17_construction of gridded population and poverty data sets from different data sources _alex de...

7/27/2019 17_construction of Gridded Population and Poverty Data Sets From Different Data Sources _Alex de Sherbinin

1/34

Alex de Sherbinin, Deputy ManagerNASA Socioeconomic Data and Applications Center

Center for International Earth Science Information NetworkThe Earth Institute, Columbia University

Palisades, New York, USA

Acknowledgements:This presentation borrows heavily frommaterial prepared by Deborah Balk, formerly of CIESIN and currentlyat Baruch College, and Gregory Yetman of CIESIN.


2/34

2

Short history of gridding population data

Why grid? Gridded Population of the World (GPW) Methodology

Global Rural Urban Mapping Project (GRUMP)

US Census Grids

Poverty Mapping

Gridded Infant Mortality Rate

Gridded Child Malnutrition


3/34

More attention to global scope

More attention to comparability

More attention to problem-orientedscience

More attention to spatial frameworks


4/34

US Census Bureaus Global Population Database (early 1990s)

Africa Population Grid (UNEP/GRID, 1991)

GPW v1 and Global Demography Project (NCGIA & CIESIN, 1994) 1 degree global grid (Environment Canada, 1995)

Europe (RIVM, 1995)

Africa update and Asia (NCGIA, UNEP/GRID & WRI, 1996)

Latin America (CIAT)

HYDE (RIVM/ Klein Goldewijk 1997, 2001 and 2006)

LandScan (ORNL, 1999 and onwards)

GPW v2 (CIESIN et al., 2000) GRUMP alpha (CIESIN et al., 2004)

GPW v3 (CIESIN et al., 2005)

In the past decade there have been far more efforts than can be

listed here


5/34


6/34

http://sedac.ciesin.columbia.edu/gpw/


7/34

!"

Find tabular information withattributes E.g., Population counts

Match to geographicboundaries Administrative

Urban footprints

Estimate Population to the target years

(1990, 1995 and 2000)

Transform to grids

Statistics South Africa

Descriptive - South Africa by Province and Municipality

Table 1

Province (PR_SA)District municipality (DC_PR_SA)Municipality (MN_PR_SA)

Main place (MP_SA)Sub-place (SP_SA)

Geography by Gender

for Person weighted

Male Female Total3 Northern Cape 401094 421636 822729

6 DC6: NAMAKWA District Municipality 53424 54687 108110

301 NC061: Richtersveld 5170 4961 10130

30101 Alexander Bay 723 729 1452

30101001 Alexander Bay Navel Base 30 12 42

30101000 Alexander Bay SP 738 675 1413

NB: Spatially matched population census (and survey) data

generally has several data providers!


8/34

#$

Population and boundary data must match Best available & matchable data are used

Matching the inputs to one another is not as easyas it might seem Boundaries change often and come in different scales

Population data may not match boundaries We may have population values for different years atdifferent levels (e.g., district-level one year, state-levelanother)


9/34

"

Clean boundaries

E.g., remove slivers

Make them consistent across borders and coasts

Use international standardthe DCWwith exceptions

Europemost spatially data supplied by one agency (SABE)

and all international boundaries are internally consistent For GPW v4 we plan to use ISciences Global Coastline v.1

Coastlines matched to DCW, except where much higherquality data are supplied

E.g., Indonesia

Data table needs to include the same variables, withthe same variable names, formats, etc.


10/34

%&'(

Places highlighted in yellow are new municipiosNeed to find where they came from & their pop size

Use on-line atlases or newer maps, when availableAdd new pop to unit of origin or allocate old population to new unit proportionally.


11/34

%'

Annual rate of change calculated:

Population estimates adjusted to target years:

Px= P2ert

Definitions

r - Annual rate of growth

P1..2 - Census estimate

t - number of years between

census enumerations

Px - Year of Estimate

Pun - UN EstimatePadj - Adjusted estimate

t

P

P

er

1

2

log


12/34

%')*

Definitionsa - Adjustment factor

Px - Year of Estimate (90 or 95)

Pun - UN Estimate

Padj - Adjusted estimate

Adjustment factor for matching national estimates to UN

estimates calculated:a = (Pun- Px) / Pun

Adjustment factor applied at the national level :

Padj= Px* a

Differences ranged from 20% under (Somalia, 1995) to

25% over (Jordon, 1990)


13/34

'

Proportional allocation used to spread the populationover grid cells

Virtually all data work completed on vector data

Gridding is the last step

National grids created, global grids assembled byadding national grids together

Country grids are created with collars so that they startand end on even degrees; therefore the assembly of

the grids without interpolation is possible

Replacement of country-specific grids feasible


14/34

'

