Deriving Sub-National Population Estimates for Sub-Saharan Africa

In previous releases of HarvestChoice 5-arc minute population (up to SChEF r12.03) we relied exclusively on population maps from CIESIN Global Rural-Urban Mapping Project, version 1 (GRUMPv1 at 30-arc second) combined to more recent national estimates from WDI. The spatial allocation of total, urban and rural populations in GRUMPv1 is based on censuses and national household surveys conducted on or before 2000. As HarvestChoice is expanding its use of sub-Saharan Africa national household surveys as a source of spatialized information, efforts are made to include more recent data. With changes in population growth rates and large migratory movements affecting the sub-continent in dramatic ways, researchers and policy-makers need the capacity to incorporate the latest available information and produce maps at much shorter intervals.

In HarvestChoice SChEF r12.04 (current release) both population and poverty estimates are anchored around the year 2005. There are more recent population censuses available (Botswana 2011, Chad 2009, Ghana 2010, Guinea 2009, etc.) and we are planning to start publishing time-series population estimates in the near future.

The major difference between r12.04 and earlier releases is in the use of survey micro-data. The steps taken and assumptions made to scale GRUMPv1 to 2005 are outlined here.

  1. Extract sub-national population data, separately for total, urban and rural populations, from 24 nationally representative household surveys and population censuses conducted in various years (see Table 1 below).
  2. Acquire or generate survey maps at level-1 (regions) and overlay these maps to HarvestChoice spatial grid.
  3. Transform GRUMPv1 30-arc second estimates to HarvestChoice 5-arc minute grid.
  4. Use a two-step allocation approach to distribute population across 5-arc minute gridcells:
    • In each country we derive regional contribution rates based on survey estimates (proportion of total population in each administrative unit)
    • In each administrative unit, we then derive gridcell contribution rates based on GRUMPv1 (proportion of regional population in each gridcell)
    • Convert gridcell contribution rates to population headcounts using WDI 2005 national population estimates.

Step 1: Derive Regional Population from Survey Micro-data

For most countries national surveys are only statistically representative at the regional level (level-1) with varying degrees of statistical robustness depending on sampling size and techniques. Table 1 below details the source of the country micro-data used since r12.03 and the number of representative administrative units.

Table 1: Countries with national household surveys and censuses used in gridded population calculations.

Step 2: Generate Survey Maps

Most surveys come with no sampling map or administrative boundaries, as such maps are generally inferred from GADM or GAUL. This is not so much an issue with level-1 boundaries as these tend to vary less over time compared to level-2 (districts) or beyond but nonetheless care is taken to account for all land area (as shown in Figure 1 below). Survey boundaries are rasterized to 5-arc minute resolution using a points-in-polygon method. In each country unmatched pixels are assigned to their nearest matched pixel.Two very small administrative units (Yaounde and Banjul) had to be transformed manually.

Step 3: Transform GRUMPv1 30-arc-second to HarvestChoice 5-arc-minute Grid

We use a resampled GRUMP population grid at 5-arc-minute.

Step 4: Distribute Population Estimates across HarvestChoice Spatial Grid

The general spatial allocation procedure is shown in Equation 1 below. This equation is applied separately to urban and rural populations.

       total population in country i (from 2005 WDI)
        population estimates in region j of country i (derived from household survey)
  population estimate in gridcell k of region j in country i (derived from GRUMPv1)

Note: in countries where no recent survey is available, we simply apply proportions from GRUMPv1. In cases GRUMPv1 reports no urban (or rural) gridcell in a given region, but the corresponding household survey reports differently, we allocate survey estimates evenly across that region. For all regions where CIESEN reports no population but the corresponding household survey reports differently, we favor the household survey and uniformly apply the regional rate across all gridcells in that region.

Next Steps

SChEF r12.05 will include updates for South Sudan (2009) and Rwanda (2005). HarvestChoice is also investigating the latest AfriPop maps to replace GRUMPv1. 


References

Center for International Earth Science Information Network (CIESIN), Columbia University; International Food Policy Research Institute (IPFRI), the World Bank; and Centro Internacional de Agricultura Tropical (CIAT), 2004. Global Rural-Urban Mapping Project (GRUMP): Gridded Population of the World, version 3, with Urban Reallocation (GPW-UR). Palisades, NY: CIESIN, Columbia University. Available at http://sedac.ciesin.columbia.edu/gpw/

 

Citation

HarvestChoice, 2012. "Deriving Sub-National Population Estimates for Sub-Saharan Africa." International Food Policy Research Institute, Washington, DC., and University of Minnesota, St. Paul, MN. Available online at http://harvestchoice.org/node/1931.

Jan 17, 2012