Presentation on defining economic boundaries of cites made at the The Potential of Satellite Imagery for Studying Human Activity - Workshop on 15 June, Liverpool, United Kingdom. Presentation by Presentation by Ana Moreno Monroy, Regional and Rural Policy, OECD. More information: http://www.oecd.org/cfe/regional-policy/

1. DEFINING THE ECONOMIC
BOUNDARIES OF CITIES
Geographic Data Science Lab, University of Liverpool,
June 15th , 2018
Ana I. Moreno-Monroy
Paolo Veneri

2. • A functional urban area (FUA): place where one person is
likely to meet any other person within the same day
• Boundaries based on administrative borders of single
municipalities do not accurately represent FUAs
• EC-OECD FUAs: Define commuting zone around urban centres
(UCs) based on commuting intensity
• Problem: Find a way to define commuting zones in places
where there is no commuting data
• Approach: make the most of the FUAs we know to learn about
the ones we do not know
Problem and approach

3. EC-OECD FUAS

4. EC-OECD Functional Urban Areas (FUAs)
Main characteristics
• Based on people and their daily behaviour instead of
administrative or purely morphological approaches
• (Partially) overcome administrative boundaries
• Facilitate international comparisons
• Support the design of urban policy and governance solutions
Method developed in cooperation with the EU in 2012
• Identified 1,197 FUAs in 33 OECD countries + Colombia
• Replicable methodology for non-OECD countries

5. STEP 3:
Identifying
the
commuting
zone
RESULTS
Two or more urban cores will belong to the same functional urban area
If more than 15% of the population
of one urban core commutes to
work to another urban core
If more than 15% of its population
commutes to work in the urban
core area
Individual municipalities will be part of the “working catchment area”
Monocentric
functional urban
areas (with only one
urban core)
Polycentric
functional urban
areas (with more
than one urban core)
STEP 1:
Identification of
urban cores
STEP 2:
Connecting non-
contiguous cores
belonging to the
same functional
area
1. Apply a threshold to identify densely populated grid cells
2. Identify contiguous high-density urban clusters
3. Identify core municipalities
If at least 50% of the population of the municipality lives within urban clusters
> 50,000 inhabitants
> 1,500 inhabitants per km2
Defining Functional Urban Areas

6. • 83 FUAs identified
• Total population in
2011 ranges from
85,000 to 11.7 million
(Paris)
• 65% of French
population live in
FUAs (Paris represents
19%)
Example: French FUAs

7. MODELLING FUAS
OUTSIDE OECD COUNTRIES

8. We use the Global Human Settlements Population Layer (GHSL)
and Population Model (SMOD) produced by JRC, containing
population by 1km2 cells to characterize points inside FUAs
Departing point: What is inside FUAs?
FUA Boundaries + SMOD, Adelaide (AU)
Medium density area, Vienna (AT)
Open!

9. 1) All cells
2) Cells in Medium
density clusters
Should we consider all cell types?
Distance to highest density cell versus population, Sydney

10. Is it OK to exclude rural cells?
02468
AT AU BE CA CH CL CO CZ DE DK EE EL ES FI FR HU IE IS IT JP KO ME NL NO PL PT SE SI SK UK US
Mean Median
Absolute differences in population in convex hull
vs. actual FUA population, (%) by country
Ljubljana, actual versus convex
hull border based on medium
density cells

11. Method steps: Estimation
1. Subset grid-cells with population
>300 inhabitants in each country
(medium density cells)
2. Identify cells falling within FUA
borders (dummy_FUA=1 (black), 0
(blue) otherwise)
3. Calculate the distance of each cell
to all urban centres, and assign to
closest4. Pool data for all countries (~ 0.5
million obs.) to estimate a logistic
regression of dummy_FUA on distance
+ size of the urban centre + size of cell +
country controls :
𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷_𝐹𝐹𝐹𝐹𝐹𝐹𝑗𝑗𝑗𝑗 𝑗𝑗 = 𝛼𝛼 + 𝜏𝜏𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖 +
𝛽𝛽1 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑖𝑖𝑖𝑖 + 𝛽𝛽2 𝑝𝑝𝑝𝑝𝑝𝑝𝑗𝑗𝑗𝑗 + 𝒙𝒙𝑪𝑪
′
𝜹𝜹 + 𝜖𝜖𝑖𝑖𝑖𝑖𝑖𝑖

