Content uploaded by Jorge Gil
Author content
All content in this area was uploaded by Jorge Gil on Jul 12, 2017
Content may be subject to copyright.
Proceedings of the 11th Space Syntax Symposium
89.1
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
#89
MODELLING BIKEABILITY;
Space syntax based measures applied in examining speeds and flows of bicycling
in Gothenburg
BENDIK MANUM
Norwegian University of Science and Technology (NTNU) and Oslo School of
Architecture and Design (AHO), Trondheim, Norway
bendik.manum@ntnu.no
TOBIAS NORDSTRÖM
Spacescape, Stockholm, Sweden
tobias.nordstrom@spacescape.se
JORGE GIL
Chalmers University of Technology, Architecture Department, Gothenburg, Sweden
jorge.gil@chalmers.se
LEONARD NILSSON
Chalmers, Gothenburg, Sweden
leonard.nilsson@chalmers.se
LARS MARCUS
Chalmers, Gothenburg, Sweden
lars.marcus@chalmers.se
ABSTRACT
For numerous reasons related to energy demand, emissions, public health as well as liveable and
attractive cities, a frequently stated aim in contemporary discussions on urban development is
to increase amount and modal share of bicycling. In recent years, space syntax based methods
have shown to be useful for providing informed premises for these discussions. Combining
space syntax analyses with data on locations of residents, workplaces and destinations opens
the door not only for predictive modelling of route choice preferences but also the potential
amount of bicycling along routes. Building on previous research, the research presented in
this paper develops space syntax based measures expected to capture bicycling and evaluates
these measures by comparing the analyses with empirical data from studies carried out in
cooperation with the City of Gothenburg. Among the variables considered essential for bicycling
and included in our GIS model are: the slope and curvature of routes, the width and surface type
of bicycle lanes and the ind and aount of trac along the route or odelling bicycling ow
potentials a easure tered riginestination etweenness betweenness is used and
tested eaining dierent cobinations of ariables and threshold distances
he epirical data consists of gate counts of bicycle trac and detailed Gtracs apping
actual bicycling speeds of ca. 900 trips along a selection of bicycle routes. Using multiple
regression analysis to odel speed data eight ariables were found signicant n addition
to slope and curature of routes the signicant ariables relate to proiity to trac signals
degree of separation from pedestrians, density of entrances along the routes and quality of
paving of the cycle lane.
Proceedings of the 11th Space Syntax Symposium
89.2
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
oncerning bicycling ow potentials the ost signicant ariables in the ultiple regression
odel were betweenness within segent angular integration within density
of residents and people at work (students included) within 1 km and network betweenness
within 3km.
ased on the results of the current proect a proposal for further research is to elaborate on the
betweenness analyses by including speeds and preferably trac safety in the betweenness
easure y using tie along segents instead of etric length for dening the analysis
threshold (radius), it should be possible to have a new and improved generation of space syntax
based accessibility analyses for bicycling studies. A working name for such a measure is “least
impedance origin destination betweenness”.
KEYWORDS
eywords ieability icycle outes icycle peeds rigin estination etweenness
1. INTRODUCTION
ong trac engineers as well as urban planners and architects there is an increasing awareness
of the positie eects of bicycling and the need to include it in planning and design of the built
environment. From personal experiences, bicyclists know that route choice as well as speed are
strongly inuenced by the character of the terrain by ode and aount of trac on route and
by type and quality of the streets, lanes and paths that together constitute the route network
for bicycling eertheless ost conteporary urban and trac planning practice handles
bicycling only schematically. Typically, current tools for analysing bicycling rely on templates
based on ed speeds paying little attention to ariations in the type of bicyclist or eplicit
properties of bicycle routes and their context. Concerning amounts of bicycling, such as modal
share of daily couting on bicycle or nubers of bicyclists on specic routes current policy
and planning is often based on assumptions of a general percentage increase over the entire
bicycle network, regardless of route location within this network and particular properties of
those routes ilsson s long as these siplied assuptions for the basis for analysis
it will be hard to make reliable comparisons of alternative proposals for bicycle infrastructure
inestents for instance by eans of cost and benet analyses urrent transport odelling
tools typically include numerous variables for transportation demand, distance measurements
and route capacities, but scarcely take into account urban form variables related to the cognitive
ease of route nding or the directness and soothness of routes ariables that hae proen
to be essential for bikeability of the built environment (de Groot, 2007). In general, analyses
that do not explicitly include urban form variables provide little support to urban planning and
design in relation to bieability herefore fro the perspectie of trac planning as well as
fro the perspectie of urban planning and design it would be useful to hae ore rened and
user friendly methods for predicting speed and amount of bicycling.
