Three presentations were given prior to the official peer review meeting to give the audience more information about the Traffic Impact Analysis process and the FMPO regional travel demand model as well as to highlight a Concurrency Districts approach done in Bellingham, WA. The presentations given were the following:
The official peer review began with an overview of the FMPO TIA process and regional model and proceeded with a more in-depth discussion of aspects of the model and TIA process. A theme through the peer review is how the regional model could be used to further the TIA process and what improvements need to be made in the model itself, the model validation, and application to support the TIA process. This section presents first the assessment of model components in their order of execution, followed by model calibration and maintenance recommendations. The TIA process and model relationships are described after the model assessment.
The FMPO model implements modified ITE vehicle trip rates applied to land-use data from the assessor's office to generate trip productions and attractions. This approach provides a more straightforward connection between trip productions and land use policy questions, such as floor area ratio and second home regulations. On the attraction side, the trips rates vary based on commercial tenant so variations between a restaurant, convenience store, and clothing store are all represented.
Forecasts are developed through scenario workshops that produce build out scenarios. Horizon year control totals are set by state projections. The scenarios are then disaggregated to TAZs.
The ITE rates are vehicle trips, and so the generated trips must be expanded to person trips using average vehicle occupancies by trip purpose. FMPO specifically requested comments on their trip generation process including whether to segment trip generation rates further to use more of the 60 land uses available in the ITE manual. Acknowledged risks of this approach are the dependency on the assessors data collection and categorization methodology, the complex forecast process, and the lack of cross-classification.
Several panelists raised concerns that this approach is unconventional and may be problematic when developing forecasts. The base year dataset can be well constructed because the information is readily available, but it is not clear how an objective forecast method can be applied. Whereas, the use of employment and demographics for trip attractions is time honored and those traditional methods are compatible with economic and land-use forecasting methodologies.
A panelist acknowledged that using ITE vehicle trip production rates are more straightforward because household and employment data do not need to be collected and forecast, but questioned the implied assumption of uniform travel behavior among all households with the same dwelling unit type. Testing this assumption would require an analysis of travel behavior pattern differences by different household types (if not also person types) using the household travel survey to see if there are statistically significant differences. For example, families within a &ldquot;single-family house&rdquot; can be different sizes and have different number of workers, vehicles, and incomes. There is also demonstrated value in segmented trip rates by vehicles and income for later steps of the model process, particularly distribution and mode choice. A panelist also mentioned that replacing the dwelling unit type based trip production with a cross-classified approach may help with some of the assignment issues where volume is high in one area and low in another.
Using employee-based attraction rates is more typical and also may be worth considering for the FMPO. Several panel members' past experiences have shown that there are a number of technical issues that may be encountered when using ITE trip generation rates to build up trip attractions. A panelist emphasized, and the panel agreed, that using employee-based trip rates from ITE are better than square footage based trip rates because they seem to have a smaller standard deviation.
Another value of the employee-based attraction rates is that the attractions can be segmented by worker's earning so that high-earning employment is associated with high-income households. This is particularly important in regions with greater income disparity than the Flagstaff region. This would require additional data collection by the FMPO and translation of employment types into income categories, for which there are methods readily available. A panelist explained the method from the Triangle Regional Model as follows:
The FMPO expressed concerns that there would be an inconsistency and/or incompatibility between a regional model using cross-classified household data and a TIA process using land use data. Several panelists asserted that this is not an incompatibility and that, moreover, the cross-classified household data are especially relevant for in-fill developments. An affordable housing infill development containing a higher percentage of low income household is an example where the expected travel behavior differs from the region average.
Audience members familiar with the Arizona TIA process agreed that there is no incompatibility between using socio-economic data within the regional model and ITE rates based on land-use for the TIA process. However, the audience emphasized the need for good communication between the data set managers to avoid inconsistencies.
University trips are generated through the same process as the other trip purposes. University students represent a large percentage of the overall Flagstaff population, and university students have different travel behavior, particularly around transit usage and non-motorized modes. These factors led the panel to recommend that FMPO collect or borrow university special generator data to better capture the large population of university students in the region. A panelist recommended that FMPO review the trip rates and study results from the NCDOT University Student Travel Survey and Modeling research project. Off-campus student address data often have quality issues because of the confusion between residence while attending school and permanent residence, e.g. parent's residence. But, FMPO should try to make use of off-campus student housing data within the model.
