Content uploaded by Willem Klumpenhouwer
Author content
All content in this area was uploaded by Willem Klumpenhouwer on Oct 13, 2020
Content may be subject to copyright.
THE IMPACT OF SHARED CORRIDORS ON INTERCITY PASSENGER RAIL1
RELIABILITY IN CANADA2
3
4
5
Willem Klumpenhouwer, PhD6
Postdoctoral Fellow7
Department of Civil and Mineral Engineering8
University of Toronto, Canada9
10
11
Word Count: 5678 words +2 table(s) ×250 =6178 words12
13
14
15
16
17
18
Submission Date: July 18, 202019
W. Klumpenhouwer 2
ABSTRACT1
This paper investigates passenger rail reliability in Canada, where long-distance and intercity pas-2
senger rail service operated by VIA Rail Canada runs almost exclusively on track that they do not3
own. As pressure for greener long-distance travel options mount, VIA is facing reliability chal-4
lenges attributed primarily to the Canadian operating model and has pushed for separate rights5
of way for passenger rail service to accommodate growing demand and higher expectations for6
reliability. To investigate the effects of mixed-traffic operations versus dedicated right of way7
operations in a Canadian context, this paper performs a comparative reliability analysis of two8
corridors - one that is owned entirely by VIA Rail, and one that is not. Real-time and scheduled9
arrival data published by VIA rail and collected over the course of a year and a half is used to de-10
velop an understanding of delay propagation on the line and to learn about the effects of reliability11
bottlenecks on the corridors. Finally, a mathematical model of stochastic movement of trains on12
a route is used to examine potential benefits of targeted reliability improvements and to provide13
direction for rail operators on where to focus improvements.14
15
Keywords: Intercity rail, Reliability, Markov chains, mixed-traffic rail operations16
W. Klumpenhouwer 3
INTRODUCTION1
Outside of commuter rail in larger cities, Canada’s passenger rail service is operated by VIA Rail2
Canada (VIA), an independent Crown corporation mandated to provide intercity rail service across3
the country. Despite increasing popularity demonstrated by steadily growing ridership(1), on-time4
performance across the intercity rail network has stayed below 75% (2). Combined with the growth5
of cities and regions and their continued struggle with congestion, road safety, and climate change,6
VIA’s passenger service is becoming increasingly vital to connect cities in a safe, sustainable, and7
efficient way. Information on the economic impact of passenger train delays is not available in8
Canada, however VIA highlights delays as an increased risk in their most recent annual report9
(3), adding that “VIA Rail’s agreements with major third-party infrastructure owners will expire10
in 2021 and at this point the Corporation has no visibility on the terms and conditions of those11
agreements.”12
Similar to Amtrak in the United States, the large majority of VIA’s service operates on a13
rail network which they do not own. Sharing a corridor with freight movements which may be pri-14
oritized by the track owner can lead to significant delays and unreliabile service. With explicit ref-15
erence to these reliability issues, VIA has pushed to expand their service in the busiest part of their16
network known as Corridor by proposing a dedicated passenger rail line in the Toronto-Ottawa-17
Montréal corridor parallel to the existing route (4). The extent of VIA’s struggle with operational18
reliability and freight operators is articulated in VIA’s 2016-2020 corporate plan summary (5):19
VIA Rail can no longer function within its existing framework. The current rela-20
tionship with freight rail operators no longer works as VIA Rail cannot control the21
fundamentals required to operate efficiently in a commercial manner in a competitive22
environment. No intercity passenger rail carrier among the G7 countries (and almost23
none in the G20), is burdened with the constraints and barriers faced by VIA Rail and24
as such, VIA Rail cannot thrive and bring the socio-economic benefits associated with25
passenger rail travel gained elsewhere in the world to Canada and to Canadians.26
Published reliability information does not include enough analysis or insight to support this27
strong statement, and it is unclear in many cases what the true impact of mixed operations is on28
the unreliability of individual trains, compared with other sources of delay such as in-station delay.29
There is little to no discussion of the cumulative impacts of delays along a route or the existence of30
areas in certain corridors where delays are common and unpredictability is high (referred to in this31
paper as “reliability bottlenecks”). Without this information, it is not clear to what extent mixed32
traffic impacts operations, and how much of VIA’s typical delays are caused by other factors such33
as delayed departures, poor scheduling, dwell times, or mechanical issues.34
Detailed information on breakdowns, individual station events, and the movement of rail35
traffic in mixed corridors is not made publicly available in Canada. This paper instead utilizes a36
more widely available set of data: Real-time and scheduled arrival and departure information. By37
collecting this data over an extended period of time, a comparative analysis of two corridors is38
possible: The Brockville-Ottawa-Coteau portion of the Toronto-Ottawa-Montréal corridor, which39
includes track owned and used exclusively by VIA, and the Toronto-Kitchener-London corridor,40
which operates entirely in mixed traffic and has a number of potential reliability bottlenecks (see41
Figure 1 for context). By performing a comparative analysis of these two corridors, the impacts of42
