Environment and Planning B: Planning and Design 1996, volume 23, pages 3 7 - 4 8

Urban convergence: morphology and attraction

R Krafta
Department of Urbanism, Federal University of Rio Grande do Sul, Porto Alegre, RS, Brazil
Received 17 August 1994; in revised form 20 March 1995

Abstract. A model, based on detailed descriptions of spatial configuration and a probabilistic

approach to the user's choice, is proposed to measure the relationship between demand and
supply locations in urban local systems. According to this model, the articulation of the public
space grid, associated to the uneven distribution of facilities, generates a powered supply network
to which demand locations are related. Choice, as well as demand satisfaction, will then be a
function of the relative position (centrality) and attractiveness of supply locations. The model
gives a simultaneous account of the spatial opportunity of demand and the spatial convergence
of supply. Concurrently it can offer a picture of the stability of space in terms of possible land-
use changes.

1 Introduction
In previous papers (Krafta, 1993; 1994) the problem of spatial differentiation has
been examined in terms of inner configurational issues and their possible role within
the urban spatial structure. The fundamental proposition in these papers was a set
of synthetic measures of urban morphology, named potential or centrality models,
which may provide the urban designer or policymaker with instruments to assess the
performance of intraurban spatial systems. The aim of this paper is to take these
models one stage further by introducing specifics which are expected to produce
finer insights into the mechanics of urban space structuring.
Several modes of capturing and describing urban spatial differentiation have
been suggested, and are better known as the various forms of accessibility measur-
ing. Arentze et al (1994) quote three groups of accessibility measures, ranging from
the simplest ones, based on travel costs faced by consumers in the process of satis-
fying their demands, to more complex measures which express accessibility in terms
of the surplus value, benefit, or utility consumers gain from facilities. In the same
paper, they propose a new, 'multistop travel model' to explore the particularities of
chain trips usually made by consumers during their shopping. Another simple form
of the use of accessibility to describe areal differentiation is proposed by Hillier et al
(1993). Their space syntax measure, named relative asymmetry, takes accessibility as
a mean, topological distance from each space to all others in the same spatial system.
A ranking of these indicators, according to the complexity of their inner struc-
ture, could show:
(a) measures that do not distinguish point hierarchy, that is, do not differentiate
origins (demand points) from destinations (supply points), for example, relative
asymmetry;
(b) measures that do not distinguish different destinations in terms of their quality,
such as the models in which the distance to the nearest supply point is considered;
(c) measures which do hierarchize supply points and do not take into account
increasing or decreasing probabilities of choice derived from chain destinations,
such as the measures of consumer welfare; and
(d) the measures which handle supply hierarchies, as well as plural destinations, for
example, the multistop travel model.
38 R Krafta

It is rather interesting that, although relative asymmetry and multistop travel

models are at the top and bottom of the list above, they have an important point in
common, which is the attention given to the spatial particularities of the system in
hand. The multistop travel models try to relate the relative position of supply points
to the probability of successful demand trips, and relative asymmetry tackles the issue
of grid configuration. The multistop travel model measures the effectiveness of attrac-
tants attached to a circulation system, whereas relative asymmetry measures the
effectiveness of the circulation system itself, with its possible effects on the location
of attractants. As space syntax is a possible measure of the urban spatial structure,
it seems clear that the absence of attractants in its models is a serious restriction; on
the other hand, it suggests that the multiple movements made possible by the urban
grid can generate complex patterns of use. In this direction, multistop-type models
could well benefit from a morphological specification of grid configuration. By adding
a dimension of route choice to users' choice of supply locations, already measured
by the multistop device, we could obtain a new model structure.
Space centrality indicators can be seen as a first bridge between these different
families of models, to the extent that they bring together grid and built-form par-
ticularities. In this case, built form works as a proxy of attractants, giving a first
dimension of their uneven distribution and effect on spatial systems. In this paper,
the idea of combining grid configuration and attractants is developed further and a
new measure, named spatial convergence, is proposed.

