Post on 04-Jan-2016
description
Global and continental population databases
“Supply side view”
• What has been done
• Related developments
• Possible next steps
Population data in raster format
• Gridding pop data is not a new idea
– Population map of West Africa (John Adams, LSE 1968)
– Statistical Offices (e.g., Japan, Sweden)
– Population Atlas of China
– ...
• Individual country or regional level
• Methods not well-documented
• Mostly not available in digital form
Continental / global data sets
• BUCEN’s CIR database
• Africa (UNEP/GRID, 1991)
• 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)
• Landscan (ORNL, 1999)
• GPW II (CIESIN, 2000)
Continental / global data sets
• Data collection focused
• Cartographic models - pycnophylactic interpolation,
dasymetric mapping
• “Smart interpolation”
– adjustment factors
based on auxiliary
GIS data layers
– accessibility based
weighting
Relationship between district-level mean accessibility and population density - India
1
10
100
1000
10000
100000
1 10 100 1000
Mean Accessibility
Po
pu
lati
on
Den
sity
Accessibility as a predictor of population density
di4
di3
di1
di2
town
currentnode i
node
transport network
V P f di k ik
k
( )
1
4
Access-based smart interpolation(population potential)
0
0.2
0.4
0.6
0.8
11 5 9 13 17 21 25 29 33
e d 2 22/
12d
1
d
we
ight
distance
Distance decay
Related developments - source data
• Initial data sets and applications have created large
demand for these types of data (gridded and small
area data)
• National statistical offices are adopting GIS for census
mapping; in developing countries supported by UNSD
and donors
• Availability of national and regional high resolution and
high quality databases; NSOs, CIESIN - China &
Mexico, ACASIAN, MEGRIN
Related developments - modeling
• Innovative modeling approaches
– Kernel estimation
– Fractal cities
– Behavioral models (settlers)
– NASA/USGS work on land cover change / urban growth
patterns
– ...
• New global data sets that can support population
modeling
– USGS elevation and land cover data
– NOAA “city lights”
– WCMC protected areas
– ...
Next steps
• Accuracy assessment of existing data sets
• User survey
– who benefits from these data?
– can we get better feedback from users?
– do current data sets address expressed needs?
– is it worth the cost?
Improve quality of source data
• Largest quality improvements will come from better
input data, not from modeling improvements
• Collection of pop figures and boundary data is a never-
ending task (e.g., 2000 round data available soon)
• Improve base pop estimates - extrapolation to
common base year, recent pop displacements
• For boundaries: focus on highest possible resolution or
on best possible positional accuracy?
• Identify new and improve existing auxiliary data sets
GPW II - Europe
Improve smart interpolation methods
• Calibration of parameters!
– currently determined ad hoc, but should be based on
observed patterns (both accessibility and other auxiliary
factors)
– adjustment factors should be determined statistically
– importance of factors unlikely to be constant across
countries
– accuracy assessment
based on districtlevel totals
based on statelevel totals
Estimated population densities
Improve smart interpolation methods
• Make more explicit use of city information
– location and size of many cities available
– urban extent approximated by “city lights” data
– may address urban / rural issue better than official
statistics
UNSD cities over 100,000 inhabitants
Resolve modeling issues
• Potential circularity
– e.g., for environmental applications, can’t use land cover
data to predict pop distribution, if users will then cross-
tabulate pop with land cover types
– but for “pop at risk” studies (e.g., health, disaster response)
we might want to use any available meaningful auxiliary
factors
– family of data sets?
Resolve modeling issues
• What is an appropriate output resolution?
– average GPW admin unit resolution is 33 km, average area is about
1070 sq. km
– pixel size is 2.5 min, or about 4.6 km at equator with an area of
about 21 sq. km
– so “modeling ratio” is about 50 output cells per admin unit
– but large variability across countries (resolution)
• Switzerland 3.7
• Luxembourg 4.7
• …
• Chad 302.8
• Saudi Arabia 374.2
• Same with population per unit (1.5 thousand to 3.4 million)
Resolve institutional issues
• Coordination between groups
– pool input data sources
– agree on coding schemes (FAO proposal)
– division of tasks
• Get endorsement from National Statistical Offices
and UN
• Determine distribution status of admin boundaries
• Funding plans
Expand scope of database
• Time series / projections or scenarios
• Rural / urban
• Demographic components (age-sex)
• Living standards
• High resolution databases for specific
regions/countries
• Work closer with application projects
45-49 50-54 55-59 60-64 65-69
70+ Tota l
P oland - U rban S ex R atios 1994
20-24 25-29 30-34 35-39 40-44
0 1-4 5-9 10-14 15-19
115110105959085
m ales per 100 females
m ore females m ore m ales
45-49 50-54 55-59 60-64 65-69
70+ Tota l
P oland - R ura l S ex R a tios 1994
20-24 25-29 30-34 35-39 40-44
0 1-4 5-9 10-14 15-19
115110105959085
m ales per 100 females
m ore females m ore m ales
Small area statistics from survey data(poverty indicators)
Poverty maps for Ecuador
Clarke and Rhind 1991
• Variety of databases with different levels of spatial
resolution
– made compatible with gridded data
– no more than a few years out of date
– time series of data for different resolutions
– ability to distribute freely for scientific purposes
GPW gridding
Administrativeunit
Admin unitdensity
(people / sq km)
Area ofoverlap(sq km)
PopEstimate
Santiago Rodriguez 64.2 5.3 340
Santiago 246.5 2.2 542
San Juan 75.9 12.8 972
Total for cell 91.3 20.3 1854
GPW gridding