The AI & Global Development Lab’s research portfolio contains studies that apply cutting‑edge artificial intelligence, machine learning, and Earth observation to pressing social and economic problems. Recent papers tackle tasks such as debiasing machine‑learning predictions for causal inference and poverty analysis using satellite imagery, optimizing multi‑scale representations to detect effect heterogeneity, and providing a scoping review of how Earth‑observation data and machine‑learning tools can support causal inference. Earlier work explores image‑based approaches to estimate treatment‑effect heterogeneity, examines IMF policies’ distributional effects, assesses post‑disaster building damage through remote sensing, and investigates the long‑term health impacts of disaster‑related home loss. Together, these interdisciplinary studies demonstrate the lab’s commitment to harnessing AI, causal inference, and remote‑sensing methods to deepen our understanding of development challenges and inform policy optimization for a better tomorrow. See also [Bibliography].
-2026-
- Adel Daoud, Connor T. Jerzak.
Planetary Causal Inference: Understanding Society, Economy, and Environment through Satellite Images.
Under contract with Cambridge University Press, 2026+.
[Overview] [First Access Notification] - Adel Daoud, Cindy Conlin, Connor T. Jerzak.
Chinese vs. World Bank Development Projects: Insights from Earth Observation and Computer Vision on Wealth Gains in Africa, 2002-2013.World Development, 202: 107328.
[PDF] [Overview] [.bib] [>] - Satiyabooshan Murugaboopathy, Connor T. Jerzak, Adel Daoud.
Platonic Representations for Poverty Mapping: Unified Vision-Language Codes or Agent-Induced Novelty?
Proceedings of the 7th International Conference on Social Computing (ICSC 2026), of Communications in Computer and Information Science, Springer Nature.
[PDF] [Overview] [Data] [.bib] - Markus Pettersson, Connor T. Jerzak, Adel Daoud.
Debiasing Machine Learning Predictions for Causal Inference Without Additional Ground Truth Data: ‘One Map, Many Trials’ in Satellite-Driven Poverty Analysis. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI-26), 40(46): 39106-39115.
[PDF] [Overview] [.bib] [>] - Kazuki Sakamoto, Connor T. Jerzak, Adel Daoud.
A Scoping Review of Earth Observation and Machine Learning for Causal Inference: Implications for the Geography of Poverty.
Hall, Ola and Ibrahim Wahab (eds.), Geography of Poverty, Edward Elgar Publishing.
[PDF] [Overview] [.bib] [>] - Mohammad Kakooei, James Bailie, Markus B. Pettersson, Albin Söderberg, Albin Becevic, Adel Daoud.
A High-Resolution Urban and Rural Settlement Map of Africa Using Deep Learning and Satellite Imagery.
Scientific Reports.
[Article] [DOI]
-2025-
- Markus B. Pettersson, Adel Daoud.
Leveraging Compact Satellite Embeddings and Graph Neural Networks for Large-Scale Poverty Mapping.
Workshop on Advances in Representation Learning for Earth Observation (REO at EurIPS)
[PDF] - Warren Zhu Fucheng, Adel Daoud, Connor T. Jerzak.
Optimizing Multi-Scale Representations to Detect Effect Heterogeneity Using Earth Observation and Computer Vision: Applications to Two Anti-Poverty RCTs. Proceedings of the Fourth Conference on Causal Learning and Reasoning (CLeaR), Proceedings of Machine Learning Research (PMLR), 275: 894-919.
[PDF] [Overview] [.bib] [>] - Francesco Grieco, Adel Daoud, Mohammad Kakooei.
Linking Socioeconomic Status and Emissions: The Predictive Power of the International Wealth Index for NO₂ column densities.
ResearchSquare Preprint.
[PDF] [DOI] - Cindy Conlin (Master’s Student of Computational Social Science, Linköping University) graduates!
Thesis: Using Machine Learning and Daytime Satellite Imagery to Estimate Aid’s Effect on Wealth: Comparing China and World Bank Programs in Africa.
[PDF] [Code] [.bib] [>] - Mikael P. Gustafsson (Master’s Student of Computational Social Science, Linköping University) graduates!
Thesis: Estimating Aid Effectiveness in Fragile and Conflict-affected States: Evidence from Satellite-based Inference in Somalia.
[PDF] [.bib] [>] - Mohammad Kakooei, Adel Daoud.
Increasing the Confidence of Predictive Uncertainty: Earth Observations and Deep Learning for Poverty Estimation.
