Cycle route network development and evaluation

using spatial multi-criteria

analysis and shortest path analysis
Michael Vorster Pr Eng
1. Define perspective and associated qualities
Principal Engineer
Aurecon South Africa (Pty) Ltd Define
[email protected] Identify analysis perspective
perspective qualities

Prof Mark Zuidgeest 2. Criteria and data identification

Department of Civil Engineering
University of Cape Town Identify Data acquisition
[email protected] assessment and (pre)
criteria processing

INTRODUCTION
3. Weightings
The planning of cycle routes should ideally
be open and transparent, involving nu-
Acquire weights
merous stakeholders, so as to adopt a holistic (stakeholders)
approach and facilitate buy-in to the plan-
ning process. Moreover, it should be focused
around the needs of cyclists, with routes
exhibiting the qualities necessary to satisfy Formulate policy
both existing cyclists, as well as to encourage vision (experts)
cycling as a viable alternative mode of
transportation. To create a cycling-inclusive
environment, cycle routes should exhibit the
4. Geo-spatial data processing
following qualities (Land Transport Safety
Authority (NZ) 2004; National Department Spatial multi- Assess existing or
criteria analysis Network predetermined
analysis
If ink/s added, redo

of Transport (RSA) 2014): (SMCA) routes

network analysis

QQ Safety and security: Routes should be

made safe by limiting conflict between
cyclists and others, including motor- 5. Cycle route directness check
ised traffic and pedestrians, and also
Propose
by providing personal security against Cycle route additional link/s
directness check
theft or attack. for evaluation
QQ Comfort: In terms of design, gentle
slopes should be provided, where pos-
sible complicated manoeuvres avoided, Sensitivity and Optimal Compare
uncertainty cycle
and frequent stopping at inter­sections analysis routes
minimised. Routes should also be
smooth and non-slip.
QQ Directness: Routes should be Figure 1 Basic flowchart of the proposed method (adapted from Keshkamat et al 2009)
direct and based on desire lines, with
minimal door-to-door delays. QQ Coherence: Cycle routes should link where possible. They should also con-
all potential origins and destinations, tribute to a pleasant cycling experience
be continuous and recognisable and by integrating with their surroundings,
The original, full-length version of this article provide consistent safety and security enhancing their security and looking
was presented at SATC 2019 where Michael
Vorster received the Runner-up Paper Award
throughout. attractive.
for professionals under 35 years of age. QQ Attractiveness: Routes should pass by The method presented in this article aims
or through aesthetically pleasing areas to provide a tool to identify optimal routes