12. We use the costDistance function of the R gdistance package (van
Etten 2017) to obtain travel times between the centroid of each
urban cluster and each medium density cell within country borders
using:
1) Global travel impedance grid (https://map.ox.ac.uk/):
– Represents time associated with moving through grid cells, quantified as a
movement speed within a “friction” grid (30 arcsec resolution). Unit of
measurement in grid is minutes required to travel one kilometre
– Information on roads (fastest type in grid takes precedence over others,
with speeds given by OSM tables), railroads, water bodies and movement
over land is used to characterize each grid cell
2) GAUL country boundaries (FAO)
Estimating travel times
Open!
Open!

13. Model validation and performance diagnosis
Model specifications:
- Including polynomial
terms on urban cluster
size, distance and country
controls
- Using different proxies for
urban cluster size (area,
population and night-time
lights)
- Cross validation using: 33 different samples, each leaving one
country out at the time and using it as test sample, and a 10%
random sample of urban clusters
- Diagnosis and model selection based on post-estimation tests using
R caret package (BIC, accuracy, … etc.)

14. IMPLEMENTATION

15. 1. Calculate distance from all
medium density cells to all
urban clusters within country
borders
2. Assign medium density cells to
most proximate urban cluster
2. Assign =1 to medium density cell
if predicted probability < 0.75
(based on estimated coefficients)
3. Draw FUA border based on cells
= 1
4. Merge borders if an estimated
FUA border crosses an urban
cluster (polycentric FUAs)
Implementation steps
Medellín (CO), actual vs estimated FUA
borders

16. Performance: Colombia
Accuracy of model using all countries except Colombia as training
set and Colombia as test set: ~80% (average all countries: ~ 75%)
Actual vs estimated FUA population, selected cities, Colombia
0 1,000,000 2,000,000 3,000,000 4,000,000 5,000,000 6,000,000 7,000,000 8,000,000 9,000,000 10,000,000
Bogotá
Medellín
Cali
Barranquilla
Bucaramanga
Pereira
Ibagué
Montería
Villavicencio
Pasto
Buenaventura
Neiva
Palmira
Tuluá
Barrancabermeja
Zipaquirá
Cartago
Aguachica
Actual population in FUA Estimated population in FUA

17. Estimated borders: Bogotá
Objective is not to reproduce actual border lines but to
approximate population in the commuting zone of urban clusters
Actual vs estimated
FUA border, Bogotá,
Colombia
Estimated MD cells = 1

18. Estimated borders: Morocco
Number of FUAs: 61
Total pop. in FUAs:
22’010.021
Out of which
4’678.803 in
commuting zones
% Population in
FUAs: 63%
Benchmark: Urban
population in
Morocco (2015):
20’439.199
Estimated population in FUAs, selected cities, Morocco
0 500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 3,500,000 4,000,000 4,500,000
Midelt
Ain Taoujdat
Azrou
Sidi Yahya El Rharb
Tiznit
Ben Guerir
Tetouan
Errachidia
Guercif
El-Kelaa-des-Sraghna
Ouazzane
Youssoufia
Al-Hoceima
Souk Sebt des Oulad Nemaa
Oulad Teima
al-Faqihi Bini Salah
Berrechid
Dar Sidi Slimane
Settat
Ksar El Kebir
Khouribga
Banii Mallal
Safi
Tetouan
Oujda
Kenitra
Meknes
Tanger
Agadir
Fes
Marrakech
Rabat
Dar-el-Beida (Casablanca)
Population in core Population in commuting zone

19. Next steps
- Implement method outside OECD  evaluate borders
- Define additional rules in case estimated FUAs become
unrealistically large (e.g. Guangzhou China)
- Use estimated FUA borders to understand suburbanization
patterns
- Use estimated FUA borders in studies on global cities

20. Thank you!
ana.morenomonroy@oecd.org
paolo.veneri@oecd.org