IEEE Transactions on Geoscience and Remote Sensing.
[IEEE Xplore] [DOI]
-2023-
- Connor T. Jerzak, Fredrik Johansson, Adel Daoud.
Image-based Treatment Effect Heterogeneity.
Proceedings of the Second Conference on Causal Learning and Reasoning (CLeaR), Proceedings of Machine Learning Research (PMLR), 2023.
[PDF] [Overview] [.bib] [>] - Markus Pettersson, Mohammad Kakooei, Julia Ortheden, Fredrik Johansson, and Adel Daoud
Time Series of Satellite Imagery Improve Deep Learning Estimates of Neighborhood-Level Poverty in Africa.
Proceedings of the International Joint Conferences on Artificial Intelligence (JIJCAI), 2023
[PDF] [.bib] - Connor T. Jerzak, Fredrik Johansson, Adel Daoud.
Integrating Earth Observation Data into Causal Inference: Challenges and Opportunities.
arXiv:2301.12985
[PDF] [.bib]
-2022-
- Adel Daoud, Anders Herlitz, S.V. Subramanian.
IMF fairness: Calibrating the policies of the International Monetary Fund based on distributive justice.
World Development, Volume 157, 2022, 105924, ISSN 0305-750X,
https://doi.org/10.1016/j.worlddev.2022.105924 - Daoud, Adel, Connor T. Jerzak, and Richard Johansson.
Conceptualizing Treatment Leakage in Text-based Causal Inference.
North American Chapter of the Association for Computational Linguistics (NAACL).
https://aclanthology.org/2022.naacl-main.413/ [.bib] [>] - Balgi, S., Peña, J. M., & Daoud, A. (2022).
Personalized Public Policy Analysis in Social Sciences Using Causal-Graphical Normalizing Flows.
Proceedings of the AAAI Conference on Artificial Intelligence, 36(11), 11810-11818.
https://doi.org/10.1609/aaai.v36i11.21437 - Kakooei, Mohammad, Arsalan Ghorbanian, Yasser Baleghi, Meisam Amani, and Andrea Nascetti.
Remote Sensing Technology for Post-Disaster Building Damage Assessment.
In Computers in Earth and Environmental Sciences, Elsevier. - Koichiro Shiba, Hiroyuki Hikichi, Sakurako S. Okuzono, Tyler J. VanderWeele, Mariana Arcaya, Adel Daoud, Richard G. Cowden, Aki Yazawa, David T. Zhu, Jun Aida, Katsunori Kondo, and Ichiro Kawachi. 2022.
Long-Term Associations between Disaster-Related Home Loss and Health and Well-Being of Older Survivors: Nine Years after the 2011 Great East Japan Earthquake and Tsunami.
Environmental Health Perspectives 130:7 CID: 077001
https://doi.org/10.1289/EHP10903 - Connor T. Jerzak, Fredrik Johansson, Adel Daoud. 2022.
Estimating Causal Effects Under Image Confounding Bias with an Application to Poverty in Africa.
arXiv:2206.06410 cs.LG.
https://doi.org/10.48550/arXiv.2206.06410 - Balgi, Sourabh, Adel Daoud, and Jose Pena.
Personalized Public Policy Analysis In Social Sciences Using Causal-Graphical Normalizing Flows.
AI for Social Impact Track in Thirty-Sixth AAAI Conference on Artificial Intelligence (Association for the Advancement of Artificial Intelligence).
-2021-
- Koichiro Shiba, Adel Daoud, Hiroyuki Hikichi, Aki Yazawa, Jun Aida, Katsunori Kondo, Ichiro Kawachi.
Heterogeneity in cognitive decline after a major disaster: a natural experiment study.
Science Advances.
DOI: 10.1126/sciadv.abj2610 - Kino, Shiho, Yu-Tien, Koichiro Shiba, Ichiro Kawachi, and Adel Daoud.
A scoping review on the use of machine learning in the research on the social determinants of health: trends and research prospects.
Social Science & Medicine – Population Health.
https://doi.org/10.1016/j.ssmph.2021.100836 - Daoud, Adel, and Devdatt Dubhashi.
“Melting together prediction and inference”, Observational Studies, Volume 7, issue 1, Commentary. - Koichiro Shiba, Jacqueline M. Torres, Adel Daoud, Kosuke Inoue, Satoru Kanamori, Taishi Tsuji, Masamitsu Kamada, Katsunori Kondo, and Ichiro Kawachi.