40 September 2019 Civil Engineering

by quantitatively evaluating all possible
route alternatives along the existing road
N
and/or cycle path network against the route
qualities defined and weighted by stake-
holders. Moreover, where the deviation of a
resulting route is calculated to be indirect,
alternative links may be added and evalu-
ated. This is achieved as the method takes
advantage of the powerful set of tools for
the manipulation and analysis of spatial
information provided by geographical
information systems (GIS), and the tech-
niques available in multi-criteria analysis
(MCA) for structuring decision problems,
and designing, evaluating and prioritising 0 1 2 4 km
alternatives. This combination, known as
spatial multi-criteria analysis (SMCA), can
be thought of as a process that transforms Figure 2 Study area including allotment areas and the road network
and combines geographical data and
the value judgements of stakeholders to to generate an optimal cycle route or institutional bodies (e.g. local govern-
obtain information for decision-making. route network as required. ment). Irrespective of this, if cycling is to
The advan­tages gained from combining 5. Compare the network route length be promoted and adopted as a viable alter-
GIS and MCA result in the development between O-D pairs as a ratio of the native means of transport, it is the user’s
of an effective spatial decision support Euclidian distance, and where the value perspective which should be prioritised.
system (SDSS). exceeds a defined threshold, add a link The case study adopts the perspective
Given the data requirements for (or links) to reduce the network length. of commuter cycling (i.e. utility network).
a comprehensive assessment of route Where additional links are required and General route qualities for this trip type
alternatives, especially at a network level, viable for inclusion, the network analysis include: (1) safety: intersection types
and the likelihood that the data will not be portion of Step 4 is to be redone. that minimise conflict with traffic and
readily available, the method is best suited facilities which give them their own
at a strategic level of planning. STUDY AREA space; (2) security: good lighting for
The case study area selected, illustrated in evening trips; (3) directness: shortest
THE METHOD Figure 2, falls within Port Elizabeth in the possible routes to minimise travel time;
The proposed method applies the princi- Eastern Cape Province of South Africa. (4) c­ oherence: continuous routes and
ples outlined within the Cycle Network and The Cycle Network and Route Planning networks linking as many destinations as
Route Planning Guide (Land Transport Guide (Land Transport Safety Authority possible; and (5) comfort: gentle gradients
Safety Authority (NZ) 2004) to evaluate (NZ) 2004) defines the primary cycle and minimal intersections. The study also
and build a network of cycle routes, using network as the most commonly used cycle includes qualities not considered key for
an adapted version of the method devel- facility, designed mainly for trips between this trip type to encourage cycling for new
oped by Keshkamat, Looijen and Zuidgeest suburbs and across town. Taking this into users (i.e. passing by attractive areas).
(2009). A flowchart of the key steps within consideration, together with the fact that The cycle route qualities, directness
the method is shown in Figure 1. As there is currently a very low demand for and coherence are not easily measured
­i llustrated in the flowchart, the five main commuter cycling, it was assumed for the spatially using the MCA part of the
steps of the bicycle route network design case study that there is an equal latent proposed method. Route directness is
method are: demand to travel between each of the later dealt with as Step 5 of the method
1. Define the perspective (goal) of the suburbs. The exact location within each and coherence is largely dealt with during
analysis and select qualities (subgoals) suburb was taken as its respective centroid, the detail design and implementation
appropriate to it. which is represented by a dot in Figure 2. of routes, where consistency in terms of
2. Identify criteria which represent the standards and materials is important.
qualities chosen, then source and pro- Step 1: Perspective and associated qualities
cess the spatial data that can be used When identifying routes there are often Step 2: Criteria and data identification
to measure these criteria. a variety of stakeholders involved with Each route quality defined in Step 1 can
3. Weight the criteria within each quality varying perspectives. Representatives be represented by one or more criteria and
and the qualities in relation to the from affected stakeholders are to meet each criterion is represented spatially by
perspective based on stakeholder and agree on the qualities to be included a map. Raster maps are used, as weighted
engagement and policy visions. for further investigation. Apart from the scores according to the criteria, can be
4. The above steps are then used to perspective of cyclists, other perspec- applied to the pixels forming the maps,
perform the SMCA. Thereafter the tives may include, for example, those of which is further discussed in the following
network analysis process is performed technical experts, environmentalists and steps. The criteria used for the case study

Civil Engineering September 2019 41

were based on the availability of geo-­spatial higher-order roads are assumed to be trips, which is the focus of the case study.
information, while still being relevant and more likely to have good street lighting. Directness and coherence are not easily
applicable to the defined qualities. Proximity to key destinations such as measured spatially using the proposed
In the SMCA process criteria are large employers, educational institutions, method and are dealt with elsewhere in
classified either as factors or constraints. community amenities and public transport the model.
Factors can be either a benefit or a cost, stations, for example, can be included as cri- For the case study in question, as stake-
depending on how they impact the quali- teria for evaluation in the selection of routes. holder engagement was excluded, arbitrary
ties which they represent. According to This was excluded from the case study given weightings were applied to the criteria and
ILWIS (2015), a criterion that contributes that an equal latent demand was assumed qualities. The weighting of qualities for the
positively to the output is a benefit and between the suburbs in the study area. commuter cyclist perspective was deter-
therefore the higher the value, the better mined in line with the relative importance
it is. A spatial cost contributes negatively Step 3: W
 eighting of route qualities shown in Figure 4. The weightings applied
to the output and therefore the lower the and criteria precede each criterion and quality in the
value, the better it is. Spatial constraints, The weighting of route qualities and cri- development of the criteria tree (Figure 5).
on the other hand, are defined as teria is an important step in the method It is worth highlighting that the environ-
complete no-go areas and these cells are and the point at which stakeholders have mental criteria do not have weightings,
therefore always represented by a zero an influence over the outcome. as these are all constraints and therefore
(ILWIS 2015). A summary of the various Prior to weighting, the relative impor- always represent a zero value.
criteria used and their relationship to the tance of criteria and qualities needs to be
respective qualities is provided in Table 1. determined by the various stakeholders Step 4: Geo-spatial data processing
The dataset used per criterion is via a ranking process. To assist with the
shown in brackets next to the criterion ranking of the cycle route qualities, it is Spatial multi-criteria analysis
name. As can be seen, the same dataset proposed that a hierarchy of qualities key The SMCA process of the method
was used for multiple criteria. With the to cycling and based on Maslow’s hierarchy combines the perspective and associated
steep costs and time often associated with of needs is used as a point of departure, qualities, the criteria and geo-spatial data-
collecting data, it is imperative that ex- following Van Hagen (2015), who did this sets and the weighting process discussed.
isting data be processed in multiple ways for public transport in the Netherlands. The software used in this study is
and used where appropriate. Where it is Figure 3 provides an adapted illustration of ILWIS 3.8.5, which is open-source and
not a perfect representation, it can be used this hierarchy. By substituting the public has a strong SMCA module. To perform
as a point of departure or proxy, which transport qualities with cycle route quali- the analysis, a criteria tree is built in
can be refined as new and more relevant ties, one is able to create a similar pyramid a hierarchical manner, commencing
data becomes available. An example of as illustrated in Figure 4. with the perspective (goal), followed
this in the case study is using road hier- The hierarchy or ranking of qualities by the qualities (sub-goals), and finally
archy as a proxy for street lighting, where in Figure 4 is better suited to commuter factors and constraints, either beneath