different rail corridor use and configurations on reliability can be better quantified and discussed.43
W. Klumpenhouwer 4
FIGURE 1: VIA’s Corridor
As with many aspects of transportation, the 2020 Coronavirus pandemic has brought the1
importance of rail travel into sharper focus. While ridership and levels of service have temporarily2
decreased in conjunction with social distancing measures and protocols, intercity passenger rail3
may see increased demand as travelers look for alternatives to air travel within the country and4
province. This, combined with a steady push to reduce transportation emissions in Canada may5
push governments to increase the viability and quality of intercity rail transportation.6
The goal and contribution of the paper is to apply a number of reliability measures and a7
stochastic process model to archived arrival and departure information to determine the location8
and impact of reliability issues on intercity rail corridors which run relatively infrequent service.9
After a brief review of literature on intercity rail reliability to conclude this section, the data col-10
lection and analysis tools are described (Methods) and the findings of these analysis tools and a11
discussion of their impact and implications are presented (Results and Discussion). Finally, some12
closing thoughts and suggestions for future work on intercity passenger rail reliability in a North13
American context are discussed (Conclusion).14
Studies on Intercity Rail Reliability15
Studies on passenger rail reliability often focus on an urban setting, where high-frequency service16
can cause significant knock-on delays (6, 7). Studies that do look at cases of long-distance travel17
and mixed freight and passenger operations focus on simulation methods to quantify performance18
of the system and aid dispatchers (8–10). While these studies offer valuable modeling results and19
advice, they are concerned with network optimization while this paper focuses on observation and20
analysis of existing conditions.21
Martland (11) investigated on-time performance of long-distance passenger trains in the22
United States, focusing on the perspective of the “host” railroad, namely the railroad that owned23
the track and did not operate the passenger rail service. In many cases it was found that scheduled24
running times on routes set by the passenger carrier Amtrak were unrealistically close to the fastest25
W. Klumpenhouwer 5
possible running times and did not leave time even for small delays. Amtrak also did not adapt1
their schedule for planned track maintenance, temporary slow orders, or other temporal and spatial2
factors, and these adjustments were suggested as a strategy for Amtrak to improve their on-time3
performance. While the study investigates a similar operating environment to that of Canada (pas-4
senger service operating on track owned by freight railways), it does not provide a detailed and5
comprehensive investigation of how and where delays propagate and whether there are specific6
links or areas where reliability is an issue. The paper also uses standard on-time performance met-7
rics as a basis for the investigation, while a more nuanced discussion of reliability made available8
with more data can provide greater insight into the locations and causes of delays. Krier et al. (12)9
took the approach a step further by implementing a more thorough statistical regression analysis of10
Amtrak performance and delays and were able to isolate a “relationship between freight control of11
the infrastructure on which Amtrak trains operate and Amtrak train delays”. They were even able12
to identify that the operations of different railroads affected the delays of Amtrak trains to different13
extents.14
These two analyses, first by Martland (11) and then by Krier et al. (12), are the most15
similar to this paper. Instead of a statistical analysis of on-time performance and a regression of16
delay factors, a quantitative investigation of delay propagation and a stochastic process model of17
train movements are used to model the observed reliability issues in VIA’s service.18
DATA EXPLORATION AND SUMMARY19
As part of their real-time arrival information system, VIA posts actual and scheduled arrival times20
on their website in a tabular format. Information on arrival and departure is accessible on a route21
and date basis, and is held on the site for approximately six months. Scheduled and actual arrival22
times were collected from these published arrivals on a monthly basis for the entire calendar year23
of 2019. A total of 55,833 arrival and departure events were collected across 16 stations in the24
Kitchener and Ottawa portions of the Corridor. For each arrival recorded at a station, a calculated25
difference between scheduled and actual arrival times as well as scheduled and actual departure26
times were calculated to the nearest minute. The distribution of these values are summarized by27
corridor and travel direction in the third and second last rows of Table 1.28
Figure 2 shows the distribution of departure delays at stations along the Kitchener and Ot-29
tawa corridors separated by direction (EB and WB). Individual stations, directions, and trains vary30
in their delays quite substantially, and in the case of the Ottawa corridor many trains enter the cor-31
ridor already significantly delayed during the portion of their trip from Toronto to Brockville (see32
Brockville (EB) in Figure 2). Terminal departure stations such as Toronto (EB) and Ottawa (EB)33
have a very narrow distribution of departure delays, indicating that in general delay is accumulated34
along the route and is a result of departure delay at the origin.35