In previous space syntax based modelling of bikeability at the neighbourhood scale, metric
distance has been the standard measure for grasping peoples’ preferences for convenient
travel (Manum and Voisin, 2010; Manum and Nordstrom, 2013; 2015). However, due to the wide
range of possible travel speeds, type and quantity of daily communing depends much more
on travel-time than on metric travel-distance. By measuring only metric distances, previous
odels do not tae the inuence of dierent bicycling speeds into account speeds that ary
a lot depending on type of bicycle and bicyclist as well as on numerous features of the built
environment. Hence, improved modelling of bikeability requires improved knowledge about
the ariation in bicycling speeds and how the built enironent inuences this ccording to
transportation research, speed along the bicycle network is also important regarding bicyclists’
route choices (Broach, Dill and Gliebe, 2012). Therefore, understanding and measuring bicycle
speed potential is a basis for understanding bicycle ow potentials esides being useful for
design of bicycle routes and related issues of urban form, improved estimations of bicycling
speeds and bicycling ows should also be applicable to traditional transport odels since
transport uantities and trael ties for dierent transport odes are basic issues in analysing
transport mode choices.
Proceedings of the 11th Space Syntax Symposium
89.3
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
he ai of this research proect has been to contribute to deeloping ethods for odelling
bikeability of the built environment. More explicitly, the aim has been twofold. First, for a better
understanding of how street properties aect speeds to deelop an epirically based odel
for estimating bicycle speeds in inner city environment. Second, for understanding how urban
structure in ters of spatial conguration and density and a cobination of the two inuences
bicycle ows to eaine the relationship between aggregated bicycle ows and a set of space
syntax based measures.
2. BACKGROUND
2.1 SPACE SYNTAX MODELS VERSUS TRAFFIC MODELS
Motorised travel is a highly technological and regulated activity, where the individual interacts
with the environment mediated by the vehicle and the technical mobility infrastructure
following strict sets of rules. Walking and bicycling, on the other hand, are shorter and slower
travel modes, sensitive to environmental conditions and closely interacting with the urban
contet his ind of interaction between built for and oeents of people is a eld where
space syntax models have proven to be highly useful.
ierently fro typical trac odels the obect of analysis in space synta odels is the
built enironent rather than obility ows his does not iply that space synta odels
are representations of the physical environment. Rather that they are representations of what
is called aordances Gibson that is what a gien enironent aords ie presents
potentials for) a certain ability in an agent (Gibson, 1986: 127). Hence, they do not model
either the physical environment or human activity, but what emerges in the meeting between
properties of the physical environment and both physical and cognitive human abilities (Marcus,
his is of principal interest to both urban and trac odelling since it presents a way
forward in oercoing the subectobect dichotoy often found at the foundations of both
urban and trac odelling e ay for instance iagine odels etending the space synta
approach to dierent trac odes where the built enironent oers particular aordances
for dierent ehicle types creating what has been called odality aordances for the dierent
locations within an urban landscape (Gil, 2016). Finally, there is reason to stress that current
space syntax-models, in comparison to most models of cities as complex systems (e.g. Batty,
2013), are static in that they do not include a time variable. They are not predictive simulations,
but rather descriptie odels preparing for analysis ne ay say that by odelling structure
as aordances in the anner described aboe they in a sense do capture process that is the
potential for particular huan actiities created by a set of aordances but they do not capture
process where these aordances in theseles change oer tie
2.2 SPACE SYNTAX BASED STUDIES ON BIKEABILITY
With the development of space syntax theory, measures and software, space syntax analyses
have proven useful for modelling the bikeability of street networks (McCahill and Garrick,
here hae been two aor space synta deelopents in this respect ne is angular
segment analysis, measuring network distance by taking into account the angles between
intersecting street segents also tered angular distance or angular depth his is dierent
from measuring network distance as topological steps of lines being either connected or not,
as is the case in traditional space syntax axial analysis (Turner, 2001; 2005; 2007; Hillier and Iida,
2005; Hillier et al., 2012). The other is the development of software combing space syntax and
G such as the lace ynta ool thle et al aford et al eained bicycling
in London by means of shortest routes, space syntax integration using angular depth and other
spatial conguration easures and found angular iniisation to be essential for bicyclists
route choice particularly for bicycle ow potentials at aggregated leel
he other deelopent eerges fro bicycling studies in the cities rondhei and slo
These studies combined space syntax choice and integration measures within metric distance
thresholds (radii) with the analysis of locations of residents, workplaces and other destinations
Proceedings of the 11th Space Syntax Symposium
89.4
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
at indiidual addresspoints applying the lace ynta ool ne result was that high alues
of street networ integration around worplaces was signicant for odal share of bicycling
while integration around home locations was not (Manum and Voisin, 2010). Furthermore, the
studies of Trondheim showed convincing correspondence between bicyclists’ route choice (as
found in the empiric study) and segment angular choice with a metric radius. These analyses
have proven useful for understanding bicycle potential of the existing bicycle route network
and for illustrating the likely performance of alternative urban planning and design proposals
(Manum and Nordstrom, 2013; 2015).