Estimating the impact of sustained social participation on depressive symptoms in older adults, Epidemiology
DOI: 10.1097/EDE.0000000000001395 - Kakooei, Mohammad, Yasser Baleghi, and Meisam Amani.
Adaptive thresholding for detecting building facades with or without openings in single-view oblique remote sensing images. Journal of Applied Remote Sensing, 15, no. 3, 036511. - Amani, Meisam, Sahel Mahdavi, Mohammad Kakooei, Arsalan Ghorbanian, Brian Brisco, Evan Delancey, Souleymane Toure, and Eugenio Landeiro Reyes.
Wetland Change Analysis in Alberta, Canada using Four Decades of Landsat Imagery.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.
-2020-
- Kakooei, Mohammad, and Yasser Baleghi.
A Two-level Fusion for Building Irregularity Detection in Post-Disaster VHR Oblique Images. Earth Science Informatics. Abstract. - Kakooei, Mohammad, and Yasser Baleghi.
Shadow detection in very high resolution RGB images using a special thresholding on a new spectral–spatial index.
Journal of Applied Remote Sensing 14, no. 1. - Kakooei, Mohammad, and Yasser Baleghi.
VHR Semantic Labelling by Random Forest Classification and Fusion of Spectral and Spatial Features on Google Earth Engine.
Journal of AI and Data Mining. - Amani, Meisam, Mohammad Kakooei, Armin Moghimi, Arsalan Ghorbanian, Babak Ranjgar, Sahel Mahdavi, Andrew Davidson, Thierry Fisette, Patrick Rollin, Brian Brisco.
Application of Google Earth Engine Cloud Computing Platform, Sentinel Imagery, and Neural Networks for Crop Mapping in Canada.
Remote Sensing. - Amani, Meisam, Brian Brisco, Sahel Mahdavi, Arsalan Ghorbanian, Armin Moghimi, Evan DeLancey, Michael Merchant, Raymond Jahncke, Lee Fedorchuk, Amy Mui, Marcelle Grenier, Thierry Fisette, Mohammad Kakooei, Seyed Ali Ahmadi, Brigitte Leblon, Amir Behnamian.
Evaluation of the First Canadian Wetland Inventory Map using Multiple Sources: Challenges of Wetland Classification using Remote Sensing.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. - Amani, Meisam, Arsalan Ghorbanian, Ali Ahmadi, Mohammad Kakooei, Armin Moghimi, S. Mohammad Mirmazloumi, Sayyed Hamed Alizadeh Moghaddam, Sahel Mahdavi, Masoud Ghahramanloo, Saeid Parsian, Qiusheng Wu, Brian Brisco.
Google Earth Engine Cloud Computing Platform for Remote Sensing Big Data Applications: A Literature Review
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. - Ghorbanian, Arsalan, Mohammad kakooei, Meisam Amani, Sahel Mahdavi, Ali Mohammadzadeh, Mahdi Hasanlou.
Improved land cover map of Iran using Sentinel imagery within Google Earth Engine and a novel automatic workflow for land cover classification using migrated training samples.
ISPRS Journal of Photogrammetry and Remote Sensing.
-2019-
- Daoud, Adel, Rockli Kim, and S V Subramanian.
Evaluating the predictive power of socioeconomic factors in capturing women’s height in 58 low- and middle-income countries: a machine learning approach, Social Science & Medicine, 238 (2019) 112486. - Daoud, Adel and Nandy, Shailen.
Implications of the politics of caste and class on child poverty in India, Sociology of Development, Vol. 5, Number 4, pp 428-451,
doi.org/10.1525/sod.2019.5.4.428 - Daoud, Adel, Bernhard Reinsberg, Alexander Kentikelenis, Thomas Stubbs, Lawrence King.
The International Monetary Fund’s Interventions in Food and Agriculture: An Analysis of Loans and Conditions. Food Policy, Vol 83, pp 204-218.
https://doi.org/10.1016/j.foodpol.2019.01.005 - Kraamwinkel, N., Ekbrand, H., Davia, S., Daoud, A.
The influence of maternal agency on child well-being in conflict-ridden Nigeria: modelling heterogeneous treatment effects with machine learning. PLoS ONE: http://doi.org/10.1371/journal.pone.0208937 (Special issue on machine learning in health and biomedicine). - Kakooei, Mohammad, and Amir Tabatabaei.