Table 1 Summary of the perspectives, qualities and criteria considered in the case study

Perspective Qualities Criteria Relationship of criteria to qualities

Spatial cost. The higher the order of the road, the greater the volume
Road class (road network)
of vehicles, number of large trucks, and speeds.

Spatial cost. The greater the intersection density, the greater the
Intersection density (road network)
Safety and number of potential crossings and points of conflict.
security Spatial benefit. The higher the order of the road, the better the street
Street lighting (road network)
lighting (assumed correlation for case study).

Spatial benefit. The presence of development means potentially greater

Urban development (urban cadastral)
Commuter visibility of cyclists.
cyclist
Gradient (grid survey) Spatial cost. Steeper gradients are less appealing to commuter cyclists.
Comfort
Spatial cost. The greater the intersection density, the more stopping
Intersection density (road network)
and starting is required.

Spatial benefit. Route attractiveness increased by passing close by

Critical biodiversity areas (as labelled above)
areas of natural beauty.
Attractiveness
Spatial benefit. Route attractiveness increased by passing close by
Recreational areas (NMBM land usage)
areas of natural beauty.

Spatial constraint. Future development may occur but it is not

Critical biodiversity areas (as labelled above)
recommended.
Environmental
Environmental Protected areas (as labelled above) Spatial constraint. No future development can occur.
conservation

Wetlands (as labelled above) Spatial constraint. No future development can occur.

42 September 2019 Civil Engineering

 is required. This is necessary as criteria
are typically unrelated and have different
classifications or units of measure and
therefore need to be standardised to utility
(Emotions)
Experience values between 0 and 1. These values then
represent a measure of suitability per pixel
Comfort for each of the raster maps created for the
(Physical effort) various criteria. A zero value means that a
Desired

cell is not suitable for an intended purpose,

Covenience in this case as a cycle route, whereas a one
(Mental effort)
means that it is highly suitable.
Essential

With the standardisation of the

Speed
(Travel time from origin to destination) individual criteria complete, these are
then aggregated to create suitability maps
Security, safety and reliability per quality, which are in turn used to
(Reliance)
produce the final suitability map for the
perspective under investigation. The cell
Figure 3 Hierarchy of PT qualities (adapted from Van Hagen 2015) suitability scores are cumulative, meaning
that overlapping cells contribute to the
final suitability score. The exception to
this is where one of the overlapping cells
is a constraint, as the resultant scores for
these cells will always be zero. The final
(Emotions)
Attractiveness suitability map produced from the case
study is shown in Figure 6.
Coherence
(Mental effort) Network analysis
Desired

To develop routes, the raster cell scores

Directness from a final suitability map are transferred
(Speed)
to a road and/or existing cycle route
Essential

network layer. This can be done using

Comfort
(Physical effort) Geospatial Modelling Environment (GME),
which is an open-source software package.
Safety and security These then need to be converted to an im-
pedance or “cost” to perform the shortest
path analysis in the GIS by subtracting
Figure 4 Proposed hierarchy of cycle route qualities the suitability score per polyline between

the qualities or standalone, whichever is

applicable. N Legend
Figure 5 illustrates the criteria tree Suitability
developed for the case study. High
Once the criteria are assigned to the Low
criteria tree and the associated geo-spatial Study area
Indian
maps linked, a process of standardisation Ocean

0 1 2 4 km

Figure 5 S creenshot of the criteria tree

developed Figure 6 Final suitability map from the perspective of commuter cyclists