n the studies of slo based on the ethods deeloped in the analyses of rondhei the
apping included seeral ariables in addition the space synta street networ conguration
easures ong these were perceied danger fro heay trac perceied social danger
safety from a lack of people and activities (particularly at night), and attractiveness of routes
fro the presence of pars seawater and other inds of natural features ased on the thorough
apping of these aspects of bieability the unicipality of slo has deeloped abitious
plans for iproing the bicycle route networ he analyses of slo showed that the choice or
betweenness centrality easure is far fro sucient for estiating bicycle ows anu and
ordstro r to put it soewhat dierent the easure grasps the potential bicycle
ows of the street segents in a bicycle route networ but due factors not captured by the
measure, this potential is often hard to achieve. The main reason is perceived safety in terms
of fear of being inured at streets craped by cars trucs buses and tras nstead any
bicyclists use less direct and longer routes that they consider safer.
conclusion fro the slo studies is that trac safety together with bicycling speeds are the
main issues regarding bicyclist’ route choice. In addition, and even more important if aiming
to increase odal share of bicycling in daily couting trac safety is the ain reason for
people interested in bicycling not to commute by bicycle. This is in particular the case for
woen ordstr n conclusion the studies of slo indicate that there is great need
for examining bicycling speeds and for including both safety and speed in bikeability modelling.
This, together with the research of Dalton (2015) and Broach et al. (2012) arguing for the
inclusion of “impedance” along routes in space syntax measures that use spatial and cognitive
distance, is the background for the bikeability modelling explored in the case of Gothenburg
presented in this paper.
3. METHODS AND MEASURES
3.1 EXAMINING SPATIAL POTENTIAL FOR BICYCLE SPEED ALONG ROUTES
For examining speed potentials, we mapped the speeds of real bicycling along a selection of
bicycle routes in Gothenburg. The routes were chosen for being representative of the bicycle
route network of Gothenburg and for being relevant references for the planning and design of
future bicycle routes. The number of routes examined was 7 and their total distance measured
in both directions was 13 km. Figure 1 shows the selected routes.
Then, 15 bicyclists were selected and recruited, representing a variety of daily commuter
bicyclists being between and years old using dierent inds of bicycles and soe
dressed for exercise while others for relaxed bicycling. In order to check the representativeness
of the sample, we carried out a survey on 2000 bicyclists in the same areas of Gothenburg,
checking for clothing, bicycle types, gender and likely age. The selected sample showed to be
fairly representative, with some bias towards too many participants in the 21 to 35 age range.
In order to capture bicycling as daily commuting, the survey was carried out between 07:30 and
09:30 and between 16:00 and 18:00. The speed measurement was done by GPS-tracking with
yelstaden a software application deeloped by the trac oce in Gothenburg together
with Clickview, their software for handling the data, mapping the routes and speeds of a total
of 875 bicycle trips.
he net part of the study consisted in apping ariables liely to inuence bicycle speeds
Since the variables reduce or increase the speed of bicycling, they can be considered speed
Proceedings of the 11th Space Syntax Symposium
89.5
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
Figure 1 - Bicycling routes of the study and the median speeds along the route segments.
impedances of the routes. Impedance is a term used in transport analysis meaning resistance to
movement, analogous to physics, where impedance measures resistance to electrical current.
The street segments, based on a road centre line data set, were processed to create a street
networ odel adapted to capture the dierent ipedances used in this study
The street segments representing the routes were modelled as a bi-directional system, i.e. with
one eleent in each direction he street sections between unctions were subdiided into a
number of segments that based on their length would give approximately constant bicycling
speeds reaing and acceleration around unctions was handled by creating a separate
segent within eters fro each unction treets were also subdiided by the ind of
bicycle route (see Table 1); in the cases where the kind of route was not constant between
unctions the street was subdiided into segents consisting of only one ind of route ased
on the bicycle speeds’ correlates with street curvature described by de Groot (2007), streets
with sharper cures than a radius of eters were subdiided into e segents the cure
adacent segents of eters pc and the reaining ends of the street pc egents
where slope varied much were subdivided into lengths with little variation of slope, using the
categories 0-2% slope, 2-4% slope and so forth. Finally, the speed impedance variables for the
individual segments were assigned, using in the categories listed in table 1.
