U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
| SUMMARY REPORT |
| This summary report is an archived publication and may contain dated technical, contact, and link information |
 |
| Publication Number: FHWA-HRT-14-054 Date: January 2015 |
Publication Number: FHWA-HRT-14-054 Date: January 2015 |
David Kuehn
Program Manager, Exploratory Advanced Research Program
Federal Highway Administration
Ed Strocko
Team Leader, Freight Analysis and Research Team
Federal Highway Administration
David Kuehn, Program Manager for the FHWA's EAR Program, opened the workshop by introducing the goals of the workshop and outlining the EAR Program's interest in advancing the state of freight modeling. Kuehn explained that scientific and engineering advances in sensor and probe data, data mining, machine learning, and other computational approaches have the potential to respond to transportation questions about freight data and model improvements to improve investment in and operation of the Nation's highways. In accordance, FHWA would like to identify a set of topics that could be incorporated into future EAR Program solicitations and ultimately inform the development of the FAF4, scheduled for release in late 2015, and beyond.
Ed Strocko, Team Leader for the Freight Analysis and Research Team at FHWA's Office of Freight Management and Operations, asked participants to help scope the problems and shortcomings surrounding past and current versions of the FAF and identify areas of possible research to improve future iterations of the FAF. FHWA organized the workshop to identify areas of improvement for future versions of the FAF, to determine the level of information required to perform multimodal assessments, and to identify whether the new FAF needs a fully routable network and predictive capabilities.
Opening Remarks
Bruce Lambert
Executive Director, Institute for Trade and Transportation Studies
Rolf Schmitt, Ph.D.
Deputy Director, Bureau of Transportation Statistics
Research and Innovative Technology Administration
Bruce Lambert, now Executive Director at the Institute for Trade and Transportation Studies, was a former manager of the FAF study at FHWA. Lambert stated that freight is an essential piece of the national economy and must be considered in transportation systems analysis. The FAF was originally created to inform Federal-level users, but over time diverse groups (e.g., planners and shipping companies) began to use the FAF. Lambert explained that users of the FAF today have a range of different requirements. As a result, he suggested that the major question FHWA needs to investigate is, "How do we make the FAF relevant to address users' –both internal and external to the U.S. Department of Transportation— many different requirements?"
Rolf Schmitt, Deputy Director of the Bureau of Transportation Statistics (formerly with the Office of Freight Management and Operations), added that the FAF went from a "what if?" tool to a "what is?" tool, meaning that it is more of a reference tool than a predictive tool. Schmitt noted the FAF needs a complete O–D matrix, based on all obtainable freight-related information, to get the clearest picture of O–D patterns possible. This O–D matrix is then converted into flows on a network, which can be used to answer forecasting "what if?" questions.
Schmitt noted that there are significant problems with data collection and aggregation on a national scale, which limit the level of detail and accuracy in the resulting O–D matrices. These problems limit the ability to disaggregate FAF flows to the county level, a problem that is counterintuitive when considering the FAF is created by the aggregate data collected at these levels.
In summary, Schmitt stated that FAF accuracy should be assessed and compared against a benchmark (and asked what that benchmark should be) to gauge how accurate the FAF needs to be. Schmitt also noted a recent National Cooperative Freight Research Program (NCFRP) report describing collection and integration of local data on the aggregate data that FAF can provide. This report, NCFRP Report 25: Freight Data Sharing Guidebook, can be viewed at http://onlinepubs.trb.org/onlinepubs/ncfrp/ncfrp_rpt_025.pdf.
Peter Bang, Ph.D.
Freight Analysis Framework Program and Data Manager Federal Highway Administration
Peter Bang, FAF Program and Data Manager and member of the FHWA Office of Freight Management and Operations' Freight Analysis and Research Team, presented background information on the FAF mission and questions to be answered. According to Bang, the FAF today provides a big picture of goods movements and truck transportation throughout the United States. The FAF supports national freight policy decisionmaking by establishing known O–D flows across the country; however, there are limitations with this tool, including:
Incorporating current data. The current FAF lacks the ability to incorporate more "real-time" data and information.
Providing insight. There are no predictive elements to address "what if ?" scenarios for national policy decision support. This limits the FAF's ability to provide multimodal, multiclass solutions and to provide insight in terms of causal freight relationships and future estimations.
