Spatial heterogeneity in machine learning-based poverty mapping: Where do models underperform?

Abstract

Accurately locating poor populations is increasingly urgent as global poverty reduction has stalled under the combined pressures of conflicts, climate shocks, rising food prices, pandemics, and growing inequality. Recent studies harnessing geospatial big data and machine learning (ML) have significantly advanced poverty mapping, enabling granular and timely welfare estimates in traditionally data-scarce regions. While much of the existing research has focused on overall out-of-sample predictive performance, there is a lack of understanding regarding where such models underperform and whether key spatial relationships might vary across places. This study investigates spatial heterogeneity in ML-based poverty mapping in East Africa, testing whether spatial regression and ML techniques produce more unbiased predictions. We find that extrapolation into unsurveyed areas suffers from biases that spatial methods do not resolve; welfare is overestimated in impoverished regions, rural areas, and single sector-focused economies, whereas it tends to be underestimated in wealthier, urbanized, and diversified economies. Even as spatial models improve overall predictive accuracy, enhancements in traditionally underperforming areas remain marginal. This underscores the need for more representative training datasets and better remotely sensed proxies, especially for poor and rural regions, in future research related to ML-based poverty mapping. For development agencies, the findings caution against treating ML-based outputs as neutral or universally reliable, highlighting instead the need to pair technical advances with investments in inclusive data collection, integration of spatial theory, and institutional strategies that address structural data inequalities.

Keywords

Poverty mapping / Machine learning / Spatial models / East Africa

Ethical statement

Ethical approval was not required for this study, as it relies exclusively on secondary data and does not involve human or animal subjects.

CRediT authorship contribution statement

Yating Ru: Writing - review & editing, Writing - original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Elizabeth Tennant: Writing - review & editing, Supervision, Software, Methodology, Investigation, Data curation, Conceptualization. David S. Matteson: Writing - review & editing, Supervision, Methodology, Conceptualization. Christopher B. Barrett: Writing - review & editing, Supervision, Resources, Project administration, Funding acquisition, Conceptualization.

Declaration of competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

This work was supported by the Cornell Atkinson Center for Sustainability. We thank Cassian D’Cunha and the Cornell Center for Social Sciences for computational resources and support. We also thank Takaaki Masaki and Arturo Jr. M. Martinez for their insightful comments and suggestions. Finally, we appreciate the constructive feedback from the editor and anonymous reviewers, which strengthened the paper.

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.geosus.2026.100413.

References

[1] Aiken E., Suzanne B., Dean K., Christopher U., Joshua B., 2022. Machine learning and mobile phone data can improve the targeting of humanitarian assistance. Nature 603, 864-870. doi: 10.1038/s41586-022-04484-9.

[2] Amara M., Mohamed A., 2013. The local geographies of welfare in Tunisia: does neighbourhood matter? Int. J. Soc. Welf. 22 (1), 90-103. doi: 10.1111/j.1468-2397.2011.00863.x.

[3] Ayush K., Burak U., Marshall B., David L., Stefano E., 2020. Generating interpretable poverty maps using object detection in satellite images. In:IJCAI’20: Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence, pp. 4410-4416. Article No. 608.

[4] Barrett C.B., Carter M.R., Little P.D., 2006. Understanding and reducing persistent poverty in Africa: introduction to a special issue. J. Dev. Stud. 42 (2), 167-177. doi: 10.1080/00220380500404587.

[5] Benson T., Chamberlin J., Rhinehart I., 2005. An investigation of the spatial determinants of the local prevalence of poverty in rural Malawi. Food Policy 30 (5-6), 532-550. doi: 10.1016/j.foodpol.2005.09.004.

[6] Bird K., Higgins K., Dan Harris D., 2010. Spatial Poverty traps: an overview. ODI Working Paper 323. Overseas Development Institute, London.

[7] Blumenstock J., Cadamuro G., On R., 2015. Predicting poverty and wealth from mobile phone metadata. Science 350 (6264), 1073-1076. doi: 10.1126/science.aac4420.

[8] Breiman L., 2001. Random forests. Mach. Learn. 45 (1), 5-32. doi: 10.1023/A:1010933404324.

[9] Browne C., Matteson D.S., McBride L., Hu L.Q., Liu Y.Y., Sun Y., Wen J.M., Barrett C.B., 2021. Multivariate random forest prediction of poverty and malnutrition prevalence. PLoS One 16 (9), e0255519. doi: 10.1371/journal.pone.0255519.