The model currently accounts for K-12 school trips as part of the Home-Based Other trip purpose. Implementing a separate Home-Based School trip purpose will require identifying the school locations with enrollment and forecasting future K-12 school locations. School bus trips are typically not included in a regional travel demand model with static assignment, but charter schools in Flagstaff have contracted with Northern Arizona Intergovernmental Public Transit Authority (NAIPTA) to use public transit service for school trips. The importance for transit led at least one panelist to recommend that FMPO add a home-based school trip purpose.
The current model does not include a freight or truck component. Trucks on non-interstate roadways in the Flagstaff region repersent less than 5% of all traffic and traffic on many roadways is less than 2% truck. The method defined in the Quick Response Freight Manual (QRFM) is straightforward and appropriate for regions this size. Calibrating the model would require expanding the number of vehicle class count locations. The panel recommended that FMPO implement a QRFM model.
University and tourist activity create seasonal traffic patterns for Flagstaff, although the similar location of both student housing and hotels means that the traffic levels are fairly similar across seasons with the fall season being slightly busier. FMPO has developed methods by which vacancy rates can be applied to represent different seasons. The rates were developed using census data and adjusted by utility and postal delivery records along with informal observations. The vacancy rates are generally not applied in forecast years. A panelist questioned the use of a consistent rate across all TAZs because vacation homes, hotels, and student housing are not uniformly distributed across the region.
The panel concluded that, given that traffic is higher during the school year, only a fall model is necessary to be maintained. The other advantage of modeling the fall season is that spring travel should be similar so a fall model actually represents half the year. If FMPO should maintain a summer model, the panel supported the use of vacancy rates to account for the smaller student population and recommended that a visitor-specific generation be included.
The FMPO model currently uses a gravity model to distribute trips. The panel noted that a destination choice model incorporates more flexibility into the model, but that it is not necessary for a region this size.
The FMPO model produces both a 24 hour and PM peak hour assignment results. However, the PM peak hour assignment has not been used much lately and was calibrated only using broad percent RMSE measures during the last model update. The 24 hour assignment is used for long range planning as well as to supply growth rates to the TIA process for large development projects that will roll out over several years. The TIA process, however, is for the peak design hour, but the growth rates based on daily travel patterns are being used from the regional model. FMPO requested advice from the panel as to how they should expand their peak hour assignments.
In a pre-meeting email, a panelist argued that no forecast should use a time interval for traffic assignment larger than 1 hour. 24 hour forecasts are not reliable and peak-period forecasts, while they might calibrate well, will not be properly sensitive to delays. Furthermore, traffic engineers have never defined 24-hour capacity. Capacity in the Highway Capacity Manual (HCM) is defined for 15 minutes, which can be extended to a full hour with the PHF (peak hour factor). Whereas any 24-hour capacity numbers are less justifiable. Regional models with static assignment, however, should not be run with shorter than one hour periods because a key assumption of the static assignment process is that all trips loaded onto the network will complete within the time period.
Another panelist noted that the model run times are so short that it would not delay the process to implement more time periods.
The panel expressed concern about the disconnect between the 24 hour model assignment and peak design hour for the TIA process and strongly recommended replacing 24-hour model with a series of shorter traffic assignment procedures that would sum to a 24-hour total. For example, there could be two peak hours (AM and PM), a midday period if necessary and the remaining off-peak periods assigned as a single block. The peak hour assignments are most important to support the TIA process.
Time of day factors can be asserted, but the assignment results will need to be validated against hourly count data. Time of day factors could be derived from the household survey and/or from neighboring regional models. A panelist noted that different time of day factors should be developed for auto and truck traffic as heavy trucks will avoid traveling during congested times.
The model currently segments External-Internal, and Internal-External trips into three trip purposes. This allows the distribution to be sensitive to trip purposes. External-External, External-Internal, and Internal-External model trips are based on an intercept survey from the 1980s combined with a hotel and truck stop survey done in 1998. Since then, FMPO has used growth factoring and comparisons with the Arizona statewide model. In the pre-meeting discussions, a panelist recommended a cordon survey to improve the external trip data.
