Publications
1.
Boo, Gianluca; Darin, Edith; Chamberlain, Heather R.; Hosner, Roland; Akilimali, Pierre K.; Kazadi, Henri Marie; Nnanatu, Chibuzor C.; Lázár, Attila N.; Tatem, Andrew J.
In: PLOS Global Public Health , 2025.
Abstract | Links | BibTeX | Tags:
@article{nokey,
title = {Tackling public health data gaps through Bayesian high-resolution population estimation: A case study of Kasaï-Oriental, Democratic Republic of the Congo},
author = {Gianluca Boo and Edith Darin and Heather R. Chamberlain and Roland Hosner and Pierre K. Akilimali and Henri Marie Kazadi and Chibuzor C. Nnanatu and Attila N. Lázár and Andrew J. Tatem},
url = {https://doi.org/10.1371/journal.pgph.0005072},
year = {2025},
date = {2025-09-04},
journal = {PLOS Global Public Health },
abstract = {Most low- and middle-income countries face significant public health challenges, exacerbated by the lack of reliable demographic data supporting effective planning and intervention. In such data-scarce settings, statistical models combining geolocated survey data with geospatial datasets enable the estimation of population counts at high spatial resolution in the absence of dependable demographic data sources. This study introduces a Bayesian model jointly estimating building and population counts, combining geolocated survey data and gridded geospatial datasets. The model provides population estimates for the Kasaï-Oriental province, Democratic Republic of the Congo (DRC), at a spatial resolution of approximately one hectare. Posterior estimates are aggregated across health zones and health areas to offer probabilistic insights into their respective populations. The model exhibits a –0.28 bias, 0.47 inaccuracy, and 0.95 imprecision using scaled residuals, with robust 95% credible intervals. The estimated population of Kasaï-Oriental for 2024 is approximately 4.1 million, with a credible range of 3.4 to 4.8 million. Aggregations by health zones and health areas reveal significant variations in population estimates and uncertainty levels, particularly between the provincial capital, Mbuji-Mayi and the rural hinterland. High-resolution Bayesian population estimates allow flexible aggregation across spatial units while providing probabilistic insights into model uncertainty. These estimates offer a unique resource for the public health community working in Kasaï-Oriental, for instance, in support of a better-informed allocation of vaccines to different operational boundaries based on the upper bound of the 95% credible intervals.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Most low- and middle-income countries face significant public health challenges, exacerbated by the lack of reliable demographic data supporting effective planning and intervention. In such data-scarce settings, statistical models combining geolocated survey data with geospatial datasets enable the estimation of population counts at high spatial resolution in the absence of dependable demographic data sources. This study introduces a Bayesian model jointly estimating building and population counts, combining geolocated survey data and gridded geospatial datasets. The model provides population estimates for the Kasaï-Oriental province, Democratic Republic of the Congo (DRC), at a spatial resolution of approximately one hectare. Posterior estimates are aggregated across health zones and health areas to offer probabilistic insights into their respective populations. The model exhibits a –0.28 bias, 0.47 inaccuracy, and 0.95 imprecision using scaled residuals, with robust 95% credible intervals. The estimated population of Kasaï-Oriental for 2024 is approximately 4.1 million, with a credible range of 3.4 to 4.8 million. Aggregations by health zones and health areas reveal significant variations in population estimates and uncertainty levels, particularly between the provincial capital, Mbuji-Mayi and the rural hinterland. High-resolution Bayesian population estimates allow flexible aggregation across spatial units while providing probabilistic insights into model uncertainty. These estimates offer a unique resource for the public health community working in Kasaï-Oriental, for instance, in support of a better-informed allocation of vaccines to different operational boundaries based on the upper bound of the 95% credible intervals.
2.
Utazi, C. Edson; Yankey, Ortis; Chaudhuri, Somnath; Olowe, Iyanuloluwa D.; Danovaro-Holliday, M. Carolina; Lazar, Attila N.; Tatem, Andrew J.
Geostatistical and machine learning approaches for high-resolution mapping of vaccination coverage Journal Article
In: Spatial and Spatio-temporal Epidemiology, vol. 54, no. 100744, 2025.
Abstract | Links | BibTeX | Tags:
@article{nokey,
title = {Geostatistical and machine learning approaches for high-resolution mapping of vaccination coverage},
author = {C. Edson Utazi and Ortis Yankey and Somnath Chaudhuri and Iyanuloluwa D. Olowe and M. Carolina Danovaro-Holliday and Attila N. Lazar and Andrew J. Tatem},
url = {https://doi.org/10.1016/j.sste.2025.100744},
year = {2025},
date = {2025-08-23},
journal = {Spatial and Spatio-temporal Epidemiology},
volume = {54},
number = {100744},
abstract = {Recently, there has been a growing interest in the production of high-resolution maps of vaccination coverage. These maps have been useful for uncovering geographic inequities in coverage and improving targeting of interventions to reach marginalized populations. Different methodological approaches have been developed for producing these maps using mostly geolocated household survey data and geospatial covariate information. However, it remains unclear how much the predicted coverage maps produced by the various methods differ, and which methods yield more reliable estimates. Here, we explore the predictive performance of these methods and resulting implications for spatial prioritization to fill this gap. Using Nigeria Demographic and Health Survey as a case study, we generate 1 × 1 km and district level maps of indicators of vaccination coverage using geostatistical, machine learning (ML) and hybrid methods and evaluate predictive performance via cross-validation. Our results show similar predictive performance for five of the seven methods investigated, although two geostatistical approaches are the best performing methods. The worst-performing methods are two ML approaches. We find marked differences in spatial prioritization using these methods, which could potentially result in missing important underserved populations, although broad similarities exist. Our study can help guide map production for other health and development metrics.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Recently, there has been a growing interest in the production of high-resolution maps of vaccination coverage. These maps have been useful for uncovering geographic inequities in coverage and improving targeting of interventions to reach marginalized populations. Different methodological approaches have been developed for producing these maps using mostly geolocated household survey data and geospatial covariate information. However, it remains unclear how much the predicted coverage maps produced by the various methods differ, and which methods yield more reliable estimates. Here, we explore the predictive performance of these methods and resulting implications for spatial prioritization to fill this gap. Using Nigeria Demographic and Health Survey as a case study, we generate 1 × 1 km and district level maps of indicators of vaccination coverage using geostatistical, machine learning (ML) and hybrid methods and evaluate predictive performance via cross-validation. Our results show similar predictive performance for five of the seven methods investigated, although two geostatistical approaches are the best performing methods. The worst-performing methods are two ML approaches. We find marked differences in spatial prioritization using these methods, which could potentially result in missing important underserved populations, although broad similarities exist. Our study can help guide map production for other health and development metrics.
