Maps of human population distributions have found use in disease burden estimation, epidemic modelling, resource allocation, disaster management, accessibility modelling, transport and city planning, poverty mapping and environmental impact assessment amongst other applications. High-income countries often have extensive mapping resources and expertise at their disposal to create such databases, but across the low income regions of the world, relevant data are either lacking or are of poor quality. The scarcity of mapping resources, lack of reliable validation data and difficulty in obtaining high resolution contemporary census statistics remain major obstacles to settlement and population mapping across the low income regions of the World. WorldPop develops methods to exploit and integrate the growing range of geospatial data on human populations, their demographics and factors relating to population distributions. Census, survey, satellite, social media, cellphone and other spatial datasets are all integrated in flexible and peer-reviewed statistical methods to produce open, fully-documented and consistent gridded maps of population distributions. Gridded population maps, whereby population numbers per 100x100m grid square are estimated, represent a more consistent representation of population distributions across a landscape than administrative unit counts, as well as enabling smooth integration with multiple other gridded datasets.
The accuracy of gridded population mapping is strongly related to the availability of contemporary and spatially detailed population census data. WorldPop works with statistics agencies, ministries of health and other organizations to construct databases of the most spatially detailed and recent population census data available, and match these to corresponding administrative boundaries. The figure below shows examples of such datasets for three countries.
Administrative units within a country typically vary substantially in their shapes and sizes, making the consistent mapping of population distributions and densities across a country, and their integration with other datasets produced on differing spatial frameworks challenging. WorldPop have developed peer-reviewed spatial statistical methods, exploiting the power of machine learning, to transform and disaggregate population counts at administrative unit levels to 100x100m grid square level, exploiting relationships with spatial covariate layers from satellites and other sources. The figure below shows an example for northern Vietnam.
WorldPop population mapping example: (Top-left) Population density from census data for each administrative level 2 unit in an area of northern Vietnam, (Top-right) Land cover dataset for the same area, (Bottom-left) Satellite image of the area at night, (Bottom-right) WorldPop population modelling methods take the census data as input, then use machine learning methods to exploit the relationship between population density and high resolution landscape features, such as those from land cover and satellite data, to predict population densities for each 100x100m grid cell on the landscape.
No population mapping method is perfect, and therefore it is important to communicate the input datasets, methods and accuracies of the output maps. WorldPop undertake this through providing comprehensive metadata with each country population distribution map, which documents which datasets were used in its production, the approaches taken, as well as measures of accuracy. The figure below shows examples of these, and metadata are available with the datasets that can be freely downloaded here.
Example WorldPop mapping statistical analytics available in the metadata for country population maps: (Left) Internal accuracy assessment showing prediction errors for each administrative unit in a country; (Right) Assessment of the importance of difference input spatial data layers in improving the accuracy of population mapping. Further info available here.
WorldPop population distribution datasets have been used in applications around the World, covering the fields of urban planning, epidemiology, humanitarian response, health metrics and impact assessments, amongst others. The output datasets depict population counts and densities for multiple years per 100x100m grid cells for individual countries, and per 1x1km grid cells for continents. The figure below shows two example outputs.
Example WorldPop population distribution map outputs: (Top) Myanmar population density per 100x100m grid cell in 2010 created for the United Nations Myanmar Information Management Unit; (Bottom) Guatemala population density per 100x100m grid cell in 2015 constructed for the World Bank. All WorldPop datasets are freely available here.
In some countries, contemporary and reliable census data are lacking, and so WorldPop are currently collaborating with the Bill and Melinda Gates Foundation, Oak Ridge National Laboratories and the UNFPA to develop methods for 'bottom-up' population distribution mapping. Here, methods for extracting human settlement patterns and buildings from very high resolution satellite imagery are used in combination with microcensus data collected in the field to transform satellite feature extractions into high resolution population distribution estimates in the absence of national census data. Work is ongoing, but initial outputs are being used in Nigeria’s vaccination tracking system
The approaches outlined above provide static snapshots of population distributions at certain timepoints. Human populations are highly dynamic however, and to complement short term dynamic population distribution methods, WorldPop is developing methods for the high resolution mapping of historical and future population change over decadal timescales. Here, census data from multiple timepoints are integrated with satellite and other spatial datasets from corresponding time periods. Example applications include collaboration with the World Bank on their recent East Asia’s Changing Urban Landscape report and collaboration with the China CDC on high resolution population distribution mapping over the 1990-2015 period (see figure below).
