You are on page 1of 21

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/270210757

Turn Pocket Blockage and Spillback Models

Article  in  Transportation Research Record Journal of the Transportation Research Board · December 2011
DOI: 10.3141/2259-10

CITATIONS READS
3 431

3 authors, including:

Nagui Rouphail Xuesong Simon Zhou


North Carolina State University Arizona State University
239 PUBLICATIONS   4,543 CITATIONS    198 PUBLICATIONS   4,681 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

DTALite 2.0 strategic and semi-dynamic transportation flow assignment and simulation View project

Warrant Development for Signalization of Superstreet Intersections with Two or Three Approaches View project

All content following this page was uploaded by Xuesong Simon Zhou on 22 February 2015.

The user has requested enhancement of the downloaded file.


Reynolds, Zhou, Rouphail and Li 1

Estimating Sustained Service Rates at Signalized Intersections with Short Left


Turn Pockets: A Mesoscopic Approach

William L. Reynolds*
Graduate Research Assistant
Institute of Transportation Research and Education (ITRE)
North Carolina State University
Centennial Campus, Box 8601
Raleigh, NC 27695-8601
Tel.: (919) 515-3427
Fax: (919) 515-8898
Email: wlreynol@ncsu.edu

Xuesong Zhou, Ph.D.


Assistant Professor
Department of Civil and Environmental Engineering
University of Utah
122 South Central Campus Dr., 210 CME
Salt Lake City, UT 84112-0561
Tel.: (801)-585-6590
Email: zhou@eng.utah.edu

Nagui M. Rouphail, Ph.D.


Director, Institute for Transportation Research and Education (ITRE)
Professor of Civil Engineering
North Carolina State University
Centennial Campus, Box 8601
Raleigh, NC 27695-8601
Tel.: (919) 515-1154
Fax: (919) 515-8898
Email: rouphail@eos.ncsu.edu

Mingxin Li
Graduate Research Assistant
Department of Civil and Environmental Engineering
University of Utah
122 South Central Campus Dr., 119 CME
Salt Lake City, UT 84112-0561
Email: m.li@utah.edu

November 2009

Revised Submission for the 89th Annual Meeting of the Transportation Research Board, January 10-14, 2010

*Corresponding Author

Word Count: 5,982 text words plus 6 figures and tables = 7,482 total
Reynolds, Zhou, Rouphail and Li 2

ABSTRACT

A computational approach for incorporating the effects of short left turn pockets on sustained service rates in
a mesoscopic modeling environment is presented. Mesoscopic models, intended to handle fairly large
networks while maintaining individual vehicle identity, provide a detailed yet efficient alternative to estimate
sustained service rates. However, mesoscopic models typically ignore mid-link perturbations and queuing,
thus limiting their reliability in the presence of a short left turn pockets at signalized intersections. The model
presented herein relies upon a gating mechanism situated at the entry point to the left turn pocket. Through a
series of logical triggers, the gating mechanism allows for the formation of a (vertical) queue of vehicles
upstream of the pocket when arrivals exceed storage capacity. The method satisfies all assumed requirements
for integrating the effects of short left turn bays, which include pocket spillback, pocket starvation and
sensitivity to signal timing and phase sequence. The approach has been implemented within the mesoscopic
modeling platform DYNASMART for both single and multiple left turn pockets, using varying pocket
lengths and demand volumes. The resulting sustained service rates favorably compared to those generated by
a representative micro-simulation model (VISSIM). Further comparisons of the proposed approach against
empirical observations are planned.
Reynolds, Zhou, Rouphail and Li 3