The panel recommended that FMPO continue the practice of segmented external trips through model updates. FMPO should also continue to work with Arizona Department of Transportation (ADOT) to improve data available through the statewide model.
The FMPO model does segment transit person trips using the &ldquot;3-D&rdquot; method, but does not use a typical logit mode choice model structure with sensitivities to path skim times. The process uses qualitative measures of level of service for walk modes, a transit level of service based on the proximity to bus stops and frequency of service, and a bike comfort index that considers the distance, bike lane connectivity, and intersections. This non-traditional method of mode split was implemented in the FMPO model based on guidance from their model development consultants.
The panel recommended that FMPO implement a more traditional logit-based mode choice model within the overall model stream as shown in Figure 4-1. FMPO could use asserted parameters based on FTA guidance and make this a straightforward process. There would need to be coding of the transit route system and non-motorized networks. Some of the existing level of service measures that FMPO has developed, such as the Bike Comfort Index (BCI) that accounts for traffic speed and volume, bike lanes and trails, and lane widths, can and should be included in the non-auto utility functions.
Figure 4-1: Traditional Mode Choice Nest Formulation
The network contains daily and peak hour link capacities by functional class and area type derived from the Highway Capacity Manual. There are manually coded intersection delays, particularly at the railroad crossing which runs through the center of town and is heavily used by freight trains and a daily Amtrak service. A panelist commented that the link capacity numbers appeared to be on the high end and expressed concern that local trips on major arterials and interstates tend to reduce capacities. Free flow speeds are based on posted speed limits, which a panelist pointed out do not necessarily represent the actual travel speed of drivers.
The FMPO defines area type by staff judgment. Area type is currently used in the model to determine link capacity and out of vehicle times (terminal times). In other regions, area type is based on a household and/or employment density within a zone, but this requires manual smoothing to avoid big changes in area types due to urban parks, for example. The panel recommended that FMPO continue their current practice of coding area types using staff judgment.
Assignment runs with the current model produce volume to capacity ratios above 1.0 for certain links. The panel advised FMPO to be cautious of links with V/C ratios significantly higher than 1.0. Links with v/c ratios over about 1.2 indicate a model scenario where demand is much higher than supply and the model results may not be valid.
The panel recommended that FMPO implement an assignment process that utilizes an intersection delay function. The intersection delay function dynamically adds delay to paths through intersections based on the signal timing. This will improve the model's usefulness in the TIA process, which is more dependent on turning movements at intersections and corridor-level volumes. The panel noted that the railroad crossings will require special tuning to account for the perceived delay that may be substantially higher than the actual delay.
The Flagstaff region has a small but heavily used bus network that serves both choice and captive riders with about half of the ridership related to NAU. The transit agency has received a Small Starts FTA grant for a BRT-light route through campus and is planning a second grant application for a BRT route.
The FMPO model does not currently include a transit network for skimming or assignment. As discussed in the Mode Choice section above, transit trips are removed from the set of person trips before highway assignment using level of service measures based on proximity and frequency of service. Transit is validated by comparing district-district flows to the observed boardings.
The panel concluded that the NAIPTA Mountain Line transit service is small enough such that coding the existing service into a transit network for skimming and assignment would be manageable. Forecast year transit will need to be represented as well, similar to the forecast highway network. The transit network skims would be used in mode choice, as described above, and the transit trips from mode choice could then be assigned to this network.
The modeled transit boardings will be easier for FMPO to validate than the current method of area to area mode shares. The validated results will be much more useful to NAIPTA for route-level planning and future FTA grant applications for BRT service.
The use of Dynamic Traffic Assignment (DTA) for the FMPO model was discussed and dismissed prior to the peer review. As a panelist explained via email, Flagstaff is most likely not big enough or busy enough to justify DTA on its technical merits. A good substitute for DTA, which is feasible for a city the size of Flagstaff, will be to implement node delays at signalized and stop-controlled intersections and run one-hour assignments in the congested times as was recommended in the Time of Day and Highway Assignment sections.