[10] Burke M., Driscoll A., Lobell D.B., Ermon S., 2021. Using satellite imagery to understand and promote sustainable development. Science 371 (6535), eabe8628. doi: 10.1126/science.abe8628.

[11] Carter M., Barrett C.B., 2006. The economics of poverty traps and persistent poverty: an asset-based approach. J. Dev. Stud. 42 (2), 178-199. doi: 10.1080/00220380500405261.

[12] Chi G.H., Fang H., Chatterjee S., Blumenstock J.E., 2022. Microestimates of wealth for all low- and middle-income countries. Proc. Natl. Acad. Sci. U.S.A. 119 (3), 2113658119. doi: 10.1073/pnas.2113658119.

[13] Christiaensen L., Lanjouw P., Luoto J., Stifel D., 2012. Small area estimation-based prediction methods to track poverty: validation and applications. J. Econ. Inequal. 10, 267-297. doi: 10.1007/s10888-011-9209-9.

[14] Corral P., Henderson H., Segovia S., 2025. Poverty mapping in the age of machine learning. J. Dev. Econ. 172, 103377. doi: 10.1016/j.jdeveco.2024.103377.

[15] Croft, Trevor, N., Marshall, A. M.J., Allen, C.K., et al., 2018. Guide to DHS Statistics. ICF Macro, Rockville, Maryland, USA.

[16] Elbers C., Lanjouw J., Lanjouw P., 2003. Micro-level estimation of poverty and inequality. Econometrica 71 (1), 355-364. http://www.jstor.org/stable/3082050.

[17] Engstrom R., Hersh J., Newhouse D., 2022. Poverty from space: using high resolution satellite imagery for estimating economic well-being. World Bank Econ. Rev. 36 (2), 382-412. doi: 10.1596/1813-9450-8284.

[18] Farr T.G., Rosen P.A., Caro E., Crippen R., Duren R., Hensley S., Kobrick M., Paller M., Rodriguez E., Roth L., Seal D., Shaffer S., Shimada J., Umland J., Werner M., Oskin M., Burbank D., Alsdorf D., 2007. The shuttle radar topography mission. Rev. Geophys. 45 (2), 65-77. doi: 10.1029/2005RG000183.

[19] Fay R.E., Herriot R.A., 1979. Estimates of income for small places: an application of James-Stein procedures to census data. J. Am. Stat. Assoc. 74 (366a), 269-277. doi: 10.1080/01621459.1979.10482505.

[20] Filmer D., Pritchett L.H., 2001. Estimating wealth effects without expenditure data-or tears : an application to educational enrollments in states of India. Demography 38 (1), 115-132. doi: 10.1353/dem.2001.0003.

[21] Fischer G., Nachtergaele F., van Velthuizen H., Chiozza F., Franceschini G., Henry M., Muchoney D., Tramberend S., 2021. Global Agro-Ecological Zones v4 —model Documentation. Food and Agriculture Organization of the United Nations, Rome [dataset].

[22] Florczyk A.J., Corbane C., Ehrlich D., Freire S., Kemper T., Maffenini L., Melchiorri M., Pesaresi M., Politis P., Schiavina M., Sabo F., Zanchetta L., 2019. GHSL data package 2019: public release GHS P2019. Vol. JRC117104. Publications Office of the European Union, Luxembourg.

[23] Fotheringham A.S., Charlton M., Brunsdon C., 1996. The geography of parameter space: an investigation of spatial non-stationarity. Int. J. Geogr. Inf. Syst. 10 (5), 605-627. doi: 10.1080/02693799608902100.

[24] Fotheringham A.S., Brunsdon C., Charlton M., 2002. Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. John Wiley & Sons, Ltd., Chichester.

[25] Franch B., Vermote E., Skakun S., Roger J.-C., Masek J., Ju J.C., Villaescusa-Nadal J.L., Santamaría-Artigas A., 2019. A method for landsat and Sentinel 2 (HLS) BRDF normalization. Remote Sens. 11 (6), 632. doi: 10.3390/rs11060632.

[26] Funk C., Peterson P., Landsfeld M., Pedreros D., Verdin J., Shukla S., Husak G., Rowland J., Harrison L., Hoell A., Michaelsen J., 2015. The climate hazards infrared precipitation with stations —a new environmental record for monitoring extremes. Sci. Data 2, 150066. doi: 10.1038/sdata.2015.66.

