1. A Framework for Measuring Urban Sprawl
from Crowd-Sourced Data
Mart´ı Bosch In collaboration with: Luciano Gervasoni
Supervised by: Serge Fenet, Peter Sturm
Grenoble, June 23, 2016
11. Timeline of An Ambiguous Term
1937 - First reference by Earle Draper in a confer-
ence of Urban Planners in Southeastern USA
12. Timeline of An Ambiguous Term
1937 - First reference by Earle Draper in a confer-
ence of Urban Planners in Southeastern USA
50s, 60s, 70s - Focus on characteristics: low den-
sity, leapfrog, ribbon development and causes: poor
planning, speculation, suburbs as “ideal place to live”
13. Timeline of An Ambiguous Term
1937 - First reference by Earle Draper in a confer-
ence of Urban Planners in Southeastern USA
50s, 60s, 70s - Focus on characteristics: low den-
sity, leapfrog, ribbon development and causes: poor
planning, speculation, suburbs as “ideal place to live”
80s-90s - New characteristic: land use mix and
consequences: car dependency, social segregation
14. Timeline of An Ambiguous Term
1937 - First reference by Earle Draper in a confer-
ence of Urban Planners in Southeastern USA
50s, 60s, 70s - Focus on characteristics: low den-
sity, leapfrog, ribbon development and causes: poor
planning, speculation, suburbs as “ideal place to live”
80s-90s - New characteristic: land use mix and
consequences: car dependency, social segregation
21st Century - More negative consequences on en-
vironment and health; new scope: GIS data science
5
16. So What is Urban Sprawl?
Without entering in subjective discussions, urban sprawl is:
6
17. So What is Urban Sprawl?
Without entering in subjective discussions, urban sprawl is:
• a dynamic process strongly attached to urban growth
6
18. So What is Urban Sprawl?
Without entering in subjective discussions, urban sprawl is:
• a dynamic process strongly attached to urban growth
• of multidimensional nature
6
19. So What is Urban Sprawl?
Without entering in subjective discussions, urban sprawl is:
• a dynamic process strongly attached to urban growth
• of multidimensional nature
• it is a loose research domain:
6
20. So What is Urban Sprawl?
Without entering in subjective discussions, urban sprawl is:
• a dynamic process strongly attached to urban growth
• of multidimensional nature
• it is a loose research domain: urban planning, geography,
economy, environmental sciences, health...
6
21. So What is Urban Sprawl?
Without entering in subjective discussions, urban sprawl is:
• a dynamic process strongly attached to urban growth
• of multidimensional nature
• it is a loose research domain: urban planning, geography,
economy, environmental sciences, health...
and increasingly computing and GIS data science
Despite that researches do no agree on a common definition, their
references to urban sprawl prove its importance
6
24. Contributions and Outline
• Cross-domain literature review of urban sprawl and its
dimensions
• Audit of proposed indicators:
• Translation of different spatial measures into a common
notation
• Extensive analysis of measures’ suitability to gauge urban
sprawl’s dimensions
7
25. Contributions and Outline
• Cross-domain literature review of urban sprawl and its
dimensions
• Audit of proposed indicators:
• Translation of different spatial measures into a common
notation
• Extensive analysis of measures’ suitability to gauge urban
sprawl’s dimensions
• Survey of data sources, appropriateness of crowd-sourced data
7
26. Contributions and Outline
• Cross-domain literature review of urban sprawl and its
dimensions
• Audit of proposed indicators:
• Translation of different spatial measures into a common
notation
• Extensive analysis of measures’ suitability to gauge urban
sprawl’s dimensions
• Survey of data sources, appropriateness of crowd-sourced data
• Presentation of a framework with a compuitng a justified
choice of urban sprawl indicators from crowd-sourced data
7
27. Contributions and Outline
• Cross-domain literature review of urban sprawl and its
dimensions
• Audit of proposed indicators:
• Translation of different spatial measures into a common
notation
• Extensive analysis of measures’ suitability to gauge urban
sprawl’s dimensions
• Survey of data sources, appropriateness of crowd-sourced data
• Presentation of a framework with a compuitng a justified
