Abstract

Understanding the level of poverty in various locations is important for the betterment of livelihoods and helps governments and organizations plan for development. Traditional approaches and data sources, such as DHS, are indeed informative but very often complex, expensive, and limited to smaller geographies. In overcoming such limitations, this research uses satellite imagery combined with deep learning, developing a system to leverage these data sources into better outcomes in the estimation of poverty on a larger scale. The system predicts local poverty by using daytime multispectral images, nighttime lights, and ground truth survey data. The estimated absolute poverty estimation framework includes a two-phase approach: Nighttime images were used in the first phase, through a deep learning model using the SatMAE architecture, which classified rural and urban areas using daytime satellite images. Nighttime images were gathered along with daytime images to avoid temporal data constraints. The nighttime model classifies rural and urban areas. During daytime images, the indicators NDVI, NDBI, and NDWI combined predictive features to estimate a poverty score. The model estimating external uncertainty joins the absolute poverty estimation framework in order to assess the confidence level of any given prediction. This model ensures that it can scale poverty estimation, that it is reliable, and that it will help policymakers in resource allocation in areas identified and believed to be underdeveloped.

References

• M. S. Islam, M. R. Uddin, and A. J. Swajoy, “Shedding Light on Poverty: Leveraging NASA’s Nighttime Data for Socioeconomic Insights in Bangladesh,” IEEE QPAIN, 2025.
• G. Rotich and S. Sarkar, “Evaluating the Robustness of Foundation Models for Satellite Imagery,” IEEE Access, vol. 13, pp. 107720–107735, 2025.
• R. Ghosh et al., “Eyes on Poverty: A Comprehensive Review of Computer Vision and Textual Data Integration for Poverty Prediction in India,” IEEE Access, 2025.
• J. Tang et al., “Poverty Estimation Using a ConvLSTM Based Model With Multisource Remote Sensing Data: A Case Study in Nigeria,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS), vol. 17, 2024.
• N. Banerji et al., “A Machine Learning Approach for Poverty Estimation Using Aerial Images,” Wiley, 2024.
• S. Ballem et al., “Machine Learning Approach with Multiple Open-source Data for Mapping and Prediction of Poverty in Myanmar,” Springer, 2024.
• V. Gupta et al., “Socio-Economic Analysis of India with High Resolution Satellite Imagery to Predict Poverty,” International Journal of Remote Sensing (IJRS), 2024.
• M. Kakooei and A. Daoud, “Increasing the Confidence of Predictive Uncertainty: Earth Observations and Deep Learning for Poverty Estimation,” IEEE Transactions on Geoscience and Remote Sensing (TGRS), vol. 62, 2024.
• A. Arif et al., “Deep Learning Approach using Satellite Imagery Data for Poverty Analysis in Banten, Indonesia,” International Journal of Remote Sensing (IJRS), 2023.
• R. Jarry et al., “Comparing Spatial and Spatio-Temporal Paradigms to Estimate the Evolution of Socio-Economical Indicators from Satellite Images,” IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2023.
• H. Zhang et al., “Satellite Imagery-based Prediction of Poverty using Faster R-CNN,” IEEE Geoscience and Remote Sensing Society (GRSS), 2023.
• Y. Ni et al., “An Investigation on Deep Learning Approaches to Combining Nighttime and Daytime Satellite Imagery for Poverty Prediction,” IEEE Geoscience and Remote Sensing Letters (GRSL), vol. 18, no. 9, 2021.
• B. Yu et al., “Poverty Evaluation Using NPP-VIIRS Nighttime Light Composite Data at the County Level in China,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS), vol. 8, no. 3, 2015.
• M. Oh et al., “Estimation of the Potential of Small scale Photovoltaic Systems in Underdeveloped Area of North Korea for Humanitarian Purposes,” IEEE Global Humanitarian Technology Conference (GHTC), 2019.
• P. Wu and Y. Tan, “Estimation of Economic Indicators using Residual Neural Network (ResNet50),” IEEE International Conference on Data Mining Workshops (ICDMW), 2019