The distribution is skewed heavily towards late departures, as indicated by the prevalence36
of outliers from the box plot and the fact that the mean sits well above the median in all corridors37
and directions (Table 1). A portion of the trains measured experience delays in excess of 6038
minutes; these trains are considered exceptions that have “run away from the schedule” and are not39
appropriate for modeling and calculating everyday delays. This 60-minute threshold was chosen40
to correspond to VIA’s fare refund policy, where delays in excess of 60 minutes typically qualify41
for a refund of some kind. Individual route trajectories that exceed this 60-minute threshold are42
excluded from the remainder of the analysis; the prevalence of these trains are reported in the last43
row of Table 1. Comparing this statistic across corridors also gives an idea of the number of major44
W. Klumpenhouwer 6
TABLE 1: Comparative Delay Statistics on Kitchener and Ottawa Corridors
Kitchener Ottawa
Eastbound Westbound Eastbound Westbound
Mean (SD) arrival delay (min) 19.85 (25.25) 7.61 (21.93) 12.58 (25.28) 5.83 (15.77)
Median arrival delay (min) 14 2 4 2
Mean (SD) departure delay (min) 18.91 (25.69) 7.4 (21.5) 10.92 (23.85) 4.75 (14.87)
Median departure delay (min) 13 1 2 0
% of arrivals ≤15 min late 54.0 88.77 74.84 90.29
% of departures ≤15 min late 56.94 89.35 78.78 92.40
Number of scheduled trains 2 2 17 19
Number of arrivals recorded 5,024 5,395 9,373 9,970
Number of departures recorded 5,413 6,141 11,403 13,084
% excluded runs 6.50 5.27 4.17 1.37
delays experienced in this corridor.1
There are two additional insights that the summary statistics in table 1 provide: Eastbound2
trains are significantly less reliable than westbound trains, and at first glance, the Ottawa corridor’s3
on-time performance is not significantly better than Kitchener’s. Directional differences stem from4
different operational issues on the corridors: On the Kitchener corridor, single-track operations,5
poor track conditions, and the infrequency of sidings for train meets mean that eastbound trains6
often have to wait for delayed westbound trains early on in their route from London to Toronto,7
compounding the amount of delays along the route. On the Ottawa corridor, trains are delayed8
upstream from the VIA-owned portion of the route between Brockville and Ottawa, meaning the9
trains often enter the corridor already delayed.10
Setting the directional differences aside, the mean and standard deviations of delay on11
the two corridors are similar. For similar operational reasons to the directional differences, these12
summary statistics fail to capture both the effects of delays on trains in the Ottawa corridor experi-13
enced between Toronto and Brockville, and schedules which are unrealistic or have not adapted to14
changing running times. A hint of these differences can be seen in comparing on-time performance15
between the two corridors; eastbound Ottawa corridor trains arrive and depart within 15 minutes16
of their scheduled time more frequently than eastbound Kitchener corridor trains. To account for17
and quantify these potential underlying causes of delay, an analysis of running time behaviour is18
needed.19
RELIABILITY IMPACT ANALYSIS AND MODELING METHODS20
This study employs three different tools to quantify and model reliability and on-time performance:21
buffer time indices, mean and standard deviation of delay growth, and a Markov chain model of22
station-to-station train movement. This section outlines these three methods in detail and the results23
are discussed in the subsequent section. On-time performance, as calculated in Table 1, is typically24
measured as a percentage of trains classified as “on-time” by reaching a destination under a certain25
threshold (e.g. 15 minutes). This does not fully capture the impact of late trains on passengers, as26
a train that is 20 minutes late is counted the same as a train that is 60 minutes late. The following27
W. Klumpenhouwer 7
FIGURE 2: Departure delay distribution at Kitchener and Ottawa corridor stations, by direction
methods are used to gain a more nuanced understanding of the delay characteristics of VIA trains.1
W. Klumpenhouwer 8
Buffer Time Indices1
To measure the impacts of travel time variation on individual links in the network, a modified
version of the buffer time index from vehicle travel is employed. For a given link ibetween a pair
of stations, the schedule buffer time index Biis defined as (13):
Bi=t95 −ts
ts
(1)
where t95 is the 95th percentile of actual travel time on the link, and tsis the scheduled travel time2
based on the difference between the scheduled arrival time at the second station and the scheduled3
departure time at the first. This formulation can produce both positive and negative buffer time4
indices, the latter indicating that the scheduled link travel time is scheduled such that the train will5
almost always run faster than normal.6
An alternative to a buffer time index is a mean buffer time index, which considers the
percent difference between the 95th percentile and the mean running time:
Ri=t95 −tµ
tµ
(2)
where tµis the mean link travel time. This measure focuses on link reliability and travel variation,7
and removes cases where the scheduled travel time does not match the mean travel time. It does8
however introduce cases where the average running time on the link is shorter than the scheduled9
time. Unlike Bi,Ri≥0 always. While the value of each of these buffer indices provide a quan-10
tifiable measure of segment reliability, the comparison of the two values can provide insight into11
whether there is a mismatch between the scheduled and mean running times.12
Mean and Standard Deviation Growth13