choice of urban sprawl indicators from crowd-sourced data
• Experimentation in real cities
7
31. Dimensional Decomposition
Based on the reviewed literature, the following dimensional
decomposition is proposed:
(i) Density: arguably urban sprawl’s most important dimension
8
32. Dimensional Decomposition
Based on the reviewed literature, the following dimensional
decomposition is proposed:
(i) Density: arguably urban sprawl’s most important dimension
(ii) Patch Distribution: how urban patches (excluding
transportation network) are distributed, regardless of such
patches’ uses
8
33. Dimensional Decomposition
Based on the reviewed literature, the following dimensional
decomposition is proposed:
(i) Density: arguably urban sprawl’s most important dimension
(ii) Patch Distribution: how urban patches (excluding
transportation network) are distributed, regardless of such
patches’ uses
(iii) Land Use Mix: how land uses are allocated over a
distribution of urban patches
8
34. Dimensional Decomposition
Based on the reviewed literature, the following dimensional
decomposition is proposed:
(i) Density: arguably urban sprawl’s most important dimension
(ii) Patch Distribution: how urban patches (excluding
transportation network) are distributed, regardless of such
patches’ uses
(iii) Land Use Mix: how land uses are allocated over a
distribution of urban patches
(iv) Accessibility: how well the transportation network serves the
urban development
8
35. Dimensional Decomposition
Based on the reviewed literature, the following dimensional
decomposition is proposed:
(i) Density: arguably urban sprawl’s most important dimension
(ii) Patch Distribution: how urban patches (excluding
transportation network) are distributed, regardless of such
patches’ uses
(iii) Land Use Mix: how land uses are allocated over a
distribution of urban patches
(iv) Accessibility: how well the transportation network serves the
urban development
Such dimensions have to be considered altogether to fully gauge
urban sprawl
8
37. Dimensional Decomposition
Density: units of a magnitude (i.e.
number of housing units, number of
residents, number of jobs...) over the
considered region’s area
9
38. Dimensional Decomposition
Density: units of a magnitude (i.e.
number of housing units, number of
residents, number of jobs...) over the
considered region’s area
Patch Distribution
9
39. Dimensional Decomposition
Density: units of a magnitude (i.e.
number of housing units, number of
residents, number of jobs...) over the
considered region’s area
Patch Distribution
Land Use Mix: allocation of uses
9
40. Dimensional Decomposition
Density: units of a magnitude (i.e.
number of housing units, number of
residents, number of jobs...) over the
considered region’s area
Patch Distribution
Land Use Mix: allocation of uses Accessibility: transportation network
9
41. Dimensional Decomposition
Density: units of a magnitude (i.e.
number of housing units, number of
residents, number of jobs...) over the
considered region’s area
Patch Distribution
Land Use Mix: allocation of uses Accessibility: transportation network
9
44. Sources of Data: OpenStreetMap
The availability of data is a huge delimiter of the scope of a study
10
45. Sources of Data: OpenStreetMap
The availability of data is a huge delimiter of the scope of a study
Several sources provide (potentially non-open) data in
heterogeneous formats that can be hard to integrate...
10
46. Sources of Data: OpenStreetMap
The availability of data is a huge delimiter of the scope of a study
Several sources provide (potentially non-open) data in
heterogeneous formats that can be hard to integrate...
OpenStreetMap (OSM) is a collaborative project to create a free
editable map of the world.
10
47. Sources of Data: OpenStreetMap
The availability of data is a huge delimiter of the scope of a study
Several sources provide (potentially non-open) data in
heterogeneous formats that can be hard to integrate...
OpenStreetMap (OSM) is a collaborative project to create a free
editable map of the world. Its API provides (among others):
10
48. Sources of Data: OpenStreetMap
The availability of data is a huge delimiter of the scope of a study
Several sources provide (potentially non-open) data in
heterogeneous formats that can be hard to integrate...
