Mapping Poverty for the Millennium Development Goals and related initiatives

1
Mapping Poverty for the Millennium Development
Goals (and related initiatives)

• Deborah Balk
• Columbia University
• 28 April 2003

2
What have we learned in 10 years?

• Basic Question Where do people live?
• Ten years ago, the answer was nationally
  delineated
• Now it is delineated by over 250,000 smaller
  areas
• Characteristics of the population
• Urban vs. rural
• Indicators of population well-being
• Mortality
• Underweight
• Poverty
• How much more will we know, in a regional or
  globally consistent fashion, in another 10 years?

3
Gridded Population of the World

• First developed in c. 1996, revised in 2000, and
  currently under revision
• Substantial improvements in inputs each revision
• Additional years of population estimation
• Best-available collection of spatially
  disaggregated population estimates
• Distributes administrative population across a
  raster surface (or grid)
• 2.5 arc-minutes (4.5 km) resolution
• Allows aggregation of population by any other
  spatial phenomenon (land cover, ecosystems, etc.)

Columbia Universityin the City of New York
4
GPW Version 1 19,000 units globally 2,200 in
Africa
Population Density
Persons per sq. km
0
0.1 - 5
5.1 - 25
25.1 - 250
250.1 - 500
500.1 - 43,086.8
5
GPW Version 2 127,000 units globally 6,000 units
in Africa
Population Density
Persons per sq. km
0
0.1 - 5
5.1 - 25
25.1 - 250
250.1 - 500
500.1 - 43,086.8
6
GPW Version 3 250,000 units globally 102,000
units in Africa (81,000 in South Africa)
Population Density
Persons per sq. km
0
0.1 - 5
5.1 - 25
25.1 - 250
250.1 - 500
500.1 - 87,157.6
7
Use in health applications in Africa

• Estimates of malaria mortality and morbidity
  (Snow et al.)
• Estimate 1 million deaths in 1995 estimated due
  to malaria and 200 million clinical events
• Combined GPW with endemicity (and other) data
• Project human and tsetse fly population
  distribution to 2040 (Reid et al.)
• Human population growth causes loss of fly
  habitat
• Pop growth affects subspecies differently
• Model human-fly interactions based on
  species-specific behavior
• Estimate that by 2040 the fly population will
  decline throughout Africa but an area as large as
  Europe will remain infested
• Combine GPW with fly pop data
• Analysis of malaria and economic growth (Gallup
  and Sachs)
• Used GPW to estimate population density within
  100 km a coastline--to proxy for access to
  transportation
• Despite the strong correlation between poverty
  and malaria, and the strong impacts of malaria on
  the economy, the causal mechanism are unclear

8
Who lives where urban and rural areas

• Combine
• GPW administrative units
• Urban areas from
• Lights at night satellite data (DMSP derived
  time-stable lights)
• Digital Chart of the World, pilotage charts for
  parts of Africa
• Population estimates for places (points) from
  census sources
• NB UN Demographic Yearbook collects these only
  to places of 100K or more
• Results in resolution about 4x higher than GPW
• 30 arc-seconds
• Three data products
• Points
• Urban-rural mask
• Population Surface (grid)

9
Rationale

• GPW is a good start, but there remains an unmet
  need for better data indicating the spatial
  distribution of human population
• Especially indicating urban areas or localized
  settlements
• Population and extent thereof
• Prior data sets do not indicate urban extent to
  this accuracy
• Approach does not need to adopt a what is urban
  definition, although users could ultimately apply
  one
• For use in analysis of
• economic development
• demographic change (including urbanization)
• hazards
• global change
• agriculture

10
Basic Steps

• Find places with their
• Population size
• Geographic co-ordinates
• Estimate
• Population to the target years (1990, 1995 and
  2000)
• Urban boundaries
• Aggregate
• Population data to urban agglomerates
• Derive population grid reallocated to urban areas
• Validate with administrative pop data

11
Input sources

• Population data
• National statistical offices for censuses, in
  most cases
• Other authoritative sources
• Coordinate data
• NIMA, in most cases
• National survey agencies or geospatial units of
  stats offices
• Other atlases (usually for validation, only)
• Extent data
• NOAAs Night-time lights (OLS) satellite data, in
  most cases
• DCW, TPC or other polygons
• Administrative boundary data
• National statistical or mapping agencies or
  regional partners
• E.g, CIAT

12
Points
Step 1 Estimate pop of places for target years
13
Night-time lights
Step 2 Derive polygons of urban extent (no
threshold applied)
14
Create circles where lights are absent

• The night-time lights disfavor economically
  poorer regions of the world
• Thus, reliance on the lights only in Africa, for
  example, will yield far fewer urban extents than
  is necessary
• We, therefore, use a regression analysis to
  predict the geographic size of an area based on
  the population size of that area
• Continental-specific regressions
• Pulls data-poor and data-rich countries alike
• Create a circle for the urban area
• Predict the circles radius