Proceedings of the 11th Space Syntax Symposium
89.6
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
Impedance variable Categories / Units
1 Kind of route • “pedestrian street, walking-speed street” (walk-
ing and bicycling merged)
• “slow bicycling-speed street”
• lane for bicycling at same level and not physical-
ly separated fro car trac
• one-way separate lane for bicycling
• two-directional separate lane for bicycling
• bicycling and walking lane (merged, but sepa-
rate fro car trac
2 Width of bicycle lane • Metres
3 Kind of bicycle lane surface material • Asphalt
• Concrete
• Natural stones
• Gravel
4 Kind of separation from pedestrians • Furniture, vegetation etc
• eight dierence dierent leel
• ierent surfaces
5 Slope • Percentage (%)
6 Horizontal curvature (radius) • Degrees
7 Length of segment • Metres
8istance between unctions • Metres
9egent connected to unction • es o
10 Entrances along segment, within 15m from
segment
• ount etres ll inds of entrances to
buildings, within straight line distance)
11 Entrances along segment, within 30m from
segment
• ount etres as preious
12 Car parking • es o
13 Bus stop • es o
Unfortunately, the GPS application failed to deliver reliable data concerning waiting times at
each intersection, making it impossible to examine the total impedance along routes at the
current stage. Therefore, the next step of the research should include a supplemental study
on speeds and waiting times at intersections. To estimate speed-models including many
dimensions such as impedances along routes and categories of bicycles and bicyclists requires
extensive GPS data (El-Geneidy et al.,2007; Romanillos et al. 2016 ;Arnesen et al. ,2017). A way to
gather detailed route specic coariates in proceeding research without laborious anual wor
is to collect sensor data such as data from an Inertial Measurement Unit (IMU), see Mohanty,
Lee et al. (2014) and the references therein, applying for instance accelerometers measuring
smoothness of road surface as well as very detailed information of the bicycling speed.
Table 1 - Impedance measures assigned to street segments
Proceedings of the 11th Space Syntax Symposium
89.7
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
he nal step in odelling consisted in assigning the alue of each ipedance ariable to eery
separate street segment. Some variables, such as slope, curvature and length of segments,
were generated autoatically fro G thers such as ind of route surface width and
separation type, required a combination of examining ortho-photos and site surveys. All the
variables of speed impedance modelled in GIS are the data to be compared with the empirical
data of bicycling speeds extracted from the GPS-tracking.
All the impedances considered were added to the statistical model for calculating their impact
on bicycle speed o nd the ost iportant independent ariables and test their signicance
a ultiple regression analysis was perfored he leel of signicance used was
Finally, the R2 value was calculated to see how much of the measured variation could be
explained by the variables in this this study.
3.2 EXAMINING STRUCTURAL SPATIAL POTENTIAL FOR BICYCLE FLOW ALONG ROUTE
he second part of the ethod odelling ow potential is based on space synta theories
and measures. Flow potential is here about predicting the amount of bicyclists along street
segments. It is not based on the impedance of the segments like the speed model, but rather
on their location in the street network relative to all other segments. Segments with higher
network centrality according to various measures are expected to have more bicyclists due to
their higher potential, which can be interpreted as being more important for the network as a
whole.
he epirical data used were gate counts of bicycle ows at points conducted by the
unicipality of Gothenburg in rlind he counting was done during rush hours
and and during lunchtie n this proect the unit applied
is the daily average of these counts, measured as number of bicyclist per hour.
The next step consisted in identifying the urban form variables to examine. The street-network
model examined was a bicycle route segment map based on an axial line map provided by
the consultancy r pacescape he selection of ariables was based on eperience fro
previous research. Altogether, 21 street-network analyses were conducted, examining 6 spatial
easures and dierent distance thresholds for each easure able
Table 2 - Spatial network measures calculated
Measure Analysis parameters
Axial integration Topological distance, topological radius along the network (7,
12, N)
Segment angular integration Angular distance (least angular change), metric radius along
the network (3000, 5000, 10000 m)
Segment angular choice Angular distance, metric radius along the network (3000, 5000,
10000 m)
Accessible population Total number of residents and workplaces, metric radius along
the network (3000, 5000, 10000 m)
Attraction betweenness Angular distance, metric radius along the network (3000, 5000,
10000 m), with accessible population as attraction weight.
betweenness From residents origins to workplaces and enrolled students
and vice-versa, angular distance, metric radius (3000, 5000,
10000 m)
Proceedings of the 11th Space Syntax Symposium
89.8
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
Besides the commonly used measure of axial integration, segment angular integration and
choice, calculated in Depthmap 10, we also examined two variations of the space syntax
choice easure recently ipleented in the lace synta tool thle et al called
ttraction betweenness and rigindestination betweenness or betweenness ttraction
betweenness, as used by Berghauser-Pont and Marcus (2015), is similar to segment angular
betweenness in ters of scoring each segent along the shortest angular route but diers
by ultiplying the score with an attraction n this study eaining potential ows of
bicycling, the “attractions” are the population accessible within a metric threshold distance.
n betweenness each segent is scored for being on the shortest routes between a set of
origins and a set of destinations, instead of routes between all nodes of the network. The score,
assigned to the segments, is a combination of the weight of the origin (number of residents)
multiplied by the normalised weight of the destination (i.e. dividing the destination weight
by the sum of all destination weights). The network analysis in this study operates on three
data sets: a set of address points of residents (the origins), a network graph of segment lines
representing all possible routes and a set of address points of workers and enrolled students
(the destinations). Every address point is linked to the nearest axial segment in the network.