OpenStreetMap (OSM) is a collaborative project to create a free
editable map of the world. Its API provides (among others):
• Transportation Network: including a street graphs, railway
graphs, bus routes and stops...
10
49. Sources of Data: OpenStreetMap
The availability of data is a huge delimiter of the scope of a study
Several sources provide (potentially non-open) data in
heterogeneous formats that can be hard to integrate...
OpenStreetMap (OSM) is a collaborative project to create a free
editable map of the world. Its API provides (among others):
• Transportation Network: including a street graphs, railway
graphs, bus routes and stops...
• Points of Interest (POIs): which can be public facilities,
amenities, shops...
10
50. Sources of Data: OpenStreetMap
The availability of data is a huge delimiter of the scope of a study
Several sources provide (potentially non-open) data in
heterogeneous formats that can be hard to integrate...
OpenStreetMap (OSM) is a collaborative project to create a free
editable map of the world. Its API provides (among others):
• Transportation Network: including a street graphs, railway
graphs, bus routes and stops...
• Points of Interest (POIs): which can be public facilities,
amenities, shops...
Existing libraries extract data from OSM API into common formats
10
51. Sources of Data: OpenStreetMap
The availability of data is a huge delimiter of the scope of a study
Several sources provide (potentially non-open) data in
heterogeneous formats that can be hard to integrate...
OpenStreetMap (OSM) is a collaborative project to create a free
editable map of the world. Its API provides (among others):
• Transportation Network: including a street graphs, railway
graphs, bus routes and stops...
• Points of Interest (POIs): which can be public facilities,
amenities, shops...
Existing libraries extract data from OSM API into common formats
Its contributions have been continuously increasing and several
quality metrics have been proposed.
10
54. Notation
Let:
• S denote the region to study
• Ω denote an areal decomposition (grid) into N = |Ω|
sub-areas of S.
11
55. Notation
Let:
• S denote the region to study
• Ω denote an areal decomposition (grid) into N = |Ω|
sub-areas of S.
• ai denote the surface of the sub-area i ∈ Ω and A the total
area of S
11
56. Notation
Let:
• S denote the region to study
• Ω denote an areal decomposition (grid) into N = |Ω|
sub-areas of S.
• ai denote the surface of the sub-area i ∈ Ω and A the total
area of S
• f
(k)
i denote the number of POIs of land use k located inside
the sub-area i ∈ Ω, and F(k) its total over all S
11
57. Notation
Let:
• S denote the region to study
• Ω denote an areal decomposition (grid) into N = |Ω|
sub-areas of S.
• ai denote the surface of the sub-area i ∈ Ω and A the total
area of S
• f
(k)
i denote the number of POIs of land use k located inside
the sub-area i ∈ Ω, and F(k) its total over all S
Given the specification of OSM’s API, two land use types are
defined: activity (act) and residential (res).
11
59. Smoothing Measures
Measures can depend greatly on the accuracy of the data (i.e.
missing data) and the used areal decomposition Ω
12
60. Smoothing Measures
Measures can depend greatly on the accuracy of the data (i.e.
missing data) and the used areal decomposition Ω
Given a set of POIs, the kernel density estimation (KDE)
interpolates a continuous surface ψ
12
61. Smoothing Measures
Measures can depend greatly on the accuracy of the data (i.e.
missing data) and the used areal decomposition Ω
Given a set of POIs, the kernel density estimation (KDE)
interpolates a continuous surface ψ
Figure 5: Set of POIs (left) and its KDE (right). Grenoble, France
12
62. Smoothing Measures
Measures can depend greatly on the accuracy of the data (i.e.
missing data) and the used areal decomposition Ω
Given a set of POIs, the kernel density estimation (KDE)
interpolates a continuous surface ψ
Figure 5: Set of POIs (left) and its KDE (right). Grenoble, France
Advantages: fair inference of POI’s density (i.e. for missing data)
12
63. Smoothing Measures
Measures can depend greatly on the accuracy of the data (i.e.
missing data) and the used areal decomposition Ω
Given a set of POIs, the kernel density estimation (KDE)
interpolates a continuous surface ψ
Figure 5: Set of POIs (left) and its KDE (right). Grenoble, France
Advantages: fair inference of POI’s density (i.e. for missing data)
Drawbacks: inaccurate inference (i.e. in rivers, sea, cliffs...)