3.
Chamberlain, Heather R; Pollard, Derek; Winters, Anna; Renn, Silvia; Borkovska, Olena; Musuka, Chisenga Abel; Membele, Garikai; Lazar, Attila N; Tatem, Andrew J
In: International Journal of Health Geographics, vol. 24, no. 1, pp. 13, 2025.
Abstract | Links | BibTeX | Tags:
@article{chamberlain2025assessing,
title = {Assessing the impact of building footprint dataset choice for health programme planning: a case study of indoor residual spraying (IRS) in Zambia},
author = {Heather R Chamberlain and Derek Pollard and Anna Winters and Silvia Renn and Olena Borkovska and Chisenga Abel Musuka and Garikai Membele and Attila N Lazar and Andrew J Tatem},
url = {https://doi.org/10.1186/s12942-025-00398-7},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {International Journal of Health Geographics},
volume = {24},
number = {1},
pages = {13},
publisher = {Springer},
abstract = {The increasing availability globally of building footprint datasets has brought new opportunities to support a geographic approach to health programme planning. This is particularly acute in settings with high disease burdens but limited geospatial data available to support targeted planning. The comparability of building footprint datasets has recently started to be explored, but the impact of utilising a particular dataset in analyses to support decision making for health programme planning has not been studied. In this study, we quantify the impact of utilising four different building footprint datasets in analyses to support health programme planning, with an example of malaria vector control initiatives in Zambia.
Using the example of planning indoor residual spraying (IRS) campaigns in Zambia, we identify priority locations for deployment of this intervention based on criteria related to the area, proximity and counts of building footprints per settlement. We apply the same criteria to four different building footprint datasets and quantify the count and geographic variability in the priority settlements that are identified.
We show that nationally the count of potential priority settlements for IRS varies by over 230% with different building footprint datasets, considering a minimum threshold of 25 sprayable buildings per settlement. Differences are most pronounced for rural settlements, indicating that the choice of dataset may bias the selection to include or exclude settlements, and consequently population groups, in some areas.
The results of this study show that the choice of building footprint dataset can have a considerable impact on the potential settlements identified for IRS, in terms of (i) their location and count, and (ii) the count of building footprints within priority settlements. The choice of dataset potentially has substantial implications for campaign planning, implementation and coverage assessment. Given the magnitude of the differences observed, further work should more broadly assess the sensitivity of health programme planning metrics to different building footprint datasets, and across a range of geographic contexts and health campaign types.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The increasing availability globally of building footprint datasets has brought new opportunities to support a geographic approach to health programme planning. This is particularly acute in settings with high disease burdens but limited geospatial data available to support targeted planning. The comparability of building footprint datasets has recently started to be explored, but the impact of utilising a particular dataset in analyses to support decision making for health programme planning has not been studied. In this study, we quantify the impact of utilising four different building footprint datasets in analyses to support health programme planning, with an example of malaria vector control initiatives in Zambia.
Using the example of planning indoor residual spraying (IRS) campaigns in Zambia, we identify priority locations for deployment of this intervention based on criteria related to the area, proximity and counts of building footprints per settlement. We apply the same criteria to four different building footprint datasets and quantify the count and geographic variability in the priority settlements that are identified.
We show that nationally the count of potential priority settlements for IRS varies by over 230% with different building footprint datasets, considering a minimum threshold of 25 sprayable buildings per settlement. Differences are most pronounced for rural settlements, indicating that the choice of dataset may bias the selection to include or exclude settlements, and consequently population groups, in some areas.
The results of this study show that the choice of building footprint dataset can have a considerable impact on the potential settlements identified for IRS, in terms of (i) their location and count, and (ii) the count of building footprints within priority settlements. The choice of dataset potentially has substantial implications for campaign planning, implementation and coverage assessment. Given the magnitude of the differences observed, further work should more broadly assess the sensitivity of health programme planning metrics to different building footprint datasets, and across a range of geographic contexts and health campaign types.
Using the example of planning indoor residual spraying (IRS) campaigns in Zambia, we identify priority locations for deployment of this intervention based on criteria related to the area, proximity and counts of building footprints per settlement. We apply the same criteria to four different building footprint datasets and quantify the count and geographic variability in the priority settlements that are identified.
We show that nationally the count of potential priority settlements for IRS varies by over 230% with different building footprint datasets, considering a minimum threshold of 25 sprayable buildings per settlement. Differences are most pronounced for rural settlements, indicating that the choice of dataset may bias the selection to include or exclude settlements, and consequently population groups, in some areas.
The results of this study show that the choice of building footprint dataset can have a considerable impact on the potential settlements identified for IRS, in terms of (i) their location and count, and (ii) the count of building footprints within priority settlements. The choice of dataset potentially has substantial implications for campaign planning, implementation and coverage assessment. Given the magnitude of the differences observed, further work should more broadly assess the sensitivity of health programme planning metrics to different building footprint datasets, and across a range of geographic contexts and health campaign types.
4.
Yankey, Ortis; Utazi, Chigozie E.; Nnanatu, Christopher C.; Gadiaga, Assane N.; Abbot, Thomas; Lazar, Attila N.; Tatem, Andrew J.
Disaggregating census data for population mapping using a Bayesian Additive Regression Tree model Journal Article
In: Applied Geography, vol. 174, 2024.