The cumulative mean and standard deviation growth along a route captures both how delay and14
randomness grows due to the cumulative random effects encountered such as boarding passengers,15
weather, and freight traffic. By plotting the mean and standard deviation of arrivals along each16
route, the downstream effects of individual links can be identified. Where buffer time indices17
capture link-level delay, this method captures the cumulative effect of individual links on the route.18
As with buffer time indices, there are two potential reference points: the schedule, and19
the mean. For the former, the mean delay is expected to fluctuate from stop to stop, typically20
growing later due to the long-tailed distribution of travel times on links, and to the holding strategy21
employed by VIA (trains do not typically depart stations early). If instead mean values are used22
as a stand-in scheduled travel time, the mean value of delay should remain close to zero while the23
standard deviation should grow relatively linearly.24
Cumulative mean and standard deviation growth shows how delay propagates along the25
route as the train progresses from stop to stop. When a certain link or set of segments are respon-26
sible for a large portion of the unreliability or delay on a route, these measurements will show a27
large jump in the standard deviation or mean departure delay. A growing mean indicates “drift”28
exists, where trains are pushed later and later behind schedule, while a growing standard deviation29
indicates that more and more unreliability is being introduced into the system.30
Markov Chain Reliability Model31
Concepts such as standard deviation growth and mean “drift” from the schedule indicate that trains32
moving from station to station as they traverse a route follow a stochastic process. If each segment’s33
travel time is assumed to be independent, this stochastic process follows a Markovian process akin34
W. Klumpenhouwer 9
to Brownian motion with drift. This concept was first introduced in an idealized setting for buses1
by Newell (14) and later adopted to optimize holding control on bus routes by Klumpenhouwer2
and Wirasinghe (15). This Markov chain model is employed here to investigate where reliability3
improvement efforts could have the greatest effect on the system.4
The Markov chain model used is a discrete state and discrete time Markov chain which uses
the deviation between actual and scheduled departure times from a stop ito the deviation at stop
i+1. These deviations δiform the state space for the Markov chain, while transition probability
matrices, which quantify the propagation of delay from one departure to the next, are typically
informed by theoretical travel time distributions based on mean and standard deviations. Once
these transition probability matrices P
i,i+1are constructed, they can be multiplied together, along
with a vector of initial delay distribution P
0, to provide a distribution at any stop. For example, a
vector showing a distribution of travel time deviation to the 4th station (indexed as 3) on a route
would involve the following operation:
P
3=P
0P
0,1P
1,2P
2,3(3)
Each individual probability transition matrix is populated using a truncated log-normal5
distribution, based both on its relatively high goodness of fit with many links, its highly late-6
skewed tail, and previous studies (16, 17). Figure 3 shows three different examples of how the7
distribution of travel time on links fits with a log-normal distribution. VIA trains do not typically8
depart early from stations, and as such a truncated log-normal distribution is created so that all9
of the probability for states where δi<0 is added to the state δi=0. This models the effects10
of holding a train that arrives early to a station until the appropriate departure time. In order to11
mitigate the potential effects of poor scheduling that has not adapted for changing track conditions12
or other factors, the model considers deviations from the mean delay instead of from the schedule.13
The model uses running time distributions on links to create a theoretical distribution and populate14
the state space.15
This Markov chain model is used to provide insight into the propagation of delays and the16
profile of reliability on a route without requiring detailed rail simulation along a corridor. These17
rail corridors are long and contain some complex and often changing elements related to track, and18
this Markovian approach allows for the incorporation of all these elements into a single stochastic19
model. Areas which are identified as reliability bottlenecks can be studied further using detailed20
rail simulation to determine how best to address the local issues.21
Passenger Demand and Delay Estimation22
Using passenger boarding data, the measure of existing conditions as well as the modeling of the
impacts of potential changes in reliability on portions of the corridor can be translated into an
estimated delay to passengers. While only boarding data is available, this can be used to provide
a relative estimate of the level of waiting time at stations. In each case, the mean adjusted delay is
weighted by the annual boardings biat a given station ito compute a total delay Don the corridor:
D=∑
i
bimax{(tµ−ts),0}(4)
This provides an annual total delay cost at a given station and on a given corridor. As Toronto23
serves multiple routes not included in this analysis and a breakdown of route-level ridership is not24
available, it is excluded from this portion of the analysis. Boardings at Malton are not reported and25
are therefore also excluded.26
W. Klumpenhouwer 10
FIGURE 3: Travel time distribution on three representative links, fit with a log-normal distribu-
tion.