12
66. Proposed Measures: Density
(i) Density: is determined for both act and res types:
ρ(act)
=
F(act)
A
; ρ(res)
=
F(res)
A
13
67. Proposed Measures: Density
(i) Density: is determined for both act and res types:
ρ(act)
=
F(act)
A
; ρ(res)
=
F(res)
A
Other measures such as density gradients or percentiles have
been empirically proved to be strongly correlated to density
itself.
13
69. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Evenness: Shannon’s Entropy
H(k)
= −
i∈Ω
¯ψ
(k)
i
F(k)
ln(
¯ψ
(k)
i
F(k)
)
14
70. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Evenness: Shannon’s Entropy
H(k)
= −
i∈Ω
¯ψ
(k)
i
F(k)
ln(
¯ψ
(k)
i
F(k)
)
substituting k for act and res.
14
71. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Evenness: Shannon’s Entropy
H(k)
= −
i∈Ω
¯ψ
(k)
i
F(k)
ln(
¯ψ
(k)
i
F(k)
)
substituting k for act and res. It operates over the average k’s
KDE density in i as in:
14
72. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Evenness: Shannon’s Entropy
H(k)
= −
i∈Ω
¯ψ
(k)
i
F(k)
ln(
¯ψ
(k)
i
F(k)
)
substituting k for act and res. It operates over the average k’s
KDE density in i as in:
¯ψi =
1
ai i
ψ(k)
da
14
73. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Evenness: Shannon’s Entropy
H(k)
= −
i∈Ω
¯ψ
(k)
i
F(k)
ln(
¯ψ
(k)
i
F(k)
)
substituting k for act and res. It operates over the average k’s
KDE density in i as in:
¯ψi =
1
ai i
ψ(k)
da
since logarithms cannot be determined for i ∈ Ω with no POIs
(i.e. a park with no residential units).
14
74. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Evenness: Shannon’s Entropy
H(k)
= −
i∈Ω
¯ψ
(k)
i
F(k)
ln(
¯ψ
(k)
i
F(k)
)
substituting k for act and res. It operates over the average k’s
KDE density in i as in:
¯ψi =
1
ai i
ψ(k)
da
since logarithms cannot be determined for i ∈ Ω with no POIs
(i.e. a park with no residential units).
It can be rescaled to yield a [0, 1] range (zero for compact
distributions and one for sprawl) as in: H (k)
= H(k)
ln(N)
14
76. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Clustering: Moran’s I (spatial autocorrelation)
I(k)
=
N
N
i=1
N
j=1 wi,j
N
i=1
N
j=1 wi,j (f
(k)
i − ¯f (k)
)(f
(k)
j − ¯f (k)
)
N
i=1(f
(k)
i − ¯f (k))2
15
77. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Clustering: Moran’s I (spatial autocorrelation)
I(k)
=
N
N
i=1
N
j=1 wi,j
N
i=1
N
j=1 wi,j (f
(k)
i − ¯f (k)
)(f
(k)
j − ¯f (k)
)
N
i=1(f
(k)
i − ¯f (k))2
weighting with the inverse distance as in:
wi,j =
1
di,j
15
78. Proposed Measures: Patch Distribution
(ii) Patch Distribution:
• Clustering: Moran’s I (spatial autocorrelation)
I(k)
=
N
N
i=1
N
j=1 wi,j
N
i=1
N
j=1 wi,j (f
(k)
i − ¯f (k)
)(f
(k)
j − ¯f (k)
)
N
i=1(f
(k)
i − ¯f (k))2
weighting with the inverse distance as in:
wi,j =
1
di,j
It ranges from -1 (negative autocorrelation) to +1 (positive
autocorrelation, high-density areas tend to be clustered
together). A value of zero indicates random scattering
15
80. Proposed Measures: Land Use Mix
(iii) Land Use Mix:
• Evenness: Dissimilarity Index (separation of land uses)
D(k,l)
=
1
2
i∈Ω
|
f
(k)
i
F(k)
−
f
(l)
i
F(l)
|
16
81. Proposed Measures: Land Use Mix
(iii) Land Use Mix:
• Evenness: Dissimilarity Index (separation of land uses)