Understanding variables. It is difficult to use the FAF to understand variables such as the effect of fluctuating gas prices, the impact of an energy policy change, the impact of a pilot or truckers' strike, seasonal fluctuations (e.g., harvest and weather), and the impact of a port shutdown. These variables tend to be multimodal, have cause and effect relationships, and can be local or national.
Ongoing maintenance. The FAF experiences a complete overhaul every 5 years to address data consistency and reproducibility issues. The maintenance and update procedure needs to be modular and requires a level of systemization and openness.
Addressing regional and local needs. The FAF cannot be used to address concerns at a level of granularity more detailed than annual statewide flows and therefore cannot be used to address growing regional and local needs.
According to Bang, there are many ways in which the FAF can be improved. For example, it could be useful to enhance the FAF to provide users with some level of understanding regarding freight movements at the regional level to inform local freight studies and projects. Creating and maintaining this level of granularity might require a transparent model, an open model and data sharing, and public participation.
Bang stated that the FAF mission for tomorrow is to provide an analysis tool that can capture details within the big picture of national goods movements. The current FAF structure is created using a survey-based, hybrid OD construction process. Bang proposed a future alternative structure for the FAF that is based on a multimodal, travel– demand model.
Bang went on to identify three categories of questions that needed to be addressed during the workshop, as follows:
O-D Data Generation. What is the future for the FAF structure? Will the FAF become a set of comprehensive travel-demand modeling tools? What kind of data need to be collected and in what format, from what, and for whom?
National Multimodal Assignment. What are the traffic assignment and mode choice techniques? What about FAF adaptation and localization?
Multimodal Routable Network. Should the model be nationwide, multimodal, or multiclass? Should it include a routable network for everyone?
Following the presentations, participants provided feedback in an open discussion. The following is a summary of the key points made:
Looking toward the future, FHWA should be mindful of the need to capture commodity flows. To address this need, commodity flow surveys (CFSs) could produce a micro-flow file based on shipment flow records (rather than the status quo of region-to-region tables by attributes).
FAF version 3 (FAF3) and FAF version 2 have been used as a data source for a number of analyses; however, they never generated a complete, free-standing freight-demand model. They could never provide the complete picture and always required additional inputs to perform meaningful analyses. For example, FAF users never had the ability to see how widening the Panama Canal would alter freight flows across the Nation.
Modeling comes down to calibration and validation. A model is only as good as the dataset from which it is constructed; therefore, any model created is only as accurate as the data collected. A supply- chain model would be a huge step forward in making use of data that is currently available. Releasing this model to the public could also give FHWA free peer data. The ability to provide a forecast based on data that are publicly available would be a significant improvement.
Opening Remarks
Ho-Ling Hwang, Ph.D.
Center for Transportation Analysis
Oak Ridge National Laboratory
Ho-Ling Hwang, of the Center for Transportation Analysis at Oak Ridge National Laboratory (ORNL), presented a high-level approach and considerations regarding the generation of O-D data in FAF3 and associated challenges. Hwang explained that FAF3 O-D data, specifically the estimates of base year tonnage and value for domestic shipments, are built on data from the 2007 CFS. A large number of CFS O-D cells are suppressed, due to disclosure issues or concerns related to publication standards, so it is necessary to use a modeling approach to fill gaps and missing information when generating the FAF3 O-D matrix for domestic movements. In addition to the CFS-based domestic shipments, Hwang explained that the FAF also integrates other data sources for many out-of-scope CFS components (e.g., farm-based agricultural shipments and imports). Hwang noted that, to maintain transparency for the FAF O-D data, these other data sources are mostly public data. These data are incorporated into models to accomplish mode share and geographic assignment needs.
Hwang explained that base year ton–mile data are estimated by using modeling approaches that disaggregate tonnage from FAF zones to counties. These disaggregated county-level tonnages are then multiplied by route distance. This is estimated by mode on the multimodal network systems to generate ton-mile estimates. Base year O-D data, including tonnage, value, and ton-mile by commodity by mode, are used for estimation of forecasts and truck network assignment.