This delay estimation is limited by the quality and availability of data on passenger move-1
ments, and should be taken as an illustrative example of the impacts of delays on overall corridor2
appeal. Boarding and alighting data, along with stop-to-stop flows of passenger would allow for3
a more comprehensive analysis of the impacts of delays. Unfortunately, this data is not publicly4
available.5
RESULTS AND DISCUSSION6
Buffer Time Indices7
Figure 4 shows the two buffer time indices Biand Rifor individual station pairs, sorted by increas-8
ing Bi. Station pairs on the west end of the Ottawa corridor trend to smaller values of both Bi
9
and Ri, signaling better overall reliability. On the Kitchener corridor, travel between Georgetown,10
Brampton, and Malton shows large relative differences between Biand Ri, with Bi>Riin all cases.11
Not only is travel time highly variable in this area, it also suffers from poor scheduling; scheduled12
travel times could be adjusted to bring Riand Bimore in line. While the exact causes of the travel13
time variation in this area is not known without delay-specific records, the area between Bramp-14
ton, Georgetown, and Guelph contains a rail crossover where freight traffic from a busy yard and15
passenger rail must cross paths. This area is the subject of the case study below.16
Kitchener-Guelph travel times are an example where running times on the link are typically17
faster than scheduled, resulting in Bi<0. This link is a specific example where travel times have18
been improved due to improvements made by the track owner, but running time schedules have not19
been adjusted. Scheduled travel times could be adjusted to accommodate these decreased travel20
times for greater reliability. In many cases, an increased scheduled travel time in the range of 10-21
20% would bring the scheduled buffer time index more in line with the mean buffer time index.22
As VIA’s schedules have remained relatively static over a number of years, it is possible that this23
required change reflects deterioration of operations in the corridor.24
W. Klumpenhouwer 11
FIGURE 4: Buffer time indices (scheduled and mean travel times) for station pairs in the Ottawa
and Kitchener corridors. Ottawa corridor station pairs are outlined in blue, Kitchener station pairs
are outlined in orange.
W. Klumpenhouwer 12
Mean and Standard Deviation Growth1
Figure 5a shows the mean departure delay by individual train number as it traverses its route. East-2
bound (even-numbered) routes on the Ottawa corridor begin with a large initial delay upon entering3
the corridor, having accumulated this delay traveling from Toronto along the shared passenger-4
freight corridor before leaving the shared corridor for VIA-owned track. In almost all cases, east-5
bound trains gain very little delay as they traverse the VIA-owned portion of the corridor.6
In contrast, trains on the Kitchener corridor experience fluctuations in their delay as they7
traverse their route. This demonstrates that the nature of the corridor is such that trains are often8
consistently running behind and recovering at different portions of the route. This points to both9
a need for potential changes in scheduling and to the presence of reliability bottlenecks. Note10
that eastbound trains (84 and 88) experience their delay accumulation early in the route, while11
westbound trains (85 and 87) experience it later in the route. Both of these spikes in mean delay12
occur between Kitchener and London. In some cases, these train delays interact due to the single-13
track configuration of the majority of the route: Train 88’s delay between stops 0 and 1 (London14
to St. Mary’s) is highly correlated with train 87’s delay, as the two trains must meet at a single15
available rail siding. As train 87 has priority, its delay growth is relatively small in that section16
(sequence 7 to 8) while train 88’s mean growth is extremely high.17
Figure 5b similarly plots the standard deviation of departure delay as a relative measure of18
departure randomness along the route. The figure tells a similar story to the mean growth. For19
Eastbound trains from Toronto almost all of the randomness is gained in the first segment, and20
a large portion of the randomness is attributable to inconsistent departures from Toronto. In the21
Ottawa corridor, some trains are able to gain time and recover from delays accumulated along the22
shared portion of the trip. This lends validity to the claim that VIA-owned exclusive passenger23
track is smore reliable than the mixed-traffic segments of its route. In contrast, Kitchener trains24
see a relatively steady and linear increase in randomness as the route is traversed, for trains in both25
directions.26
Markov Chain Modeling27
Transition probability matrices were calculated for each scheduled train by obtaining the mean and28
standard deviation of travel time distributions for each link, and generating a log-normal distribu-29
tion for each stop-to-stop transition. This log-normal distribution was discretized into transition30