15
The results Population Density, 1995

• Compare GPW with the urban rural output for a
  part of W. Africa

GPW
Urban-Rural
16
Urban Rural Points
Populated Places
3,830 places (36 1 million )
38 - 1,000,000
1,000,001 - 7,645,936
17
Lights from DMSP (this is what youd get if you
used only the night time lights)
Lights with population
18
Urban Areas (includes areas for which we
predicted areas)
Population Density
Persons per sq. km
0 - 1,000
1,001 - 10,000
19
(No Transcript)
20
(No Transcript)
21
Urban Rural Surface
Population Density
Persons per sq. km
0
0.1 - 5
5.1 - 25
25.1 - 250
250.1 - 500
500.1 - 87,157.6
22
Improvement from GPW

• Starting resolution is pretty good
• Level 2, although that varies
• But the number of polygons increase considerably
• The lower the starting resolution the greater the
  room for improvement
• Up to 5x (in the number of polygons) for Cameroon

23
Quantifying the improvement

• Thinner curve represents GPW thicker curve
  represents U-R
• 40 percent of the population lives on only 6
  percent of the land area rather than occupying
  closer to 15 percent of the land area
• The inset histogram shows a multimodal
  distribution of population density, with roughly
  a third of Ecuadors population living at
  densities below 100 persons per square kilometer,
  another third living between 100 to 1000 persons
  per square kilometer, and the remaining third
  above 1000 persons per square kilometer. Ecuador
  has a relatively small proportion of the
  population living at densities considered highly
  urban by international standards (i.e., 10,000
  persons per square kilometers).

Lorenz curves with the distributions of
cumulative population and land areas, in Ecuador
24
Are the places in the correct location? Use roads
overlay with urban areas

• Greyish blobs are urban areas from our database
• Irregular/rougher shapes are derived from
  night-time lights
• Circles are estimated from a regression
• Green lines are roads
• Use VMAP to overlay the roads
• Good match for places found
• Unknown for places not found

25
Estimates of coastal populations
GPW shown with coastal buffers of 50 and 100 km
26
Improved estimates of coastal population
Urban-rural shown with coastal buffers of 50 and
100 km
27
Estimates of coastal populations

• At 50 km there are some substantial differences
  in the population estimates
• This will increase with coarser admin data
• Urban-rural does much better here (regardless of
  the difference)

28
Does estimating the coastal population matter?

• At risk of certain hazards
• Better access to ports and markets
• Coastal dwellers tend to be less poor
• Coastal areas grow differently (face more
  constraints)
• Are urban dwellers disproportionately coastal
  dwellers?
• Well be able to answer that soon.

29
Mapping Development Indicators

• First two MDG indicators selected
• Infant mortality
• Underweight (Hunger)
• Integrate data from surveys
• Demographic Health Surveys (DHS)
• UNICEFs Multiple Indicator Cluster Surveys
  (MICS)
• African Nutrition Database Initiative (ANDI)
• UNDP Human Development Reports
• Construct geographies to match the surveys
• Although these surveys have been around for a
  while, use in a GIS is (more or less) new
• Between the several survey types coverage is
  becoming increasing complete, spatially and
  temporally

30
(No Transcript)
31
(No Transcript)
32
What do these all have in common?

• Underlying spatial data infrastructure that makes
  the transformation of census or survey data
  possible
• Significant Institutional collaborations (among
  others!)
• GPW
• UC Santa Barbara/NCGIA
• UN Environmental Program
• World Resources Institute IFPRI
• US Census Bureau
• Urban Rural
• International Food Policy Research Institute
  (IFPRI)
• Center for International Agriculture of the
  Tropics (CIAT)
• WCMC Millennium Ecosystem Assessment (MA)
• Poverty Maps
• World Bank
• World Resources Institute
• UNEP-GRID
• CG centers (IFPRI, CIAT, ILRI)
• Macro International

33
Whats next

• Poverty maps
• Health infrastructure maps
• These will allow for analysis of mortality or
  malnutrition outcomes with health service
  provision, access to services, and underlying
  poverty

34
Poverty maps

• Supplement the small-area estimation approach
  based on income which require extensive income
  and expenditure survey data
• Construct asset indicators
• First at the same geographic-level as the IMR and
  Hunger maps shown
• Compare these with income indicators
• In data rich and data poor countries
• Determine replicability with a variety of
  assets since surveys vary in measurements of
  assets
• Determine whether scale of output is useful as it
  will be coarser than the small-area estimate
  approach

35
Recent Poverty Mapping Workshop

• Agreement that a confluence of approaches will
  help us to better
• scrutinize poverty profiles (geographic,
  occupational, demographic) to identify
  characteristics of the poor that are common
  across countries
• evaluation of definitions of poverty transient
  vs. chronic, absolute versus relative
• Fitness-of-use guidelines
• Determine which approaches work where and which
  are appropriate for what types of analysis and
  policy inferences

36
In less than ten years

• We went from a map of global population
  distribution at the national level (with about
  250 units) to one with 1000 times as many
• Many institutional partners and data sharing
  model
• Many uses in health and environmental studies and
  policy
• Thus, with the same model and commitment, poverty
  mappingalthough more complexwill look much
  different 10 years from now.
• For more info, contact dbalk_at_ciesin.columbia.edu