models that could be used?
San Francisco is an analogous example; ABM's have been used for transit New Starts forecasting in San Francisco. ABM has also been used in Columbus for transit New Start forecasting and analysis. The case needs to be made that the tool used for New Starts analyses is useful for the need at hand (whatever tool that is). FTA does not approve models; they approve forecasts. However, they scrutinize model structure and parameters to ensure that they are reasonable and logical.
What procedures can be implemented to conform to FTA requirements?
An MPO model should be a solid foundation for New Starts forecasting. This requires accurate network coding, socioeconomic data, and aggregate travel patterns. Often times the biggest concern with transit forecasts is related to the underlying person trip/activity demand. Models are often not well calibrated for specific corridors, and users of the model should expect that many corridors have unique issues. A proactive approach to New Starts forecasting would involve collecting corridor specific data and making corridor specific improvements in the context of studies. In this respect, NYMTC should incorporate improvements incrementally. The panel feels strongly that NYMTC should not have an expectation the MPO can foresee all needs and get everything right within a single framework.
What recommendations are there to develop a non-motorized network that would support modeling of non-motorized trips?
This depends on whether NYMTC wants a model that is sensitive to non-motorized networks. If so, it is going to be expensive, time-consuming and complicated, but potentially quite useful. This will include special networks for pedestrians and bicycles.
How to improve the non-motorized mode and destination choice model?
The existing NYBPM has a basic capability and a special sub-model for non-motorized travel; however, this sub-model is greatly simplified. This is an important issue for non-motorized and transit mode choice, as well as destination choice.
What strategies can NYMTC implement to incorporate urban design factors such as walkability, bike friendliness, density, etc?
The panel recommends that NYMTC incorporate more explanatory variables (walkability, bike friendliness, density, etc) into the demand models; approaches depend on GIS data availability and quality.
What are the recommendations for a visitor model and what are the data needs?
A non-resident model could be approached incrementally, but you need some basic data. Basic data needs include relevant land-use data (hotel rooms, tourist attractions) and ideally a hotel survey distinguishing business/leisure travel. A non-resident model can be activity-based and with a synthesized (daily) population at the individual level, or aggregate.
How to forecast non-resident visitors? What are the key variables for visitor profiles?
The panel feels it is first necessary to get a handle on current visitation, and consider macro trends. Sometimes visitor/convention bureaus do surveys/forecasts, so NYMTC should research any readily available data on overall travel, trip purpose (business/leisure) and mode usage, for example. Macro trends can be linked to the national and regional economic and employment forecasts.
What are the recommendations for the development of an airport access model?
This is important to include since the airports are major generators and there have been and will continue to be major studies of ground access infrastructure associated with the airports. Remember that airport choice is also important since there are 3 major competing airports in the NY region. Thus, the choice structure should include two levels: airport choice and ground access mode choice (conditional upon the chosen airport). Data exists for this from a variety of recent studies by the Port Authority, and the Port Authority is considering investing in an airport choice and access mode choice model for its planning purposes that would interface with the NYBPM.
What are the pros and cons of a discrete choice model and a continuous choice model for our purposes?
The panel feels that it is essential to work towards a fine temporal resolution (at least approaching continuous), particularly on the demand side. A fine-grained temporal resolution is important to properly consider time-of-day pricing and dynamic pricing, along with representing the fact that both the road and transit systems are at capacity in the peak periods. Therefore, the goal is to ensure model sensitivity (whether via discrete or continuous methods) to congestion and pricing. Currently continuous time-of-day modeling is difficult to achieve, particularly on the supply-side, although the corresponding theory and numeric methods are improving rapidly.
It is common for ABM's to have finer temporal resolution in the activity schedules than in traffic assignment. The more you sub-divide the static assignment process by time-of-day, the trickier traffic assignment becomes due to large percentages of trips spanning time periods. The panel recommends that NYMTC consider a planning-level DTA option at perhaps a 1-hour resolution as an intermediate solution, but feels that a full DTA implementation is impractical given other needs.
As for the demand side, the panel recommends methods to model consistent daily schedules for individuals in a few recently developed advanced ABMs (e.g. San Diego, Phoenix, Chicago, and Los Angeles).
What recommendations are there to improve the commercial vehicle model?
Modeling techniques and data availability vary by truck type (freight, local distribution, commercial vehicles, service). Long-haul freight is best modeled with an economic, commodity flow framework. The Calgary tour-based microsimulation local freight model offers a potential framework for local distribution.
What data can be obtained?
What freight data can be obtained in the establishment survey?
Basic trip rates for the surveyed establishments, by type of truck. An establishment survey will only provide some basic validation data. Keep in mind that freight/commercial vehicles may access surveyed sites in a different location from employees/visitors.
Is Dynamic Traffic Assignment a feasible option given the size of the model? How about simulation of transit vehicles in Dynamic Traffic Assignment?
The panel recommends consideration of a planning-level DTA, if available and feasible. Caliper recently released a planning-level DTA, which is a 30-60 minute static assignment that is linked across time increments so that trips "bridge" time periods. This is distinct from a regional meso or micro-simulation model (e.g. TransModeler). That type of DTA is currently infeasible at the regional scale. Transit vehicles (buses) can be included in traffic simulations but only some of the DTA packages provide this option.
What is a feasible strategy for the regional model to interface with mesoscopic and simulation models?
The NYMTC model already has a sub-area extraction capability. Some other comparable models, such as the Atlanta Regional Commission model, do not have a sub-area extraction capability. Model run-time is not prohibitive, so the panel advises NYMTC not to over-think the need to develop an intricate sub-area extraction capability and recognize it may be easier to just run the whole model. The panel recommends that NYMTC consider opportunities to bring corridor details into the regional model if appropriate from a regional context.