probabilities to determine appropriate probabilities for each δi→δi+1pair. Each row of the matrix31
P
i,i+1therefore becomes a distribution of transition probabilities from a given δito any accessible32
state δi+1.33
Modeled results from the application of the Markov chain model, analogous to those in34
Figure 5, are shown in Figure 6.1As the model uses the mean travel time as a basis for stop-to-35
stop stochastic delay modelling, it is not dependent on a scheduled time which may not follow36
average travel times. The model also smooths out jumps in delay or travel times due both to its use37
of a theoretical travel time distribution (log-normal) and to its reliance on cumulative delay effects.38
This can also account for the model’s higher overall standard deviation for stops later in the route.39
The shape of the model’s cumulative mean and standard deviation of delay is what provides40
insight as to the nature of delays on the corridor. Figure 6b demonstrates, similar to Figure 5b, that41
1The model considers departure to departure times, which incorporate both dwell times and travel times. This,
combined with the truncated route for the Ottawa corridor north of Belleville, means that the number of stops will be
fewer than those in Figure 5.
W. Klumpenhouwer 13
(a) Mean Delay Growth (b) Standard Deviation of Delay Growth
FIGURE 5: Mean and standard deviation of departure delay as a function of stop sequence on the
route. Routes are labelled by route number; Ottawa corridor trains are in blue, Kitchener corridor
trains are in orange. Even numbered trains are Eastbound, odd numbered trains are Westbound.
the initial randomness incurred along the corridor both east of Ottawa for westbound trains and1
west of Brockville for eastbound trains is recovered somewhat in the corridor. Overall, the shape2
of the growth in the model reflects the trends present in the data presented in Figure 5.3
Train 88 is an important outlier in the modeled data. The train experiences frequent and4
substantial delays on its way from London to Kitchener, as indicated by its starting delay and5
standard deviation being quite high and growing quickly. Closer to Toronto, the train is able to6
maintain its status due in part to its late-night arrival time. Occasionally, the train is so delayed that7
it encounters increased freight traffic at night which can cause further delays.8
Incorporating Passenger Boarding9
Figure 7 shows boarding counts and resulting estimated delay for the stations studied, These sta-10
tions combined with the measured trains for an estimated 163,000 hours of annual delay at stations11
due to late trains, with just over half of the delay occurring on the Ottawa corridor. Comparing12
figures 7a and 7b, the delays on the Kitchener corridor are enough to affect the overall rankings of13
delays, despite Ottawa’s significantly higher boarding volume. The discrepancy between the mean14
and scheduled travel time buffer time indices along the Ottawa corridor, particularly on the Smiths15
Falls - Brockville and Fallowfield - Ottawa segments provides a good starting point for reducing16
overall delay on the corridor.17
Case Study: Improving the Georgetown-Brampton Link18
The portion of track between VIA’s Georgetown and Brampton stations contains a crossover heav-19
ily used by rail freight operations (18). Freight typically travels on the Halton subdivision from the20
W. Klumpenhouwer 14
(a) Mean Delay Growth (b) Standard Deviation of Delay Growth
FIGURE 6: Modeled mean and standard deviation of departure delay as a function of stop se-
quence on the route. Routes are labelled by route number; Ottawa corridor trains are in blue,
Kitchener corridor trains are in orange. Even numbered trains are Eastbound, odd numbered trains
are Westbound.
(a) Annual boardings by station (b) Annual boarding delay estimate
FIGURE 7: Annual boardings and associated estimated delay by stations, 2018
Concord intermodal yard to the east before continuing south just west of Georgetown (see Figure1
W. Klumpenhouwer 15
8). VIA and Metrolinx’ GO rail service operates on the portion of the Halton subdivision that runs1
through Bramalea, Brampton, and Georgetown. VIA and GO train service continue west on the2
Guelph subdivision. The Halton subdivision is owned and controlled by the freight railway Cana-3
dian National Railway (CN). This creates a business incentive for CN to prioritize their freight4
traffic ahead of third-party passenger rail.5
Operational issues in the area are well known by all parties. A plan was underway to6
construct a freight bypass in the area to allow for improved passenger rail reliability and expanded7
service, however this plan was scrapped in favour of continued negotiations with CN to manage8
shared railway service (19). While this study does not include Metrolinx trains, the link reliability9
between Georgetown and Brampton as well as Brampton and Malton to the east of Bramalea is10
one of the lowest in the study area (see Figure 4).11
FIGURE 8: Context map of Halton subdivision freight crossover. Track used by VIA is in orange,
other track is in black.