[27] Gelaro R., McCarty W., Suárez M.J., Todling R., Molod A., Takacs L., Randles C.A., Darmenov A., Bosilovich M.G., Reichle R., Wargan K., Coy L., Cullather R., Draper C., Akella S., Buchard V., Conaty A., da Silva A., Gu W., Kim G.K., Koster R., Lucchesi R., Merkova D., Nielsen J.E., Partyka G., Pawson S., Putman W., Rienecker M., Schubert S.D., Sienkiewicz M., Zhao B., 2017. The modernera retrospective analysis for research and applications, version 2 (MERRA-2). J. Clim. 30 (14), 5419-5454. doi: 10.1175/JCLI-D-16-0758.1.

[28] Georganos S., Grippa T., Gadiaga A.N., Linard C., Lennert M., Vanhuysse S., Mboga N., Wolff E., Kalogirou S., 2021. Geographical random forests: a spatial extension of the random forest algorithm to address spatial heterogeneity in remote sensing and population modelling. Geo. Int. 36 (2), 121-136. doi: 10.1080/10106049.2019.1595177.

[29] Georganos S., Kalogirou S., 2022. A forest of forests: a spatially weighted and computationally efficient formulation of geographical random forests. ISPRS Int. J. Geo-Inf. 11 (9), 471. doi: 10.3390/ijgi11090471.

[30] Ghosh M., Rao J.N.K., 1994. Small area estimation: an appraisal. Stat. Sci. 9 (1), 55-76. http://www.jstor.org/stable/2246284.

[31] Gollini I., Lu B.B., Charlton M., Brunsdon C., Harris P., 2015. GWmodel : an R package for exploring spatial heterogeneity. J. Stat. Softw. 63 (17), 1-50. doi: 10.18637/jss.v063.i17.

[32] Gruber M., 1998. Improving Efficiency by Shrinkage: The James-Stein and Ridge Regression Estimators. CRC Press, Boca Raton doi: 10.1201/9780203751220.

[33] Gwatkin D., Rutstein S., Johnson K., Suliman E., Wagstaff A., Amouzou A., 2007. Socio-economic differences in health, nutrition, and population within developing countries: an overview. Niger. J. Clin. Pract. 10 (4), 272-282. https://pubmed.ncbi.nlm.nih.gov/18293634/.

[34] Hancock J.T., Khoshgoftaar T.M., 2020. Survey on categorical data for neural networks. J. Big Data 7 (1), 28. doi: 10.1186/s40537-020-00305-w.

[35] Head A., Manguin M., Tran N., Blumenstock J.E., 2017. Can human development be measured with satellite imagery? In: Proceedings of the Ninth International Conference on Information and Communication Technologies and Development (ICTD’17). New York, NY, USA. Association for Computing Machinery doi: 10.1145/3136560.313657, Article..

[36] Hoerl A.E., Kennard R.W., 1970. Ridge regression: biased estimation for nonorthogonal problems. Technometrics 12 (1), 55-67. doi: 10.1080/00401706.1970.10488634.

[37] Huang Q.X., Yang X., Gao B., Yang Y., Zhao Y.Y., 2014. Application of DMSP/OLS nighttime light images: a meta-analysis and a systematic literature review. Remote Sens. 6 (8), 6844-6866. doi: 10.3390/rs6086844.

[38] Jalan J., Ravallion M., 2002. Geographic poverty traps? J. Appl. Econ. 17 (4), 329-346. doi: 10.1002/jae.645.

[39] Jean N., Burke M., Xie M., Alampay Davis W.M., Lobell D.B., Ermon S., 2016. Combining satellite imagery and machine learning to predict poverty. Science 353 (6301), 790-794. doi: 10.1126/science.aaf7894.

[40] Kanbur R., Venables A.J., 2005. Spatial Inequality and Development. Oxford University Press, Oxford.

[41] Kennedy P., 2003. A Guide to Econometrics. The MIT Press, Cambridge.

[42] Khan S.N., Li D.P., Maimaitijiang M., 2022. A geographically weighted random forest approach to predict corn yield in the US corn belt. Remote Sens. 14 (12), 2843. doi: 10.3390/rs14122843.

[43] Kopczewska K., 2022. Spatial machine learning: new opportunities for regional science. Ann. Reg. Sci. 68, 713-755. doi: 10.1007/s00168-021-01101-x.