Abstract | Links | BibTeX | Tags:
@article{nokey,
title = {Disaggregating census data for population mapping using a Bayesian Additive Regression Tree model},
author = {Ortis Yankey and Chigozie E. Utazi and Christopher C. Nnanatu and Assane N. Gadiaga and Thomas Abbot and Attila N. Lazar and Andrew J. Tatem},
url = {https://doi.org/10.1016/j.apgeog.2024.103416},
year = {2024},
date = {2024-09-14},
journal = {Applied Geography},
volume = {174},
abstract = {Population data is crucial for policy decisions, but fine-scale population numbers are often lacking due to the challenge of sharing sensitive data. Different approaches, such as the use of the Random Forest (RF) model, have been used to disaggregate census data from higher administrative units to small area scales. A major limitation of the RF model is its inability to quantify the uncertainties associated with the predicted populations, which can be important for policy decisions. In this study, we applied a Bayesian Additive Regression Tree (BART) model for population disaggregation and compared the result with a RF model using both simulated data and the 2021 census data for Ghana. The BART model consistently outperforms the RF model in out-of-sample predictions for all metrics, such as bias, mean squared error (MSE), and root mean squared error (RMSE). The BART model also addresses the limitations of the RF model by providing uncertainty estimates around the predicted population, which is often lacking with the RF model. Overall, the study demonstrates the superiority of the BART model over the RF model in disaggregating population data and highlights its potential for gridded population estimates.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Population data is crucial for policy decisions, but fine-scale population numbers are often lacking due to the challenge of sharing sensitive data. Different approaches, such as the use of the Random Forest (RF) model, have been used to disaggregate census data from higher administrative units to small area scales. A major limitation of the RF model is its inability to quantify the uncertainties associated with the predicted populations, which can be important for policy decisions. In this study, we applied a Bayesian Additive Regression Tree (BART) model for population disaggregation and compared the result with a RF model using both simulated data and the 2021 census data for Ghana. The BART model consistently outperforms the RF model in out-of-sample predictions for all metrics, such as bias, mean squared error (MSE), and root mean squared error (RMSE). The BART model also addresses the limitations of the RF model by providing uncertainty estimates around the predicted population, which is often lacking with the RF model. Overall, the study demonstrates the superiority of the BART model over the RF model in disaggregating population data and highlights its potential for gridded population estimates.
5.
Chamberlain, Heather R.; Darin, Edith; Adewole, Wole Ademola; Jochem, Warren C.; Lazar, Attila N.; Tatem, Andrew J.
Building footprint data for countries in Africa: To what extent are existing data products comparable? Journal Article
In: Computers, Environment and Urban Systems, vol. 110, 2024.
Abstract | Links | BibTeX | Tags:
@article{nokey,
title = {Building footprint data for countries in Africa: To what extent are existing data products comparable?},
author = {Heather R. Chamberlain and Edith Darin and Wole Ademola Adewole and Warren C. Jochem and Attila N. Lazar and Andrew J. Tatem},
url = {https://doi.org/10.1016/j.compenvurbsys.2024.102104},
doi = {10.1016/j.compenvurbsys.2024.102104},
year = {2024},
date = {2024-03-22},
journal = {Computers, Environment and Urban Systems},
volume = {110},
abstract = {Growth and developments in computing power, machine-learning algorithms and satellite imagery spatiotemporal resolution have led to rapid developments in automated feature-extraction. These methods have been applied to create geospatial datasets of features such as roads, trees and building footprints, at a range of spatial scales, with national and multi-country datasets now available as open data from multiple sources. Building footprint data is particularly useful in a range of applications including mapping population distributions, planning resource distribution campaigns and in humanitarian response. In settings with well-developed geospatial data systems, such datasets may complement existing authoritative sources, but in data-scarce settings, they may be the only source of data. However, knowledge on the degree to which building footprint data products are comparable and can be used interchangeably, and the impact of selecting a particular dataset on subsequent analyses remains limited. For all countries in Africa, we review the available multi-country building footprint data products and analyse their similarities and differences in terms of building area and count metrics. We explore the variation between building footprint data products across a range of spatial scales, including sub-national administrative units and different settlement types. Our results show that the available building footprint data products are not interchangeable. There are clear differences in counts and total area of building footprints between the assessed data products, as well as considerable spatial heterogeneity in building footprint coverage and completeness.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Growth and developments in computing power, machine-learning algorithms and satellite imagery spatiotemporal resolution have led to rapid developments in automated feature-extraction. These methods have been applied to create geospatial datasets of features such as roads, trees and building footprints, at a range of spatial scales, with national and multi-country datasets now available as open data from multiple sources. Building footprint data is particularly useful in a range of applications including mapping population distributions, planning resource distribution campaigns and in humanitarian response. In settings with well-developed geospatial data systems, such datasets may complement existing authoritative sources, but in data-scarce settings, they may be the only source of data. However, knowledge on the degree to which building footprint data products are comparable and can be used interchangeably, and the impact of selecting a particular dataset on subsequent analyses remains limited. For all countries in Africa, we review the available multi-country building footprint data products and analyse their similarities and differences in terms of building area and count metrics. We explore the variation between building footprint data products across a range of spatial scales, including sub-national administrative units and different settlement types. Our results show that the available building footprint data products are not interchangeable. There are clear differences in counts and total area of building footprints between the assessed data products, as well as considerable spatial heterogeneity in building footprint coverage and completeness.
6.
Sanchez-Cespedes, Lina Maria; Leasure, Douglas Ryan; Tejedor-Garavito, Natalia; Cruz, Glenn Harry Amaya; Velez, Gustavo Adolfo Garcia; Mendoza, Andryu Enrique; Salazar, Yenny Andrea Marín; Esch, Thomas; Tatem, Andrew J.; Bohórquez, Mariana Ospina
Social cartography and satellite-derived building coverage for post-census population estimates in difficult-to-access regions of Colombia Journal Article
In: Population Studies, pp. 1-18, 2023.
Abstract | Links | BibTeX | Tags:
@article{nokey,
title = {Social cartography and satellite-derived building coverage for post-census population estimates in difficult-to-access regions of Colombia},
author = {Lina Maria Sanchez-Cespedes and Douglas Ryan Leasure and Natalia Tejedor-Garavito and Glenn Harry Amaya Cruz and Gustavo Adolfo Garcia Velez and Andryu Enrique Mendoza and Yenny Andrea Marín Salazar and Thomas Esch and Andrew J. Tatem and Mariana Ospina Bohórquez},
url = {https://doi.org/10.1080/00324728.2023.2190151},
doi = {10.1080/00324728.2023.2190151},
year = {2023},
date = {2023-03-28},
journal = {Population Studies},
pages = {1-18},
abstract = {Effective government services rely on accurate population numbers to allocate resources. In Colombia and globally, census enumeration is challenging in remote regions and where armed conflict is occurring. During census preparations, the Colombian National Administrative Department of Statistics conducted social cartography workshops, where community representatives estimated numbers of dwellings and people throughout their regions. We repurposed this information, combining it with remotely sensed buildings data and other geospatial data. To estimate building counts and population sizes, we developed hierarchical Bayesian models, trained using nearby full-coverage census enumerations and assessed using 10-fold cross-validation. We compared models to assess the relative contributions of community knowledge, remotely sensed buildings, and their combination to model fit. The Community model was unbiased but imprecise; the Satellite model was more precise but biased; and the Combination model was best for overall accuracy. Results reaffirmed the power of remotely sensed buildings data for population estimation and highlighted the value of incorporating local knowledge.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Effective government services rely on accurate population numbers to allocate resources. In Colombia and globally, census enumeration is challenging in remote regions and where armed conflict is occurring. During census preparations, the Colombian National Administrative Department of Statistics conducted social cartography workshops, where community representatives estimated numbers of dwellings and people throughout their regions. We repurposed this information, combining it with remotely sensed buildings data and other geospatial data. To estimate building counts and population sizes, we developed hierarchical Bayesian models, trained using nearby full-coverage census enumerations and assessed using 10-fold cross-validation. We compared models to assess the relative contributions of community knowledge, remotely sensed buildings, and their combination to model fit. The Community model was unbiased but imprecise; the Satellite model was more precise but biased; and the Combination model was best for overall accuracy. Results reaffirmed the power of remotely sensed buildings data for population estimation and highlighted the value of incorporating local knowledge.