First the calculation uses metric distance for the radius threshold, then, it uses angular distance
least angular change for the shortest route calculation as specied in the paraeters of the
spatial measures in Table 2.
hereas the rst regression odel deals with speed potential the second regression odel
deals with ow potential ainly testing structural properties of the street segents related to
all the other segents in the networ iilar to the rst ultiple regression analysis is
used to test arious predictor ariables and nd their signicance and iportance in eplaining
the ariation in the obsered bicycling ows
4. RESULTS
4.1 THE SURVEY DATA
Table 3 shows a summary of the GPS speed data, whereas Table 4 shows the results for each of
the 7 routes. Figure 1 maps the speeds along the routes by colour range, showing speeds in both
directions he results include all segents ecept the segents closest to unctions hese
segments are excluded due to an automatic functionality of the GPS-unit causing unreliable
speed data close to or in combination with full stops. As expected, due to the slope, bicycling
speeds at the hill north of the river are among the fastest as well as the slowest, depending on
direction of travel. Not surprisingly, we also see that speeds are very low on routes including
nuerous traclight unctions such as parts of stra and stra angatan see igure
he aerage speeds dier signicantly across dierent routes being slower at stra
angaten than at Gta lbron and h respectiely his illustrates the need for
developing models handling bicycling speeds as a measure dependent on route properties at
a detailed scale. The range of speeds is similar to former studies dealing with bicycle speeds.
ost studies show free ow speed arying between h and h in urban contets l
Geneidy et al., 2007; Cheng Xu et al., 2015).
Proceedings of the 11th Space Syntax Symposium
89.9
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
Median speed 85 percentile
Gta lbron 21 30
Lindholmsallén 21 27
Vasagatan 18 24
Nya allén 16 26
Kyrkogatan 16 20
Östra Hamngatan 14 24
stra hangatan 13 23
Unstand. Stand.
Coef Correlations Collinearity
Statistics
Model Coef. Std.
Error Beta t Sig. Zero
order Partial Part Toler-
ance VIF
(Constant) 14,491 ,905 16,007 ,000
Connected signal
intersection -5,430 ,454 -,486 -11,972 ,000 -,452 -,573 -,470 ,934 1,070
uber of entranes
30 meter -,025 ,008 -,147 -3,014 ,003 -,317 -,173 -,118 ,652 1,535
Slope downhill ,999 ,159 ,256 6,297 ,000 ,353 ,345 ,247 ,932 1,073
Pedestrian street -2,399 ,796 -,148 -3,013 ,003 -,272 -,173 -,118 ,640 1,562
Double sided bicycle
track 1,350 ,354 ,171 3,813 ,000 ,249 ,217 ,150 ,764 1,308
Horizontal curvature ,029 ,009 ,134 3,331 ,001 ,062 ,191 ,131 ,960 1,042
Length of segment ,012 ,003 ,210 4,294 ,000 ,309 ,243 ,169 ,644 1,554
Slope uphill*segment
lenght -,005 ,001 -,171 -3,684 ,000 -,040 -,210 -,145 ,716 1,396
Natural stone -1,193 ,470 -,120 -2,537 ,012 -,236 -,147 -,100 ,687 1,455
Table 3 - Summary of speed tracking on the bicycle routes, averaged for both directions
able he results of the ultiple regression analysis for edian speed potential
4.2 THE BICYCLING SPEED MODEL
he rst analysis eaining ipedances epected to aect the edian speed on the street
segents shows that nine predictors are signicant and contribute to the eplanation see
table heir eplanatory usefulness aries but not to a large etent he ariance ination
factors (VIF) show that the variables do not covariate to any considerable amount. The F-value
for the whole odel is signicant which shows that at least soe of the prediction ariables
contribute to the explanatory power of the model.
Proceedings of the 11th Space Syntax Symposium
89.10
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
Model Summary j
Model R R Square dusted
Square
Std. Error of
the Estimate
Change Statistics
R Square
Change
F
Change df1 df2 Sig. F
Change
Table 5 ,740i ,548 ,534 2,68861 ,010 6,434 1 293 ,012
i. Predictors: (Constant)
, Connected signal intersection
uber of entrances eter
, Slope downhill, Pedestrian street
, Double sided bicycle track
, Horizontal curvature
, Length of segment
, Slope uphill*segment length
, Natural stone
ependent ariable edian speed
Figure 2 - Residual plot for speed model.