Hwang highlighted that there are major challenges regarding the generation of FAF O-D flows. For example, there is currently insufficient information regarding domestic movements of foreign goods (i.e., imports and exports), specifically in terms of mode choice, mode-sharing, and the geographic detail of U.S. origin (exports) and U.S. destination (imports). In addition, Hwang noted that a level of detail and validity is a concern when FAF users try to look beyond FAF zones and attempt to "cut and slice" the FAF for local applications. A challenge for FAF users is how to integrate locally collected information, either with FAF or to supplement FAF. An additional challenge is how to seamlessly stitch together data from various sources. Hwang stated that the approach of simply combining data could create numerous problems, because local datasets are collected by local agencies and may include varying levels of accuracy, levels of detail' and specific definitions of data. Other data needs featured during Hwang's presentation included the distribution of shipment distance by commodity carried (e.g., for out-of-scope shipments).
Following the presentation, the presenters asked participants to provide their feedback in an open discussion. The group discussed problems and potential solutions, which are summarized below.
Problem A: Privacy and intellectual property concerns create data restrictions for data included in a publicly released model. These data restrictions limit the capabilities of a model.
Participants discussed several specific concerns within this problem area, as follows:
Confidentiality constrictions. Privacy concerns may require that publicly available datasets are reduced to protect privacy or to avoid intellectual property issues.
Restricting data. Participants noted that the agencies that currently collect data could provide insight regarding what datasets can be released to the public (based on the governing rules laid out when the data were collected). A potential first step could be to look at agency regulations to see what data can be released. Questions should include, Is the data internal? Can it be released? As an example, participants highlighted that business income tax data with restrictive regulations might provide some insight and be applicable to a host of highway and railway datasets.
Protecting trade secrets. Karen McClure indicated that these issues also exist within the Federal Railroad Administration (FRA). Details relating to the business operations of a particular shipper cannot be released to the public (to protect their business information). Without this information, however, the remaining dataset has limited capabilities. To account for these privacy and intellectual property issues while maximizing the usefulness of the remaining dataset, McClure explained that the FRA developed some tools that make some of this processed information available to the States. These tools are designed to help States identify the major points for each State without revealing business-specific information.
Disaggregating data. The FAF ultimately needs to be able to disaggregate data to the county level, maintain a quality and level where it would be useful, and remain mindful of privacy and intellectual property concerns.
Participants discussed the following solutions to address privacy and intellectual property concerns.
Solution A1: Create and distribute different datasets or different models for different user groups.
Providing different models, with different datasets, for different user groups could be an optimal solution to the privacy problem, thereby limiting the distribution of sensitive information. During discussion, participants suggested that there could be two sets of data: one for Federal internal use and one that would consist of input data for different external user groups. To make this work, an internal Federal freight group could be established; however, a meta-data framework and a mechanism to perform meta-data review would be needed.
Models and input data ultimately could be created separately and distributed separately, catering to the specific needs (e.g., level of detail, datasets included, and model capabilities) and addressing the potential privacy concerns of each FAF user group.
Solution A2: Synthetic Data.
By using a synthetic dataset, participants confirmed that it might be feasible to capture the required level of detail at an appropriate level of accuracy in a publicly released model, without violating privacy and intellectual property policy issues.
One of the biggest concerns regarding synthetic data relates to whether data appear realistic and what level of accuracy a synthetic dataset needs to have to be considered useful. Participants also questioned at what level are the data wrong. During discussion, it was noted by participants that there are ways to generate data synthetically. For example, synthetic data are currently being used in California; however, the most difficult aspect is validating and calibrating such data. The validity of such a dataset is affected directly by the quality of the data from which the synthetic dataset is constructed.
Participants commented that synthetic data would need to be validated against local data and local understanding of freight activity.
Problem B: Quality and type of data collected and data available are inadequate to serve current needs.
Participants discussed several specific concerns within this problem area, as follows:
Improving available data. One of the recurring themes throughout this workshop was that the quality, type, availability, and granularity of the data collected across the Nation are inadequate to serve the Government needs. Workshop participants stated that a higher quality national-level model cannot be developed without first improving available data. Areas for improvement include:
Improving accuracy. Some local, regional, and State agencies adjust the data included in the FAF; however, these data adjustments remain at the local level and often are not passed back up to the Federal level. As a result, the base model remains inaccurate. Additional information is available, but there are no frameworks or incentives currently available to feed this information back up the chain. As a result, an opportunity is lost to make the model more accurate.