2 A note on space centrality

The first idea structuring space centrality is that every built-form unit (BFU) is
reachable from every other BFU through a sequence of public spaces. Consequently,
public spaces fall on the path between every pair of BFUs and so they are central in
relation to them, as illustrated in figure 1. Considering an uneven distribution of
BFUs and a multiplicity of paths between each pair, we can then assume that some
public spaces will be more central than others in the whole linkage of all possible
pairs of BFUs in a given system.

Figure 1. The reachability scheme proposed by the theory, in which every built-form unit is
reachable from every other through a multiplicity of public space paths, is exemplified here.

The second idea which is essential to spatial centrality is the indissolubility

between built form and public space, in the sense that tension is created by built form
and channelled through public spaces; no tension is generated without built form and
no tension is distributed without public spaces. With the two basic system compo-
nents (built form and public space) and one fundamental relationship between them,
adjacency, it is possible to represent such a system as an urban graph, in which each
Urban convergence 39

dot is a unit of space (built and open) and each edge is an adjacency. The result is a
galaxy (Kruger, 1979) with a constellation of public spaces dotted with small con-
stellations of BFUs, as suggested in figure 2. Such a graph is then processed in the
following way.
(1) All shortest paths between all possible pairs of BFUs are identified. The shortest
paths are topologically determined, that is, spaces linking BFUs are taken irrespec-
tive of their actual length and counted as one step. The shortest path between two
BFUs will be the one with the least steps.
(2) A unit of weight is assigned to each space belonging to the shortest path(s)
between each pair of BFUs, signalling its role in providing reachability from one to
another.
(3) All units of weight assigned to each space are summed up after all pairs of BFUs
are processed. The result shows the relative centrality of all spaces of the system;
high figures report spaces heavily assigned to paths between pairs and are more
central than ones ranked low.
It is clear that axial lines support a varied number of BFUs throughout the
system, according to their length, number of land plots, density and height of build-
ings, etc. Consequently, even if the actual activities housed inside each building are
not considered at this stage, the dimensions of spatial differentiation are considered.
Spatial centrality measured through such a procedure can be consistently correlated
to some aspects of urban spatial structure, as described in Krafta (1994).

7 . 8 9 10 11 12
II

HI •
P i.

1 4 i
i
-I -
r
1• l —

Figure 2. Construction of an urban graph, expressing the reachability scheme, from usual
urban maps. Each hollow circle represents a public space, each dot represents a built-form
unit, and each edge represents an adjacency.

3 The measure of spatial convergence

Spatial convergence is defined as a particular form of space centrality, in which
BFUs are distinguished by their demand or supply nature (residential or service), as
well as by the attractiveness each of those activities attains in the urban system. In
this version, each dot of the urban graph, representing a BFU, will be labelled as an
'origin' or a 'destination' which restricts the choice of possible pairs between which
to search for shortest paths and allows us to proceed with any other calculations.
The objectives are listed below.
(a) To identify the supply points, which are most central in respect of their position
in relation to demand points and to their size or variety, to which the urban system
converges.
40 R Krafta

(b) To rank the demand points in relation to their spatial opportunities, that is, their
relative accessibility to facilities located at supply points. Such a rank will reflect
the position of each demand point in relation to the convergence cones of the
system.
(c) Complementary to the above, the model may be required to identify any
potential new supply points, that is, those demand points which, because of their
relative position, also have spatial privileges and are thus potentially qualified as
supply points.
3.1 Assumptions
The following assumptions are made.
(a) Each BFU houses one activity.
(b) Residential activity is rated at attractiveness 'zero', whereas the attractiveness of
the several service activities will be parametrized by the indicators of service size
(within specific groups) and kind (among diverse groups). Weighted in this way,
BFUs will orient relationship between demand and supply points and form spatial
clusters of supply activities. It is then assumed that demand satisfaction is a proba-
bilistic function of the size and variety of supply points (Arentze et al, 1992).
(c) The functional link between a demand point 'F and a supply point '/' is mediated
by the reachability of j from / and the eventual presence of another supply point in
the shortest path between / and /', in the same way as it is mediated by the size and
variety of supply points, as described above. Consequently, distance weakens the
functional relationship between / and j in the same way as does the eventual
occurrence of another supply point located in the shortest path between them, as
illustrated in figure 3.