Table 5 - Summary of the multiple regression analysis for median speed potential.
Proceedings of the 11th Space Syntax Symposium
89.11
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
Figure 2 shows the residual plot for the speed model. Even though the plot seems fairly random,
it is not completely ruled out that the residual plot can hide some variable not taken into
account, for example prevailing wind directions and delays caused by congestion. The R2 of the
model is 0.54, which mean that the selected variables explain 54% of the speed variations (table
aing the copleity of bicycling ows in ind this is an acceptable result particularly
when also having in mind the potential improvements that can be made to the model in the
future ne eaple is to the eect of slope which currently is odelled in interals but with
continuous odelling the eect ight iproe the odel igure shows the edian speed
from the GPS data (left) compared to the median speeds estimated by the regression model.
Figure 3 - Speed from GPS-tracking (left) and estimated by the model (right)
4.3 THE BICYCLING ROUTE MODEL
The second model, dealing with network measures expected to predict the potential for
bicycle ows also eplains the easured ariations to a fair etent ll ariables are signicant
and contribute to the explanatory capability of the model. Their explanatory power varies,
according to the coecients seen in table but not to a large etent and betweenness is
the ost signicant his can be eplained by the fact that betweenness can be considered
to measure the potential amount of bicycle trips to work, which according to travel survey data
is the most frequent bicycle trip in Sweden (Saxton, 2015).
t rst glance surprisingly accessible population within correlates negatiely with
bicycle ows ooing closer the ariable has a positie correlation up to a certain accessible
population, and is negative in the densest parts of Gothenburg. This explains that accessible
population is a proxy for low speed potential, which implies that bicyclists choose alternative
routes in less dense parts of the city. This corresponds to the results related to route choices in
slo anu and ordstro
Proceedings of the 11th Space Syntax Symposium
89.12
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
Unstand. Stand.
Coef Correlations Collinearity
Statistics
Model Coef. Std.
Error Beta t Sig. Zero
order Partial Part Toler-
ance VIF
(Constant) ,648 ,585 1,108 ,270
betweenness least
angular within 5000 m 2,882E-05 ,000 ,258 3,709 ,000 ,507 ,286 ,220 ,725 1,380
Segment angular
integration within 10
000 m
,002 ,000 ,635 5,763 ,000 ,550 ,421 ,341 ,289 3,461
Attraction density
(population within 1
000 m)
-1,525E-05 ,000 -,326 -3,209 ,002 ,239 -,250 -,190 ,339 2,953
Network betweenness
(shortest route within
3000 m)
4,080E-07 ,000 ,141 2,030 ,044 ,416 ,161 ,120 ,728 1,373
a ependent ariable logy
able ultiple regression analyse icycle ow potential
closer loo at ariance ination shows larger alues than the rst analysis up to seen
in table although they are udged to be acceptable in this analysis he alue for the whole
odel is large and signicant which indicates that at least soe of the predictors contribute to
the explanatory power of the model. The residual plot from the analysis is random.
Model Summary j
Model R R Square dusted
Square
Std. Error of
the Estimate
Change Statistics
R Square
Change
F
Change df1 df2 Sig. F
Change
Table 7 ,679i,460 ,446 ,61644 ,460 32,853 4 154 ,000
i. Predictors: (Constant)
beal
ialacoenderbetandeal
dbstuddeg
ntegrationetric
ependent ariable logy
able odel suary for bicycle ow potential
Proceedings of the 11th Space Syntax Symposium
89.13
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
inally the of the odel is which ean that of the ariations in the easured ow
can be explained by the selected predictors. This is lower than the speed model result (0.45
compared to 0.54), and can be explained by the large number of relevant issues not included in the
model. Some variables that according to research are essential for route choice not yet included
in the odel are the dierences in speed and the feeling of safety and cofort pacescape
2015). For example, some fast commuter routes along the water have low betweenness
alues while they hae any bicyclists n the other hand any of the busiest streets hae
high betweenness alues but few bicyclists his conrs patterns found in preious research
anu and ordstro he discrepancy of the betweenness easures and ows of
bicycling at particular inds of routes can be eplained by bicycling speeds and trac safety
The separate commuter routes allow for convenient and fast bicycling, implying that bicyclist
choose these routes for being the quickest and easiest despite unfavourable metric distance.
