Presentation by John Östh, Aura Reggiani
& Laurie Schintler
Advanced Brainstorm Carrefour (ABC): ‘Smart People in Smart Cities’
Matej Bel University, Banská Bystrica, Slovakia (August, 2016)
1. Resilience in Spatial and Urban
Systems 2
John Östh, Aura Reggiani
& Laurie Schintler
Smart People in Smart Cities
Faculty of Economics, Matej Bel University &
Regional Science Academy & The City of
Banská Bystrica
2. Presentation
• The main idea
• Theoretical framework
– Central Place Theory
– Self-Organization Theory
• Questions
• Data
– Mobile phone data
– GIS data
• Methods
– Setting up a a self-organizing BigData dataset
• Results
• Conclusions
3. The main idea
• There is an increasing amount of papers
discussing urban and regional resilience.
• However, most times the geography of urban
areas and regions are taken for granted – i.e. the
spatial administrative organization of urban areas
and regions may or may not be mismatching the
functional regions.
• The main idea is to make use self-organization
methods to trace the spatial patterns of the
urban and regional fabric
4. Central Place Theory
• Invented the study of systems of cities and the
interrelationship between cities.
– Assuming that:
• Space is flat, population and resources evenly
distributed.
• Competition, cost and direction for transport, etc.
identical throughout space
– Concepts
• Threshold – minimum population needed for x
• Range – maximum distance population is willing to
commute
Christaller, W (1933), Die zentralen Orte in Süddeutschland. Gustav Fischer, Jena.
5. Central Place Theory and Sweden
• Year 1962-1971, a municipality reform redrew the borders
of Sweden
• Central for the process was Christaller and the CPT –
especially the idea about the administrative principle (k=7)
• This means that between 1962 and 1971, all Swedish
municipalities were redrawn so that:
– Central places became municipalities and gained control over
smaller urban areas and rural areas being near.
– Metropolitan areas were set aside due to the administrative
complexity and population size (became too populous to
administer as “local”)
– Some very remote areas were also set aside (threshold not met
but municipalities needed for administrative reasons).
Set aside ~ regions not determined on the basis of threshold and range
6. Self-Organization theories
”…finding that in certain situations external
forces acting on the system do not
determine/cause its behavior, but instead trigger
an internal and independent process by which
the system spontanelosuly self-organizes itself.”
(Portugali, 2000)
7. Self-Organization of Regions
• There is a large body of literature working on
self-organization – the amount of self-
organization literature that deals with regions
is smaller.
• However, using a wide definition…
8. Self-Organization of Regions
• Has been studied for a very long time:
– Von Thünen and the annuluses of economic
activities
– Alonso – bid/rent
– Christaller (1933) and Lösch (1940) – hierarchies
of activities
– Burgess (1925) and Hoyt (1939) – the morphology
of the urban landscape
9. Self-Organization of Regions
• Self-organizing methods are borrowed from
chemistry, physics, computer science and
math including:
– Fractals and related – i.e. sand pile cities, cellular
automata,…
– Game-related methods (see for instance Schelling)
(further reading Portugali; Batty)
10. Our approach to Self-Organization
• Starts with inspiration from Kohonen (1982, 2001) and
Self-Organizing Maps – where (at least) two interacting
subsystems are used to reposition neurons using a
spatially restricted and iterative learning process.
• We set up a method where mobile phones are
clustered using an iterative learning process where a
hypothetical gravitational force determines the spatial
realms of influence
• Why is this smart?
– Ai, factual flows, responsive and dynamic (not historical
data)…
11. Questions
• Overarching questions:
– Since CPT was used for the construction of
Swedish municipalities - can SO methods be
employed to determine CP?
– Can the Self-Organization of Phones be used to
delineate functional regions of today…tomorrow?
– Can regions of scales be constructed?
12. Data
• Comes from one of the major Swedish mobile
phone operators (among the largest 5)
• Network Driven Records (NDR) stored at the
MIND database at Uppsala University.