Vromans et al. (20) showed that moving from a heterogeneous mixture of railway traffic12
to a homogeneous one can result in a dramatic improvement in delays, upwards of 66-77 percent,13
in large part due to decreases in secondary delays as a knock-on effect from large primary delays.14
This reduction informs the value used in this case study. The goal of this case study is to examine15
the downstream effects of reliability improvements on a single link and to demonstrate how the16
model can provide insight into the relative improvement of reliability. To simulate a move towards17
a more homogeneous mixture of traffic (either by construction of a bypass or by relegating the18
movement of freight trains to times when passenger rail is not running), the standard deviation of19
travel time on the Georgetown-Brampton link was halved, and the original model is compared with20
the updated one. While this choice of reliability improvement on the link is somewhat arbitrary, it21
W. Klumpenhouwer 16
is based on the improvement seen by Vromans et al. (20) and is chosen as an illustrative example1
of how the effects of improvements can be captured by the Markovian model.2
(a) Mean growth (b) Standard deviation growth
FIGURE 9: Comparative modeled mean and standard deviation growth. Dashed lines are the
original model (Figure 6), solid lines are with improvements on the Georgetown-Brampton link.
TABLE 2: Comparative Model Delay Statistics on Kitchener Corridor
Train Original Reliability Improved Link % change
Maximum Mean (SD) Delay (min) Minimum Mean (SD) Delay
84 4.61 (4.74) 4.49 (4.67) 2.67 (1.26)
85 3.73 (4.13) 3.49 (3.92) 6.57 (5.22)
87 5.88 (6.07) 5.78 (5.99) 1.69 (1.22)
88 15.69 (13.00) 15.69 (13.00) 0 (0)
Discussion and Policy Implications3
Taken together, Figures 4 and 5 indicate that, when accounting for cumulative reliability, reliabil-4
ity in the Ottawa corridor is better than that of the Kitchener corridor. By and large, delays are5
accumulated on sections of track that are shared with freight rail. Given the high boarding values6
at the Ottawa station leading to the majority of the delay experienced by passengers on both of the7
study area corridors, efforts should be focused on improving upstream reliability between Toronto8
and Brockville. This is in line with VIA’s current efforts.9
Table 2 shows that the downstream effects of single-link improvements can be significantly10
less than the improvement applied to the link, although they still provide an improvement in the11
W. Klumpenhouwer 17
final delay and the final reliability of the train at the end of the route. Even a two-minute improve-1
ment per train can make a difference over the course of a year and over tens of thousands of annual2
boardings and alightings.3
The case study highlights an important conflict point in the Kitchener line rail network.4
This, combined with the study area reliability analysis points to the need for changes in the way5
that freight and passenger railway are mixed. In cases where the configuration of the track, existing6
urban form, and cost prohibit separation of passenger and freight rail traffic, competing business7
priorities will likely favour the company that is responsible for rail traffic control on the track. It8
is possible that separating the operational control of trains, the ownership of tracks, and the train9
operators themselves would improve fairness and allow for improved passenger rail reliability.10
Further study on the impacts of the current rail operational policy on reliability is needed.11
CONCLUSION12
Employing reliability-focused statistical analysis of passenger rail travel times on two segments of13
VIA’s Corridor can provide a comprehensive comparison of operations between different operating14
conditions. The Kitchener and Ottawa corridors differ in track conditions, rail traffic mixture, and15
dispatching control. This paper demonstrates some of the insights that are available by doing such16
an analysis using only real and scheduled arrival and departure time data. This analysis was done17
both to demonstrate the type of analysis possible with limited data as well as to provide motivation18
for further study on the impacts of mixed freight and passenger rail operations in Canada.19
These initial results indicate that despite initial descriptive statistics indicating otherwise,20
trains on VIA’s owned corridor between Brockville and Ottawa experience better reliability than21
those in the Kitchener corridor. It also indicates that a large portion of the randomness in travel22
times occur outside of the Ottawa corridor, where VIA trains operate mixed with freight. By23