Civil Engineering September 2019 43

nodes from one (the maximum suitability),
and then multiplying each by the length N Legend
KORSTEN
of the respective polyline to obtain the Cycle route/s
NORTH Existing road network
impedance per polyline between nodes. By END
COTSWOLD
multiplying the impedance by the length of Study area
MOUNT
a segment between nodes, the total route NEWTON ROAD
PARK
length continues to play an important role, SUNRIDGE
PARK FERNGLEN
although not a dominant one. The total
MANGOLD MILL
impedance per route is calculated by sum- PARK PARK CENTRAL
ming the individual polyline impedances FAIRVIEW
SOUTH
forming the route using the shortest path END
HUMEWOOD
analysis in ArcMap 10.3.1.
With the impedance values assigned CHARLO
to the road network, it is then possible to WALMER
solve either individual routes or a network MOUNT
PLEASANT
of routes between defined O-D pairs
using the Network Analyst extension in
0 1 2 4 km
ArcMap. Figure 7 illustrates the optimal
route network generated between all
the neighbourhood centroids contained Figure 7 Cycle route network developed using proposed method
within the study area.
The total route impedance can be The cells highlighted in red represent assigned to routes allow for an objective
used when comparing alternative routings ratios in excess of 1.6 and are therefore comparison of alternatives. The intended
between a common O-D pair. For example, defined as indirect. Of the 105 O-D links, requirements not completely met are that it
Figures 8 and 9 show the optimal route (de- 24 exceed 1.6, resulting in 22.9% of the needs to be uncomplicated and user-friendly,
veloped by the shortest path analysis) and network links being indirect. In order to and that it must be cost-effective and time-
an alternative route between the suburbs reduce these, additional links are required. efficient. A fair level of skill is required to set
of Charlo and Newton Park respectively. the model up, but it is easy to understand
The total route length and impedance for CONCLUSIONS from a stakeholder perspective when taken
the two routes are included in the figures. The study set out to develop an objective through the process. The cost and time to
It is important to note that although the method for the development and evaluation collect geo-spatial data can also be extensive
alternative route is shorter in length, its of cycle route networks as its primary aim. if the data is not readily available. This can
impedance according to the qualities and The case study undertaken demonstrates in part be overcome by reprocessing datasets
weightings considered is greater. that this was achieved, as stakeholder for different criteria, as was demonstrated in
engagement is facilitated, it is universal the case study. However, it is recommended
Step 5: Cycling route directness and adaptable and can include a variety of that further research should be done into the
Cycle route directness (CRD) is a ratio qualities and criteria, it is transparent and relationships between criteria and data, and
of the actual route distance versus the back-traceable, and the quantitative scores the pre-processing required to get them into
straight line or Euclidean distance between
two points. A ratio of one is the best pos-
sible and occurs where the Euclidean and N Legend
KORSTEN
route distances are the same. Cycle route/s
NORTH Existing road network
According to Dill (2004), the INDEX END
COTSWOLD
PlanBuilder Users Guide developed by Study area
MOUNT
Criterion Planners Engineers recom- NEWTON ROAD
PARK Total Route Length:
mends pedestrian route directness (PRD) SUNRIDGE
PARK FERNGLEN 6 305 m
ratios of 1.2 to 1.5, with values in excess MANGOLD MILL Total Route Impedance:
of 1.6 being considered as indirect. As PARK PARK CENTRAL 3 105
FAIRVIEW
similar guidelines were not found for SOUTH
END
CRD in the literature review undertaken,
HUMEWOOD
the PRD thresholds have been adopted
for CRD until further research has been CHARLO
conducted in this field for cycling. The WALMER

CRD values calculated at a network level MOUNT

PLEASANT
can be found in Table 2.
As with any matrix, the values in 0 1 2 4 km
Table 2 are mirrored either side of the
diagonal line, which represents intra-zonal
travel (i.e. same origin and destination). Figure 8 Optimal cycle route between Charlo and Newton Park, Port Elizabeth