Improving decisionmaking. In some instances, there is already State-level aggregation of data to help inform local agencies, short-line railroads, and others to improve coordination and decisionmaking on a local level. Releasing information could create data misuse or abuse concerns (e.g., companies who misuse the data to gain an unfair advantage from a competitor); however, the participants made the point that data released at any level could raise these concerns to one degree or another, and that the threat of misuse should not impede the Federal effort in creating a more efficient model to improve the freight network. Participants ultimately noted that the optimal solution would involve communicating this information to the public without putting any agency, company, or business at risk. In addition, the goal is to optimize the system by improving decisionmaking abilities.
Incorporating real-time updates. Participants highlighted that there is no mechanism in place to facilitate the timely reporting of data. If data could be updated in the model more frequently, then the model could prove more useful.
Other data-related problems and concerns that participants voiced during this session included:
Incorporating geographical data. The lack of these data in the current version of the FAF means that users are not able to identify areas of high activity and what percentage of that freight activity is inbound or outbound.
Incorporating O-D data. Another problem highlighted by the group is the overall lack of sufficient O-D data for public use.
Participants discussed the following short- and long-term solutions to address data collection and quality concerns.
Solution B1: Short-term research.
Develop a tool in which users would be responsible for approaching agencies and obtaining the data directly (this would remove restrictions based on data-sharing and privacy).
Establish an agreement or terms of use to restrict misuse of data. This could be accomplished by implementing a fine or other penalty designed to deter misuse of data and misuse of the system.
Emulate California's aggregated data and create a database to distribute it statewide. The Online California Freight Data Repository can be viewed at http://freight.its.uci.edu/calfred/.
Solution B2: Long-term research.
Develop new techniques for disaggregation for modeling purposes. Participants noted that this was key to addressing the problem.
Develop a methodology or system in which peers submit data to allow agencies to share data at a level at which they feel comfortable.
Capture levels of activity in an area and make adjustments to the network based on localized freight volumes.
Opening Remarks
Kenneth ("Ned") Mitchell, Ph.D.
U.S. Army Corps of Engineers
Karen McClure
Federal Railroad Administration
Ned Mitchell of the U.S. Army Corps of Engineers (USACE) shared his experiences regarding freight and port activity. To identify and monitor port activity (for port dredging and maintenance), Mitchell worked with Doug McDonald and the USACE Navigation Data Center. The Navigation Data Center details waterway data, allowing users to look at the system as a whole.
To answer questions regarding multimodal travel, Mitchell has worked with Ho-Ling Hwang at ORNL, where they were able to aggregate domestic flows and other modes. Mitchell explained that this aggregate dataset will give USACE insight into the system function as a whole and will help determine the impact that U.S. waterways play on freight. He noted that this aggregate dataset will allow USACE to maintain and create waterway infrastructure to suit the needs and demands of the region. Mitchell went on to highlight that to assess multimodal freight movement accurately, it is vital to maintain detailed waterway data and to reconcile this information with land-based datasets, making connections into a workable analysis.
Karen McClure of FRA talked about an existing O-D matrix maintained by FRA. McClure explained that the matrix is accurate, contains commodity O-Ds, and provides insight regarding the flow of these commodities along a rail network. This matrix gives the most accurate look along well-established and heavily traveled corridors.
Current issues noted by McClure involve assigning an O-D matrix to a rail network. Although it is known that commodity flows may occur along the links of a well-traveled corridor or major corridor, existing datasets only include reported car volumes and tons of freight on an annual level (total values per year). There is no monthly, seasonal, daily, or hourly breakdown of these data. McClure noted that seasonal fluxes in commodity flows are not captured, and therefore fluxes in demands cannot be considered when attempting to alleviate highway congestion caused by freight movements. For logistics planners to effectively divert fluxes in truck volume, it is important for national freight data to identify modal supply costs.
In summary, McClure stated that to plan and facilitate multimodal freight shipments using the FAF, the next generation should provide some insight regarding the size of these multimodal nodes and facilities. This information should include the number of tracks and cars that can be used.
McClure also noted that including only yard sizes in the next-generation FAF would not provide sufficient insight for planning multimodal travel; however, yard sizes would be ideal for classification purposes. At a bare minimum, the location and size of intermodal yards and any highway access to a yard (not just rail access to a yard) would be required to use the FAF for any level of planning. From a modeling standpoint, McClure claimed that this could begin to answer some of the high- priority questions, particularly about how to link the rail network to the highway network.