How to develop a guideline to facilitate the application of the regional model for different types of studies?
The panel advises against overly prescriptive guidelines for forecasting work, since there is truly not a "one size fits all" approach to forecasting work. Instead, consider ways to maintain documentation in "real-time" (e.g. wiki). Ideally, any application work in a sub-area or corridor would be informed by useful data that NYMTC could leverage for its own use. Ideally any forecasts would tell a clear story and be grounded in data regardless of the technique used to arrive at that forecast.
What about transit considerations in a sub-area context? The current model does not allow transit analysis after sub-area extraction.
The way NYBPM's sub-area extraction process works, you have to run the entire model first (so all modes are taken into consideration on a regional basis). Demand can be extracted covering all modes, but the assignment process is particularly complicated on the transit side (requiring ultimate origin and destination data for accurate path-finding, both of which will not often exist within the sub-area). Other MPOs do not have sub-area transit extraction capabilities, but in the San Francisco Bay Area MTC region, there are 2 counties that have their own models with transit capabilities.
What strategy can be implemented to achieve flexible TAZ structure?
Hierarchical geographic structures should be explored as a solution to flexible TAZs and improved geographical resolution. Simulated demand can be parcel-based, but supply is often TAZ-based (these geographies should nest). Transit and walk accessibility can be modeled at a finer resolution that nests within aggregate TAZs (e.g. Sacramento, San Diego, Chicago is in process).
What are the complexities involved?
With hierarchical structures, land-use data needs to be allocated to the lowest spatial resolution units (MGRAs or parcels). Special procedures have to be developed and applied to model transit access/egress and non-motorized modes.
Manhattan, for example, has much more detail than other parts of the model, but still the calibration is "not very good". What can be done?
It is relatively easy to look at the model output in intelligent ways. The challenge in making a better model is how effective your data are for the questions at hand, and identifying why the model output is unreasonable. Let the data and intelligent comparisons guide necessary model improvements. Start at the macro level and get regional flows right, then work down into more detailed contexts. Complexity of a model should not be confused with accuracy. Do not use crude techniques such as k-factors simply to match observed data.
What are the advantages and disadvantages of cloud computing (feasibility, the platform and other relevant issues)?
All large agencies with ABMs have invested in a server environment as opposed to the cloud; this server environment can cost up to $100K. Cloud computing technology is changing rapidly, so explore the feasibility of the cloud, and talk to vendors about how this product works (as members of the panel have in the past). Keep in mind that modeling servers can be used for many purposes (not just running models).
There are numerous model users at NYMTC and in the consulting/stakeholder community, and these entities have already spent "lots" of money on hardware/software.
It is acknowledged that it is difficult to install new versions of the NYBPM (hardware/software compatibility issues).
While the hardware costs seem big, the panel has found they are small considering the operating costs, which are the basis for cloud computing costs. Keep in mind a model has to be run many times for a given application need. The cloud may be the best way to make the model available more broadly, however. The panel felt that from the point of view of an MPO, the first priority is to take care of the MPO, and the MPO is best served by a server and not the cloud. One possibility to explore in lieu of a cloud might be a cost-sharing arrangement where NYMTC staff would apply models for other agencies. Additionally, setting up the NYBPM in a cloud environment is bound to a special agreement with Caliper to provide a cloud TransCAD license.
What can be done to reduce the perception that the model is a black box?
NYMTC's source code is available; it makes sense to first read the model documentation thoroughly. The models are incredibly complicated, which admittedly affects transparency.
How to address sustainability within the context of the regional model?
The panel feels that NYMTC must define sustainability first, but then travel models estimate an extensive amount of data that can be recast as aspects of sustainability (emissions, multi-modal mobility and accessibility, VMT, etc.)
How to incorporate fuel price?
The NYBPM already considers auto operating cost (fixed coefficient per mile by vehicle type). Real-world dynamics are more complicated however. For example, do transit fares rise due to cost of operating buses? How does vehicle type choice change? There are different choices that can be made in the short-, mid- and long-term in response to gas prices. The panel recommends that as gas prices change significantly, NYMTC should monitor observed elasticities. There may be some opportunity to analyze current travel survey data, given recent large fluctuations in gas prices. Comparisons between the Household Travel Surveys in 1996 and 2010 could be also useful.
How important is a land-use model for NYMTC?
So many aspects of BPM 2.0 require detailed land-use data, and a land-use model is one way to get at that, but it is probably not feasible in the timeframe (2017). Generally speaking, NYBPM requires an intelligent way to inform Population Synthesis and populate small zones (MGRAs).
Induced demand is another aspect to consider with respect to land-use modeling. There is near unanimous agreement that an integrated model is the right theoretical approach for a variety of reasons (two-way interactions), but not an immediate need - the rates of land-use, population, and employment change in the New York metropolitan area are relatively modest in the aggregate.
NYMTC stakeholders have serious concerns about the need to forecast land-use detail. The panel feels that this effort should be more systematic, but not necessarily more complicated than what NYMTC currently does without a model. The database you need for a land-use model, or detailed demand modeling, is very valuable independent of land-use forecasting.
Some recommendations from the panel might lead to simplifications (e.g. transit network representation, and overlaid networks). The panel feels strongly that theoretical elegance (e.g. more temporal resolution in assignment) results in a model that is easier to explain in some instances. There is an ongoing need to make the model easier to install and quicker to run (acknowledging that a lot has changed in 10 years in this regard with respect to general software and hardware).
Finally, complexity is neither inherently good nor inherently bad. Generally speaking, NYMTC should strive for "meaningful" complexity, rather than "arbitrary" or "excessive" complexity.
