A

E
S
O
P

A
E
S
O
P


RECONSTRUCTING URBAN COMPLEXITY
From Temporal Descriptions of Historical Growth to Synthetic Design Models
16/01/2014

Kinda Al_Sayed

k.sayed@ucl.ac.uk




Space Group

B a r t l e t t S c h o o l o f G r a d u a t e S t u d i e s
UNIVERSITY COLLEGE LONDON
A
E
S
O
P

A
E
S
O
P

Understanding
Decoding urban complexity
Design
Reconstructing urban complexity
A
E
S
O
P

A
E
S
O
P

TOWARDS MODELLING SPATIAL DYNAMICS IN URBAN SYSTEMS
Add hoc models in architecture and urban design
Cellular automaton with agent modelling (Batty, 1991)
Multi-layer agent model (Krafta et. al., 2003)
Centrality as a process (Hillier, 1999)
Generative experiments (Hillier and Hanson, 1984)
Integration is static and choice is dynamic (Hillier et. al., 1987)
Changes in the shape of cities (Hillier & Hanson, 1993)
Centrality and extension (Hillier, 2002)
Selforganisation in organic grid (Hillier, 2004)
Demand and supply agent model (Krafta, 1994)
2
0
1
1





































































































1
9
8
0


U
N
D
E
R
S
T
A
N
D
I
N
G

M
O
D
E
L
L
I
N
G

MODELLING SPATIAL DYNAMICS IN URBAN SYSTEMS
Self-Organization and the City (Portugali, 2000)
A
E
S
O
P

A
E
S
O
P

Space Syntax versus Complexity Science
Cities are Complex! Cities are Simple!
http://otp.spacesyntax.net/methods/urban-methods-2/interpretive-models/

Hillier, B. (2009). The genetic code for citiesis it simpler than we thought?. in proceedings
of complexity theories of cities have come of age at tu delft september 2009
http://en.wikipedia.org/wiki/File:Logistic_Bifurcation_map_High_Resolution.png

Strogatz, Steven H. Nonlinear Dynamics and Chaos. Cambridge MA: Perseus, 1994.
Devaney, Robert L. An Introduction to Chaotic Dynamical Systems. Menlo Park, CA: Benjamin/Cummings, 1986.
The tension between chaos and order often keeps cities on the edge of chaos _ a
situation that enables cities to be adaptive complex systems and withstand
environmental changes. (Portugali, 2012)

"Chaos is aperiodic long-term behavour in a deterministic system that exhibits sensitive
dependence on initial conditions" (Strogatz: 323).
Let V be a set. The mapping f: V V is said to be chaotic on V if:
1. f has sensitive dependence on initial conditions,
2. f is topologically transitive (all open sets in V within the range of f interact under f),
3. periodic points are dense in V. (Devaney 50)

"A chaotic map possesses three ingredients: unpredictability, indecomposability, and an
element of regularity "(Devaney: 50).
A city is a network of linked centres at all scales set into a
background network of residential space. We then show that this
universal pattern comes about in two interlinked but conceptually
separable phases: a spatial process through which simple spatial
laws govern the emergence of characteristically urban patterns of
space from the aggregations of buildings; and a functional process
through which equally simple spatio-functional laws govern the way
in which aggregates of buildings becomes living cities. It is this dual
process that is suggested can lead us in the direction of a genetic
code for cities. (Hillier, 2009)
A
E
S
O
P

A
E
S
O
P

Premise
Cities show an autonomous behaviour, where
local processes appear to reinforce natural
patterns of growth and differentiation.


+ -
addition deletion edges
Urban
System
Subdivision mergence middle


Positive feedback Reinforcing feedback
A
E
S
O
P

A
E
S
O
P

On order, structure and randomness: where do urban systems fall?