D(k,l)
=
1
2
i∈Ω
|
f
(k)
i
F(k)
−
f
(l)
i
F(l)
|
with k = act, l = res. D(act,res)
is symmetric w.rt. act, res:
D(act,res)
= D(res,act)
16
82. Proposed Measures: Land Use Mix
(iii) Land Use Mix:
• Evenness: Dissimilarity Index (separation of land uses)
D(k,l)
=
1
2
i∈Ω
|
f
(k)
i
F(k)
−
f
(l)
i
F(l)
|
with k = act, l = res. D(act,res)
is symmetric w.rt. act, res:
D(act,res)
= D(res,act)
The range of D(act,res)
starts at zero indicating high mix of act
and res uses, and approaches one when no mix
16
83. Proposed Measures: Land Use Mix
(iii) Land Use Mix:
• Evenness: Dissimilarity Index (separation of land uses)
D(k,l)
=
1
2
i∈Ω
|
f
(k)
i
F(k)
−
f
(l)
i
F(l)
|
with k = act, l = res. D(act,res)
is symmetric w.rt. act, res:
D(act,res)
= D(res,act)
The range of D(act,res)
starts at zero indicating high mix of act
and res uses, and approaches one when no mix
Advantage: D(act,res)
implicitly weights the sub-areas i ∈ Ω in
proportion to their density of activities f
(act)
i and residential
units f
(res)
i
16
88. Considered Dataset
´Avila Spanish medieval town with its historical center surrounded
by prominent walls
• Area: 40.93 km2
• POIs: 291 activities, 326 residential
18
89. Considered Dataset
´Avila Spanish medieval town with its historical center surrounded
by prominent walls
• Area: 40.93 km2
• POIs: 291 activities, 326 residential
18
91. Considered Dataset
Chandigarh Indian city planned by Le Corbusier to evenly
distribute residential units and facilities
• Area: 474.34 km2
• POIs: 825 activities, 32848 residential. Observation:
potentially missing activity POIs
19
92. Considered Dataset
Chandigarh Indian city planned by Le Corbusier to evenly
distribute residential units and facilities
• Area: 474.34 km2
• POIs: 825 activities, 32848 residential. Observation:
potentially missing activity POIs
19
94. Considered Dataset
Dresden German city reconstructed after World War II, with a
posterior growth around the city’s satellites
• Area: 605.40 km2
• POIs: 21654 activities, 96645 residential
20
95. Considered Dataset
Dresden German city reconstructed after World War II, with a
posterior growth around the city’s satellites
• Area: 605.40 km2
• POIs: 21654 activities, 96645 residential
20
97. Considered Dataset
Grenoble French city set up in a valley of the Alps with several
residential towns scattered in the surrounding mountain ranges
• Area: 376.47 km2
• POIs: 11260 activities, 84443 residential
21
98. Considered Dataset
Grenoble French city set up in a valley of the Alps with several
residential towns scattered in the surrounding mountain ranges
• Area: 376.47 km2
• POIs: 11260 activities, 84443 residential
21
100. Considered Dataset
Raleigh North-American city that encountered a massive
sub-urban growth from the 60s that connected it to the
neighbouring cities of Durham and Chapel Hill
• Area: 1575.28 km2. Observation: huge surface.
• POIs: 6718 activities, 126871 residential
22
101. Considered Dataset
Raleigh North-American city that encountered a massive
sub-urban growth from the 60s that connected it to the
neighbouring cities of Durham and Chapel Hill
• Area: 1575.28 km2. Observation: huge surface.