INTRODUCTION AND PROBLEM DEFINITION

Capacity, as defined in the Highway Capacity Manual (HCM) (1), refers to the “maximum hourly rate” at
which vehicles can be expected to traverse a roadway segment during a given time under “prevailing
roadway, traffic, and control conditions.” At signalized intersections, capacity is typically measured at the
stop bar and incorporates factors such as geometry, signal timing, and distribution of vehicle types. However,
there exist several cases in which throughput will never reach the signal capacity values due to geometric
limitations upsteam or downstream of the stop bar. The Sustained Service Rate (SSR) of a facility can be
defined as “the highest rate of flow that can be sustained over a peak demand period under prevailing
conditions.” By focusing on sustainable rates of throughput, SSR provides a mechanism for incorporating a
variety of factors that affect stop bar discharge not necessarily included in capacity calculations.
Spillback from a short turn pocket, and blockage of the entrance to a short turn pocket, are two
examples of upstream effects that may reduce SSR relative to signal capacity. In both cases, one movement
impedes the other, and leads to wasted green time. Although the HCM provides a warning regarding the
assumption of adequate pocket storage, its analytical model does not have the ability to quantify turn pocket
spillback effects. Since both signal timing and intersection design processes rely heavily on accurate
projections of the carrying capacity of a signalized intersection, failure to include the sensitivity of SSR to the
storage capacity of left turn pockets has the potential to significantly impact operation during periods of heavy
demand.
Using queuing theory as a starting point, Kikuchi (2) developed a macroscopic method for predicting
the probability of spillback from a double-left turn pocket, as well as from a three-branch fork. Although
useful for initial intersection design, the models are unable to incorporate the complicating effects of lane
switching within the through lanes. As a result, the models are highly unreliable when multiple through lanes
are present due to the fact that vehicles have the ability to change lanes to avoid queue buildup or pocket
spillback.
Similar to Kikuchi, Wu (3) developed a series of equations to predict intersection discharge based on
turn percentages, pocket storage area, and signal capacity. Relying heavily on micro-simulation for
calibration, the models provide a simple method of predicting throughput analytically in the case of a single
through lane. As with Kikuchi, however, through demand within the leftmost through lane must be treated as
a constant, preventing reliable application of the models when more than on through lane is present.
aaSIDRA (4), based on analytical models of intersection operations has the ability to predict the
probability of spillback from a short turn pocket based on the amount of storage space available. aaSIDRA
also considers a lane-by-lane analysis to demonstrate lane imbalance. However, in its current form, no
blockage model has been implemented to reliably reduce through lane throughput upon spillback from a short
turn pocket.
As can be seen, short turn pocket effects can be very difficult to reliably predict using macroscopic
models. Lane utilization plays an important role, both upstream and downstream of the pocket diverge point,
and assumptions must be made that may not hold during periods of heavy demand. Microscopic simulation
provides an attractive option, and sophisticated car following models allow vehicles to utilize storage space as
efficiently as possible. But microscopic models are computationally intensive and require a large amount of
time to set up trials and generate results. Within large regional networks, microscopic simulation to examine
turn pocket effects is often simply infeasible, and may still require lane distribution assumptions. As such, a
mesoscopic approach is likely the only reasonable method of predicting turn pocket effects in more than a
select few intersections.
This paper first provides a detailed description of mesoscopic models, discussing each of the
improvements needed to accurately examine short turn pocket effects within their framework. The
methodology outlines specific modifications to the modeling platform DYNASMART (5, 6) followed by a
description of the applied simulation algorithm. A results section provides a sensitivity analysis of a few
select trials and a discussion of the extent to which each of these enhancements allow for an accurate
mesoscopic approach to modeling short turn pocket effects. Concluding remarks detail methods of moving
forward with this approach and specific needs for additional research.
Reynolds, Zhou, Rouphail and Li 4

MESOSCOPIC MODELS

Depending on the level of representation detail, flow models embedded in traffic simulators can be classified
as macroscopic, mesoscopic or microscopic. In macroscopic models (e.g. the classical first-order model by
Lighthill and Whitham (7) and Richards (8)), traffic flow is described as one-dimensional compressible fluid
through partial differential equations, and the vehicular flow on discretized highway segments are moved
according to a speed-density relationship. Microscopic models, on the other hand, offer a more detailed
representation by considering stimuli and responses among individual drivers, including both car-following
and lane changing behavior.
Focusing on the effect of commuter route choice decisions, Chang, Mahmassani and Herman (9)
developed an early mesoscopic simulation model to characterize traffic flow as discrete vehicle
groups/particles, and individual vehicle positions are updated by a macroscopic flow-density relationship.
Compared to the fluid-based representation in macroscopic models, mesoscopic models keep track of
individual vehicles and their origin, destination and path trajectory data to better simulate travelers’ behavior
in a network. Without considering a sophisticated lane-changing mechanism, mesoscopic models are able to
use a longer simulation time interval than microscopic models, for example, 6 seconds vs. 0.1 seconds. This
leads to significant computational savings, especially when searching for dynamic traffic user equilibrium or
day-to-day learning solutions on large-scale regional networks.
In existing mesoscopic simulators such as DYNASMART and DYNAMIT (10), the traffic flow
model on a segment or link includes a moving part and a queuing part, where the moving part applies a
macroscopic flow-density function to move vehicles forward; a vehicle joins the queue after it reaches to the
end of the link or the tail of the physical queue. At each simulation interval, queued vehicles are discharged
to downstream links according to an allocated transfer capacity, which is determined by specific merge or
diverge characteristics on freeway facilities, or available green time for a protected phase on a signalized
intersection. For a permissive phase or at non-signalized intersections, the average discharge rate can be
calculated based on gap acceptance models or established tables, e.g. from the Highway Capacity Manual or
similar sources. To model left-turn bays on arterial links, existing mesoscopic models typically consider
separate queues for different movements (i.e. left, through and right), and vehicles join and are dissipated
from different movement-specific queues according to a first-in-first out rule. A queue is considered to spill
back from the current link to upstream links if its density exceeds jam density.