Metric Mean Depth MMD Radius 1000metric Normalised Angular integration Radius n
A
E
S
O
P

A
E
S
O
P

Goal/purpose/rule Assumption
Y = 99.877e
0.1622x

R = 0.9657
A model can be outlined from the process of growth and structural differentiation in cities











Condition






Expansion affordances Space to expand people to occupy
Will determine whether positive or negative dynamic changes
Input




Mapping and externalising
growth dynamics in
historical growth patterns

spatial system


Output






simulations and short term
predictions
Searching for clues in the historical growth patterns of Barcelona and Manhattan
A
E
S
O
P

A
E
S
O
P

Method/Product

Looking for invariants in the transformations of street networks
Temporal Mapping of Historical Growth
Mapping transformations in-between synchronic
states of the growing system


A dynamic model
That implements generative rules
A
E
S
O
P

A
E
S
O
P

Method/Product

DECODING AND ENCODING GROWTH DYNAMICS
Temporal mapping of historical
growth
Mapping transformations in-between
synchronic states of the growing system


Extract an invariant that marks
growth patterns
State A
State A+1
State A+2
Infer invariants from
urban growth patterns
State T = transformation (A, A+1)
Make assumptions on how
they contribute to urban
growth
A
E
S
O
P

A
E
S
O
P

Emergence
Generative growth is a bottom up activity. Given the condition of spatio-
temporal configurations in the street network, a generative mechanism
operates to allow for the emergence of new elements and patches.
A
E
S
O
P

A
E
S
O
P

PREFERENTIAL ATTACHMENT
where configurational increase in the network accessibility is likely to
occur, new elements/patches tend to attach to existing street structures
A
E
S
O
P

A
E
S
O
P

A matrix of maps plotting changes in integration (radius 500m) over time
CHANGE WILL NOT LEAD TO CHANGE
A
E
S
O
P

A
E
S
O
P

Waves of change in integration values transferring from the core of Barcelona towards the edges
CHANGE TRANSFERS

1806-1855 1855-1891 1891-1920


1920-1970 1970-2010

A
E
S
O
P

A
E
S
O
P

Preferential attachment
Angular choice is generative globally
A
E
S
O
P

A
E
S
O
P

PRUNING
Once the growing structure reaches its maximum boundaries, patches
with low local integration will tend to disappear
A
E
S
O
P

A
E
S
O
P

Colour range 3colours at 130
PRUNING
Weak local structures are trimmed down
Manhattan (current state) Manhattan (gaps filled)
A
E
S
O
P

A
E
S
O
P

Self-organisation
Self-organisation mechanisms are likely to have a role in maintaining a
part-whole structural unity. As a side effect of this process, a fractal
structure emerges in the form of monocentric patchwork patterns that
have certain metric proximity. the overall distance between patches
approximates one and half the radius that defines them.
A
E
S
O
P

A
E
S
O
P

Manhattan Barcelona




Distances between each two neigbouring patches linking their peaks
higher MMD R 1000 values marking patchwork patterns in the physical street
network

Manhattan Barcelona




MMD Radius 1000

MMD Radius 2000
1780 2000 1780 2000



Manhattan Barcelona




Distances between each two neigbouring patches linking their peaks
higher MMD R 1000 values marking patchwork patterns in the physical street
network

Manhattan Barcelona




MMD Radius 1000

MMD Radius 2000
1780 2000 1780 2000



Manhattan Barcelona




MMD Radius 1000

MMD Radius 2000
1780 2000 1780 2000



DISTANCE CONSERVATION BETWEEN PATCHES
Clusters were derived directly from MMD radius 1000 metric
Al_Sayed K. (2013). The Signature of
Self-Organisation in Cities: Temporal
patterns of clustering and growth in
street networks, International Journal
of Geomatics and Spatial Analysis
(IJGSA), Special Issue on
Spatial/Temporal/Scalar Databases
and Analysis, In M. Jackson & D.
Vandenbroucke (ed), 23 (3-4).
A
E
S
O
P

A
E
S
O
P

SPATIAL BEHAVIOUR IN CITIES
Overlaying two maps; Angular choice map R 6000 metres and Patchwork map, metric mean depth within radius 1000
Barcelona
Distinguishing two layers in the spatial structure: a background & a foreground
Cities grow naturally wherever an emergent bottom-up
activity is possible
Cities deform to differentiate the uniform grid either by
intensifying the grid where more through-movement is
expected or by pruning weak local structures.
In a process of preferential attachment, city structure
records a certain memory wherever integration change
takes place and recalls this memory to attach to new
elements.
This process is continuously updated once the system
reconfigures its local settings.
The system is apt to to fit within a certain distribution
and tends to conserve metric distance between
patches.
Structural differentiation aims to adapt the grid to
match organic city structures.
Spatial structures in cities can be considered as
independent systems that are self-generative and self-
organised.