Figure 3. Graphical representation of the hierarchy of supply points (S), based on the distance
from the demand location (D) and on the relative positioning relationship to other supply
locations.
Urban convergence 41

3.2 Specification
Reachability is determined from every demand point to every supply point, with the
usual identification of all shortest paths, and spaces engaged in them. The first step
is to process the usual reachability of the system; this is done by the procedure of
identifying all possible pairs of demand or supply points, determining the shortest
paths between them, and listing all spaces belonging to those paths. Such a proce-
dure differs from that usually performed for centrality calculations only in that the
points are labelled (demand, supply) and reachability is considered only from demand
to supply points.
The second step is related to weight assignment. For centrality calculation, a
unit of weight is assigned to every pair of BFUs and then distributed among every
space belonging to the shortest paths, each one obtaining a fraction of that unit. In
this case, however, two specifications are introduced. First, because supply points
have different statuses, there is no unit of weight to be shared among spaces, but a
figure reflecting the size and variety of the supply point. Second, the assignment as
such should also be labelled in such a way that the statuses of the points 'origin',
'destination', and 'link' are distinguished. The first two are the pair of points being
processed; the third is related to all other points which make up the shortest paths,
as illustrated in figure 4.

8 9 10 11 12
1I , , I• .
II
1 1
7 1 1 •

4
1 4(D)
L_c
» g ~i) I)

D S

(c)
Figure 4. From the urban map (a) an urban graph is constructed (b) in which spaces are
labelled as demand (D) and supply (S). A third graph (c) shows, for a randomly selected pair
of demand-supply points, the diversified statuses of the points (origin, destination, and link),
as well as the privileges of supply points derived from their centrality.

The third step implies the selective accounting of all those figures credited to
spaces in the process of describing reachability. Objective (a) can be reached by
summing up the figures labelled 'origin' for all demand points of the system. In the
resulting list, demand points will be marked according to their spatial opportunities
in relation to existing facility points. Objective (b) can be reached by summing up
all figures credited to supply points. In this case, not only the intrinsic convergence
power of each point, but also their relative dominance over other supply points, as a
result of relative position and particular configurational characteristics of the public
space grid, are considered. The objective (c) can be reached by including the
figures credited to all points in the calculation. The list containing demand points
will point out those that hold highly favourable positions in the spatial system and
can be seen as possible alternative supply points.
 42 R Krafta

READ
List of demand points B l l l l l l l "*wm>,
List of supply points B ^ P 5 ^
Weights of supply points "*'"'
Connectivity matrix

( Start 1 Pick up a pair of demand - supply points

mm

1
Find shortest path(s)

List spaces belonging to shortest paths

Divide the weights of supply points by the

1
number of spaces in the shortest paths

Assign the result to demand pomts and label

]
'spopporf

Assign the result to supply points and label

'converg'
1
Assign the result to other demand points and

•fa*
label 'link'
J
^^^^^mmm^mmm^^^M
WMMMMM8&®®**

List all scores labelled 'spopport'

Sum up all supply points'

scores labelled 'converg'

List all scores labelled 'converg'

j
Sum up all the scores of supply
points labelled 'link'
J
List all scores labelled 'link'

Sort out scores top-down

Figure 5. The model flow chart illustrated.