In addition to this historical population mapping, WorldPop is developing methods for the spatiotemporal prediction of future population distributions, using urban spatial growth models as part of the Modelling and forecasting African Urban Population Patterns for vulnerability and health assessments (MAUPP) project funded by BELSPO.
The accuracy of existing population distribution datasets is limited within African urban areas and urban sprawl is currently not taken into account in population projections. We are developing a high-resolution, generalizable urban expansion model using built-up density layers for 48 African cities and 5 different time periods (1995, 2000, 2005, 2010 and 2015). In contrast to a discrete model (typically urban/non-urban; 0 or 1), the state of a cell takes continuous values, ranging from 0 to 1, to represent the transition between non-urban to urban. Random Forest (RF) models are used to calculate a probability of rural to urban conversion for each non-urban 30m pixel based on a set a covariates such as the accessibility, the topography, or the proportion of urban pixels in the neighbourhood. Cross-validation techniques are used to calibrate and validate the model. Example outputs are shown in the figure below.
The age-group composition of populations varies considerably across the world, and obtaining accurate, spatially detailed estimates about key groups, such as numbers of children under 5 years is important in designing vaccination strategies, educational planning or maternal healthcare delivery, for example. In addition to the high resolution mapping of total population counts and densities, WorldPop works to produce similar outputs broken down by age and sex to provide detailed spatial mapping of population pyramids. The construction of spatially detailed databases of boundary-matched census data (outlined above) enables construction of such mapped outputs. Further details are provided here , the datasets are available to download here , and an example output for women of childbearing age is shown below.
Just as is the case for population counts, in some countries, contemporary and reliable census data are unavailable for supporting detailed mapping of population pyramids. WorldPop have therefore developed methods for the production of high resolution age-structure mapping from geolocated household survey data, together with full quantification of model uncertainty. The figure below shows an example for Nigeria.
WorldPop high resolution mapping of population age structures in the absence of census data: (top-left) geolocated household survey cluster data coloured by the proportion of the population surveyed that was under 5 years of age; (top-right) example of one of the geospatial covariate layers shown to be strongly correlated with population age-structures, travel time to the nearest large settlement; (bottom-left) predicted proportion of the population under 5 years of age per 1x1km grid square using model-based geostatistics in a Bayesian framework; (bottom-right) map of per grid square uncertainty measure showing the level of confidence in each prediction made in the under 5yr proportion map.
The population of Africa is predicted to double over the next 40 years, driving exceptionally high urban expansion rates. Existing population distribution data for urban areas in low and middle-income regions is often limited by (i) a lack of spatially detailed data to capture within-urban heterogeneities, (ii) a lack of multi-temporal data to support improved understanding and forecasting of urban growth. Both of these limitations can be significantly improved with advanced remote sensing data and techniques. Through collaboration with the Modelling and forecasting African Urban Population Patterns for vulnerability and health assessments (MAUPP) project, WorldPop is working toward improved spatial understanding and forecasting of urbanization and urban population distributions in Africa. The performance of urban expansion models have thus far been limited by the quality and type of data available, reducing the confidence and the applications of models for Africa. Through the use of high resolution and very high resolution remote sensing data and spatial modelling, we develop methods to predict urban expansion at moderate spatial resolution, improve understanding and prediction of intra-urban variations in human population density, and integrate results into human population distribution models and forecasts.
Population censuses taken every decade provide a single snapshot of the distributions of populations across a country. People rarely stay still however, and knowing where they are is of primary importance for accurate impact assessments, particularly those focused on population health, food security, climate change, conflicts and natural disasters. Substantial shifts in population distributions occur between day and nighttime, weekdays and weekends, holidays and workdays and across seasons and years. Previously we have had no way to capture and quantify these variations at fine spatial and temporal scales across entire countries, but the advent of new digital datasources is changing this.