egarding the busy central streets these are often craed with trac in soe cases large
aounts of cars as well as tras and buses and in other cases pedestrians he rst cases are
dangerous for the bicyclist; the second cases force the bicyclist to slow down and give priority
to pedestrians. In both cases, it is often more convenient, safer or quicker for the bicyclist to
choose alternative routes, even though they might be longer in metric distance or cognitively
less direct. To conclude the discussion of the results, a possible issue with the bicycle network
representation should be mentioned. An important measure in the analysis is angular change
at unctions his research proect used an aial ap produced for other purposes without
coparing the angles at unctions in this aial ap with the geoetries of real bicycling routes
through unctions uch coparisons should be part of future research liely resulting in ore
detailed odelling of lines at unctions and an iproed odel
5. CONCLUSIONS
his proect illustrates the ariety of bicycling speeds along urban routes and sheds light on
the relatie inuence of soe particular bicycle route ariables signicant for bicycle speeds
In addition to the obvious result that downhill slopes correlate with higher speeds whereas
signal crossings correlate with lower speed at adacent segents the ost signicant of the
variables examined were: many entrances along the segment (-), mixed use with walking (-),
twoway bicycle lanes radius of curature and length of segent s entioned
earlier in the paper, there is a need to handle bicycle speed and route properties at a detailed
scale. In the work of Arnesen et al. (2017), a Markov model for predicting bicycle speed along a
route with high resolution considering vertical and horizontal curvatures is being developed for
this purpose. In this Markov model, the speed in the current road segment is dependent of the
speed in preious and future segents proiding ore realistic speed proles suggestion for
further work is therefore to include the larger variety of covariates presented in this paper into
this more advanced methodology of speed modelling.
egarding bicycle ows the proect has eained a selection of space synta based spatial
measures, measures that can be mapped directly from GIS. The latter is important to apply
the analysis tools on large urban systes he ost signicant ariables regarding bicycle
ows are betweenness least angular change within segent angular integration
within accessible population within and networ betweenness as shortest
distance within en though bicycle ows are inuenced by any personal social and
econoic issues to eer be fully grasped by space synta odels and Ganalyses this proect
shows that seeral of the easures eained particularly the betweenness conincingly
capture the ain patterns of bicycle ows ue to the iportance of bicycle speeds and
trac safety on route choice and these issues not being included in the current odel adding
ariables inuencing those factors should signicantly iproe the correlation with ows of
real bicycling.
Based on this conclusion, the main issues for future research are to examine how speed
dierences perceied safety and conenience can be analysed in G based tools that include
space synta easures ne way of achieing this is to conert trac safety and conenience
into added travel time. This method has been discussed in transport research (Ellis, 2015) and
Proceedings of the 11th Space Syntax Symposium
89.14
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
is currently wor in progress within the research proect tratod orhei and rset
where impedance values from the aforementioned report is used to calculate generalised time
for each road segment. Another option, suggested by Dalton (2015) is to add impedances into a
spatial conguration odel by conerting ipedances for instance the eect of trac signals
into weights added to topological distance ased on the results of the current proect our
approach to further research will be to elaborate on the betweenness analyses by including
speeds and ideally trac safety into the betweenness easure he rst step in addition to
iproing the speed odel to include the range of speeds caused by dierent inds of bicycles
and bicyclists and by impedances along routes, will be to convert speeds on the segments into
tie and then apply trip tie rather than etric distance as the radiusthreshold unit in the
spatial analyses. By measuring time along segments (intersection impedances included), it
should be possible to develop a new and improved generation of space syntax based accessibility
analyses - analyses where the bicycling potential of a bicycle route network is based on spatial
congurations but also on tie and conenience of real bicycling at the routes woring
title for the new measure is “least impedance origin destination betweenness”.
ACKNOWLEDGEMENTS
his research proect on easuring bieability has been supported by the rac ce in
Gothenburg together with Chalmers Architecture and Chalmers Transport.
Proceedings of the 11th Space Syntax Symposium
89.15
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
REFERENCES
rnesen alin and ahl aro odel for predicting bicycle speed ubitted for
publication.