Urban convergence 43

3.3 Implementation
In order to calculate scores for the spatial convergence, as outlined above, a
computer program has been created. As described in the flowchart (figure 5), the
algorithm that
(a) matches all pairs of demand-supply points,
(b) goes through the connectivity network of the system,
(c) finds all shortest paths between each pair,
(d) identifies all spaces belonging to those shortest paths,
(e) calculates the fraction of weight to be assigned to each space,
(f) assigns the calculated fractions of weight to demand, supply, as well as to inter-
mediate demand and supply points according to their specific labels,
(g) sums up all partial scores, according to objectives (a), (b), and (c), after all pairs
have been processed and produces different score sheets, ranking separately demand
points according to their spatial opportunity situation, supply points according to
their spatial convergence, and demand points according to their spatial convergence,
can be run on personal computers and can process fairly large networks.
The program receives data related to demand points, supply points, and the
connectivity characteristics of the spatial system. On the demand side, data refer to
the number of points only; on the supply side, data include the number of points and
an indicator of the size of attractiveness of each. For the whole system, a connec-
tivity matrix, reflecting the characteristics of the public space network, is included in
the data.
3.4 Interpretation of results
The list labelled 'spopport' ranks demand points according to their relative spatial
opportunity. This means that the accessibility to all supply points in the system is
measured for each demand point. The scoring procedure, described above, assigns
higher scores to demand points which are nearer to large and varied supply points.
The nearness, referred to above, is measured in terms of topological (adjacency) dis-
tance and can be represented by different maps, as explained in the next section. The
results are positive figures which compare all demand points within the same system.
A relative measure, which accounts for different size systems, can be introduced.
The list labelled 'converg' ranks supply points according to their spatial conver-
gence. This means that each supply point is measured in terms of its (a) size and
variety, (b) capacity of being nearer to demand points, and (c) capacity of falling on
the shortest path between the demand and other supply points. In this sense,
supplypoints which are in close spatial proximity to demand points, take spatial
precedence over other supply points, and are large and varied, will be better
scored.
The list labelled 'link' ranks demand points according to their spatial position in
the system. It is a classic measure of centrality, in which spaces are scored for their
capacity to provide reachability between two other points; in this case, between a
demand and a supply point. As they have high scores, demand points could be
interpreted as having 'supply potential', that is, having the spatial condition of being
transformed, over time, into new supply points for the system.

4 Spatial and operational representations

Among the several grid representation possibilities—axial (Hillier et al, 1993), point-
axial (Kruger, 1989), link (Krafta, 1991)—the axial one has been the most suitable in
urban morphology cases because of its capacity to retain essential quality of public
space, connectivity. In effect, by representing a public space grid through a set of
44 R Krafta

axial lines, it is possible to reduce the complexity of its configuration to a basic

characteristic—its linear dimension. Axial lines in a grid intercept each other, so
they can be further reduced to a graph representation in which an axial line is
expressed by a dot and its connections to other lines by lines.
Applications of centrality measures on axial maps (Krafta, 1994) have proved
effective for the simulation of areal differentiation in intraurban—that is, very
detailed—situations. Results have proved to be consistently correlated to relevant
aspects of urban spatial structure. For spatial convergence cases, however, axial
maps present a few problems. First, demand and supply locations are not precisely
determined in axial representation, notably in those cases of regular, Barcelona
ensanche type grids. Second, axial lines could vary their built form, as well as their
activity contents intensely along their actual extension. Service facilities tend to be
located at key points, as far as local urban situations are concerned; similarly,
demand locations are better evaluated in terms of their spatial opportunities if a
precise spatial definition of their locations is given. Point-axial maps seem to be an
acceptable answer to these operational demands.
A point-axial map, also called a 'bunch' map (Kruger, 1989), is a combination of
axial and link maps. From the first, the point-axial map retains the property of axial
connectivity, and from the latter it retains the property of precise definition of
corners. In axial maps, the basic spatial entity is a line (represented by a dot in a
graph) whereas point-axial maps have a complex basic spatial entity, which is the
intersection of two axial lines (the corner), also represented by a dot in a graph, as
exemplified in figure 6 and 7. It is complex to the extent that it combines a precise
spatial location with axial connectivity: each corner is taken as adjacent to all others
along the segments of axial lines intersecting at it.

