Predicting nitrogen dioxide column density from the International Wealth Index with a convolutional neural network in Africa
Journal article, 2026

In the pursuit of improving socioeconomic living conditions of their populations, many nations struggle to meet their nationally determined contributions (NDCs) under the Paris agreement, and that shortfall makes accountability for emissions reductions critical. The long atmospheric lifetimes of key gases such as carbon dioxide (CO2) and methane (CH4) hinder traditional greenhouse gas (GHG) monitoring, because what any one place emits adds only a small excess to a large, well-mixed background. Nitrogen dioxide (NO2) behaves differently: it is short-lived and highly variable in space and time, so satellite instruments resolve its patterns close to source regions. Across the African continent, mapped temporal correlations between NO2 and CO2 are moderate in magnitude but spatially structured and consistently positive. Although we do not use that association to mechanically convert NO2 observations into CO2, we report it as contextual evidence towards NDCs. Given this context, this study asks the following question: to what extent can socioeconomic development at the neighbourhood scale predict NO2 column densities (the amount of NO2 in the air column above that neighbourhood) across Africa? We apply a convolutional neural network (CNN) to gridded International Wealth Index (IWI) estimates and to NO2 observations from the tropospheric Monitoring Instrument (TROPOMI), and obtain an average predictive accuracy of = 0.56 across the African continent. Because IWI can be projected forward, we also present an illustrative 2030 scenario as a methodological demonstration of the relationship between socioeconomic development and NO2 trajectories. These findings indicate that gridded wealth estimates carry predictive information about NO2 column density in Africa. Such estimates offer a lightweight and scalable screening tool that complements process-based atmospheric monitoring rather than replacing it.

Spatial autocorrelation

NO2 monitoring

International Wealth Index (IWI)

Machine learning in environmental policy

Africa

Author

Francesco Grieco

Linköping University

Chalmers, Computer Science and Engineering (Chalmers), Data Science and AI

Mohammad Kakooei

Linköping University

Karlstad University

Adel Daoud

Linköping University

Chalmers, Computer Science and Engineering (Chalmers), Data Science and AI

Scientific Reports

2045-2322 (ISSN) 20452322 (eISSN)

Vol. 16 1 32439

Analysis of satellite data on NO2 emissions to evaluate development in Africa

Swedish Research Council (VR) (2023-06608), 2024-01-01 -- 2026-12-31.

Subject Categories (SSIF 2025)

Environmental Sciences

Climate Science

DOI

10.1038/s41598-026-73707-0

Related datasets

Data [dataset]

URI: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DV N/9DINV4 , https://disc.gsfc.nasa.gov/datasets/S 5P_L2__NO2___HiR_2/summary?keywords=TROPOMI, https://www.gu.se/en/quality-government/qog-data/data-downloads/standard-dataset

More information

Latest update

10/27/2026