7.
Qader, Sarchil; Chamberlain, Heather; Kuepie, Mathias; Hunt, Freja K.; and Andrew J. Tatem, Attila Lazar
Field testing of pre-Enumeration Areas created using semi-automated delineation approach, Democratic Republic of Congo Technical Report
2023.
Abstract | Links | BibTeX | Tags:
@techreport{nokey,
title = {Field testing of pre-Enumeration Areas created using semi-automated delineation approach, Democratic Republic of Congo},
author = {Sarchil Qader and Heather Chamberlain and Mathias Kuepie and Freja K. Hunt and Attila Lazar and Andrew J. Tatem},
url = {https://eprints.soton.ac.uk/475327/},
doi = {10.5258/SOTON/WP00759},
year = {2023},
date = {2023-03-15},
urldate = {2023-03-15},
abstract = {This report details the main outcomes of the field testing of pre-Enumeration Areas (EAs) created from WorldPop and Flominder’s semi-automated EA approach that took place across three test sites in the provinces of Kinshasa and Kongo-Central, Democratic Republic of the Congo in December 2019. The field testing was conducted over four days by the BCR technical staff with participation from UNFPA and WorldPop staff.
Generally, EA boundaries from one census will form the basis for the EAs in the next census, with updates needed to account for new settlements and changes in population density. However, in countries where there hasn’t been a census for many years, often due to conflict or insecurity, EA boundaries can be incomplete, outdated, or missing altogether. The delineation of EAs is, therefore, a crucial pre-census activity but can often be particularly challenging and highly resource intensive. Creating EAs requires consideration of population and area size within each unit to ensure that they have approximately equal-sized populations and are a manageable size to be covered by census enumeration staff. To respond to this challenge, WorldPop has developed a semi-automatic approach of delineating pre-EAs to support census cartography. This approach utilises high-resolution gridded population estimates and digitised geographic features, including administrative boundaries, and natural and man-made features, such as rivers and roads, to divide the regions into small areas which are then merged to meet criteria specified for population size and geographic area.
The last census in DRC was conducted in 1984; consequently, a recent, national, digital EA dataset which can be used for cartography planning does not exist. GRID3 is supporting the realisation of a fully digital 2020 round census in the DRC and is working closely with the National Institute of Statistics and the DRC Census Bureau (Bureau Central de Recensement, BCR) to provide technical guidance regarding options for incorporating geospatial methodologies into census planning and census cartography. As the DRC Census Bureau prepares for the 2nd National Population and Housing Census (RGPH2), a new dataset of EA boundaries is needed. As part of GRID3’s work with the BCR, a field test was conducted to assess the feasibility of using a semi-automated approach for the delineation of pre-EA boundaries.
A preliminary pre-EA dataset was produced for the three test sites (Site 1: Quartier Kingu, Kinshasa (urban), Site 2: Quartier Dumi, Kinshasa (sub-urban), Site 3: Secteur Kasangulu, Kongo-Central (rural)) that span both rural and urban contexts. The geographic area covered by the three sites totalled 1,190 km2 and was sub-divided into approximately 312 pre-EAs. The pre-EAs created for the three test sites were classified as classes 1-3 depending on the degree to which the pre-EA boundaries followed visible features (e.g. roads). Class 1 being those pre-EAs with boundaries which fully followed visible features, class 2 boundaries followed visible features in part, and class 3 which didn’t follow visible features at all. A visual assessment was carried out by comparing the pre-EA boundaries with recent high-resolution satellite imagery. A subset of the pre-EAs (15 pre-EAs), covering classes 1, 2 and 3 were selected, and assessed in the field to check how the boundaries related to ground features and their feasibility as units for population enumeration. Class 1 pre-EAs were only found in urban contexts and tended to be bounded fully by roads, which were found to be simple for the field teams to follow. In class 2 and class 3 pre-EAs, the field teams were generally able to follow roads or tracks throughout the pre-EA to reach settlements, and ascertain when they had reached the boundary of the pre-EA using the maps and GPS location indicator on the tablets. The pre-EA boundaries were also created to avoid splitting settlements and therefore even in rural areas, the field teams were able to know where housing units needed to be enumerated.
A range of limitations with this work have been identified, both with the methods and equipment used in the field data collection and the methods and input data used to produce the pre-EA boundaries. Despite the identified limitations and the challenges encountered in the field, the findings from the field test were generally consistent, with the pre-EAs created by the semi-automated approach found to be suitable for population enumeration in the field. Overall the fieldwork was successfully conducted and expectations were met and even exceeded: the BCR found that the pre-EA outputs were found to help facilitate enumeration, as the BCR team could navigate within the pre-EA boundaries and know which housing units to enumerate. The findings of the field test indicate this semi-automated approach to creating pre-EAs has the potential to be used by the BCR to create pre-EAs in preparation for census cartography, and offers large savings in terms of time, labour and cost. Nonetheless, it would be expected that the pre-EA outputs created in the approach are carefully reviewed in the lab, and manually edited as needed prior to census cartography. Then whilst in the field, the pre-EA boundaries should be validated. Limitations associated with input datasets can be addressed through a comprehensive review of existing datasets, incorporating newly available feature extraction datasets as appropriate. Further development of the approach and potential solutions and suggestions to overcome the identified limitations are outlined and discussed in detail in the report.