McClure informed participants that intermodal terminal locations and the rail network are available on the ORNL Web site in various formats. In addition, an intermodal facility database exists (including highway- to-rail, water-to-rail, and air-to-rail); however, McClure noted that this database needs to be updated and that the current update procedure is to use the Official Railway Guide and to convert text into updates.
Open Discussion
Following the presentation, presenters asked participants to provide their feedback in an open discussion and to highlight several problems. The following is a summary of these problems.
Problem A: Multimodal, trans-modal, intermodal, and mode-split analysis need to be assessed and addressed in the next generation of FAF.
Introducing multimodal, trans-modal, inter-modal, and mode-split analysis capabilities into the FAF requires the incorporation of additional datasets. These will include commodity-specific variables, the impacts of other modal users (and potential impacts created by a mode shift of these modal users), a detailed financial breakdown of each system and mode (the cost of each system), and the timeframe for each of these modes.
Workshop participants highlighted that the Government cannot model these impacts without first understanding them; however, the Government cannot fully understand them without thorough data collection and aggregation. Participants also noted that State-by-State variances in some reported data creates fluxes in data quality; thus, the Government actually cannot get a clear picture of the status quo by using currently available datasets.
An attendee inquired whether the FRA had publicly available information regarding rail yards in terms of flows, size, and ability to serve as a multimodal node. McClure identified a quarterly publication that provides insight into which rail yards were built to service intermodal operations, including detailed rail yard information. The publication, Official Railway Guide (https://ubm-sub.halldata.com/site/UBM000455LUnew/init.do), could theoretically be turned into a database, although McClure indicated that this would take an incredible effort. McClure noted there is a rail yard database compiled for security purposes; however, this database is not integrated with other modes.
Problem B: To model multimodal, trans- modal, intermodal, and mode-split shipments, a better understanding of modal operations across the country needs to be established. To develop the required understanding, a higher quality dataset needs to be collected from regions across the country.
To model multimodal, trans-modal, and intermodal shipments, or the mode-split choices of different shipments accurately, the variables that impact modal decisionmaking in freight must first be known and understood. To develop a working understanding, a national, uniform, detailed dataset must be collected from across the country. These data should include and provide insight into the following items:
Commodity flows along the network and commodity-specific variables including:
Seasonal, monthly, weekly, and daily.
This includes items that can or cannot be transported together and transport container requirements.
Any seasonal variations in the time- sensitive nature regarding the transport of these commodities.
This includes perishable goods and shipments that must be delivered to their destination prior to a major deadline (e.g., "Black Friday").
Seasonal and time-specific variations of the flows along each link, including:
Cost and variations within the cost, including:
Labor costs and shipping costs for each mode.
Time-based variations in the cost (e.g., time of day, day of week, month of year, season).
Establishment of a freight-specific economic model that could grant the user forecasting capabilities.
Granularity of this information is vital to establish a true working model.
Timeframes for transport along the links of each mode, including:
Turnover times and total delivery time.
This would include any "dwell time" or time required to transfer goods to another mode and any delays that might be incurred along the trip.
Includes the location, capacity, and abilities of multimodal transfer nodes throughout the country.
Understanding how these different elements interact, including:
Participants noted that with modal decisions, the forecasting abilities of the model, or "what if?" scenario analysis capabilities of this model, hinge on accurately capturing all of the variables that impact mode split and their interdependencies.
Problem C: The model should grant insight into the short- and long-term issues.
Workshop participants noted that, at the micro-level, the model should bring in operational models that railroads themselves use to simulate movement (e.g., a FRA simulator that integrates multiple variables; the Uniform Rail Costing Model). These models, unfortunately, have high data requirements and a high level of expertise is required to use them. Again, one of the major problems here is the quality and quantity of data required.
At the macro-level (long term or strategic), planners need to understand the network and network interactions to make informed decisions geared toward optimizing freight movement across the country. An example of this would be estimating how much money railroads have to invest in infrastructure to handle increases in demand and how this estimate might change to reflect the widening of the Panama Canal. On the basis of current standards, the existing condition of a rail line is not a high concern and therefore is often overlooked in modeling.
Opening Remarks
Katherine Hancock, Ph.D.
