Fig 1.
Zones used in analysis of coastal proximity of PICT populations.
Zones of 1km, 5km, and 10km (black lines) overlaid onto local census enumeration areas in south-west Santo Island, Vanuatu. Enumeration areas that crossed boundaries were assigned a single zone based on the centroid of the enumeration area.
Fig 2.
Differences in allocation of houses to 1 km zone using Landscan and GAUL boundaries illustrated using ArcGIS.
(A) LandScan zone boundaries (brown) and associated 1 km zone miss-allocated many houses (green points) around inlet bays. (B) Allocations using GAUL land boundaries correctly allocated houses to zones.
Table 1.
Summary of coastlines, populations and data sources for PICTs. Population estimates from https://sdd.spc.int/topic/population.
Coastline lengths from the World Vector Shoreline database at 1:250.000 scale and The World Factbook (https://www.cia.gov/library/publications/resources/the-world-factbook/index.html). Census unit abbreviations: EA = Enumeration Area and LLGA = Local Level Government Area. Analysis methods vary depend on the data availability (see text for further details).
Fig 3.
Proportions of households within 1, 5 and 10 km from the coast in 22 PICTSs.
Proportions of households in each of the three buffers (see legend). See Table 1 for interpretation of the three digit ISO country codes.
Table 2.
Percentage of populations within 1, 5 and 10 km zones in PICTS, sub-regions and region.
Estimates are derived from census, GPWv4 and Landscan. Where estimates differed by less than 5% only the census estimate is provided, otherwise the census estimate is bracketed as ‘census (GPWv4, Landscan)’. A complete listing of estimates is provided in S1 Table.–indicates that all of the land area is within the boundary of the next smaller zone. * Population as per most recent census (see Table 1).
Fig 4.
Comparison of methods for disaggregating populations to enumeration areas and zones.
Known household locations shown in the top row (green) translated to high-precision estimates of population (second row), when population was known. LandScan estimates of population distribution overestimated populations in urban centres (row 3). GPWv4 did not favour urban populations, instead, the population was averaged across areas (row 4). The maps show the islands of Tongatapu (Tonga), Viti Levu (Fiji), and part of Upolu (Samoa).
Fig 5.
Differences in allocation between global population datasets.
Percent differences between estimates of the population in (a) 1 km, (b) 5 km, and (c) 10 km buffer zone estimated by GPWv4 and LandScan and estimates using the census method.