Batty, M. (2013), The new science of cities, Cambridge, MA, USA: MIT Press
rlind Inventering av fotgängare och cyklister i centrala Göteborg, Gteborg raontoret
erghauser and arcus hat can typology eplain that conguration cannot n arii
Vaughan, L., Sailer, K., Palaiologou, G. and Bolton, T. (eds.) Proceedings of the Tenth International Space Syntax
Symposium, London: University College London, paper 043
Broach, J., Dill, J. and Gliebe, J. (2012),´Where Do Cyclists Ride? A Route Choice Model Developed with Revealed
Preference GPS Data´. In Transportation Research Part A: Policy and Practice, Vol. 46 (10), p. 1730–1740
Cheng Xu, Qiangwei Li, Zhaowei Qu, and Sheng Jin, (2015),´Predicting Free Flow Speed and Crash Risk of Bicycle
rac low sing rticial eural etwor odels n Mathematical Problems in Engineering, Vol. 2015, Article
ID 212050
de Groot, R. (2007), Design manual for bicycle trac, entru oor egelgeing en nderoe in de Grond
Water-, en Wegenbouw en de Verkeerstechniek),The Netherlands
Dalton, N., S. (2015),´Momentum integration: The syntax of cycling´. In: Karimi, K., Vaughan, L., Sailer, K.,
Palaiologou, G. and Bolton, T. (eds.) Proceedings of the Tenth International Space Syntax Symposium, London:
University College London, paper 067
lGeneidy rie and acono redicting bicycle trael speeds along dierent facilities using
GPS data: A proof concept model´. Paper at Transportation Research Board 86th Annual Meeting, Paper #07-
2971
Ellis, I. (2015) Målrettede sykkeltiltak i re byområder, slo rbanet nalyse
Gibson, J. (1986), The ecological approach to visual perception, New York, NY, USA: Psychology Press
Gil, J. (2016), Urban modality - Modelling and evaluating the sustainable mobility of urban areas in the city-region, PhD-
thesis, TU Delft, Delft, The Netherlands
illier and ida etwor and sychological eects in urban oeent n an es ed Proceedings
of the Fifth International Space Syntax Symposium, Delft: University of Technology, Vol. I, p. 553
Hillier, B., Yang, T. and Turner, A. (2012),´Normalising least angle choice in Depthmap´. In: The Journal of Space
Syntax, Vol 5 (2), p.155-193
McCahill, C. and Garrick, N., W. (2008),´The Applicability of Space Syntax to Bicycle Facility Planning’. In Journal of
the Transportation Research Board, Vol. 2074, p.46–51
anu and ordstr ccessible density a condition for sustainable urban deelopent aper at
uropean etwor for ousing esearch arragona une ailable at httpbroed
leswordpresscoenhrpaperanunorstropdf
anu and ordstr ntegrating bicycle networ analysis in urban design iproing bieability in
rondhei by cobining space synta and Gethods using the lace ynta ool n i ar
and Seo, K. W. (eds.), Proceedings of the Ninth International Space Syntax Symposium, eoul eong niersity
Vol. 28, p.1-14.
Manum, B. and Voisin, D. (2010),´Urban form and daily travel; non-motorised transport in Trondheim examined
by combining space syntax and GIS-based methods´. Paper at: CITTA Bringing City Form Back Into Planning,
orto ay ailable at httpsbroedleswordpresscocittapaperanuoisinpdf
Marcus, L. (2015),´Ecological space and cognitive geometry – linking humans and environment in space syntax
theory´. In: Karimi, K., Vaughan, L., Sailer, K., Palaiologou, G. and Bolton, T. (eds.), Proceedings of the Tenth
International Space Syntax Symposium, London: University College London, paper 124
Mohanty, S., A. Lee, T. Carvalho, L. Dias and G. Lovegrove (2014). ´A global review of current instrumented probe
bicycle (IPB) technology and research´. At The International Cycling Safty Conference, Gothenburg, Sweden.
Nilsson, A. (2013), Samhällsekonomisk analys av ett snabbcykelstråk mellan Malmö och Lund, und riectorordstr
slosylisten tochol slo oune
Proceedings of the 11th Space Syntax Symposium
89.16
MODELLING BIKEABILITY; space syntax based measures applied in examining
speeds and ows of bicycling in Gothenburg
ordstr and anu easuring bieability pace synta based ethods applied in planning for
iproed conditions for bicycling in slo n arii aughan ailer alaiologou G and olton
T. (eds.), Proceedings of the Tenth International Space Syntax Symposium, London: University College London,
paper 077
orhei and rset eeloping a policy odel for sustainable urban transport aper presented at the
European Transport Conference, Frankfurt. The Association for European Transport.
Raford, N., Chiaradia, A. and Gil, J. (2007),´Space syntax: The role of Urban Form in Cyclist Route Choice in
entral ondon t nnual eeting ashington ailable at httpsescholarshiporguc
itef
oanillos G ustwic ttea and de ruif ig ata and ycling ransport eiews
pp 114-133.
aton erige tochol raanalys
thle arcus and arlstr lace ynta Geographical ccessibility with aial lines in G n
van Nes, A. (ed.), Proceedings of the Fifth International Space Syntax Symposium, Delft: University of Technology,
Vol. I, p. 131
Turner, A. (2005),´Could a road-centre line be an axial line in disguise?´. In: van Nes, A. (ed.), Proceedings of the Fifth
International Space Syntax Symposium, Delft: University of Technology, Vol. I, p. 145
Turner, A. (2007),´From axial to road-centre lines: a new representation for space syntax and a new model of route
choice for transport network analysis´. In Environment and Planning B: Planning and Design, 34(3), p.539–555
Turner, A. (2001),´Angular analysis´. In: Peponis, J., Wineman, J. and Bafna, S. (eds.), Proceedings the Third International
Space Syntax Symposium, Atlanta, U.S.A: Georgia Institute of Technology, p. 30.1–30.11
