KidSat: satellite imagery to map childhood poverty

Makkunda Sharma, Fan Yang, Duy-Nhat Vo, Jack Gidney, Matthew Sutcliffe, Mengyan Zhang, Esra Suel, H Juliette T Unwin, Swapnil Mishra, Samir Bhatt, Oliver Fiala, William Rudgard, Seth Flaxman

Keywords: satellite imagery, remote sensing, social science, global health, economic, health and development indicators

TL;DR: We present a new dataset and benchmark consisting of satellite images and corresponding child poverty indicators in Eastern and Southern Africa

Abstract: Satellite imagery has emerged as an important tool to analyze demographic, health, and development indicators. While various deep learning models have been built for these tasks, each is specific to a particular problem, with few standard benchmarks available. We propose a new dataset pairing satellite imagery and high-quality survey data on child poverty to benchmark satellite feature representations. Our dataset consists of 33,608 images, each 10 km × 10 km, from 16 countries in Eastern and Southern Africa in the time period 1997-2022. As defined by UNICEF, multidimensional child poverty comprises six fundamental factors—housing, sanitation, water, nutrition, education, and health (UNICEF, 2021)—which can be calculated from geocoded, face-to-face Demographic and Health Surveys (DHS) Program data. Using our dataset we benchmark multiple feature representations for encoding satellite imagery, from low-level satellite imagery models such as MOSAIKS (Rolf et al., 2021), to deep learning foundation models, which include both generic vision models such as DINOv2 (Oquab et al., 2023) and specific satellite imagery models such as SatMAE (Cong et al., 2022). As part of the benchmark, we test spatial as well as temporal generalization, by testing on unseen locations, and on data beyond the training years. We provide open source code to reproduce and extend our entire pipeline: building the satellite imagery dataset, obtaining ground truth data from DHS, and comparing the various models considered in our work.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6598