We expect the findings of the field test in DRC to be transferable to other similar contexts, with the approach having applicability in countries with no recent digital EAs. We also expect the approach could be adapted to update digital EA boundaries in contexts with outdated EA datasets, but this should be explored through further research and testing in such contexts.
Worth noting that in close collaboration with GeoData at the University of Southampton, UNFPA and multiple national statistical offices around the world, WorldPop has now converted the automatic delineation script to a user-friendly tool which require minimal GIS skill to run.},
howpublished = {eprints Soton},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
This report details the main outcomes of the field testing of pre-Enumeration Areas (EAs) created from WorldPop and Flominder’s semi-automated EA approach that took place across three test sites in the provinces of Kinshasa and Kongo-Central, Democratic Republic of the Congo in December 2019. The field testing was conducted over four days by the BCR technical staff with participation from UNFPA and WorldPop staff.
Generally, EA boundaries from one census will form the basis for the EAs in the next census, with updates needed to account for new settlements and changes in population density. However, in countries where there hasn’t been a census for many years, often due to conflict or insecurity, EA boundaries can be incomplete, outdated, or missing altogether. The delineation of EAs is, therefore, a crucial pre-census activity but can often be particularly challenging and highly resource intensive. Creating EAs requires consideration of population and area size within each unit to ensure that they have approximately equal-sized populations and are a manageable size to be covered by census enumeration staff. To respond to this challenge, WorldPop has developed a semi-automatic approach of delineating pre-EAs to support census cartography. This approach utilises high-resolution gridded population estimates and digitised geographic features, including administrative boundaries, and natural and man-made features, such as rivers and roads, to divide the regions into small areas which are then merged to meet criteria specified for population size and geographic area.
The last census in DRC was conducted in 1984; consequently, a recent, national, digital EA dataset which can be used for cartography planning does not exist. GRID3 is supporting the realisation of a fully digital 2020 round census in the DRC and is working closely with the National Institute of Statistics and the DRC Census Bureau (Bureau Central de Recensement, BCR) to provide technical guidance regarding options for incorporating geospatial methodologies into census planning and census cartography. As the DRC Census Bureau prepares for the 2nd National Population and Housing Census (RGPH2), a new dataset of EA boundaries is needed. As part of GRID3’s work with the BCR, a field test was conducted to assess the feasibility of using a semi-automated approach for the delineation of pre-EA boundaries.
A preliminary pre-EA dataset was produced for the three test sites (Site 1: Quartier Kingu, Kinshasa (urban), Site 2: Quartier Dumi, Kinshasa (sub-urban), Site 3: Secteur Kasangulu, Kongo-Central (rural)) that span both rural and urban contexts. The geographic area covered by the three sites totalled 1,190 km2 and was sub-divided into approximately 312 pre-EAs. The pre-EAs created for the three test sites were classified as classes 1-3 depending on the degree to which the pre-EA boundaries followed visible features (e.g. roads). Class 1 being those pre-EAs with boundaries which fully followed visible features, class 2 boundaries followed visible features in part, and class 3 which didn’t follow visible features at all. A visual assessment was carried out by comparing the pre-EA boundaries with recent high-resolution satellite imagery. A subset of the pre-EAs (15 pre-EAs), covering classes 1, 2 and 3 were selected, and assessed in the field to check how the boundaries related to ground features and their feasibility as units for population enumeration. Class 1 pre-EAs were only found in urban contexts and tended to be bounded fully by roads, which were found to be simple for the field teams to follow. In class 2 and class 3 pre-EAs, the field teams were generally able to follow roads or tracks throughout the pre-EA to reach settlements, and ascertain when they had reached the boundary of the pre-EA using the maps and GPS location indicator on the tablets. The pre-EA boundaries were also created to avoid splitting settlements and therefore even in rural areas, the field teams were able to know where housing units needed to be enumerated.
A range of limitations with this work have been identified, both with the methods and equipment used in the field data collection and the methods and input data used to produce the pre-EA boundaries. Despite the identified limitations and the challenges encountered in the field, the findings from the field test were generally consistent, with the pre-EAs created by the semi-automated approach found to be suitable for population enumeration in the field. Overall the fieldwork was successfully conducted and expectations were met and even exceeded: the BCR found that the pre-EA outputs were found to help facilitate enumeration, as the BCR team could navigate within the pre-EA boundaries and know which housing units to enumerate. The findings of the field test indicate this semi-automated approach to creating pre-EAs has the potential to be used by the BCR to create pre-EAs in preparation for census cartography, and offers large savings in terms of time, labour and cost. Nonetheless, it would be expected that the pre-EA outputs created in the approach are carefully reviewed in the lab, and manually edited as needed prior to census cartography. Then whilst in the field, the pre-EA boundaries should be validated. Limitations associated with input datasets can be addressed through a comprehensive review of existing datasets, incorporating newly available feature extraction datasets as appropriate. Further development of the approach and potential solutions and suggestions to overcome the identified limitations are outlined and discussed in detail in the report.
We expect the findings of the field test in DRC to be transferable to other similar contexts, with the approach having applicability in countries with no recent digital EAs. We also expect the approach could be adapted to update digital EA boundaries in contexts with outdated EA datasets, but this should be explored through further research and testing in such contexts.
Worth noting that in close collaboration with GeoData at the University of Southampton, UNFPA and multiple national statistical offices around the world, WorldPop has now converted the automatic delineation script to a user-friendly tool which require minimal GIS skill to run.
Generally, EA boundaries from one census will form the basis for the EAs in the next census, with updates needed to account for new settlements and changes in population density. However, in countries where there hasn’t been a census for many years, often due to conflict or insecurity, EA boundaries can be incomplete, outdated, or missing altogether. The delineation of EAs is, therefore, a crucial pre-census activity but can often be particularly challenging and highly resource intensive. Creating EAs requires consideration of population and area size within each unit to ensure that they have approximately equal-sized populations and are a manageable size to be covered by census enumeration staff. To respond to this challenge, WorldPop has developed a semi-automatic approach of delineating pre-EAs to support census cartography. This approach utilises high-resolution gridded population estimates and digitised geographic features, including administrative boundaries, and natural and man-made features, such as rivers and roads, to divide the regions into small areas which are then merged to meet criteria specified for population size and geographic area.