Angular choice R 6000 metres against MMD R 500

Angular choice Rn metres against MMD R 1000

Angular choice Rn metres against MMD R 2000
A
E
S
O
P

A
E
S
O
P

INVARIANTS OF GROWTH OR RULES FOR URBAN PATTERN RECOGNITION
FIRST : Skewed distribution of angular depth

SECOND : street structures are moderately intelligible

THIRD : Choice overcomes the cost of depth

FOURTH : Conserved distance between patches

A
E
S
O
P

A
E
S
O
P

Research questions
What type of mechanism is needed to convert an
explanatory reading of architectural phenomena into a
synthetic and yet creative design approach?

How far can we automate an urban design process?


Is a systematic design approach counter-creative?

A
E
S
O
P

A
E
S
O
P

In search for a sensible approach

DESIGN FILTERS IN SPACE SYNTAX (HILLIER, 1996)
Genotype
Generic function of space (movement
and occupation)
Phenotype Cultural identity (locality & time)
Phenotype of Phenotype Individual cultural identity
Cultural identity?
A
E
S
O
P

A
E
S
O
P

Generic function/
genotype
The affordances of a spatial structure as a
physical domain for movement and occupation
Direct Phenotype parameters that can be interpreted numerically
Indirect Phenotype Qualitative properties (aesthetic, identity)
REDEFINING DESIGN FILTERS
A
E
S
O
P

A
E
S
O
P

A PRIORITISED-STRUCTURED MODEL OF DESIGN THINKING
1. The first set of design filters or rules are mainly
based on quantifiable spatial attributes of street
networks.
2. The second set of design filters is dependent on the
first set of spatial measures. It accounts for the
relationships between street spaces and general
formal and functional attributes of urban regions.
3. The third set of design filters will have weak
dependencies on the first set but will be constrained
by quantifiable and well-defined variables
(environmental measures, natural lighting, etc).
4. The fourth design filter is where singularities can be
presented to reflect on design idealism, individual or
communal cultures.
Space



Space-dependant parameters




Other quantitative parameters



Qualitative criteria
A
E
S
O
P

A
E
S
O
P

To inform our synthetic urban design model we
need to extract rules from real urban systems


We need rules that define
The first filter
The second filter

We will only talk about the first two filters, and probably
explore the role of a designer in defining the final solution
A
E
S
O
P

A
E
S
O
P

First filter
For the first filter we define rules that capture the growth of street
networks and rules to filter best performing street configurations
Temporal mapping
Mapping transformations in-between
synchronic states of the growing system
Invariants
that help recognising urban
growth patterns

A generative model
That implements simple
generative rules
State A
State A+1
State A+2
Extract invariants
State T = transformation (A, A+1)
Use the invariants to assess
structures generated by the
model
Rules for defining the invariant patterns that characterise street networks
A
E
S
O
P

A
E
S
O
P

A SIMPLE GENERATIVE GROWTH MODEL
















Simple generative rule (Centrality and extension, Hillier, 2002)
Do not block a longer alignment if it is possible to block a shorter alignment
A
E
S
O
P

A
E
S
O
P

Generative Networks: comparing growth iterations to real and random systems
Iteration 1 Iteration 2 Iteration 3 Iteration 4 Barcelona random structure
Iterations



Choice Rn



TD/NC 15726.2 11366.29 7747.37 8109.46 1756.161 3564.98
MMD R500metric





Distribution

KSL test 0.09 0.15 0.088 0.1 0.035 0.04
Skewness -1.32 -1.15 -1.46 -1.5 0.28 -0.18
Intelligibility R
2
=0.12 R
2
=0.13 R
2
=0.17 R
2
=0.1 R
2
=0.33 R
2
=0.56
Synergy R
2
=0.37 R
2
=0.35 R
2
=0.36 R
2
=0.38 R
2
=0.62 R
2
=0.80

I
A
E
S
O
P

A
E
S
O
P

FIRST : Skewed distribution of angular depth

SECOND : street structures are moderately intelligible

THIRD : Choice overcomes the cost of depth

FOURTH : Conserved distance between patches

Observed invariants that help recognising urban patterns
A
E
S
O
P

A
E
S
O
P

Evaluating growth iterations
Iteration 1 Iteration 2 Iteration 3 Iteration 4 Barcelona random structure
Iterations