MODEL REQUIREMENTS

To capture the effects of short turn pockets on SSR within this mesoscopic model framework, four general
requirements must be met. The model must (1) characterize and account for the effects of pocket spillback,
(2) capture the wasted green time associated with turn pocket starvation, (3) appropriately measure
differences in SSR due to phase order, and (4) account for dynamic lane assignment and the ability of through
vehicles to utilize available lane space to bypass queues from pocket spillback. As discussed, although
several macroscopic models are sensitive to one or more of these requirements, none account for the effects of
pocket spillback in a comprehensive manner.
Pocket spillback (1) occurs when turn demand per cycle exceeds storage capacity, forcing left turning
vehicles to queue within the leftmost through lane. In the case of fully exclusive left and through phases,
spillback into the leftmost through lane can reduce service flow by an entire lane for the duration of the
blocked state. If the movement phases overlap, the effect is less severe, but SSR reductions occur in both the
turn pocket and the leftmost through lane since both rely on a single feeder lane after the initial queues clear
the stop bar.
Pocket starvation (2) occurs when queues from through vehicles in the leftmost through lane blocks
the entrance to the turn pocket. Such a scenario leads to wasted green time as well as excess demand placed
on the turn pocket during the following cycle. Poorly timed signals may generate alternating cycles of
spillback and starvation, leading to reductions in SSR for both movements.
Reynolds, Zhou, Rouphail and Li 5

Phase order (3) is a critical component that is extremely difficult to analyze macroscopically. While
Wu’s models (3) take into account phase exclusivity and overlap, the models are insensitive to phase order
due to the assumption of a continuous queue (d/c >1.0). Lead or lag left phasing, with or without overlap, can
have a significant effect on SSR, however, and measuring these differences is an important added benefit of
mesoscopic analysis.
Lane Utilization (4) is the most difficult factor to comprehensively analyze, and microscopic
simulation is currently the only environment that is able to dynamically account for lane switching. Most
analytical models currently rely on a lane utilization constant to adjust demand in the leftmost through lane as
a best approximation for queue jumping. As mesoscopic models do not perform lane-by-lane analysis, some
assumptions will need to be made to allow vehicles to efficiently utilize available lane space.

METHODOLOGY

To accurately capture the propagation of queues induced by short left-turn bays, this research extends the
existing link-based mesoscopic simulation model by adding a gating mechanism at the entry point to the left-
turn pocket. Through a series of logical triggers, the gating mechanism allows for the vertical queuing of
vehicles upstream of the turn pocket when arrivals exceed storage capacity.
Conceptually, a link with left-turn bays can be partitioned into 3 parts: (I) left-turn pocket with K bays
and a length of L, (II) through pocket (adjacent to the left-turn pocket) with M lanes and a length of L, and
(III) upstream segment before the gate.
Fig.1 shows a representative approach link with two through lanes (M=2) and a double left-turn
pocket (K=2). The length of the minor pocket is denoted as H.
A clock-based simulation scheme is used in this study. The simulation time interval is denoted as ∆T,
which should not be shorter than the shortest free-flow link travel time in the network, e.g. 6 seconds, so that
a vehicle does not jump across two links during a simulation time interval. At each simulation time interval t,
the position of vehicle i x(i, t) is updated according to its speed v(t). Similar to the modified Greenshields
model used in DYNASMART, a minimal moving speed (e.g. 6 mph) is imposed to ensure vehicles can move
forward even at jam density. Without loss of generality, the position of the downstream end of a link is
assumed to be 0, so a vehicle’s position starts with the link length and moves decreasingly toward 0.
The simulation model uses a vertical queue or a point queue representation scheme, which leads to
two important properties: (I) if the gating condition is not triggered, vehicles can always move to the end of
the link and join the vertical queue, and (II) only vehicles in the vertical queue can be discharged to the
downstream links. With the additional gating mechanism, if either a through or left-turn vehicle is blocked at
the gate, the vehicle cannot reach the end of the link (i.e. stop bar) and cannot join the vertical queue. In this
case, even if the green phase is displayed for the corresponding movement at a simulation time interval, a
vehicle stopping at the gate is unable to be discharged, leading to wasted green time and a capacity loss due to
blockage thus giving rise to the concept of sustained flow rate or SSR.
For simplicity, the following discussion focuses only on left-turn and through queues, as right-turn
vehicles typically have sufficient permissible time and storage space to be dissipated at all times. NL and NT
represent the maximum numbers of vehicles that can be stored in the left-turn and through pockets,
respectively, at any given time. These two parameters can be viewed as the space capacity of each pocket,
which is different from flow rate-based capacity (e.g. number of vehicles passing through a point during a
certain given interval).
Consider the average vehicle length as AVL. The space capacity is determined by:

L H L
NL , NT M
AVL AVL AVL (1)

where represents rounding down to the nearest integer.


Reynolds, Zhou, Rouphail and Li 6

Accordingly, counters nL and nT are used to record the numbers of vehicles stored in the left-turn
pocket and adjacent through lanes, respectively.
There are four major events to be triggered, and the following binary flags are set to “false” by
default:

fL = true when the left-turn pocket is full, i.e. nL = NL


fT = true when the through pocket is full, i.e. nT = NT
bL = true when a left-turn vehicle blocks the gate.
bT = true when a through vehicle blocks the gate.

The conceptual discussion below aims to thoroughly describe the sequence of pocket full and
blocking events and the interactions of left-turn and through vehicles at the gate.
In Fig.2(a), the left-turn bay is occupied by two left-turn vehicles, so fL = true, while the second
through vehicle can still travel through to the gate. In Fig.2(b), the left-turn blockage occurs when the third
left-turn vehicle arrives at the gate, and the vehicle has to stop there. Algorithmically bL is triggered to true by
an additional left-turn vehicle after fL = true. The fourth through vehicle in Fig.2(c) then cannot use the
leftmost through lane to reach the downstream through pocket. Because only M-1 lane(s) are available for
those following through vehicles, the following approximation formula is used in this study to determine the
reduced through flow capacity at the gate:

CT MFR ( M 1) T (2)

where MFR is the maximum flow rate (i.e. the number of vehicles can traverse a roadway segment) per lane
per second, equivalent to the saturation flow rate.
In Fig.3(a), four through vehicles occupy the downstream through pocket, so fT = true, while the left-
turning vehicle can still enter the left-turn bay. Fig.3(b) shows how the through blockage event is triggered
when the fifth through vehicle arrives at the gate. That is, flag bT is set to true by an incoming through
vehicle when fT = true. After the through blockage event occurs, no following left-turn vehicles can pass
through the gate, as illustrated in Fig.3(c).
It should be noted that, in a real world situation, if the fifth through vehicle queues within the
rightmost lane instead, then blockage occurs later when another through vehicle enters the leftmost lane. This
study adapts a simplistic deterministic assumption: the next through vehicle will block the leftmost lane.
Probabilistic analysis results can be found in a study by Tian and Wu (11) about how through vehicles are
distributed across different lanes.

SIMULATION ALGORITHM

Given the green time allocation for different movements at each simulation time interval, the following
algorithm is periodically performed for the associated arterial link with left-turn bays. Thus, the proposed
mesoscopic algorithm is sensitive to signal phase orders.

Step 1: Initialization
L T
Initialize counters n and n to 0. Reset f L , b L , f T and bT to false.
T
Initialize reduced through capacity at the gate C MFR ( M 1) T.
Calculate moving speed v(t ) according to a speed-density relationship.

Step 2: Vehicle Movement Simulation


For each vehicle i in link l, perform the following steps.

Step 2.1: Moving


Reynolds, Zhou, Rouphail and Li 7

Calculate the new tentative position of vehicle i at time t+1 according to link speed v(t ) :
x(i, t 1) x(i, t ) v(t ) T (3)

Step 2.2: Gating Mechanism


If x(i, t 1) L x(i, t ) , then the current vehicle i is about to pass through the gate,
According to the movement type of this vehicle, switch to Step 2.2.1 or Step 2.2.2.