The last census in DRC was conducted in 1984; consequently, a recent, national, digital EA dataset which can be used for cartography planning does not exist. GRID3 is supporting the realisation of a fully digital 2020 round census in the DRC and is working closely with the National Institute of Statistics and the DRC Census Bureau (Bureau Central de Recensement, BCR) to provide technical guidance regarding options for incorporating geospatial methodologies into census planning and census cartography. As the DRC Census Bureau prepares for the 2nd National Population and Housing Census (RGPH2), a new dataset of EA boundaries is needed. As part of GRID3’s work with the BCR, a field test was conducted to assess the feasibility of using a semi-automated approach for the delineation of pre-EA boundaries.
A preliminary pre-EA dataset was produced for the three test sites (Site 1: Quartier Kingu, Kinshasa (urban), Site 2: Quartier Dumi, Kinshasa (sub-urban), Site 3: Secteur Kasangulu, Kongo-Central (rural)) that span both rural and urban contexts. The geographic area covered by the three sites totalled 1,190 km2 and was sub-divided into approximately 312 pre-EAs. The pre-EAs created for the three test sites were classified as classes 1-3 depending on the degree to which the pre-EA boundaries followed visible features (e.g. roads). Class 1 being those pre-EAs with boundaries which fully followed visible features, class 2 boundaries followed visible features in part, and class 3 which didn’t follow visible features at all. A visual assessment was carried out by comparing the pre-EA boundaries with recent high-resolution satellite imagery. A subset of the pre-EAs (15 pre-EAs), covering classes 1, 2 and 3 were selected, and assessed in the field to check how the boundaries related to ground features and their feasibility as units for population enumeration. Class 1 pre-EAs were only found in urban contexts and tended to be bounded fully by roads, which were found to be simple for the field teams to follow. In class 2 and class 3 pre-EAs, the field teams were generally able to follow roads or tracks throughout the pre-EA to reach settlements, and ascertain when they had reached the boundary of the pre-EA using the maps and GPS location indicator on the tablets. The pre-EA boundaries were also created to avoid splitting settlements and therefore even in rural areas, the field teams were able to know where housing units needed to be enumerated.
A range of limitations with this work have been identified, both with the methods and equipment used in the field data collection and the methods and input data used to produce the pre-EA boundaries. Despite the identified limitations and the challenges encountered in the field, the findings from the field test were generally consistent, with the pre-EAs created by the semi-automated approach found to be suitable for population enumeration in the field. Overall the fieldwork was successfully conducted and expectations were met and even exceeded: the BCR found that the pre-EA outputs were found to help facilitate enumeration, as the BCR team could navigate within the pre-EA boundaries and know which housing units to enumerate. The findings of the field test indicate this semi-automated approach to creating pre-EAs has the potential to be used by the BCR to create pre-EAs in preparation for census cartography, and offers large savings in terms of time, labour and cost. Nonetheless, it would be expected that the pre-EA outputs created in the approach are carefully reviewed in the lab, and manually edited as needed prior to census cartography. Then whilst in the field, the pre-EA boundaries should be validated. Limitations associated with input datasets can be addressed through a comprehensive review of existing datasets, incorporating newly available feature extraction datasets as appropriate. Further development of the approach and potential solutions and suggestions to overcome the identified limitations are outlined and discussed in detail in the report.
We expect the findings of the field test in DRC to be transferable to other similar contexts, with the approach having applicability in countries with no recent digital EAs. We also expect the approach could be adapted to update digital EA boundaries in contexts with outdated EA datasets, but this should be explored through further research and testing in such contexts.
Worth noting that in close collaboration with GeoData at the University of Southampton, UNFPA and multiple national statistical offices around the world, WorldPop has now converted the automatic delineation script to a user-friendly tool which require minimal GIS skill to run.
8.
Chamberlain, Heather R.; Lazar, Attila N.; Tatem, Andrew J.
High-resolution estimates of social distancing feasibility, mapped for urban areas in sub-Saharan Africa Journal Article
In: Scientific Data, vol. 9, no. 711 (2022), 2022.
Abstract | Links | BibTeX | Tags:
@article{nokey,
title = {High-resolution estimates of social distancing feasibility, mapped for urban areas in sub-Saharan Africa},
author = {Heather R. Chamberlain and Attila N. Lazar and Andrew J. Tatem },
url = {https://doi.org/10.1038/s41597-022-01799-0
},
doi = {10.1038/s41597-022-01799-0},
year = {2022},
date = {2022-11-18},
urldate = {2023-11-18},
journal = {Scientific Data},
volume = {9},
number = {711 (2022)},
abstract = {Social distancing has been widely-implemented as a public health measure during the COVID-19 pandemic. Despite widespread application of social distancing guidance, the feasibility of people adhering to such guidance varies in different settings, influenced by population density, the built environment and a range of socio-economic factors. Social distancing constraints however have only been identified and mapped for limited areas. Here, we present an ease of social distancing index, integrating metrics on urban form and population density derived from new multi-country building footprint datasets and gridded population estimates. The index dataset provides estimates of social distancing feasibility, mapped at high-resolution for urban areas across 50 countries in sub-Saharan Africa.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Social distancing has been widely-implemented as a public health measure during the COVID-19 pandemic. Despite widespread application of social distancing guidance, the feasibility of people adhering to such guidance varies in different settings, influenced by population density, the built environment and a range of socio-economic factors. Social distancing constraints however have only been identified and mapped for limited areas. Here, we present an ease of social distancing index, integrating metrics on urban form and population density derived from new multi-country building footprint datasets and gridded population estimates. The index dataset provides estimates of social distancing feasibility, mapped at high-resolution for urban areas across 50 countries in sub-Saharan Africa.
9.
Chamberlain, Heather R.; Lazar, Attila N.; Tatem, Andrew J.
High-resolution estimates of social distancing feasibility, mapped for urban areas in sub-Saharan Africa Journal Article
In: Scientific Data, vol. 9, no. 711, 2022.