Anonimised mobile phone call data records offer a new datasource for understanding and measuring the dynamics of population distributions across countries. With more mobile phones on the planet than people and phone subscriptions reaching large percentages of the population, even in the poorest countries, such data represent a valuable resource for understanding population dynamics. Every time a communication is made or received, it is routed through the nearest tower to the user, indicating the presence of a person within the radius of that tower. With millions of users and billions of communications every day across thousands of towers, unprecedented data on population level activities are produced.Watch our video below to get more information:
WorldPop, in collaboration with the Flowminder Foundation, have developed methods to translate the billions of anonymous mobile communications into population maps that show similar levels of spatial accuracy as census data. Moreover, not only can accurate population maps be produced in the absence of census data, but the continual record of the mobile network data enables production of these maps for any time period, with examples shown below and in the video.
Map of population distribution changes in France between holidays and workdays
Moreover, the addition of satellite and GIS based covariates can overcome some of the limitations of mobile network data in highly rural areas where cell towers are widely spaced. By integrating both the mobile call data records and satellite/GIS layers within the WorldPop population modelling framework, further improvements in the spatial accuracy of population mapping can be achieved, as in the example below.
Map of predicted population distribution in Portugal through integration of mobile network data and satellite/GIS covariates built using the WorldPop random forests modelling framework
The analyses undertaken in France and Portugal enabled testing of methods against detailed and reliable census validation data. It is in the poorest areas of the World though where reliable data on population distributions and dynamics are most lacking and needed. Here, WorldPop and Flowminder are working to extend the mobile data methods to many low and middle income settings, along with assessments of alternative and complimentary datasources, such as dynamic satellite data. An example output for Namibia can be seen below, built in collaboration with the National Vector-borne Disease Control Program, MTC and the Clinton Health Access Initiative, in order to assess changing health facility catchment denominators.
Map of predicted population distribution in Portugal through integration of mobile network data and satellite/GIS covariates built using the WorldPop random forests modelling framework
Human movement typically has a central role in economics and development, the delivery of services, and the spread of infectious diseases. The progression of epidemics and maintenance of endemic diseases are strongly linked to human movement patterns, while access to markets and efficient transportation to increase workforce mobility and the flow of goods can drive economic development. Planning, intervention, mitigation, and development policies can be better informed through the incorporation of spatial information on human movement and connectivity across spatiotemporal scales, but reliable data for human mobility mapping have often been lacking, particularly in resource-poor settings. WorldPop works closely with its partner the Flowminder Foundation to fill these gaps, providing data and models on human mobility and connectivity in low and middle income settings.
Globalization and the expansion of transport networks has transformed migration into a major policy issue because of its effects on a range of phenomena, including resource flows in economics, urbanization, as well as the epidemiology of infectious diseases. Quantifying and modeling human migration can provide a better understanding of the nature of migration and help develop evidence-based interventions for disease control policy, economic development, and resource allocation. WorldPop is constructing a database on international and domestic migration, as well as modelling approaches, to enable a consistent mapping of migration flows and connectivity across all low and middle income countries. Example outputs are shown below, and further details are available on modelling approaches and data comparisons.
Mapping subnational migration in Latin America through integrating census microdata on migration patterns into a spatial interaction model framework. Close-up box shows Central America.
Perhaps the most promising new source of data on population movement patterns is that derived from anonimized mobile phone call detail records (CDRs). The time of each call or text made by an individual and the location of the tower it is routed through are recorded by mobile network operators for billing purposes. Through analyzing sequences of communications and their locations, the movement patterns of an anonimized individual can be inferred. Across the full set of phone users subscribed to the network, the movement patterns of millions of individuals across time periods of years can therefore be quantified at the spatial scale of phone tower reception areas. With estimates suggesting that there are now more mobile phones than people on the planet and high ownership levels, even in some of the poorest and most remote places, such data offer an unprecedented source of information on human mobility. In partnership with the Flowminder Foundation and multiple mobile network operators across the globe, WorldPop is developing methods for understanding and modelling population movements at unprecedented spatial and temporal scales, as well as feeding into improving population and poverty mapping. The figure below shows example outputs, with applications already shown in the fields of malaria elimination, cholera modelling, rubella dynamics, mapping of treatment seeking, Ebola control, mobility prediction, disaster response, poverty mapping and dynamic population modelling. WorldPop is also providing ongoing support in mapping population displacements following the recent Nepal earthquakes, with data and reports available here.
Human mobility mapping through mobile phone usage data: (left) Namibia population movement totals between locations in 2011; (centre) Haiti population flows with high rates of movement coloured in red, and low rates in blue; (right) West Africa movement predictions built on mobile phone data from Senegal and Cote d’Ivoire (further details here).