Associate Professor, Transportation Infrastructure and Systems Engineering, The Charles E. Via, Jr. Department of Civil and Environmental Engineering, Virginia Polytechnic Institute and State University
Katherine Hancock, Associate Professor at Virginia Polytechnic Institute and State University, focused on freight operations and planning, transportation safety, and geospatial solutions to transportation problems. Hancock stated that, from a modeling standpoint, it is impossible to have multimodal networks without multimodal assignment. In addition, it is impossible to assess multimodal assignment without multimodal networks. Hancock also noted that the FAF network has only a few attributes, and what is necessary for a true flowable network is not available in the publicly available version of FAF.
According to Hancock, the publicly available public transportation network is not where it needs to be. For example, 12-million values need to be filled for highway alone, and these values are currently missing, with feasibility issues associated with obtaining them. Moreover, assignments are different across modes. Hancock noted that FHWA has not captured these differences in the FAF and currently does not know what attributes would be required, because the volume-capacity ratio does not work as it does for other modes. Hancock stated that the Government needs to identify what the accuracy requirements are for mode choice.
Open Discussion
Following the presentation, the presenters asked participants to provide their feedback in an open discussion. What follows is a summary of this discussion.
One possible solution to the issues discussed during this workshop would be to develop agent-based models and to assign demand to a multimodal, routable network.
Workshop participants suggested that the Government needs to focus on mode choice or needs to break this into manageable parts. It would be optimal to create a detailed multimodal network for which FAF users can assign demand to each link. This can be achieved through agent-based modeling.
Agent-based modeling could capture and evaluate the dynamics and variables that need to be evaluated to create a model with a multimodal, routable network. From a systems-level perspective, there is no single owner of the national highway system; however, from a shipper's perspective, they must have some concept of a network. Participants noted that if something needs to move from one place to another, a shipper needs to be able to move it across the network by using the vehicles that he or she has available (or any combination of the vehicles he or she possesses and other modal vehicles available). From the shipper's perspective, it is not comprehensive, but it gives the planner a path (similar to the mental model that drivers have to make about routing decisions).
A set of variables exists that impact freight planning decisions, yet these are not in use by freight planners. An agent-based model could therefore be designed to incorporate the vital elements that need to be included in network analysis. These could include information and variables that influence local-level decisionmaking in freight but that are not currently factored into decisionmaking processes. An agent-based model could then show how this additional information might impact the freight network.
Providing tools that allow more informed decisions will inherently alter how decisions are made within the system. The network model for assignment does not have to be the same as the model for trip generation.
Workshop participants noted that different levels of granularity are required for groups who use the models for different purposes. Rather than a multimodal network from the top down, the Government could look at creating this network from the ground up. Participants suggested that the model should focus on heuristic decisionmaking by shippers and supply chains, rather than on other decisionmaking.
Additional questions and comments discussed during this session.
Workshop participants asked whether existing regional models should be stitched together to provide a fuller picture or whether data should be collected and created by a national-level agency to create a national model. Participants noted that the FAF cannot provide insight regarding local or regional impacts of a major project without this level of detail. For example, how would the Panama Canal expansion project impact the network? Participants suggested that to answer this, planners would have to understand the resolution of this project among various modes, across various links, and within the entire network.
Additional questions posed during this discussion are outlined as follows:
Should different models be distributed for different user groups? Distributing different models would address some of the concerns expressed during the workshop, including:
Should a synthetic dataset be created to provide the required level of detail to FAF users without creating privacy issues?
Should the new FAF include all variables or simply those variables that may have the biggest impact on freight-planning decisions?
Would a highly detailed model have the capacity to also include an intermodal network? Should this model be created to answer all questions or only the big questions? Will it answer these questions at a high level or on a detailed level?
Would an overly detailed model lose a level of usefulness? Would it run the risk of becoming unwieldy? Would an all- inclusive model actually be used? If so, by whom?
Should the Government invest in a model that includes all variables versus different models to serve different users for different purposes? Before this question can be addressed, workshop participants suggested that the Government should consider the market needs for modeling, specifically that the Government should be able to answer the following questions:
For what are the models being used? What is important to include? What would be a poor use of public resources to investigate?
What level of detail is sufficient? Could the next generation of FAF incorporate loosely coupled or connected data?
Should all of the required data be gathered first, or should the model be created and then data used afterward? Participants questioned which would be the optimal approach. In addition, calibration requires full datasets—does the Government have the luxury to wait for a complete set of data, or do they need to address this issue now?