Step 2.2.1: Left-turn


If flag f L =true or bT =true, // left-turn pocket is full or through blockage occurs
reset vehicle position x(i, t 1) L // stop at gate
set b L = true if b L is false
Else
Move vehicle i through the gate, x(i, t 1) is not changed.
Endif

Step 2.2.2 Through


If (f T =true) // through pocket is full
reset x(i, t 1) L // stop at gate
T T
set b = true if b is false
L
Elseif ( b = true)
// through pocket is not full and left-turn blockage occurs
T
If ( C 0)
// remaining through capacity exists
Move vehicle i through the gate.
T
Decrease remaining through capacity C CT 1
Else // no remaining through capacity
Reset x(i, t 1) L // stop at gate
Else // through pocket is not full and no left-turn blockage
Move vehicle i through the gate.
Endif

Step 2.3: Vertical Queue Counting


If x(i, t 1) 0 , then vehicle i reaches the stop bar, and joins the left-turn or though queue. Depending on the
movement type of vehicle i, switch to Step 2.3.1 or Step 2.3.2
Step 2.3.1: Left-turn
L
Increase left-turn vehicle counter n nL 1
L L
If ( n N L ), then left-turn pocket is full and f = true

Step 2.3.2: Through


T
Increase through vehicle counter n nT 1
T T
If ( n N T ), then through pocket is full and f = true
EndFor // each vehicle i

1. Vertical queue representation


To calculate nL and nT, we count the number of vehicles passing through the gate, or reaching the
stop bar. In order to be consistent with the overall vertical queue representation in a link, the queue at the
Reynolds, Zhou, Rouphail and Li 8

gate is also implemented as a vertical queue. That is, as a result of a full pocket or a blocked lane, vehicles
always stop at the gate L, other than waiting at the tail of the physical queue.
The values of nL and nT are constantly changing, as some vehicles are leaving the queue and some
vehicles join the queue at each simulation time interval. In the proposed algorithm, all vehicles on the link are
scanned at each simulation time interval, including vehicles already in the vertical queue, to recount nL and
nT. One implementation can also explicitly update these two counters throughout the simulation horizon.
That is, when a vehicle is dissipated from or joins the corresponding vertical queue, the counter is
automatically decreased or increased by 1.

2. Approximation error associated with remaining through capacity at gate


When left-turn blockage occurs while there is still space in the through pocket, the following through
vehicles will use the remaining through capacity at the gate. When each through vehicle travels through the
gate, the remaining capacity value is deducted by 1. When no remaining capacity is available, the through
vehicle stops at the gate. Note that CT is estimated for the entire simulation time interval (e.g. ∆T = 6
seconds), but the left-turn blockage event could happen at any time in the middle of the simulation interval
(say 2 seconds), so the actual remaining capacity might be lower than the value estimated from ∆T.

3. Comparison with explicit link decomposition scheme.


Using a network representation scheme, one can explicitly split a link into the above-mentioned 3
parts (left-turn pocket, upstream through pocket and upstream cell) as cells/segments/links, and accordingly
use a diverge bottleneck model (12) to describe the interaction between left-turn and through vehicles.
However, this cell-based representation leads to a very short simulation time interval, which equals the length
of the short left-turn bay times the corresponding speed limit. In a case of a 20-foot left-turn bay with a speed
limit of 40 mph, the simulation time interval should be set to about 0.3 seconds, requiring a dramatic increase
in computational burden.

4. Discharge rate for very short left-turn bays


In the proposed simulation logic, for each left-turn lane, the number of left-turn vehicles that can be
discharged within a simulation interval needs to be constrained by the available green time and physical space
capacity. As shown in Fig.1, the minor left-turn lane can only accommodate 1 vehicle. Let us assume the
maximum service flow rate is 1,800 vehicles per hour per lane. When the simulation time interval equals 6
seconds, the minor and major left-turn lanes can discharge no more than 1 and 3 vehicles, respectively, during
the interval. Under this modeling assumption, a vehicle within the major left-turn lane cannot switch to the
minor left-turn lane to utilize unused green time within a single time interval. Instead, a vehicle must enter
the minor left-turn lane at one simulation step and wait to be discharged at the next time interval. This
mechanism essentially throttles the discharge rate once the initial minor pocket queues have cleared,
constraining throughput to a single left turn stream upon completion of the initial period of dual-stream
throughput.

RESULTS

Experimental Design

To demonstrate the sensitivity of the model to a number of explanatory variables, it is helpful to examine a
simple case study. For the purpose of this analysis, discussion will focus on an intersection approach with a
single 100’ left turn pocket (storage = 4 vehicles), 2 through lanes, operating in a 120-second cycle with an
exclusive, protected left phase (g/C = 0.2) followed by a through-only phase (g/C = 0.4) on the approach of
interest. All SSR values are generated by performing 20 separate trials in DYNASMART and averaging the
observed throughput during a 60-minute time interval following a 60-minute network loading period. Results
are presented as the ratio of the observed throughput, or SSR, divided by signal capacity (c=Sg/C), where the
Reynolds, Zhou, Rouphail and Li 9

saturation flow rate (S) is set at 1,900 vph/lane for all movements. This is plotted against the X axis d/c ratio
by movement, where d is the upstream unconstrained movement demand.