The mobile phone data represent a valuable and unique datasource on human movements, but they also represent sensitive and biased datasets. WorldPop is therefore working to produce models that replicate the patterns seen in mobile data, while accounting for biases, as well as integrating data on human mobility from other sources. Where feasible, WorldPop and Flowminder are working to distribute freely these mobility estimates and models. Examples in support of the West Africa Ebola outbreak containment efforts can be found on the dedicated WorldPop page here, and for the Nepal earthquake response here, while the figures below show further examples.
Human mobility spatial models: (left) Southern Africa modelled population flows per month parameterized using Nambia mobile phone data; (middle, right) Modelled seasonal population density changes in cities in Niger and Nigeria using fluctuations in satellite nightlight brightness (further details here)
The expanding global air network provides rapid and wide-reaching connections accelerating both domestic and international travel. To understand human movement patterns on the network and their socioeconomic, environmental and epidemiological implications, information on passenger flow is required. However, comprehensive data on global passenger flow remain difficult and expensive to obtain, prompting researchers to rely on scheduled flight seat capacity data or simple models of flow. WorldPop has collaborated with researchers at the Vector-Borne Disease Airline Importation Risk tool (VBD-Air, www.vbd-air.com) to construct open access modelled passenger flow datasets. Full details on the dataset construction methods for annual flows are here, and monthly flows here. The datasets are available to download on the WorldPop site, and a visualisation of annual flows is shown below.
Improved understanding of geographic variation and inequity in health status, wealth, and access to resources within countries is increasingly being recognized as central to meeting development goals. The lack of spatial datasets to aid in identifying the magnitude of inequities for women and newborns, both in outcomes and services, are now a key constraint to progress. Moreover, as international funding for health and development comes under pressure, the ability to target limited human resources and health services to underserved groups becomes crucial. Despite some progress in very recent years, insufficient global attention has been paid to disaggregating national data by geographical units and the maternal and newborn health community has yet to fully capitalize on the emerging capacity of GIS. In collaboration with UNFPA, NORAD, Integrare, the World Health Organization and the Open Health Initiative of the East African Community, WorldPop is working to improve the spatial evidence base for maternal and newborn health.
Planning for safer births and healthier newborns can be improved firstly by providing more accurate estimations of population distributions – targeting particularly women of childbearing age. WorldPop has developed methods for the high spatial resolution mapping the distributions of women of childbearing age (see example below), and details are provided here. Output age and sex-structured population distribution datasets are available to download through the data link above.
(Left) Tanzania census data showing proportions of the population who are women of childbearing age (15-49 yrs old) for each enumeration area; (Right) the census data are integrated with WorldPop’s Tanzania population distribution maps to produce estimates of numbers of women of childbearing age per 100x100m grid cell.
Spatially disaggregated projections of the approximate numbers of pregnancies and births that are likely to occur in the short and long-term are needed for more effective strategies on human resources and infrastructure. WorldPop has developed and applied methods for the high resolution mapping of pregnancies and births, across 75 of the highest burden countries in terms of maternal and newborn health. The integration of census, survey, satellite and GIS datasets is undertaken to account for the spatial and temporal heterogeneities in population distributions, age structures and fertility rates (see figure below).
WorldPop births and pregnancies distribution datasets were used as the basis for recent analyses presented by the UNFPA, World Health Organization and the International Confederation of Midwives in the State of the World’s Midwifery 2014 report (see figure below). Ongoing WorldPop analyses are extending and updating this work.
There is a need to link information on pregnancies and births to better information on health facilities in districts and regions so that distance to services and their coverage can be assessed. WorldPop projects in collaboration with NORAD, Integrare and the Open Health Initiative of the East African Community, are undertaking such analyses (example in figure below), as well as constructing new high resolution data layers on factors such as skilled birth attendance and adolescent birth rates.
(Left) estimated number of pregnancies per 100x100m grid cell in 2012 for Ethiopia, with the location of comprehensive EmONC facilities overlaid; (Right) Estimated percentage of pregnancies within 50km of a comprehensive EmONC facility in Ethiopia in 2012 by woreda administrative unit.
Finally, the potential of mobile phone call data records in improving measurement of access to maternal and newborn health services are being explored.