• POIs: 6718 activities, 126871 residential
22
103. Results
City Density ρ Shannon’s H Moran’s I Dissimilarity D
act res act res act res
´Avila 7.11 7.98 0.8654 0.7726 0.0563 0.0211 0.7302
Chandigarh 1.74 69.25 0.7964 0.8293 0.0249 0.0597 0.9620
Dresden 35.77 159.64 0.9409 0.9696 0.0649 0.0399 0.6375
Grenoble 29.91 224.30 0.8697 0.9354 0.0547 0.0862 0.7725
Raleigh 4.26 80.54 0.9389 0.9522 0.0147 0.1053 0.9520
23
104. Results
City Density ρ Shannon’s H Moran’s I Dissimilarity D
act res act res act res
´Avila 7.11 7.98 0.8654 0.7726 0.0563 0.0211 0.7302
Chandigarh 1.74 69.25 0.7964 0.8293 0.0249 0.0597 0.9620
Dresden 35.77 159.64 0.9409 0.9696 0.0649 0.0399 0.6375
Grenoble 29.91 224.30 0.8697 0.9354 0.0547 0.0862 0.7725
Raleigh 4.26 80.54 0.9389 0.9522 0.0147 0.1053 0.9520
• Densities vary greatly: missing data directly influences ρ
23
105. Results
City Density ρ Shannon’s H Moran’s I Dissimilarity D
act res act res act res
´Avila 7.11 7.98 0.8654 0.7726 0.0563 0.0211 0.7302
Chandigarh 1.74 69.25 0.7964 0.8293 0.0249 0.0597 0.9620
Dresden 35.77 159.64 0.9409 0.9696 0.0649 0.0399 0.6375
Grenoble 29.91 224.30 0.8697 0.9354 0.0547 0.0862 0.7725
Raleigh 4.26 80.54 0.9389 0.9522 0.0147 0.1053 0.9520
• Densities vary greatly: missing data directly influences ρ
• ´Avila, Chandigarh have more even distribution of POIs (lower
H s).
23
106. Results
City Density ρ Shannon’s H Moran’s I Dissimilarity D
act res act res act res
´Avila 7.11 7.98 0.8654 0.7726 0.0563 0.0211 0.7302
Chandigarh 1.74 69.25 0.7964 0.8293 0.0249 0.0597 0.9620
Dresden 35.77 159.64 0.9409 0.9696 0.0649 0.0399 0.6375
Grenoble 29.91 224.30 0.8697 0.9354 0.0547 0.0862 0.7725
Raleigh 4.26 80.54 0.9389 0.9522 0.0147 0.1053 0.9520
• Densities vary greatly: missing data directly influences ρ
• ´Avila, Chandigarh have more even distribution of POIs (lower
H s). Grenoble has a more even distribution of act than res.
23
107. Results
City Density ρ Shannon’s H Moran’s I Dissimilarity D
act res act res act res
´Avila 7.11 7.98 0.8654 0.7726 0.0563 0.0211 0.7302
Chandigarh 1.74 69.25 0.7964 0.8293 0.0249 0.0597 0.9620
Dresden 35.77 159.64 0.9409 0.9696 0.0649 0.0399 0.6375
Grenoble 29.91 224.30 0.8697 0.9354 0.0547 0.0862 0.7725
Raleigh 4.26 80.54 0.9389 0.9522 0.0147 0.1053 0.9520
• Densities vary greatly: missing data directly influences ρ
• ´Avila, Chandigarh have more even distribution of POIs (lower
H s). Grenoble has a more even distribution of act than res.
• No strong clustering of POIs (spatial autocorrelation): values
of I are close to 0 indicating random scattering
23
108. Results
City Density ρ Shannon’s H Moran’s I Dissimilarity D
act res act res act res
´Avila 7.11 7.98 0.8654 0.7726 0.0563 0.0211 0.7302
Chandigarh 1.74 69.25 0.7964 0.8293 0.0249 0.0597 0.9620
Dresden 35.77 159.64 0.9409 0.9696 0.0649 0.0399 0.6375
Grenoble 29.91 224.30 0.8697 0.9354 0.0547 0.0862 0.7725
Raleigh 4.26 80.54 0.9389 0.9522 0.0147 0.1053 0.9520
• Densities vary greatly: missing data directly influences ρ
• ´Avila, Chandigarh have more even distribution of POIs (lower
H s). Grenoble has a more even distribution of act than res.