[44] Kummu M., Taka M., Guillaume J.H.A., 2018. Gridded global datasets for gross domestic product and human development index over 1990-2015. Sci. Data 5, 180004. doi: 10.1038/sdata.2018.4.

[45] Lakner C., Mahler D.G., Negre M., Espen Beer Prydz E.B., 2022. How much does reducing inequality matter for global poverty? J. Econ. Inequal. 20 (3), 559-585. doi: 10.1007/s10888-021-09510-w.

[46] Lee K., Braithwaite J., 2022. High-resolution poverty maps in sub-Saharan Africa. World Dev. 159, 106028. doi: 10.1016/J.WORLDDEV.2022.106028.

[47] Li X.C., Zhou Y.Y., Zhao M., Zhao X., 2020. A harmonized global nighttime light dataset 1992-2018. Sci. Data 7 (1), 168. doi: 10.1038/s41597-020-0510-y.

[48] Lloyd C., Sorichetta A., Tatem A., 2017. High resolution global gridded data for use in population studies. Sci. Data 4, 170001. doi: 10.1038/sdata.2017.1.

[49] Lu B.B., Charlton M., Harris P., Fotheringham A.S., 2014. Geographically weighted regression with a non-euclidean distance metric: a case study using hedonic house price data. Int. J. Geogr. Inf. Sci. 28 (4), 660-681. doi: 10.1080/13658816.2013.865739.

[50] Luo Y.W., Yan J.G., McClure S., Li F., 2022. Socioeconomic and environmental factors of poverty in China using geographically weighted random forest regression model. Environ. Sci. Pollut. Res. 29 (22), 33205-33217. doi: 10.1007/s11356-021-17513-3.

[51] Masaki T., Newhouse D., Silwal A.R., Bedada A., Engstrom R., 2020. Small area estimation of non-monetary poverty with geospatial data. Policy Research Working Paper 9383. World Bank, Washington, D.C.

[52] Michler J.D., Josephson A., Kilic T., Murray S., 2022. Privacy protection, measurement error, and the integration of remote sensing and socioeconomic survey data. J. Dev. Econ. 158, 102927. doi: 10.1016/j.jdeveco.2022.102927.

[53] Nikparvar B., Thill J.-C., 2021. Machine learning of spatial data. ISPRS Int. J. Geo-Inf. 10 (9), 600. doi: 10.3390/ijgi10090600.

[54] Pokhriyal N., Jacques D.C., 2017. Combining disparate data fources for improved poverty prediction and mapping. Proc. Natl. Acad. Sci. U.S.A. 114 (46), E9783-E9792. doi: 10.1073/pnas.1700319114.

[55] Rao C.R., 1965. Linear Statistical Inference and Its Applications. John Wiley & Sons Inc., New York.

[56] Rao J.N.K., 1999. Some recent advances in model-based small area estimation. Surv. Methodol. 25 (2), 175-186.

[57] Ratledge N., Cadamuro G., de la Cuesta B., Stigler M., Burke M., 2022. Using machine learning to assess the livelihood impact of electricity access. Nature 611, 491-495. doi: 10.1038/s41586-022-05322-8.

[58] Roe D., 2008. The origins and evolution of the conservation-poverty debate: a review of key literature, events and policy processes. Oryx 42 (4), 491-503. doi: 10.1017/S0030605308002032.

[59] Ru Y.T., Blankespoor B., Wood-Sichra U., Thomas T.S., You L.Z., Kalvelagen E., 2023. Estimating local agricultural gross domestic product (AgGDP) across the world. Earth Syst. Sci. Data 15 (3), 1357-1387. doi: 10.5194/essd-15-1357-2023.

[60] Rutstein S.O., 2008. The DHS wealth index: approaches for rural and urban areas. DHS Working Papers. Macro International, Calverton, Maryland, USA.

[61] Rutstein S.O., Johnson K., 2004. The DHS wealth index. DHS Comparative Report 6. ORC Macro, Calverton, MD doi: 10.13140/2.1.2806.4809.

[62] Sirko W., Kashubin S., Ritter M., Annkah A., Bouchareb Y.S.E., Dauphin Y., Keysers D., Neumann M., Cisse M., Quinn J., 2021. Continental-scale building detection from high resolution satellite imagery. http://arxiv.org/abs/2107.12283

[63] Smits J., Steendijk R., 2015. The international wealth index (IWI). Soc. Indic. Res. 122 (1), 65-85. doi: 10.1007/s11205-014-0683-x.