The Millennium Development Goals (MDGs) established global poverty eradication as a central focus of the international community. In the new Sustainable Development Goals (SDGs) era, targets for poverty eradication are being designed, with an increasing focus on subnational measurement of progress. Poverty mapping is a valuable approach for effectively characterizing sub national and multi-national disparities in the distribution of poverty. By providing a detailed description of the spatial distribution of poverty and inequality, poverty maps have the potential to play a key role in highlighting inequalities and providing a tool for crafting more effective policies aimed at targeting poverty interventions at detailed levels of spatial disaggregation. Previous approaches to poverty mapping have typically been based on census data, providing difficulties in terms of access to detailed and contemporary data that can be updated regularly, and limiting the ability to provide high resolution, up-to-date poverty maps.
The WorldPop project, in collaboration with researchers at the University of Oxford’s Malaria Atlas Project, and the Bill and Melinda Gates Foundation’s Financial Services for the Poor Team, have developed and applied novel approaches to poverty mapping built on geolocated household survey data. Full details of the methodologies used to construct the datasets depicting estimates of the proportion of the population living in poverty in each grid cell are available in this report. In brief, geolocated national household survey data were obtained through either the Demographic and Health Surveys (DHS) program or the Living Standards Measurement Study (LSMS) program and either $1.25 and $2 a day consumption-based poverty metrics or the Multidimensional Poverty Index (MPI) were calculated for each survey cluster. A Bayesian geostatistical modeling framework, following approaches constructed for the Malaria Atlas Project was then established to exploit spatiotemporal relationships within the data, leverage ancillary information from an extensive set of covariates, and rigorously handle uncertainties at all stages to generate output surfaces with accompanying confidence intervals. The figures below show example outputs for Nigeria, and further outputs can be found here.
(Top left) Geolocated household survey clusters from the 2011 Living Standards Measurement Survey (LSMS) showing the proportion of people below the $1.25 a day poverty line; (Top right) Example geospatial covariate layer showing an index of ‘accessibility’ measured through estimated travel time to the nearest large settlement; (Bottom left) Output poverty map showing predicted proportion of people below the $1.25 a day poverty line per 1x1km grid cell, available to download here; (Bottom right) Measurement of per-grid cell mapping uncertainty, available to download here.
These maps, as well as many more available here on the WorldPop site, were used as the basis for the Bill and Melinda Gates Foundation’s FSP maps tool, supporting governments in tracking progress against National Financial Inclusion Strategies and making evidence-based decisions. The approach is applicable for asset-based indices (e.g. Multidimensional Poverty Index, DHS wealth index) and consumption measures (e.g. <$2 a day) and recent analyses has shown value also for rapid measurement tools (e.g. Progress out of Poverty Index).
WorldPop continues to collaborate with a range of Financial Service Providers, the Bill and Melinda Gates Foundation and the Grameen Foundation on meeting the needs of governments and other stakeholders in undertaking and scaling up such high resolution poverty mapping. However, while such survey and satellite mapping overcomes some of the problems associated with census-based poverty mapping, a number of challenges remain, such as (i) a lack of mapping detail within urban areas, (ii) difficulties in updating these maps in the absence of new survey or census data, and (iii) substantial variation in poverty rates still remaining uncaptured by the approach.
With funding from the Bill and Melinda Gates Foundation, and in collaboration with the Flowminder Foundation, researchers at the University of Washington and multiple mobile network operators, the WorldPop project is working on overcoming the drawbacks of the satellite-based approach through the use of anonimised mobile phone network data. Anonimised data on mobile calling patterns, phone user mobility and credit top ups amounts and frequencies have been shown in a range of studies to correlate strongly with income and poverty metrics. Moreover, such data are produced in near real-time and are spatially detailed in urban areas (see figure below), providing valuable and complimentary additions to the mapping approach outlined above.
Spatial structure of mobile phone tower configuration in Bangladesh. The zoom window shows the spatial detail of Dhaka, where a high number of mobile towers enable the estimation of important within-city variations.
Results from Bangladesh provide the first research outputs produced by combining anonymised data from mobile phones and satellite imagery data to create high resolution maps of poverty. The results demonstrate that models integrating these data sources provide improvement in predictive power and lower error. These results are promising as the CDR data here produce accurate, high-resolution estimates in urban areas not possible using RS data alone. This research is now expanding to other countries and Bangladesh outputs can be downloaded here.