The FMPO model consists of 340 traffic analysis zones although zones are sometimes subdivided in the TIA process. Smaller zones can be useful to analyze pedestrian and bicycle travel, but additional zones increases computational intensity. The run time of the current model is very short so that is not a concern, but panelists concluded that the number of zones was appropriate for a region this size.
The FMPO model does not implement a speed-feedback loop whereby congested travel times and distances are used to redistribute trips. Adding a speed-feedback loop will cause average trip lengths to be shorter because congested travel times are longer than free flow travel times so the friction factor curve will resolve to closer zones. This will typically result in lower vehicle miles traveled (VMT) in the forecast year and the reduction or elimination of links with high volume to capacity rations. However, it is much more behaviorally sound to include congested travel times in the distribution model, particularly for the peak travel times.
The panel strongly recommended that FMPO implement a speed feedback model, particularly with the peak travel periods. The panel recommended that a composite impedance that considers both distance and time be used in distributions and potentially in assignment. Including distance as an impedance measure will mitigate the reduction in VMT under speed feedback iterations.
The panel recommended that data collection be focused in the fall to be consistent with the recommendation that FMPO calibrate and maintain a fall season model, as discussed in the Trip Generation section above. The panel also recommended that FMPO conduct year round counts at 10-12 locations to gain more insight into seasonal traffic variations.
FMPO conducts a household survey every six years, which the panel supported and recommended that the survey results be incorporated into more aspects of the model validation. As a complement to the household survey, a panelist recommended that FMPO conduct a transit on-board survey to collect better mode share data and estimate intra-zonal transit trips.
FMPO requested more information about collecting data on non-motorized modes. One panelist noted that their region uses volunteers to conduct manual counts along bike paths. Another panelist provided the following references for more information on non-motorized data collection:
In general, the panel recommended that FMPO invest more time on calibration of the demand models.
For trip generation, the household survey can be used to validate district level trip generation by purpose. Trip generation can also be submitted to reasonableness checks as identified in NCHRP-716 and the TMIP Model Validation and Reasonableness Checking Manual.
The panel recommended that the household survey and Census Transportation Planning Products (CTPP) Journey to Work data be used to calibrate trip length frequencies and district to district flows where the districts are a small number of key travel areas. Suggested measures include average trip length and the coincidence ratio[1], which compares how well the distribution shape fits. It is also important to validate that the trips are being distributed between the right locations. Although there are insufficient data to analyze flows at a zonal level, the FMPO could aggregate to the district level, using professional judgment to identify key areas and validate the district to district flows in the model. District to district flows differ from screenlines in that they can be used to identify flows from areas that are not adjacent.
For assignment, the panel suggested that the FMPO continue their robust count collection and enhance it with vehicle class counts to validate the truck model, but that the average annual weekday count (AAWDT) be used rather than average annual daily count (AADT), which include weekend data. The count data are currently used to validate overall model VMT as well as several screen-lines. FMPO expressed concern that some minor screen line locations are not supported by counts. A panelist advised that the FMPO review the reasoning for each screen lin. If the screen line represents a major or key flow in the region, it should be counted and the counts should be complete. The panelist explained that the best practice is to also count traffic on roadways across the screen line that are not part of the model network. But, screen lines do not need to be maintained for their own sake. Some panelists thought that a major corridor link level validation is also necessary, given that the region is dominated by a single corridor and divided by the railroad and interstate.
FMPO reported cases of over/under issues in the highway assignment where volume is too high in one area but too low in another. A panelist suggested that FMPO use 'select link' type analyses to gain insight into the origin and destination of trips through the problem areas.
In addition to the count data, travel times are also important for assignment validation of peak periods. The panel recommended that HERE and/or INRIX data be used to validated travel times. Travel times should be validated by facility type and volume group and may be used to calibrate the volume-delay function.
FMPO had incorporated the maximum desirable deviation curve in the validation report, although it was not used as a primary determinant of model validation. Panelists agreed that the maximum desirable deviation curve is an antiquated measure and should not be used in validation.
Finally, validation standards from states such as Ohio, Michigan, and Florida should be reviewed for guidance as to the specific validation measures and for ideas on best ways to organize validation tables, graphics and reports.