Figure 6. The Czech new medieval city of Unicov: its survey map (a); axial map (b); point-
axial map (c); and the 'bunches' for locations ' 1 ' and '2' (d and e).
Urban convergence 45

In-place activity aggregation is conducted at corners, in accordance with the

influence radius of each one.
Quantification of supply activities is related to two factors:
(a) the actual size of each corner, measured in terms of selling area, number of
employees, or other relevant indicators; and
(b) the variety of activities, measured by a multiplier, a parameter applied on the
first indicator.
Parameters could be regulated by the usual calibration processes and oriented by
correlations with independent variables of urban spatial structure, such as pedes-
trian or vehicular flows.

H [ J K L
1 2 3 4 5

6 7 8 9 10
B-
(11 12 13 14 15 16
C
17 18 19
E
20 21
F< i 1
>

Figure 7. An axial or point-axial map and the corresponding axial and point-axial graph
representations.

5 Limitations and consistencies of the proposed indicators

Comparatively, spatial convergence calculations based on axial and point-axial maps
have proved to be similar, in the sense that the relative distribution of points in a
relief map describes a similar relief. However, point-axial convergence exhibits a
much more precise spatial description, as axial lines are detailed in their inner
differentiations and displayed accordingly. Figure 8 and 9 show such a relative
distribution: for the axial version of the convergence measure, lines representing
entire, long streets are selected; for the point-axial version of it, each dot represents
an intersection of two lines, and these lines are selected according to their relative
condition in the system. The spatial opportunity measure for demand areas, in
which points representing precise locations are ranked according to their relative
accessibility to supply locations, is described in a similar way.
The indicators have shown equal acute responsiveness to discrete changes in the
spatial, as well as the functional, configuration of systems. Applications of measures
to settings, varying the composition of the grid and the profiles of the supply loca-
tions, have produced diverse results, demonstrating the ability of the model to
respond to discrete changes. In a test area of more than 100 spaces, a 1% increase
in the size of a supply point was enough to produce altered results throughout the
system, with changes detected in the scores of several points. For the same system,
an increase of 2% in the size of the same supply point was enough to change the
relative positions of some other points. Similarly, changes in grid configuration have
been detected by the model, provided that an altered connectivity matrix was
produced because of such a change.
Tests on the consequences of whether or not the deterrent effect of distance
should be considered in the model were performed for several test areas. The results
both for dimensionless (adjacency based) and for cartesian (metric, straight-line
46 R Krafta

distance between the pair of points) grid systems were similar and did not alter the
ranking of spaces within the system considered. Tests were performed on areas
configured as orthogonal, slightly deformed grids, in which the distribution of lines
and corners is fairly regular. In these cases, the sequential ordering of lines and
corners could reproduce roughly the distribution of distances between points and so

Figure 8. A series of sliced 'relief maps showing the results of a test area (a, c, e, g) and
point-axial (b, d, f, h) representations. The correspondence between the two representations,
with finer detail given by the latter, can be seen.
Urban convergence 47

produce correlated results. However, it is expected that grids with a greater degree
of deformation will distort such a correlation.
Limitation of the use of the model may derive from the grid representation. In
effect, point-axial maps tend to configure very large graphs, with hundreds of dots
and thousands of edges. Although the algorithm itself has shown the capacity to
handle such large amounts of data with economy (hundreds of pairs in an ordinary
personal computer, and several thousands in a Cray supercomputer), the prepara-
tion of files is time consuming and affords plenty of opportunities for error.
Another limitation in the use of the model derives from the spatial representa-
tion. Handling intraurban, very detailed, systems poses difficulties to the delimitation
and boundary identification of the system. In centrality measures (Krafta, 1991) the
effects of boundaries and systems of different sizes were explored with satisfactory
results. The model did not present problems in representing areas of different sizes,
whereas boundaries were achieved by considering natural barriers (connectivity
cutoff) and axial extensions outside the area. In point-axial maps and spatial con-
vergence models, tests for system size effects need to be conducted; boundary
definitions should also be revised.