Land Area: 458.4 square km


15/34

Area 2.6 kmPop = 628.5 persons

per sq. km * 2.6 =1,634.1persons

Area 16.1 km

Pop = 628.5 personsper sq. km * 16.1 =10,118 persons

Area 0.05 kmPop = 628.5 personsper sq. km * 0.05 =31.4 persons


16/34

Population 2000

PersonsHigh: 10,123.5

Low: 7


17/34

!"# $%"# $%"# $%"#

& ' ''

( ' ' )'

#+,$"$-./

)*


18/34

01)01*

http://sedac.ciesin.columbia.edu/gpw/


19/34

01)01*

Objective: To delineate urban and rural extents and

populations

Collaboration between CIESIN, IFPRI, World Bank, & CIAT Builds on GPW infrastructure

Adds urban areas from Nighttime lights satellite data

Three databases: Settlement Points (>70,000 w/ pop of 1k+)

Urban Extents (>23,500 w/pop of 5k+)

Pop Grid at 1 km resolution


20/34

&01)2*

Stand alone model GRUMPe written in C

Combines the following pieces of information: Population and boundaries of each urban area based on NTLs

Boundaries sometimes based on buffered points where no NTL signature

Population and boundaries of each admin area

Size of the intersect areas where urban and admin areas overlap UN national estimates for percentage of population in urban and

rural areas

20


21/34

&01)3*

The algorithm reallocates the total pop in each admin

unit into rural and urban areas based on UN estimates,with six contraints:

1. Total admin pop remains constant

2. Urban pop density in any admin unit must be > rural density

3. Rural pop density cannot be lower than national mininimumrural pop density threshold for country/region

4. Rural pop density cannot be higher than the national maximumrural pop density threshold for country/region

5. Urban pop density cannot be lower than national minimumurban pop density threshold for country/region

6. Urban pop density cannot be higher than the nationalmaximum urban pop density threshold for country/region

21


22/34

&01)4*

The algorithm is trivial where only one urban area iscontained within an admin unit

It is more complex when:

there are multiple urban areas overlapping an admin unit

Urban areas overlap more than one admin area

Large urban areas contain more than one admin area

These are common situations and require successiveiterations to meet all constraints

22


23/34

(

23


24/34

1

Close upof Brazilusing the100K

person cutoff

Note thevariety ofshape

Much morethan pointsconvey


25/34

1"

http://sedac.ciesin.columbia.edu/usgrid/


26/34

1"

Uses proportional allocation algorithm

Higher resolution: Resolution is 1km (30 arc-sec) for the country as a whole

Metropolitan areas are available at 250m (7.5 arc-sec)

More census variables:

Individual data: age distribution, race, ethnicity, income,poverty, educational level, and immigrant status

Household data: household size, one-person households,female-headed households with children under 18, and

linguistically isolated households

Housing unit data: occupied housing units without a vehicle,and year of construction

26


27/34

1""

27


28/34

http://sedac.ciesin.columbia.edu/povmap/


29/34

Infant mortalityrates (IMRs):

Serve as a usefulproxy for overallpoverty levelsbecause they arehighly correlated

with metrics suchas income,education levels,and health statusof the population

This metric isparticularly goodfor distinguishingpoverty levels atthe lower end ofthe income ladder


30/34

$0

Sources Demographic and Health Surveys (39 countries)

Multiple Indicator Cluster Surveys (5 countries)

National Human Development Reports (14 countries) National Statistical Offices (18 countries)

6,494 spatial units in global data base Brazil and Mexico 5,372 units

74 other countries with subnational data 22 units per country on average

115 countries national level data only (UNICEF) 36 countries no data

Calibration Subnational IMR values adjusted to be consistent with national UNICEF

2000 IMR values

Gridding using proportional allocation algorithm We also converted rates to counts

For each subnational unit, estimates of live births, infant deaths calculatedbased on griddedpopulation, nationalfertility data, and subnationalIMR.


31/34

31


32/34

Use anthropometric data found in household surveys DHS and MICS data were aggregated to the spatial units at which the

surveys report, based on raw data where it was available, and publishedreports otherwise.

These spatial units are typically equivalent to first level administrativeregions or aggregations thereof.

Geospatial boundary files that match those spatial units werelocated or created in order to match the reporting regions ofthe surveys as closely as possible.

In many cases, the survey reports contained maps detailing the survey

regions. Elsewhere, matches were purely name-based. Map percent of children underweight

Underweight defined as being two standard deviations or more belowthe mean weight for a given age, as compared to an international

reference population. 32


33/34

5!

Continued emphasis on higher resolution inputs

Effort to collect and grid more census variables Age and sex distribution Urban/rural distribution

Proposed output resolution: 1km grids

May create time series back to 1980

Looking for data sharing partners

33


34/34

#6'-7+180+16

'("

&9..

34

17_construction of gridded population and poverty data sets from different data sources _alex de...

Documents