Choice Rn



TD/NC 15726.2 11366.29 7747.37 8109.46 1756.161 3564.98
MMD R500metric





Distribution

KSL test 0.09 0.15 0.088 0.1 0.035 0.04
Skewness -1.32 -1.15 -1.46 -1.5 0.28 -0.18
Intelligibility R
2
=0.12 R
2
=0.13 R
2
=0.17 R
2
=0.1 R
2
=0.33 R
2
=0.56
Synergy R
2
=0.37 R
2
=0.35 R
2
=0.36 R
2
=0.38 R
2
=0.62 R
2
=0.80

I
A
E
S
O
P

A
E
S
O
P

Second filter
Rules for defining the relationship between street networks and Form-Function
A
E
S
O
P

A
E
S
O
P

> ==

Pixelmapper* binning Spatial and urban data

* Al_Sayed, Space Syntax as a parametric model (2011)
For the second filter we define latent variables that capture the
relationship between spatial structure and form-function parameters
Method for mapping space-form-function
A
E
S
O
P

A
E
S
O
P

ANNs model to forecast form-function attributes by means of spatial factors






Input
Activation functions stored
in 3 hidden nodes
Output

S
p
a
t
i
a
l

f
a
c
t
o
r
s


F
o
r
m
-
f
u
n
c
t
i
o
n

r
e
s
p
o
n
s
e
s


A
E
S
O
P

A
E
S
O
P

ANNs model to forecast form-function attributes by means of spatial factors





Applied and validated against Barcelona

Tested against Manhattan

Used to forecast form-function attributes

Streets wider than 30m
Street width predictions
Block centroids
Block density predictions

Building heights above 35m
Estimated building height
Superstores
Estimated retail activity


Streets wider than 30m
Street width predictions
Block centroids
Block density predictions

high-rise density above 100m
Estimated building height
Superstores
Estimated retail activity

A
E
S
O
P

A
E
S
O
P

Predicted form-function attributes
Block density High-rises Commercial zones Street width



Higher values of each measure ___ Routes with high choice values [SLW]

A
E
S
O
P

A
E
S
O
P

Reconstructing urban
complexity
from predictions to
formalised design
propositions
A
E
S
O
P

A
E
S
O
P

Design variations
A
E
S
O
P

A
E
S
O
P

Publications

Al_Sayed, K. (2014) Thinking systems in Urban Design: A prioritised structure model. In Explorations in
Urban Design. M. Carmona (ed), (Farnham: Ashgate, 20XX), Copyright 2014
Al_Sayed K. (2013). Synthetic Space Syntax: A generative and supervised learning approach in urban
design, In Proceedings of the 9th International Space Syntax Symposium, Edited by Y O Kim, H T Park, K
W Seo, Seoul, Korea.
Al_Sayed K. (2013). The Signature of Self-Organisation in Cities: Temporal patterns of clustering and
growth in street networks, International Journal of Geomatics and Spatial Analysis (IJGSA), Special Issue
on Spatial/Temporal/Scalar Databases and Analysis, In M. Jackson & D. Vandenbroucke (ed), 23 (3-4).
Al_Sayed, K. (2012) A systematic approach towards creative urban design. In Design Computing and
Cognition DCC12. J.S. Gero (ed), pp. xx-yy. Springer 2012.
Al_Sayed, K., Turner, A. (2012) Emergence And Self-Organization In Urban Structures, In Proceedings of
AGILE 2012, Avignon, France.
Al_Sayed, K. (2012) Urban Pattern Recognition In Generative Structures, In Proc. of AGILEs workshop on
Complexity Modelling for Urban Structure and Dynamics, Avignon, France.
Al_Sayed K., Turner A., Hanna S. (2012). Generative Structures In Cities, In Proc. of the 8th International
Space Syntax Symposium, Edited by Margarita Greene, Jos Reyes, Andrea Castro, Santiago de Chile:
PUC, 2012..
Al_Sayed K., Turner A., Hanna S. (2010). Spatial Morphogenesis in Cities: A Generative Urban Design
Model, Pro. of the 10th International Conference On Design And Decision Support Systems In
Architecture And Urban Planning, Edited by In: Harry JP Timmermans, Eindhoven.
Al_Sayed K., Turner A., Hanna S. (2009). Cities as emergent models: The morphological logic of
Manhattan and Barcelona, In Proc. of the 7th International Space Syntax Symposium, Edited by Daniel
Koch, Lars Marcus and Jesper Steen, Stockholm: KTH, 2009.
A
E
S
O
P

A
E
S
O
P

THANK YOU!