Demand Sensitivity

Fig.4 presents the results of the scenario described with the total approach demand consisting of 15% lefts,
20% lefts, and 25% lefts in parts (a), (b), and (c), respectively. The X axis represents d/c by movement,
keeping the fraction of left turns in the traffic stream fixed in each of the 3 graphs. With no pocket spillback
or starvation, each movement’s SSR would be expected to follow the dotted line, representing signal capacity
by movement. As the signal is timed based on 20% left turn demand, both movements in part (b) follow a
similar trend, maximizing SSR at just over 80% of signal capacity. With fewer lefts (a), however, the through
movement SSR improves while left SSR never rises above 70% of the movement’s signal capacity. With
25% lefts (c), left turn SSR nearly reaches signal capacity at the expense of through vehicles within the
approach.
These trends demonstrate the sensitivity of the model to both spillback and starvation effects, which
begin to affect throughput for a 100’ pocket when demand levels reach approximately 80% of the
movements’ signal capacity. As shown, the loss in throughput can be significant with maximum SSR values
10-30% below expected approach capacity. It should be noted that as demand begins to significantly exceed
capacity, results become highly variable due to the mesoscopic approach.

Pocket Length Sensitivity

Fig.5 presents SSR by movement for 5 different pocket lengths, with a fixed signal timing plan as indicated
above, and 20% left turn demand. The lines marked by diamonds in (a) and (b) represent the 100’ pocket
case displayed in Fig.4 (b), with maximum SSR values less than 85% of signal capacity. The extent of
throughput reduction due to spillback and pocket starvation is even more severe in the case of a 50’ pocket,
with both left and through SSR values peaking around 65% of signal capacity. With longer pockets, the
blockage effects are much less significant, and the through movement for the 500’ pocket essentially tracks
signal capacity in under-saturated conditions.
As noted previously, when demand begins to exceed capacity, the results generated using a
mesoscopic approach become highly variable. The point of departure from the signal capacity line therefore
provides the most reliable estimate of SSR in each case, as increased demand will lead to essentially
continuous queuing due to the onset of severe spillback and blockage effects. In the case of extremely short
turn pockets, this peak occurs well below a d/c ratio of 1.0, as very little demand per cycle is necessary to
generate spillback and blockage. Very long pocket lengths capable of storing the entire per-cycle left-turn
demand, on the other hand, essentially prevent the onset of turn pocket effects until demand begins to exceed
capacity. Once a d/c ratio of 1.0 is reached, repetitive cycle failures cause the gradual buildup of queues,
eventually causing pocket spillback, lane blockage, and highly variable discharge estimates.
Clearly, the signal timing plan plays a significant role in the trends observed. With an effective g/c
ratio of 0.2 at a cycle length of 120 seconds, the left turn signal capacity is slightly less than 13 vehicles per
cycle. As such, the two very long pockets have available space to store this left turn demand each cycle with
limited spillback. Only after demand exceeds signal capacity does spillback begin to significantly affect
operation. If the signal is timed in such a way that turn capacity per cycle exceeds available turn pocket
storage, the effects of spillback and starvation can be expected to occur below a d/c ratio of 1.0, with the
extent of the capacity reduction correlated to total turn pocket length. It should be noted, however, that SSR
estimates may be affected by a combination of the signal timing plan and the 6-second discrete time interval
used within DYNASMART. A multi-second interval between updates is necessary to ensure computational
efficiency, but care should be taken in interpreting results if a combination of the signal timing plan, gate
location, and update period appear to generate severe drops in SSR.
Reynolds, Zhou, Rouphail and Li 10

Sensitivity to Phasing Sequence

One final application of the model is the ability to compare the effects of phasing sequence and movement
overlap. This research has focused on model development, and none of the simple cases studied thus far
demonstrated significant differences in SSR based on phase order. However, by nature of the design of the
priority gating mechanism, this mesoscopic approach is sensitive to phase order and overlapping sequences.
Additional research is needed to quantify these effects.