Abstract | Links | BibTeX | Tags: Africa, covid-19, NPIs
@article{nokey,
title = {High-resolution estimates of social distancing feasibility, mapped for urban areas in sub-Saharan Africa},
author = {Heather R. Chamberlain and Attila N. Lazar and Andrew J. Tatem },
doi = {10.1038/s41597-022-01799-0},
year = {2022},
date = {2022-11-18},
journal = {Scientific Data},
volume = {9},
number = {711},
abstract = {Social distancing has been widely-implemented as a public health measure during the COVID-19 pandemic. Despite widespread application of social distancing guidance, the feasibility of people adhering to such guidance varies in different settings, influenced by population density, the built environment and a range of socio-economic factors. Social distancing constraints however have only been identified and mapped for limited areas. Here, we present an ease of social distancing index, integrating metrics on urban form and population density derived from new multi-country building footprint datasets and gridded population estimates. The index dataset provides estimates of social distancing feasibility, mapped at high-resolution for urban areas across 50 countries in sub-Saharan Africa.},
keywords = {Africa, covid-19, NPIs},
pubstate = {published},
tppubtype = {article}
}
Social distancing has been widely-implemented as a public health measure during the COVID-19 pandemic. Despite widespread application of social distancing guidance, the feasibility of people adhering to such guidance varies in different settings, influenced by population density, the built environment and a range of socio-economic factors. Social distancing constraints however have only been identified and mapped for limited areas. Here, we present an ease of social distancing index, integrating metrics on urban form and population density derived from new multi-country building footprint datasets and gridded population estimates. The index dataset provides estimates of social distancing feasibility, mapped at high-resolution for urban areas across 50 countries in sub-Saharan Africa.
10.
Boo, Gianluca; Darin, Edith; Leasure, Douglas R; Dooley, Claire A; Chamberlain, Heather R; and Lázár, Attila N; Tschirhart, Kevin; Sinai, Cyrus; Hoff, Nicole A; Fuller, Trevon
High-resolution population estimation using household survey data and building footprints Journal Article
In: Nature Communications, vol. 13, no. 1330, 2022.
Abstract | Links | BibTeX | Tags: Bayesian inference, Demographic and Health Surveys, Population
@article{nokey,
title = {High-resolution population estimation using household survey data and building footprints},
author = {Boo, Gianluca and Darin, Edith and Leasure, Douglas R and Dooley, Claire A and Chamberlain, Heather R and and Lázár, Attila N and Tschirhart, Kevin and Sinai, Cyrus and Hoff, Nicole A and Fuller, Trevon},
doi = {https://doi.org/10.1038/s41467-022-29094-x},
year = {2022},
date = {2022-03-14},
urldate = {2022-03-14},
journal = {Nature Communications},
volume = {13},
number = {1330},
abstract = {The national census is an essential data source to support decision-making in many areas of public interest. However, this data may become outdated during the intercensal period, which can stretch up to several decades. In this study, we develop a Bayesian hierarchical model leveraging recent household surveys and building footprints to produce up-to-date population estimates. We estimate population totals and age and sex breakdowns with associated uncertainty measures within grid cells of approximately 100 m in five provinces of the Democratic Republic of the Congo, a country where the last census was completed in 1984. The model exhibits a very good fit, with an R2 value of 0.79 for out-of-sample predictions of population totals at the microcensus-cluster level and 1.00 for age and sex proportions at the province level. This work confirms the benefits of combining household surveys and building footprints for high-resolution population estimation in countries with outdated censuses.},
keywords = {Bayesian inference, Demographic and Health Surveys, Population},
pubstate = {published},
tppubtype = {article}
}
The national census is an essential data source to support decision-making in many areas of public interest. However, this data may become outdated during the intercensal period, which can stretch up to several decades. In this study, we develop a Bayesian hierarchical model leveraging recent household surveys and building footprints to produce up-to-date population estimates. We estimate population totals and age and sex breakdowns with associated uncertainty measures within grid cells of approximately 100 m in five provinces of the Democratic Republic of the Congo, a country where the last census was completed in 1984. The model exhibits a very good fit, with an R2 value of 0.79 for out-of-sample predictions of population totals at the microcensus-cluster level and 1.00 for age and sex proportions at the province level. This work confirms the benefits of combining household surveys and building footprints for high-resolution population estimation in countries with outdated censuses.
11.
Chamberlain, H. R.; Lazar, A. N.; and Tatem, A. J
An index to map feasibility of social distancing within urban areas Conference
29th Annual GIS Research UK Conference (GISRUK), 2021.
Abstract | Links | BibTeX | Tags:
@conference{nokey,
title = {An index to map feasibility of social distancing within urban areas},
author = {Chamberlain, H.R. and Lazar, A.N. and and Tatem, A.J},
doi = {https://doi.org/10.5281/zenodo.4670091},
year = {2021},
date = {2021-05-07},
booktitle = {29th Annual GIS Research UK Conference (GISRUK)},
abstract = {The COVID-19 pandemic has brought factors affecting disease transmission into the spotlight, and required widespread use of public health measures to reduce transmission and contain outbreaks. Urban areas inherently have large concentrations of people, providing high potential for large outbreaks and rapid disease spread, necessitating extensive use of measures to reduce transmission. Social distancing, also called physical distancing, is a public health measure intended to reduce infectious disease transmission, by maintaining physical distance between individuals or households. In the context of the COVID-19 pandemic, populations in many countries around the world have been advised to maintain social distance, with distances of 6ft or 2m commonly advised. The feasibility of social distancing is affected by the availability of space and the number of people, which varies geographically. In locations where social distancing is difficult, a focus on alternative measures to reduce disease transmission should be prioritised. To help identify and map such locations at high spatial resolution, this paper describes an index to quantify ease of social distancing, applied to urban areas across sub-Saharan Africa.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
The COVID-19 pandemic has brought factors affecting disease transmission into the spotlight, and required widespread use of public health measures to reduce transmission and contain outbreaks. Urban areas inherently have large concentrations of people, providing high potential for large outbreaks and rapid disease spread, necessitating extensive use of measures to reduce transmission. Social distancing, also called physical distancing, is a public health measure intended to reduce infectious disease transmission, by maintaining physical distance between individuals or households. In the context of the COVID-19 pandemic, populations in many countries around the world have been advised to maintain social distance, with distances of 6ft or 2m commonly advised. The feasibility of social distancing is affected by the availability of space and the number of people, which varies geographically. In locations where social distancing is difficult, a focus on alternative measures to reduce disease transmission should be prioritised. To help identify and map such locations at high spatial resolution, this paper describes an index to quantify ease of social distancing, applied to urban areas across sub-Saharan Africa.
12.
Lazar, A. N.; Nicholls, R. J.; Hutton, C. W.; Payo, A.; Adams, H.; Haque, A.; Clarke, D.; Salehin, M.; Hunt, A.; Allan, A.; Adger, W. N.; Rahman, M. M.
Potential social-ecological development of coastal Bangladesh through the 21st century Conference
EGU General Assembly 2021, 2021.
