San Francisco County Transportation Authority (SFCTA) Travel Model Peer Review Report
6.0 Panel Discussion and Recommendations
The following text summarizes the panel's discussion on the
topics of interest to SFCTA and the audience. The summary of this discussion
follows the panel's final presentation back to SFCTA. The discussion begins
with some general observations made by the panel on the overall modeling
process followed by short and long-term recommendations. Section 6 includes a
point-by-point response to the questions the SFCTA posed to the panel at the
outset of the meeting. The summary in this section is a consistent summary of
the detailed responses in the form of a recommended action plan.
Observations: General Comments
The panel applauded the overall effort made by SFCTA to
systematically develop and calibrate a DTA model, and specifically on the
SFCTA's focus on the calibration of the traffic flow model parameters using
observed data. They understand that this is a non-trivial endeavor and are not
aware of many efforts that have been made in this direction and at such detail.
The panel recognizes that this has been a truly collaborative effort with
committed staff from SFCTA, the consultant team, and partner agencies which has
laid a foundation for further research and application in this area of travel modeling. The panel also commended SFCTA's commitment to the development of an
open-source code-base that can help other agencies who would like to develop
DTA models in the future. The data driven approach taken by SFCTA towards this
project has in the panel's collective experience been known to be a very
effective approach, rather than synthesizing inputs or using a default set of
parameters. The panel suggested that it would be useful to always keep the applications
of this tool in perspective, which would then guide the expansion of its
capabilities and sophistication.
Recommended Shorter-Term Priorities
SFCTA requested that the panel's recommendation focus on
what can be achieved in the next few months before the end of the current
project. The panel feels that the SFCTA should focus on the following items in
the relatively near term:
- Improve subarea extraction: The panel thought that
the demand carving for a subarea was a non-trivial process and there
should be some more focus on that. External geographical information could
be preserved and the demand from external stations could be offset
temporally based on the travel time to the model subarea.
In addition, a temporal profile at
external gateways and in internal zones could be created to avoid a flat or
uniform loading profile to the DTA model. Since traffic counts are available,
they may be used to create the temporal profile. Alternatively, departure times
from the household survey could also be processed to obtain a temporal profile.
- Expand DTA simulation period: The panel felt that a
longer warm-up period may be needed to improve the accuracy of the
simulated traffic in the current modeling period (4:30 - 6:30 PM). This
could also incorporate a portion of mid-day demand to create background
traffic which would already be present before the model period. Further,
there might be some value in simulating the entire 3-hour period (3:30 -
6:30 PM) in the DTA model to facilitate a straightforward comparison to
the static model.
- Devise and perform more
sensitivity tests: To aid calibration and validation, the panel
recommended conducting more sensitivity tests from the perspective of
future and alternative policy scenarios. It might help to devise some
future scenarios tests and targeted policy tests. Since SFCTA already has
an idea about the initial policies that it would like to use this model
for, it might be useful to build some tests around them and evaluate the
results qualitatively first. Sensitivity tests on both demand and traffic
flow parameters would help understanding of which parameters affect which,
and to what extent. This in turn could help identify the parameters on
which to focus.
- Adjust traffic flow parameters for local streets:
For the overestimation issue on local streets, the panel thought imposing
twice the free flow time may not be ideal. The panel indicated that there
might also be an aversion component in addition to the time component. It
might be better to try and tweak the traffic flow parameters such as
reaction time. This factor could act as a "perception penalty" and could
be a function of the number of stop signs.
Alternatively, a separate facility
type for residential low-volume streets could be created. In addition to this,
a reaction time factor that includes friction due to pedestrian traffic on such
streets could be tested.
- Inclusion of distance term in generalized cost function: The
panel felt that including distance in the generalized cost function for routing
behavior might help towards the traffic overestimation issue on local streets.
It is probably more applicable for truck traffic than autos.
- Adjust traffic flow parameters to reflect influence of
pedestrians: The panel indicated that there might be value in
developing link/node specific reaction time factors targeted at specific
areas in the network where this could be important. The factors could also
be derived in a systematic way by using information on aggregate
pedestrian demand within a radius of a node (buffering). For intersections
with a high concentration of pedestrian-generating land uses within a
short distance (say one-quarter mile), it was suggested that delay due to
pedestrian volumes be increased.
In addition to area type and
facility type currently being used to classify the various parameters, there is
potential to use a third dimension - "intersection type" that would allow for
- Calibration of turn movements may be overkill: It was
mentioned during the presentation that turn movements were important. The panel
thought that calibrating and validating turn movements at a system level might
not be needed. It could be done for targeted areas based on specific corridor
level applications of the model.
- Traffic flow parameters for bus-lane modeling: The
modeling team described a method of splitting a bus-lane link and making
one-half of the link right-turn only to deal with bus lanes in the model. The
panel encouraged SFCTA to pursue that link splitting method but also suggested
that targeted reaction time factors adjustments could potentially be used to
accommodate more throughput going through the general purpose lanes and better represent
the traffic flow on these lanes.
Recommended Longer-Term Priorities
The panel urged SFCTA to focus on the above priorities and to
undertake other potentially good ideas over the longer term. These longer-term
- Collecting more data: The panel felt that there may
not be enough traffic count data available. Counts from 200 locations may
be a bit low for a city of this size. The SACOG model with about a million
people uses counts from about 400 locations. The panel recommended that SFCTA
collect more counts that are not correlated with the existing ones and
also that are geographically distributed.
For freeways, effective length was
derived from data collected on local and arterial streets. The panel suggested
that SFCTA could try using aerial photos for estimating effective length on
The panel noted that data on local
streets would also be useful since at present, it may not be very clear as to
why the predictions from the model are not validating very well against the
counts. It is possible that the traffic volumes being simulated on local
streets by the DTA model are reasonably accurate but are not being supported by
the limited data currently available.
- Demand model information for DTA: The panel
suggested that distinction between parking location and activity location
may need to be made in the longer term. This might not be very important
during the initial phase, but will probably be essential during the
evaluation of various congestion pricing policies.
The panel felt that market
segmentation of the demand being passed on to the DTA model could be refined
further to include an additional dimension based on value of time (VOT). The
panel acknowledged that there will possibly be run time implications of this
The panel recommends that
reliability only be considered after model linkage has been well established at
a reasonable level of temporal detail.
- DTA information for demand model: As a first step,
the panels recommends that the modeling team work on developing a hybrid
approach that combines static skims regionally and DTA skims locally to be
fed back into the demand model.
The DTA model could eventually be
temporally expanded to simulate a full day. The panel felt that it may be
better to prefer temporal over spatial detail while increasing the scope and
sophistication of the model. A fully disaggregate DTA-ABM integration could
then be considered at a later stage, which would require a significant amount
of restructuring to both ABM (SF-CHAMP) and the DTA model. The panel felt that
a regional level DTA model might involve a huge amount of work and may even be
- SF-CHAMP: The panel recommends that temporal
resolution of SF-CHAMP be increased from the current 5 time periods.
Again, it may be better to give priority to adding temporal detail to
SF-CHAMP before spatial detail. The panel noted that changes to the
resolution of skims would probably require a number of changes to other
model components in SF-CHAMP such as tour time of day and trip departure
- Person-based transit assignment (Transit DTA): The
panel recommends that only after adding temporal and spatial detail to the
model should SFCTA consider transit DTA using packages like FAST-TrIPs.
The panel feels the SFCTA may not choose to implement some
of the long term recommendations because they may be beyond the scope of the
current DTA project.