Presentation by Ana Moreno Monroy, OECD at the OECD Workshop on Spatial Dimensions of Productivity, 28-29 March 2019, Bolzano.
More info: https://oe.cd/GFPBolzano2019
Ana Moreno Monroy - Global definition of cities and their areas of influence
1. A global definition of cities
and their areas of influence
Ana I . More no Monroy, OE CD/CFE /RDT
Workshop on Spat ial Dim e nsions of Produ ct iv it y
Bol zano, March 2 8 th , 2 0 1 9
2. Many cities do not match their
respective administrative boundaries
3. Note: Metro large regions performed above national levels
(1.13) and metros close to national levels (0.98)
Spatial productivity comparisons require
a consistent definition of cities and their
area of influence
0.84
0.86
0.88
0.9
0.92
0.94
0.96
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
GDPperworkerrelativetocountry(weightedaverage)
Non-Metro Close toMetro Non-Metro Close toSmall & Medium Non-Metro Remote
4. Functional Urban Areas (FUA) or urban agglomerations
demarcate the space encompassed by daily commuting
flows
Extending the definition of urban
agglomerations globally
Number OECD-EC FUAs Estimated FUAs
FUAs 1 191 ~10 000
Countries 34 178
Data Global grid + National census
(travel-to-work flows)
Global grids
Method Commuting intensity Probabilistic
6. 1. Identification of densely inhabited
and large places (cities or “urban
centres”)
2. Definition of commuting zone (area
of influence) linked by commuting
flows to cities
3. The sum of city area and
surrounding commuting zone area is
the Functional Urban Area
How to define OECD-EC FUAs
City
Commuting zone
City
FUA
7. • 83 FUAs
• FUA population
(2011) 85,000 -
11.7 million
• 65% people live
in FUAs (19% in
Greater Paris)
Example OECD-EC FUAs - France
Core
Commuting zone
8. Visit the updated web-site: https://measuringurban.oecd.org/#
The OECD Metropolitan eXplorer
11. Global FUA method in two steps
Country with commuting
flows data
Country without commuting
flows data
Step 2: CLASSIFY each pixel as
within or outside FUA based on
actual FUA borders (left) or
estimated probability (right)
City 1
City 2
Based on predicted probabilities
from logistic regression (on 500K
obs.) on travel time to closest city,
city size and country characteristics
City 1
City 2
Estimate travel time from each
pixel to every city in the country
using travel impedance grid.
Choose smallest
Step 1: ASSIGN each pixel with
300+ people to the closest city
12. Good/Excellent predictive
capacity and no notable
biases by country
characteristics
Robust to alternative proxies
and acceptable predictive
power (100 training and test
sets based on random
samples of ~1 400 cores)
Model performance and validation
Proxy for core size AUROC
Population 0.862924
Area 0.861503
Night-time lights 0.862459
AUROC : 0.5 = no distinction cells
inside/outside FUAs; 1= perfect distinction
13. Draw borders based on pixels
with predicted probability >
optimal threshold (~0.73)
Merge FUAs with touching
boundaries within 5km from
each other into polycentric
FUAs
Implementation and borders
Actual vs estimated FUA border, Bogotá,
Colombia
14. Administrative estimated FUAs and
external validity tests
Jaccard = 1 Two maps are the same. Threshold = 15% of
population in municipality falls in estimated FUA(s)
16. Global comparison results
0%
10%
20%
30%
40%
50%
60%
70%
Least Developed Less Developed,
excluding least
developed
More Developed
Commuting Pop. % in FUA
FUA % in Total Pop (millions of persons)
0%
10%
20%
30%
40%
50%
60%
70%
80%
Africa Asia Europe Latin
America &
Caribbean
North
America
Oceania
Commuting Pop. % in FUA
FUA % in Total Pop (millions of persons)
54% of the world’s population live in FUAs (3.6 billion). 12% of
them live in commuting zones
17. The ratio of people living
in commuting zones over
people living in urban
centres is highest for the
richest countries
Amongst large countries,
USA has the largest share
of population in
commuting zones (30%)
Suburbanisation is higher in high-
income countries