10 15 20 25 30 35 40
Figure 9. Measurements of two test areas plotted on the basis of axial (curve 1) and point-
axial (curve 2) representations for corresponding system spaces. The horizontal and vertical
axes represent ordered spaces and convergence measure values, respectively.

6 Conclusions
The proposed model differs from previous ones in the following ways.
(a) A probabilistic approach is taken to the user's choice among alternative supply
locations. No supply point is excluded a priori from the user's choice, although they
are considered in a context in which proximity and relative position increase the
chance of being chosen. Being located near demand locations and on the route
between demand and other supply locations are essential for the success of supply
locations.
(b) In the probability of the supply location satisfying facility demand related to the
size and variety of that supply location is considered, similarly to the method
suggested by Arentze et al (1994). To this extent, the attractiveness of supply loca-
tions associated to the relative positions constitutes a powered network to which
demand locations converge unevenly. The model computes trips from each demand
location to all supply locations, weighting each one according to relative position
and attractiveness.
48 R Krafta

(c) Spatial configuration is described carefully and given great importance, in the
sense that all routing possibilities are examined and weighted accordingly.
(d) The algorithm enables simultaneous measures of convergence, opportunity, and
potentiality to be taken into account. They correspond to a systematic evaluation of
supply locations, through which attractors associated to spatial configurations are
related to demand locations, to an evaluation of demand locations, which measures
their relative position to supply locations, and to the supply potentiality of demand
locations, respectively.
As a decision-support instrument, spatial convergence seems to be able to offer
support to planning-management problems, give a systematic evaluation of urban
situations and assess the expected performance of virtual ones, from the point of
view of the qualified distribution of urban facilities in relation to residential areas.
The proposed indicators are responsive to other aspects of urban morphology, such
as road conditions and capacity, transport facilities, etc. To this extent, it can
interact with other indicators within a complex geographic information systems
environment, and give a comprehensive assessment of urban investment alternatives.
References
Arentze T A, Borgers A W J, Timmermans H J P , 1994, "Multistop-based measurement
of accessiblity in a GIS environment" International Journal of Geographical Information
Systems 8 343-356
Hillier B, Penn A, Hanson J, Grajewski T, Xu J, 1993, "Natural movement: or, configuration
and attraction in urban pedestrian movement" Environment and Planning B: Planning and
Design 20 2 9 - 6 6
Krafta R, 1991 A Study of Intra-urban Configurational Development in Porto Alegre—Brazil
unpublished PhD thesis, The Martin Centre for Architectural and Urban Studies,
University of Cambridge, Cambridge
Krafta R, 1993, "Avaliacao do impacto de alguns instrumentos urbanisticos sobre o
desenvolvimento configuracional urbano", Anais do Seminario Internacional sobre
avaliacao dos instrumentos de intervencao urbana, (magnetic) Universidade de Sao
Paulo, FAU; copy available from the author
Krafta R, 1994, "Modelling intraurban configurational development" Environment and
Planning B: Planning and Design 21 6 7 - 8 2
Kruger M J T, 1979, "An approach to built-form connectivity at urban scale: system
description and its representation" Environment and Planning B: Planning and Design 6
67-88
Kruger M J T, 1989 On Node and Axial Maps: Distance Measures and Related Topics Second
European Conference on Management and Representation of Urban Change,
Cambridge/ United Kingdom; copy available from the Bartlett School of Architecture,
University College, London

p © 1996 a Pion publication printed in Great Britain