Comparison with Microsimulation Estimates

Validation of any model that explicitly models the effects of turn pocket spillback on total approach SSR is
challenging due to the difficulty of identifying appropriate field data. Messer and Fambro (13) were among
the first to model short pocket effects in a simulation environment, calibrating vehicle spacing through the use
of field observations. Wu, Oppenlander (14), Zhang (15) and others have based analytical models on
simulation due to field data limitations. While Qi et al. (16) warned against using the use of VISSIM (17) as
a proxy for real world observations in the case of short turn pocket effects, they dealt primarily with
probability of spillback and queue length predictions rather than comprehensive SSR analysis that includes
through lane interactions. VISSIM results are presented here as the most representative micro-simulation
model for turn pocket analysis.
Fig.6 compares the average observed throughput over 20 hours for each of 180 different trials
simulated in DYNASMART (x-axis) and VISSIM (y-axis). In VISSIM, very long approach links were
provided to accommodate queuing, and the turn pocket length was adjusted based on vehicle storage rather
than length in order to appropriately match results between modeling platforms. Each trial includes one turn
pocket of varying storage capacity (2, 4, 10, 15, and 20 vehicles), two through lanes, a 120-second cycle with
effective g/c ratios for the left and through movements of 0.2 and 0.4, respectively, and varying demand
levels. The y=x line is provided for visual comparison. With R-squared values ranging from 0.8590 for the
sum of both movements to 0.9393 for the left turning movement only, the models demonstrate very similar
trends. Overall, VISSIM tends to produce higher SSR values than the mesoscopic model developed here,
which may be due in part to DYNASMART’s multi-second update period as well as the lack of a capacity
constraint within VISSIM, resulting in some values exceeding the theoretical signal capacity predictions.

CONCLUSIONS

The mesoscopic model presented above represents an attempt to move beyond some of the simplifying
assumptions necessary for a macroscopic approach while providing a more computationally efficient and
consistent alternative to microscopic simulation. As demonstrated, the proposed model incorporates the four
major components identified as necessary to accurately account for short turn pocket effects: (1) turn pocket
spillback, (2) turn pocket starvation, (3) phasing sensitivity, and (4) lane utilization adjustments. The lane
utilization adjustment, while simplified and insufficient to accurately model vehicles on lane-by-lane basis,
represents an improvement over the required static assumptions within analytical models. Although
macroscopic models may be sufficient for intersection design (18), where the primary focus is to limit the
probability of spillback, a mesoscopic approach may be more appropriate when evaluating strategies to
maximize the operational efficiency of an existing intersection. In some extreme cases, short turn pocket
effects can reduce the effective operating approach capacity of an intersection by 10-20%, which may
significantly impact operation at the corridor level.
Several additional variables will need to be included within the model before a final version can be
adopted. Short right turn lanes, for example, may experience spillback and blockage as well, and although the
effects are typically less drastic due to right-turn-on-red and overlapping phasing, the effects will need to be
addressed. Permitted left turns were ignored here as well, although HCM-based capacity adjustments may
prove to be an effective method of inclusion within the model. Finally, due to the scattered nature of turn
pocket effects research, a comprehensive comparison of all available models, including macroscopic,
Reynolds, Zhou, Rouphail and Li 11

mesoscopic, microscopic, and cell transmission approaches, will be useful moving forward. Certainly a
mesoscopic approach represents a potential improvement over existing macroscropic methods while
providing a simplified and computationally efficient alternative to microscropic analysis.

ACKNOWLDEGMENTS

This research was conducted under the sponsorship of the Strategic Highway Research Program (SHRP2) as
part of C05: Understanding the Contribution of Operations, Technology, and Design to Meeting Highway
Capacity Needs. The authors would like to thank the National Academies for the opportunity to participate in
this research and for the permission to share this modeling framework with the research community. The
authors would also like to thank the other members of the research team, whose contributions and insight
made this research project possible.
Reynolds, Zhou, Rouphail and Li 12

REFERENCES

1. Highway Capacity Manual. Transportation Research Board of the National Academies, Washington
D.C., 2000.

2. Kikuchi, S., M. Kii, and P. Chakroborty. Lengths of Double or Dual Left-Turn Lanes. In
Transportation Research Record: Journal of the Transportation Research Board, No. 1881,
Transportation Research Board of the National Academies, Washington, D.C., 2004, pp. 72-78.

3. Wu, N. Total Approach Capacity at Signalized Intersections with Shared-Short Lanes – A


Generalized Model Based on Simulation Study. In Transportation Research Record: Journal of the
Transportation Research Board, No. 2027, Transportation Research Board of the National
Academies, Washington, D.C., 2008, pp. 19-26.

