This report describes field observations of multi-lane freeway traffic upstream of an oversaturated off-ramp. It is based on empirical evidence from freeway I-880 (northbound) near Oakland, California. Two diagnostic tools reveal hidden features of the traffic stream. Based on these findings, we propose congestion mitigation strategies that could work for similar locations with little or no construction.
The first tool (oblique
plots) was used to identify a “moving bottleneck” that occurred before
the rush hour and to characterize, for the first time, the traffic stream directly
upstream and downstream of it. The fine detail of these plots affords a unique
glimpse into the realm of driver psychology. The data show that the first hundred
drivers in the queue initially accepted relatively short average spacings, and
that a little later the same drivers allowed themselves to relax into larger
spacings. Later drivers also adopted the new spacings. The empirical observations
are compared to the predictions of kinematic wave (KW) theory. It is also found
that once the queue caused by the off-ramp spill-back fills the study section,
the system’s flows and speeds at all locations within the section stabilize
in a steady state that lasts for over 30 minutes.
The second diagnostic tool, cross-correlation of synchronized cumulative counts (also known as N-curves), reveals that during this time, and close to the congested off-ramp, small disturbances in the shoulder lanes move upstream while those in the median lanes move forward. This indicates that the shoulder lanes are in a congested/queued state while the median lanes are not, and that the freeway should not be modeled as a “single pipe” KW stream near the off-ramp.
Lane-specific behavior is not found beyond the 2.2-km mark, however. At these locations, it was found that: (i) the KW model with a wave speed of about 20 km/hr holds well in congested (queued) traffic, (ii) the onset of queuing is gradual and nearly simultaneous across all lanes and (iii) the transition zone where vehicles decelerate toward the end of the queue is about 1 km long, and propagates like a shock. Because the difference between the queued and unqueued flows was small, the transition zone moved slowly (at about 3.5 km/hr) and took almost 20 minutes to pass over a detector. Flow-occupancy scatter plots exhibit macroscopic patterns consistent with the above conditions.The report also explains in a simple way some strange events that occurred during the congestion dissipation process, most notably, a large and sustained increase in the through flow that took place with no appreciable change in other traffic conditions.
The report concludes
with an assessment of the vehicle-hours of delay caused by the bottleneck, and
various proposals for eliminating a large fraction of these. Three alternative
improvements are proposed, which can be implemented alone or in combination:
(a) modifying the off-ramp flow, (b) modifying the lane assignments, and (c)
modifying the HOV lane.
(a) Off-ramp actions: The off-ramp would be able to carry more flow if
(i) the competing flow from Hesperian Blvd onto I-238 was reduced, and/or if
(ii) the merge area of our off-ramp onto I-238 were re-striped so as to give
priority to the vehicles from our off-ramp. The first suggestion can be field-tested
by changing the timing of the signals on Hesperian Blvd, or more drastically
by closing access to I-238 from Hesperian altogether. An increase in the off-ramp
flow as small as 100 veh/hr is estimated to reduce the delay by more than 50%.
(b) Lane assignments: Our research shows that traffic upstream behaves as a single pipe; i.e., that through vehicles are entrapped in the queue of exiting vehicles. Our research also shows that when the exit lanes flow faster, through drivers do not encroach on them; i.e., for the most part drivers are disciplined enough to stay on their lanes. This suggests that “through only” lane markings could be effectively used on three through lanes to encourage exiting vehicles to stay on the exit lanes, even if these lanes are queued. These markings should be strictly enforced, and applied for a distance upstream of the off-ramp equal to the maximum extent of the off-ramp queue. If lane markings were used for the three km upstream of the diverge, we estimate that total delay would be reduced by more than 80%.
(c) HOV actions: If “through only” lane assignments are used,
the HOV designation should be terminated well upstream of the location where
the lane assignments begin. (Otherwise HOV’s are encouraged to break the
rules and generate pernicious weaving.) In our case, if a static strategy were
to be used, then the HOV lane would have to be terminated about 4 km upstream
of the off-ramp. The termination point, like the lane-assignments could be changed
dynamically, perhaps in response to traffic conditions.
The next step is to test the above ideas at different locations, including our
site. There are many congested diverges in US cities, and many could probably
benefit from policies (b) and (c) above.
A pdf of the complete research report is available at
http://www.path.berkeley.edu/PATH/Publications/PDF/PWP/2000/PWP-2000-13.pdf