The typical MPO procedure calls for a model update for each transportation plan, which occurs every five years in Flagstaff. FMPO currently updates the model networks and inputs on a three year cycle and would like to systematically incorporate information from the assessor's office and TIA process enhancements into the model. On average, there is one major project per year that requires changes to the model inputs to support a TIA process.
A new development may be completed within the three year model update cycle and the panel expressed concern that this presents the opportunity for TIAs to be conducted with obsolete versions of the model. To support the TIA process and other local uses, the panel recommended that the network of existing and committed projects be updated on an annual basis. The land use dataset should also be updated on an annual basis. FMPO may explore more frequent updates as new projects are funded or implement an Existing and Under-Construction dataset to represent multi-year developments.
FMPO requested guidance on how the regional model could be officially approved for use by member agencies. A panelist offered the following opinion:
Approaches vary widely by jurisdiction. One of the most effective methods I have seen is creation of a committee that participates in model update and development processes. The committee will usually meet 3-4 times over the course of a model update. An example meeting list might include:
Results of this process include a broader understanding of the model's purpose and better review by a more varied audience. Example committee names include &ldquot;Model Development Team,&rdquot; &ldquot;Model Validation Task Force,&rdquot; or &ldquot;Model Improvement Team.&rdquot;
The model currently consists of a base year and 20 year forecast year. The panel agreed with the planned development of 10-year interim forecasts (e.g., 2020, 2030, and 2040). The panel advised to also consider 5-year increments to avoid the need to interpolate over long periods of time. Where interpolation is appropriate, the process should be applied to model inputs (i.e., socio-economic/land use data), not outputs. Very short term forecasts (e.g., 3-5 years) may be done with simple growth forecasting.
Panelists cautioned against running model forecasts with a no-build network as it may be stressing the assignment procedure beyond its applicability. Forecasts should be run with a fiscally constrained development pattern and perhaps include projects that are beyond the Existing and Committed list but are reasonably affordable.
There was a robust discussion about working with the model outputs to support the TIA process. The implications for the model structure are discussed in the previous sections and recommended model applications are discussed below.
Currently, the model is used in TIA studies to baseline background traffic and derive growth factors in the forecast year. The panel concluded that the model can be used in other aspects of the TIA process, given that certain enhancements are made to the model to ensure better results.
The audience emphasized the need for a clear process. Audience members with experience developing TIAs cited experiences where the unknown time to complete model runs and additional processing were onerous for a development project. The panel agreed with a threshold of 100 trips per hour as a prerequisite for involving the regional model in the TIA process.
The TIA process uses trip rates from the ITE manual to derive the number of vehicle trips generated by a particular development. A panelist explained the method for how the ITE manual rates are developed and distinguished between averaging across the rates and recalculating the average rate using the total land use and trips, which is effectively a weighted average. Different ranges of square footage require different approaches, i.e. plotted points in adtion to the convetional average rate or fitted curve rate, to determining an appropriate trip rate.
Another key concern with the TIA process around the adjustment of generated trips is handling mixed used developments. A panelist and audience members agreed that the term &ldquot;mixed use&rdquot; has been misapplied to developments that do not truly reduce trips outside of the development. The Trip Generation Handbook, 3rd Edtion provides clear guidance on the definition of mixed-sue developments. A well-designed regional travel model with cross-classified household or person-based trip rates is well suited to forecast background traffic growth, particularly in-fill type travel that is not well represented by ITE rates.
The TIA process uses a fairly low level of information (e.g. link volumes and turning movements) while a regional model is primarily calibrated at the corridor, screen-line, and regional volume level. So great caution should be exercised before using these outputs from the regional travel demand model. Panelists recommended a careful validation beyond what is currently being conducted and, even with the additional effort, still concluded that post-processing and refinement of the output is necessary. Moreover, the TIA may require splitting of model zones and/or adding links to zoom in on a smaller area. Therefore, the regional model results cannot be used directly, but the indirect applications still need to be carefully calibrated.
NCHRP-255 / NCHRP-765 define a standard method for intersection and link volume post-processing. For example, link volumes can be adjusted using difference method according to the ratio or average volume and iterative proportional fitting (IPF) applied to the turning movements to fit the adjusted link volumes. The panel recommended that FMPO follow a process similar to that recommended in NCHRP-765.