• No strong clustering of POIs (spatial autocorrelation): values
of I are close to 0 indicating random scattering
• Higher land use mix in European cities (lower Ds). However
missing data for Chandigarh might be very determinant to D
23
112. Conclusions
Urban Sprawl is a very complex phenomena...
• there is a need to bring its research to a commonspace
24
113. Conclusions
Urban Sprawl is a very complex phenomena...
• there is a need to bring its research to a commonspace
• its multidimensionality must be assessed with multiple
indicators
24
114. Conclusions
Urban Sprawl is a very complex phenomena...
• there is a need to bring its research to a commonspace
• its multidimensionality must be assessed with multiple
indicators
• ranking cities is very subjective, clustering them might be
more appropriate
24
115. Conclusions
Urban Sprawl is a very complex phenomena...
• there is a need to bring its research to a commonspace
• its multidimensionality must be assessed with multiple
indicators
• ranking cities is very subjective, clustering them might be
more appropriate
When working with crowd-sourced data...
24
116. Conclusions
Urban Sprawl is a very complex phenomena...
• there is a need to bring its research to a commonspace
• its multidimensionality must be assessed with multiple
indicators
• ranking cities is very subjective, clustering them might be
more appropriate
When working with crowd-sourced data...
• The definition of the study region is crucial to the measures’
values (actually not only with crowd-sourced data)
24
117. Conclusions
Urban Sprawl is a very complex phenomena...
• there is a need to bring its research to a commonspace
• its multidimensionality must be assessed with multiple
indicators
• ranking cities is very subjective, clustering them might be
more appropriate
When working with crowd-sourced data...
• The definition of the study region is crucial to the measures’
values (actually not only with crowd-sourced data)
• Missing data is very determinant: OSM is very complete in
Europe but not reliable in other regions
24
120. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation.
25
121. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
25
122. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
Improve the reliability of the results:
25
123. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
Improve the reliability of the results:
• Evaluate OSM POIs coverage metrics
25
124. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
Improve the reliability of the results:
• Evaluate OSM POIs coverage metrics
• Automatic detection of an appropriate grid size for Ω
25
125. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
Improve the reliability of the results:
• Evaluate OSM POIs coverage metrics
• Automatic detection of an appropriate grid size for Ω
• Determination of the appropriate bandwidth of the KDE
25
126. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
Improve the reliability of the results:
• Evaluate OSM POIs coverage metrics
• Automatic detection of an appropriate grid size for Ω
• Determination of the appropriate bandwidth of the KDE
Assess urban sprawl globally:
25
127. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
Improve the reliability of the results:
• Evaluate OSM POIs coverage metrics
• Automatic detection of an appropriate grid size for Ω
• Determination of the appropriate bandwidth of the KDE
Assess urban sprawl globally:
• Compute indicators for a global city dataset and cluster them
25
128. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
Improve the reliability of the results:
• Evaluate OSM POIs coverage metrics
• Automatic detection of an appropriate grid size for Ω
• Determination of the appropriate bandwidth of the KDE
Assess urban sprawl globally:
• Compute indicators for a global city dataset and cluster them
• Correlate the indicators it to other environmental and
socio-economic metrics.
25
129. Future Work
Implement the computation of more indicators:
• Work on accessibility measures, considering public
transportation. Easy within the framework!
Improve the reliability of the results:
• Evaluate OSM POIs coverage metrics
• Automatic detection of an appropriate grid size for Ω
• Determination of the appropriate bandwidth of the KDE
Assess urban sprawl globally:
• Compute indicators for a global city dataset and cluster them
• Correlate the indicators it to other environmental and
socio-economic metrics. Potentially from crowd-sourcing sites:
Numbeo, Transitland, Livehoods 25