• Record all events (silent handovers, text, Internet,
Calls, etc.) and codes each event temporarily to
the nearest 5min interval – 288 temporal units in
24h
• Geography is restricted to mast-level
• Data drawn from a Tuesday in January in 2016
13. Data
• Used dataset contains:
– The average position of each phone and hour
(allowing for positions between masts)
– Each phone can appear in the dataset 24 times -
this is however unusual – in most cases phones
are idle for at least a few hours per 24h.
• Since we don’t want to introduce spurious locations
(i.e. back-tracking and assuming that phones are at the
same location at time t as at time t-1) – we only
position active phones.
– No data of activity or holder is included
14. Data
• To make handling of data easier, all average
coordinates are aggregated to the nearest
100m x 100m coordinate. The dataset still
contains of more than 1.6 million unique
locations of which the majority have more
than one phone
15. Data
• GIS data used to validate our SO-results
– GIS-layers depicting the distribution of urban
areas, municipality borders and of major water-
bodies
16. Methods: - setting up a SO dataset
• Assumption:
– Each phone exerts gravity.
– The gravitational force is modelled to decay
exponentially
– Decay parameter is derived mathematically using
a HLM design on observed mobility
(see Östh et al. 2016)
– Decay parameter value in this case = 0.00166
– Gravity is used as weight at distance dij
Alternative assumption: using Boolean k-borders (0|1) for the construction of thresholds
proved not to work – images available in the post-presentation section
17. Methods: - setting up a SO dataset
• The iterations are conducted using EquiPop
– K-nearest neighbour “contextualizer” for very large
datasets.
– In this study we set up EquiPop to retrieve the distance-
decay weighted average Y-coordinate (first) from the k
nearest neighbours, than the X-coordinate (second) from
the k nearest neighbours.
– We manipulate the outdata, constructing a new file with
updated Y and X coordinates and iterate the procedure
– In our studies, iterations were terminated at iteration 20
because there was no significant difference in cluster
mobility from previous state*
*for k = 50 000, the rule was thereafter applied to all ks
18. Methods: - setting up a SO dataset
• Determining k-values.
– Doubling sequences of k can
roughly be associated with
varying neighbourhood
functions
(Östh 2014; Östh et al. 2015)
– By applying the same
strategy to our SO regions
dataset, CP hierarchies can
be defined crudely
We constructed the following
k-phone regions:
6 250 phones
12 500 phones
25 000 phones
50 000 phones
100 000 phones
19. Methods – setting up a SO dataset
• Next slides will show how the 20 iterations
clustered the phones in the greater Stockholm
region
46. Creating phone areas surrounding each phone at initial
position is conducted using Thiessen polygon
techniques.
Using each area as a
building-block, and by
keeping trace of its
mobility over
iterations we may
piece together (dissolve)
areas that contribute to
a self-organized cluster
for each k at iteration 20
47. Results
• First section
– Self-organization of phones compared to the
spatial distribution of urban areas
• Second section
– Comparison of the spatial realms of municipalities
and the spatial realms of phone-origins for the
creation of self-organized clusters.
49. Self-organization of phones compared to
the spatial distribution of urban areas
• How many of the phone clusters end up
within urban areas?
– After iteration 20 and k=6250
(the most wide spread), including both clusters
reaching k and not reaching k:
• 8.3% of all phones end up in locations being more than
1000m from the nearest urban area
• 91.7% end up within or close to urban areas.
– Using only clusters reaching k:
• 100% of all phones end up in urban areas.
Since CPT was used for the construction
of Swedish municipalities - can SO
methods be employed to determine CP?
50. Comparing spatial realms
• The 1962 municipality delineation idea means
that very rural and very urban areas will not
match SO regions.
• Midsized municipalities will display strong
similarities with SO regions
• Can the Self-Organization of Phones be
used to delineate functional regions of
today?
• Can regions of scales be constructed?
54. Conclusion
• Self-organization of phones can be used to
create functional regions.
• Using phones of specific hours or using the
trajectories of phones could help to construct
different functional regions