Abstract | Links | BibTeX | Tags:
@conference{nokey,
title = {Potential social-ecological development of coastal Bangladesh through the 21st century},
author = {Lazar, A. N. and Nicholls, R. J. and Hutton, C. W. and Payo, A. and Adams, H. and Haque, A. and Clarke, D. and Salehin, M. and Hunt, A. and Allan, A. and Adger, W. N. and Rahman, M. M.},
doi = {https://doi.org/10.5194/egusphere-egu21-1404},
year = {2021},
date = {2021-04-28},
urldate = {2021-04-28},
booktitle = {EGU General Assembly 2021},
abstract = {Deltas occupy only 1% of global land surface area, but contain 7% of the global human population (ca. 500 million). The influence of changing and interacting climates, demography, economy, land use and coastal/catchment management on deltaic social-ecological systems is complex and little understood. We apply a new and innovative integrated assessment model: The Delta Dynamic Integrated Emulator Model (ΔDIEM) to coastal Bangladesh to explore a range of plausible future scenarios and quantify the sensitivities of selected environmental and socio-economic outcomes to key external and internal drivers. ΔDIEM is a tightly coupled integrated assessment platform considering climate and environmental change, demographic changes, economic changes, household decision making and governance, and designed to support the delta planning in Bangladesh. ΔDIEM allows the testing of a large number of water-based structural and policy interventions within a robust scenario framework, as well as quantify different development trajectories and their trade-offs. In this sensitivity analysis, we quantified the impact of (i) climate (precipitation, temperature and runoff), (ii) relative sea-level rise, (iii) cyclone frequency, (iv) embankment maintenance, (v) population size, (vi) economic changes at household level such as selling price of crops, cost of food, etc., (vii) land cover, and (viii) farming practices on trajectories of inundated area, soil salinity, rice productivity, poverty, income inequality and GDP/capita, assuming two contrasting scenarios in a more Positive and a more Negative World. Trajectories of these plausible futures showed a clear separation and the long-term trends are greatly influenced by the combinations of scenario assumptions. Our systemic results indicate a diverse potential set of futures for coastal Bangladesh, where good governance and adaptation could effectively mitigate the threat of sea-level rise-induced catastrophic inundation and other adverse impacts of the changing climate. However, societal inequality requires special attention otherwise climate-sensitive population groups may be left behind.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Deltas occupy only 1% of global land surface area, but contain 7% of the global human population (ca. 500 million). The influence of changing and interacting climates, demography, economy, land use and coastal/catchment management on deltaic social-ecological systems is complex and little understood. We apply a new and innovative integrated assessment model: The Delta Dynamic Integrated Emulator Model (ΔDIEM) to coastal Bangladesh to explore a range of plausible future scenarios and quantify the sensitivities of selected environmental and socio-economic outcomes to key external and internal drivers. ΔDIEM is a tightly coupled integrated assessment platform considering climate and environmental change, demographic changes, economic changes, household decision making and governance, and designed to support the delta planning in Bangladesh. ΔDIEM allows the testing of a large number of water-based structural and policy interventions within a robust scenario framework, as well as quantify different development trajectories and their trade-offs. In this sensitivity analysis, we quantified the impact of (i) climate (precipitation, temperature and runoff), (ii) relative sea-level rise, (iii) cyclone frequency, (iv) embankment maintenance, (v) population size, (vi) economic changes at household level such as selling price of crops, cost of food, etc., (vii) land cover, and (viii) farming practices on trajectories of inundated area, soil salinity, rice productivity, poverty, income inequality and GDP/capita, assuming two contrasting scenarios in a more Positive and a more Negative World. Trajectories of these plausible futures showed a clear separation and the long-term trends are greatly influenced by the combinations of scenario assumptions. Our systemic results indicate a diverse potential set of futures for coastal Bangladesh, where good governance and adaptation could effectively mitigate the threat of sea-level rise-induced catastrophic inundation and other adverse impacts of the changing climate. However, societal inequality requires special attention otherwise climate-sensitive population groups may be left behind.
13.
Lloyd, Christopher T.; Sturrock, Hugh J. W.; Leasure, Douglas R.; Jochem, Warren C.; Lázár, Attila N.; Tatem, Andrew J.
Using GIS and Machine Learning to Classify Residential Status of Urban Buildings in Low and Middle Income Settings Journal Article
In: Remote Sensing, vol. 12, no. 23, 2020, ISSN: 2072-4292.
Abstract | Links | BibTeX | Tags:
@article{rs12233847,
title = {Using GIS and Machine Learning to Classify Residential Status of Urban Buildings in Low and Middle Income Settings},
author = {Christopher T. Lloyd and Hugh J. W. Sturrock and Douglas R. Leasure and Warren C. Jochem and Attila N. Lázár and Andrew J. Tatem},
url = {https://www.mdpi.com/2072-4292/12/23/3847},
doi = {10.3390/rs12233847},
issn = {2072-4292},
year = {2020},
date = {2020-01-01},
journal = {Remote Sensing},
volume = {12},
number = {23},
abstract = {Utilising satellite images for planning and development is becoming a common practice as computational power and machine learning capabilities expand. In this paper, we explore the use of satellite image derived building footprint data to classify the residential status of urban buildings in low and middle income countries. A recently developed ensemble machine learning building classification model is applied for the first time to the Democratic Republic of the Congo, and to Nigeria. The model is informed by building footprint and label data of greater completeness and attribute consistency than have previously been available for these countries. A GIS workflow is described that semiautomates the preparation of data for input to the model. The workflow is designed to be particularly useful to those who apply the model to additional countries and use input data from diverse sources. Results show that the ensemble model correctly classifies between 85% and 93% of structures as residential and nonresidential across both countries. The classification outputs are likely to be valuable in the modelling of human population distributions, as well as in a range of related applications such as urban planning, resource allocation, and service delivery.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Utilising satellite images for planning and development is becoming a common practice as computational power and machine learning capabilities expand. In this paper, we explore the use of satellite image derived building footprint data to classify the residential status of urban buildings in low and middle income countries. A recently developed ensemble machine learning building classification model is applied for the first time to the Democratic Republic of the Congo, and to Nigeria. The model is informed by building footprint and label data of greater completeness and attribute consistency than have previously been available for these countries. A GIS workflow is described that semiautomates the preparation of data for input to the model. The workflow is designed to be particularly useful to those who apply the model to additional countries and use input data from diverse sources. Results show that the ensemble model correctly classifies between 85% and 93% of structures as residential and nonresidential across both countries. The classification outputs are likely to be valuable in the modelling of human population distributions, as well as in a range of related applications such as urban planning, resource allocation, and service delivery.