4. Ackelik & Associates. aaSIDRA User Guide. Akcelik and Associates Pty Ltd., Melbourne, Australia,
2004.

5. Jayakrishnan, R., H. S. Mahmassani and T. Y. Hu. An Evaluation Tool for Advanced Traffic
Information and Management Systems in Urban Networks. In Transportation Research Part C, Vol.
2, No. 3, 1994, pp. 129-147.

6. Mahmassani, H. S. Dynamic Traffic Simulation and Assignment: Models, Algorithms, and


Application to ATIS/ATMS Evaluation and Operation. In Labbé, M., Laporte, M.G., Tanczos, K.,
Toint, P. (Eds.), Operations Research and Decision Aid Methodologies in Traffic and Transportation
Management, NATO ASI Series F: Computer and Systems Sciences (166), Springer, Berlin, 1998,
pp. 104-135.

7. Lighthill, M.J. and G.B. Whitham. On Kinematic Waves II: A Theory of Traffic Flow on Long
Crowded Roads. In Proceedings of the Royal Society, London Ser. A., Vol. 229, No. 1178, 1955, pp.
317-345.

8. Richards, P. I. Shock Waves on the Highway. In Operation Research, Vol. 4, 1956, pp. 42-51.

9. Chang, G.L, H. S. Mahmassani, and R. Herman. Macroparticle Traffic Simulation Model to


Investigate Peak-Period Commuter Decision Dynamics. In Transportation Research Record: Journal
of the Transportation Research Board, No. 1005, Transportation Research Board of the National
Academies, Washington, D.C., 1985, pp. 107- 121.

10. Ben-Akiva M., M. Bierlaire, H. Koutsopoulos and R. Mishalani. DynaMIT: A Simulation-based


System for Traffic Prediction and Guidance Generation. Presented at TRISTAN III, San Juan, Puerto
Rico, 1998.
11. Tian, Z. and N. Wu. Probabilistic Model for Signalized Intersection Capacity with a Short Right-Turn
Lane. In Journal of Transportation Engineering, Vol. 132, No. 3, 2006, pp. 205-212.
Reynolds, Zhou, Rouphail and Li 13

12. Daganzo, C. The Cell Transmission Model. Part II: Network Traffic. In Transportation Research B,
No. 29, 1995, pp. 79–93.

13. Messer, C.J. and D.B. Fambro. Effects of Signal Phasing and Length of Left-Turn Bay on Capacity.
In Transportation Research Record: Journal of the Transportation Research Board, No. 644,
Transportation Research Board of the National Academies, Washington, D.C., 1977, pp. 95-101.

14. Oppenlander, J.C. and J.E. Oppenlander. Storage Requirements for Signalized Intersection
Approaches: Uniform Arrivals. In ITE Journal, Vol. 69, No. 11, 1999, pp. 84-88.

15. Zhang, Y. and J. Tong. Modeling Left-Turn Blockage and Capacity at Signalized Intersection with
Short Left-Turn Bay. In Transportation Research Record: Journal of the Transportation Research
Board, No. 2071, Transportation Research Board of the National Academies, Washington, D.C.,
2008, pp. 71-76.

16. Qi, Y., L. Guo, L. Yu and H. Teng. Estimation of Design Lengths of Left-Turn Lanes. Presented at
the 88th Annual Meeting of the Transportation Research Board, Washington, D.C., 2009

17. PTV AG. VISSIM. CD-ROM. Karlsruhe, Germany.

18. Gattis, J. L. Turn Lane Storage Length Design: Theory for the Practitioner. In Transportation
Research Record: Journal of the Transportation Research Board, No.1737, Transportation Research
Board of the National Academies, Washington, D.C., 2000, pp. 84-91.
Reynolds, Zhou, Rouphail and Li 14

LIST OF FIGURES

Figure 1 - Illustration of a link with dual left-turn pockets


Figure 2 - Illustration of left-turn pocket blockage events
Figure 3 - Illustration of through blockage events
Figure 4 - SSR by movement (100’ pocket)
Figure 5 - Sensitivity to turn pocket length (20% left turn demand, 120 second cycle)
Figure 6 - Mesoscopic vs. microscopic comparison of 180 trials
Figure 1 - Illustration of a link with dual left-turn pockets
Figure 2 - Illustration of left-turn pocket blockage events
Figure 3 - Illustration of through blockage events
Figure 4 - SSR by movement (100’ pocket)
Figure 5 - Sensitivity to turn pocket length (20% left turn demand, 120 second cycle)
Figure 6 - Mesoscopic vs. microscopic comparison of 180 trials

View publication stats

You might also like