The model can be useful to identify the radius of impact for a development project. Rather than placing a simple radius around the development project, there may be intersections farther away that are more adversely impacted than closer intersections not on a key link for the particular development. For example, a development that will house university students would impact the intersections between the development and university to a greater degree than intersections between the development and downtown.
Thresholds that define a significant impact, constraints on the total intersections to be considered and the minimum development size need to be defined. Example thresholds are:
Again, these were offered as example thresholds and constraints and they should be defined based on the FMPO staff judgment.
As discussed in the Seasonality and Count Data section, a fall model is sufficient for the FMPO's needs. If the agency were to maintain a summer model, for example to test a proposed development consisting primarily of hotels or summer homes that would have peak travel during the summer, the panel recommended that a hotel-based trip purpose be added to the model. Visitor travel is not well represented by the standard purposes that are currently in the model.
A major concern raised by the FMPO and their consultants is how to assess the proportional share of new developments. The concern stems from the situation where developments are not restricted while there is ample capacity in the network, but if the available capacity is limited, developments end up with the burden of substantial improvements even though they would represent a minority of the total traffic. This is an issue for urban developments because it discourages infill type development and promotes development in outlying areas, i.e. sprawl, which results in increased congestion (real or perceived) in outlying areas and on critical connections to outlying areas. Sprawl also increases the potential of early developments providing no transportation improvements, and later developments required to provide substantial improvements.
Allocation of proportional share responsibilities to developers is primarily a political issue, but the regional model can be used to assess impacts based on the proportion of existing capacity used by the new development. The process would be to run the model with and without each development and compare the base and/or forecast year assignment results. Some other tools that can be used are a comparison of delays to examine where the impact is on roads and intersections that are near capacity. A select link/zone analysis can also be used identify critical points in the network.
An audience member explained that many cities in Arizona have an impact fee program that takes effect even if the degradation in level of service does not exceed any threshold. A key aspect of the impact fee program is that the growth rate is clearly defined and agreed upon because developers need to be able to determine in advance what the potential impact fee might be.
Flagstaff has a high non-auto mode share because of the large university population and dense, walkable downtown. Pedestrian traffic is at a level where it is creating congestion to key roadways during the pedestrian crossing cycle. Several of the bus lines are over capacity. Therefore, the impact of new developments analyzed needs to be more than just the vehicle trips generated.
A panelist noted in a pre-meeting email that the margin of error in transit trip prediction is greater than the predicted volume, particularly for smaller regions such as the FMPO planning area. Therefore sensitivity tests should be run to produce a range of non-auto results from the model. The sensitivity tests should be constructed to determine the magnitude of error on the transit predictions that results in a different conclusion of transportation network improvements. For example, what is the possibility that the development will create severely crowded buses?
The panelist went on to explain that, in terms of auto trip reduction, there should be robustness thresholds that the transit system meets to justify any reduction. For example: an auto trip reduction greater than one percent can only be justified by bus stops or stations within one-quarter mile of a development, with service frequencies of 15 minutes or less, direct connections to destinations, and travel times less than 50% greater than private vehicle. The transit system must have characteristics that would compel choice riders to regularly utilize that system if they worked, lived, shopped, or recreated in the proposed development.
The entire panel acknowledged these threshold recommendations and proposed the following method to use the model to support the non-auto process. First, deficiencies by mode would be identified as part of an off-model process. Next, the model would be used to identify development traffic interacting with areas where deficiencies have been noted. Model tools to support this analysis are the select zone analysis and examining proximity in the network. Implementation of a transit network for skimming and assignment is particularly important to make the model useful for non-auto analyses.
A presentation of a concurrency district analysis process in Bellingham, WA referenced during the peer review calls for looking at network &ldquot;completeness&rdquot; rather than the level of service range. The panel agreed that this process would be appropriate for Flagstaff given the connectivity issues caused by the railroad and interstate. This type of approach would be complementary to the model results and help balance the risk of using model predictions with known large error terms.
[1] For a full description of coincidence ratios, please see section 8 of the Quick Response Freight Manual: http://www.ops.fhwa.dot.gov/freight/publications/qrfm2/sect08.htm