44 September 2019 Civil Engineering

 solving for a network of routes as opposed
N Legend to a single route, (2) suggesting a method
KORSTEN
Cycle route/s of measuring route directness using CRD
NORTH Existing road network
END to identify missing links, and (3) utilising
COTSWOLD
Study area CRD as an additional measure in the
MOUNT
NEWTON ROAD ­prioritisation of cycle facility upgrades.
PARK Total Route Length:
SUNRIDGE
PARK FERNGLEN 6 261 m
MANGOLD MILL Total Route Impedance:
REFERENCES
PARK PARK CENTRAL Dill, J 2004. Measuring network connectivity for
3 148
FAIRVIEW
SOUTH cycling and walking. Portland, OR: School
END of Urban Studies and Planning, Portland
HUMEWOOD
State University.
CHARLO ILWIS 2015. ILWIS (Version 3.8.5) (software).
WALMER 52°North ILWIS Community.
MOUNT Keshkamat, S, Looijen, J M & Zuidgeest, M H
PLEASANT
P 2009. The formulation and evaluation
0 1 2 4 km of transport route planning alternatives:
A spatial decision support system for the
Via Baltica project, Poland. Journal of
Figure 9 Alternative route between Charlo and Newton Park, Port Elizabeth Transport Geography, 17: 54–64.
Land Transport Safety Authority (New
the correct format. A guideline document of their importance, which are ultimately Zealand) 2004. Cycle network and route
detailing how data can be used as a proxy used to define an impedance per road or planning guide. Wellington (NZ): Land
where information is missing is then drawn path segment. The length of the route is Transport Safety Authority.
up to assist planners in the development of controlled in that the impedance per seg- National Department of Transport (South
the model. ment is calculated by using the length of a Africa) 2014. NMT Facility Guidelines.
The proposed method improves on segment as a multiplier, and therefore “total Pretoria: National Department of
existing methods as it explicitly considers route length” continues to play an impor- Transport.
both the route qualities and length in the tant role, although not a dominant one. Van Hagen, M 2015. Effect of station
development of routes. The route qualities Lastly, this study adds innovation to improvement measures on customer
can include requirements from a variety the field of SMCA for cycle route planning satisfaction. Journal of Traffic and
of stakeholders and be weighted in terms and prioritisation by: (1) simultaneously Transportation Engineering, 3(1): 7–18.

Table 2 Matrix containing CRD ratios between all O-D pairs

Humewood

South End
North End
Cotswold

Sunridge
Fernglen

Mangold

Mill Park
Fairview

Newton
Korsten

Walmer
Central

O/D
Mount
Charlo

Road
Park

Park

Central 0.00 1.28 1.07 1.29 1.10 1.65 1.17 1.49 1.09 1.41 1.04 1.28 2.30 1.12 1.52

Charlo 1.28 0.00 1.22 1.97 1.67 1.16 1.29 1.33 1.58 1.40 1.35 1.35 1.17 2.19 1.49

Cotswold 1.07 1.22 0.00 1.42 1.31 1.23 1.36 1.16 1.33 1.19 1.17 1.33 1.30 1.27 1.38

Fairview 1.29 1.97 1.42 0.00 2.56 1.31 1.36 1.18 1.36 1.35 1.46 1.40 1.36 3.41 1.51

Fernglen 1.10 1.67 1.31 2.56 0.00 1.28 1.30 1.79 1.48 1.26 1.21 1.37 1.37 1.30 1.65

Humewood 1.65 1.16 1.23 1.31 1.28 0.00 1.22 1.61 1.42 1.52 1.26 1.38 1.36 1.28 1.22

Korsten 1.17 1.29 1.36 1.36 1.30 1.22 0.00 1.57 1.58 1.46 1.21 1.53 1.27 1.28 1.63

Mangold Park 1.49 1.33 1.16 1.18 1.79 1.61 1.57 0.00 1.84 1.66 1.80 1.61 1.77 1.64 1.87

Mill Park 1.09 1.58 1.33 1.36 1.48 1.42 1.58 1.84 0.00 1.44 1.50 1.22 1.66 1.42 2.06

Mount Road 1.41 1.40 1.19 1.35 1.26 1.52 1.46 1.66 1.44 0.00 1.36 1.60 1.71 1.23 1.56

Newton Park 1.04 1.35 1.17 1.46 1.21 1.26 1.21 1.80 1.50 1.36 0.00 1.46 1.37 1.18 1.81

North End 1.28 1.35 1.33 1.40 1.37 1.38 1.53 1.61 1.22 1.60 1.46 0.00 1.42 1.32 1.33

South End 2.30 1.17 1.30 1.36 1.37 1.36 1.27 1.77 1.66 1.71 1.37 1.42 0.00 1.35 1.21

Sunridge 1.12 2.19 1.27 3.41 1.30 1.28 1.28 1.64 1.42 1.23 1.18 1.32 1.35 0.00 1.71

Walmer 1.52 1.49 1.38 1.51 1.65 1.22 1.63 1.87 2.06 1.56 1.81 1.33 1.21 1.71 0.00

Civil Engineering September 2019 45