[64] Tang B., Liu Y.Y., Matteson D.S., 2022. Predicting poverty with vegetation index. Appl. Econ. Perspect. Policy 44 (2), 930-945. doi: 10.1002/aepp.13221.

[65] Tarozzi A., 2011. Can census data alone signal heterogeneity in the estimation of poverty maps? J. Dev. Econ. 95 (2), 170-185. doi: 10.1016/j.jdeveco.2010.05.003.

[66] Tarozzi A., Deaton A., 2009. Using census and survey data to estimate poverty and inequality for small areas. Rev. Econ. Stat. 91 (4), 773-792. doi: 10.2139/ssrn.997829.

[67] Tennant E., Ru Y.T., Sheng P.Z., Matteson D.S., 2025. Microlevel structural poverty estimates for Southern and Eastern Africa. Proc. Natl. Acad. Sci. U.S.A. 122 (6), e2410350122. doi: 10.1073/pnas.2410350122.

[68] Tingzon I., Orden A., Sy S., Sekara V., Weber I., Fatehkia M., García-Herranz M., Kim D., 2019. Mapping poverty in the Philippines using machine learning, satellite imagery, and crowd-sourced geospatial information. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 42 (4/W19), 425-431. doi: 10.5194/isprs-archives-XLII-4-W19-425-2019.

[69] United Nations General Assembly. 2015. Transforming our world: the 2030 agenda for sustainable development. A/RES/70/1. Resolution adopted by the General Assembly on 25 September. https://sdgs.un.org/2030agenda

[70] Vermote E. NOAA CDR Program, 2019. NOAA Climate Data Record (CDR) of AVHRR Normalized Difference Vegetation Index (NDVI), Version 5. NOAA National Centers for Environmental Information [Dataset].

[71] Wang Y.G., Lin Z., Wu C.Y., 2022. Comparison of the spatial and temporal variability of cloud amounts over China derived from different satellite datasets. Remote Sens. 14 (9), 2173. doi: 10.3390/rs14092173.

[72] Weiss D.J., Nelson A., Gibson H.S., Temperley W., Peedell S., Lieber A., Hancher M., Poyart E., Belchior S., Fullman N., Mappin B., Dalrymple U., Rozier J., Lucas T.C.D., Howes R.E., Tusting L.S., Kang S.Y., Cameron E., Bisanzio D., Battle K.E., Bhatt S., Gething P.W., 2018. A global map of travel time to cities to assess inequalities in accessibility in 2015. Nature 553 (7688), 333-336. doi: 10.1038/nature25181.

[73] Winn R.A., Tossas K.Y., Doubeni C.A., 2023. Commentary: some water in the data desert: the cancer intervention and surveillance modeling network’s capacity to guide mitigation of cancer health disparities. J. Natl. Cancer Inst. Monogr. 2023 (62), 167-172. doi: 10.1093/jncimonographs/lgad032.

[74] World Bank, 2022. Poverty and shared prosperity 2022: correcting course. World Bank, Washington, D.C doi: 10.1596/978-1-4648-1893-6.

[75] Xie M., Jean N., Burke M., Lobell D., Ermon S., 2016. Transfer learning from deep features for remote sensing and poverty mapping. In:Proceedings of the 30th AAAI Conference on Artificial Intelligence ( AAAI-16), pp. 3929-3935. doi: 10.48550/arXiv.1510.00098.

[76] Yeh C., Perez A., Driscoll A., Azzari G., Tang Z., Lobell D., Ermon S., Burke M., 2020. Using publicly available satellite imagery and deep learning to understand economic well-being in Africa. Nat. Commun. 11 (1), 2583. doi: 10.1038/s41467-020-16185-w.

[77] Zhao X.Z., Yu B.L., Liu Y., Chen Z.Q., Li Q.X., Wang C.X., Wu J.P., 2019. Estimation of poverty using random forest regression with multi-source data: a case study in Bangladesh. Remote Sens. 11 (4), 375. doi: 10.3390/rs11040375.

[78] Zhou Y., Liu Y.S., 2022. The geography of poverty: review and research prospects. J. Rural Stud. 93, 408-416. doi: 10.1016/j.jrurstud.2019.01.008.