modeling the Kitchener corridor using a stochastic Markov chain model of train movement, the24
downstream effects of single link reliability improvements can be captured and demonstrated. In-25
corporating the data into a model of train movement can also help policy makers and analysts26
pinpoint where reliability (and policy) improvements can do the most good.27
There is more work to be done establishing the effects of Canada’s rail policies on the28
successful operations of passenger rail. With additional information on trip patterns, a more com-29
prehensive quantification of the impacts of unreliable service is possible. Expanding this analysis30
to other areas including commuter rail services in Toronto, Montreal, and Vancouver would provide31
a more detailed insight into these corridors. Additional, detailed modeling of specific reliability32
improvements that capture train movements and dynamics through an area can also provide a more33
focused understanding of the opportunities for rail improvement.34
VIA’s operations are impacted daily by its need to run the majority of its operations on35
track owned and dispatched by a major freight railway. These impacts of long hindered VIA’s36
ability to improve its reliability in key corridors. By providing detailed insight into reliability on37
these corridors, a more nuanced conversation of rail reliability can occur.38
REFERENCES39
1. VIA Rail Canada, VIA Rail Reports Ridership and Revenue Increase for a Fourth Consec-40
utive Year, 2019.41
2. VIA Rail Canada, Annual Report 2018. VIA Rail Canada, 2019.42
3. VIA Rail Canada, Annual Report 2019. VIA Rail Canada, 2020.43
W. Klumpenhouwer 18
4. VIA Rail Canada, Proposal for High Frequency Rail in the Québec City - Toronto Corri-1
dor, 2020.2
5. VIA Rail Canada, Summary of the 2016-2020 Corporate Plan and 2016 Operating and3
Capital Budgets. VIA Rail Canada, 2016.4
6. Saidi, S., N. H. M. Wilson, H. N. Koutsopoulos, and J. Zhao, Mesoscopic Modeling of5
Train Operations : Application to the MBTA Red Line. In 2019 IEEE Intelligent Trans-6
portation Systems Conference (ITSC), 2019, pp. 98–103.7
7. Higgins, A. and E. Kozan, Modeling Train Delays in Urban Networks. Transportation8
Science, Vol. 32, No. 4, 1998, pp. 346–357.9
8. Huisman, T. and R. J. Boucherie, Running times on railway sections with heterogeneous10
train traffic. Transportation Research Part B: Methodological, Vol. 35, No. 3, 2001, pp.11
271–292.12
9. Büker, T. and B. Seybold, Stochastic modelling of delay propagation in large networks.13
Journal of Rail Transport Planning and Management, Vol. 2, No. 1-2, 2012, pp. 34–50.14
10. Murali, P., M. Dessouky, F. Ordóñez, and K. Palmer, A delay estimation technique for15
single and double-track railroads. Transportation Research Part E: Logistics and Trans-16
portation Review, Vol. 46, No. 4, 2010, pp. 483–495.17
11. Martland, C. D., Improving On-Time Performance for Long-Distance Passenger Trains18
Operating on Freight Routes. Journal of the Transportation Research Forum, Vol. 47,19
No. 4, 2010, pp. 63–80.20
12. Krier, B., C. M. Liu, B. McNamara, and J. Sharpe, Individual freight effects, capacity uti-21
lization, and Amtrak service quality. Transportation Research Part A: Policy and Practice,22
Vol. 64, 2014, pp. 163–175.23
13. Gittens, A. and A. Shalaby, Evaluation of bus reliability measures and development of a24
new composite indicator. Transportation Research Record, Vol. 2533, No. 416, 2015, pp.25
91–99.26
14. Newell, G. F., Unstable Brownian Motion of a Bus Trip. In Statistical Mechanics and27
Statistical Methods in Theory and Application, Plenum Press, New York, 1977, pp. 645–28
667.29
15. Klumpenhouwer, W. and S. C. Wirasinghe, Optimal Time Point Configuration of a Bus30
Route – A Markovian Approach. Transportation Research Part B: Methodological, Vol.31
117, No. A, 2018, pp. 209–227.32
16. Kaparias, I., M. G. H. Bell, and H. Belzner, A New Measure of Travel Time Reliability33
for In-Vehicle Navigation Systems. Journal of Intelligent Transportation Systems, Vol. 12,34
No. 4, 2008, pp. 202–211.35
17. Bergström, A. and N. A. Krüger, Modeling Passenger Train Delay Distributions - Evidence36
and Implications. Centre for Transport Studies, Stockholm, 2013.37
18. Metrolinx, Kitchener GO Rail Service Expansion Initial Business Case Update. Metrolinx,38
2019.39
19. Bueckert, K., Province says they’ve developed alternative to bypassing 30-kilometre track40
owned by CN, 2018.41
20. Vromans, M. J., R. Dekker, and L. G. Kroon, Reliability and heterogeneity of railway42
services. European Journal of Operational Research, Vol. 172, No. 2, 2006, pp. 647–665.43