A Fast Parallel GPS Acquisition Algorithm Based on Hybrid GPU and Multi-core CPU. Wireless Personal Communications 104, no. 4, pp. 1355-1366. Abstract. - Joakim Åkerström, Adel Daoud, and Richard Johansson.
Natural Language Processing in Policy Evaluation: Extracting Policy Conditions from IMF Loan Agreements, Published in the 22nd Nordic Conference on Computational Linguistics (NoDaLiDa’19). - Kakooei, Mohammad, and Yasser Baleghi.
Spectral Unmixing of Time Series Data to Provide Initial Object Seeds for Change Detection on Google Earth Engine.
In 2019 27th Iranian Conference on Electrical Engineering (ICEE), pp. 1402-1407. IEEE. Abstract. - Ban, Yifang, Andrea Nascetti, and Mohammad Kakooei.
Sentinel-1 SAR and Sentinel-2 MSI Dense Time Series for Urban Extraction in Support of Urban Sustainable Development Goal, Dragon 4 Symposium.
-2018-
- Daoud, Adel, and Bernhard Reinsberg.
Structural adjustment, state capacity, and child health: Evidence from IMF programs, International Journal of Epidemiology, Volume 48, Issue 2, Pages 445–454.
https://doi.org/10.1093/ije/dyy251 - Kakooei, Mohammad, and Yasser Baleghi.
Leaf-Less-Tree feature for semantic labeling applications on Google Earth Engine.
In 2018 9th International Symposium on Telecommunications (IST), pp. 609-613. IEEE. Abstract. - Kakooei, Mohammad, Andrea Nascetti, and Yifang Ban.
Sentinel-1 Global Coverage Foreshortening Mask Extraction: An Open Source Implementation Based on Google Earth Engine.
In IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium, pp. 6836-6839. IEEE. Abstract. - Nascetti, Andrea, Mohammad Kakooei, and Yifang Ban.
Urban Extraction Using Sentinel-1 and Sentinel-2 Dense Time Series with Google Earth Engine. In 2nd mapping urban areas from space, ESA. Abstract.
-2014-
- Kakooei, Mohammad, and Hadi Shahriar Shahhoseini.
A parallel k-means clustering initial center selection and dynamic center correction on GPU.
In 2014 22nd Iranian Conference on Electrical Engineering (ICEE), pp. 20-25. IEEE, 2014. Abstract. - Zeinali, Behnam, Ahmad Ayatollahi, and Mohammad Kakooei.
A novel method of applying directional filter bank (DFB) for finger-knuckle-print (FKP) recognition.
In 22nd Iranian Conference on Electrical Engineering (ICEE), pp. 500-504. IEEE, 2014. Abstract.
TL;DR: What is the AI & Global Development Lab’s research about?
We fuse Earth observation (satellite imagery), machine learning, and causal inference to measure and explain human development across time and space. Our work focuses on building reliable, reusable data products (like poverty and wealth maps) and the methods needed to turn those products into credible evidence for policy and social impact.
- If you want the papers: they’re listed above by year (with PDFs, overviews, and BibTeX).
- If you want to use the research: start with the quick guide below, then visit Data, Code, and Tools.
Quick navigation
- What are the lab’s core research themes?
- Satellite imagery vs. household surveys: how do we measure development?
- Prediction vs. causal inference: which one do you need?
- What is “planetary-scale causal inference”?
- Best way to explore this research page
- How to use our data, code, and tools (without guesswork)
- Key terms (quick definitions)
- FAQ
What are the lab’s core research themes?
We work at the intersection of measurement (mapping development outcomes) and inference (estimating what causes change), using AI systems that scale to planetary data.
- Satellite-driven poverty and wealth analysis: creating fine-grained maps where surveys are infrequent or incomplete.
- Debiasing ML predictions for causal inference: making “one map, many trials” possible so downstream impact evaluations aren’t quietly attenuated by prediction shrinkage.
- Effect heterogeneity and targeting: identifying where and for whom interventions work best using multi-scale representations.
- Methods + open infrastructure: releasing datasets, code, and interactive tools to support reproducible research and policy learning.
Satellite imagery vs. household surveys: how do we measure development?
Surveys remain essential ground truth, but they can be expensive and sparse. Satellite imagery is frequent and wide-coverage, but it’s an indirect proxy—so the strongest workflows combine surveys, remote sensing, and careful validation.
| Approach | Best for | Main limitation |
|---|---|---|
| Household surveys | High-confidence measurement and benchmarking | Costly, infrequent, and geographically uneven |
| Satellite + ML | High-resolution, scalable mapping across space/time | Proxy signals can be biased or miscalibrated |
| Hybrid (survey + satellite + inference) | Usable maps and credible evaluation workflows | Requires careful assumptions, transparency, and robustness checks |
Practical takeaway: When you see a map on this site, think “actionable measurement”—then check whether the downstream question is predictive (“what is happening?”) or causal (“what changes outcomes?”).
Prediction vs. causal inference: which one do you need?
Prediction estimates outcomes (e.g., local wealth) from observed signals like satellite imagery. Causal inference estimates the impact of an intervention (e.g., an aid program) by comparing what happened to what would have happened otherwise.
- Use prediction when you need timely, fine-grained measurement (mapping, monitoring, targeting).
- Use causal inference when you need defensible answers about policy effectiveness (impact evaluation, counterfactual comparisons, heterogeneous effects).
- Use debiasing when you want to reuse ML-predicted outcomes in causal studies without quietly shrinking estimated effects.
If you’re specifically interested in making ML outcomes safer for evaluation, see: Debiasing ML Predictions (“One Map, Many Trials”). For multimodal poverty mapping at scale, start here: Platonic Poverty Mapping.
What is “planetary-scale causal inference”?
It’s our umbrella term for doing causal analysis with planet-scale data—Earth observation imagery, geolocated interventions, and modern ML—while keeping the core causal questions (identification, bias, uncertainty) explicit.
- Why it matters: many development questions are spatial, long-run, and hard to measure with traditional data alone.
- What it enables: comparing interventions, tracing change over time, and studying heterogeneous effects across geography.
- What it requires: transparent assumptions, validation against ground truth, and methods designed for bias correction and reuse.
Best way to explore this research page
- Scan the most recent year at the top of this page for the newest projects.
- Click “Overview” when available for a fast understanding (problem → method → results → limitations).
- Grab assets (datasets + code) from Data and Code.
- Try an interactive demo on Tools to see outputs in context.
- Watch a talk on Videos if you prefer walkthroughs and tutorials.
How to use our data, code, and tools (without guesswork)
Pick your goal, then use the matching entry point below—each is designed to be scannable by humans and extractable by AI systems.
- I want datasets: go to Data for curated releases (often with links to Hugging Face and Dataverse).
- I want reusable packages: go to Code (e.g., causal inference with images; post-hoc debiasing for ML predictions).
- I want interactive outputs: go to Tools for visual exploration and validation.
- I want to collaborate or follow along: see About and Partners, or check Jobs for opportunities.
Key terms (quick definitions)
- Earth observation (EO)
- Satellite and remote-sensing data used to measure human and environmental conditions across large areas and long time periods.
- Poverty / wealth mapping
- Using EO and machine learning to estimate economic well-being at fine spatial resolution, often to fill gaps between survey waves.
- Causal inference
- Methods for estimating the impact of interventions by reasoning about counterfactuals (what would have happened otherwise).
- Treatment-effect heterogeneity
- How an intervention’s impact varies across places, people, or contexts—often the difference between “works on average” and “works in practice.”
- Debiasing ML predictions
- Correcting systematic prediction distortions (like shrinkage toward the mean) so predicted outcomes can be reused in downstream evaluation without attenuating effects.
FAQ: AI & Global Development Lab research
What kinds of problems do you study?
We study development questions where measurement is hard and interventions matter—especially poverty and wealth dynamics, policy effectiveness, and how impacts vary across places. Our methods are designed to scale across geography and time while staying transparent about uncertainty.
Do you release data and code?
We aim to make our work reproducible with open datasets, code, and benchmarks whenever permitted. Start with Data and Code; note that some upstream sources (e.g., certain survey microdata) may have independent access policies.
Can I use your research without being an ML expert?
Yes. Many pages include TL;DRs, direct answers, and practical takeaways, and the Tools section lets you explore outputs interactively. If you want a fast overview, use the “Overview” links next to papers on this page.
How do you avoid overstating causal claims from satellite-based models?
We treat satellite and AI outputs as measurement tools that must be calibrated, validated, and paired with explicit causal designs. When ML predictions are reused for evaluation, we focus on bias correction and uncertainty so downstream decisions reflect real-world impact rather than model artifacts.
Where should I start if I want to cite or replicate a paper?
Use the paper’s “PDF” for the full method and results, “.bib” for citation metadata, and “Code” when available for replication. For step-by-step explainers, open the internal overview pages linked from this research list.