In a world where more than 700 million people live in extreme poverty, assessing the true extent of economic hardship remains a daunting global challenge. Remarkably, 53 countries have not conducted agricultural surveys in the past 15 years, and 17 nations have not even published a population census. Addressing this data gap, innovative technologies are emerging to estimate poverty levels by harnessing alternative sources, including street views, aerial photographs, and satellite imagery.

A recent paper published in Nature Communications presents a groundbreaking approach that utilizes artificial intelligence (AI) to analyze economic conditions from daytime satellite images. What sets this technology apart is its applicability to the most underdeveloped countries, including the notoriously secretive North Korea, which lack reliable statistical data typically needed for machine learning training.

The researchers leveraged Sentinel-2 satellite images from the European Space Agency (ESA), which are publicly available. These images were divided into small six-square-kilometer grids, allowing the quantification of economic indicators through visual cues such as buildings, roads, and vegetation. Consequently, the team succeeded in creating the first-ever fine-grained economic map of regions like North Korea. The same algorithm was applied to several other underdeveloped Asian countries, including Nepal, Laos, Myanmar, Bangladesh, and Cambodia.

A distinguishing feature of this research model is its “human-machine collaborative approach,” enabling researchers to combine human assessments with AI predictions in areas with limited data. In this study, ten human experts evaluated satellite images and assessed economic conditions, with the AI learning from this human input to assign economic scores to each image. The results demonstrated that this Human-AI collaboration outperformed machine-only learning algorithms.

The interdisciplinary research team behind this groundbreaking work included computer scientists, economists, and a geographer from institutions such as KAIST, IBS, Sogang University, HKUST, and NUS. The paper underwent rigorous peer review at Nature Communications, with Dr. Charles Axelsson, Associate Editor at the journal, overseeing the process.

One of the key findings of the study was that the scores derived from the AI model exhibited a strong correlation with traditional socio-economic metrics such as population density, employment rates, and the number of businesses. This underscores the versatility and scalability of the approach, especially in countries with limited data availability. Furthermore, the model’s capability to detect annual changes in economic conditions at a granular geospatial level without relying on survey data is particularly noteworthy.

This AI model holds immense potential for monitoring progress toward Sustainable Development Goals, including poverty reduction and the promotion of equitable and sustainable growth on a global scale. Additionally, it can be adapted to measure various social and environmental indicators, such as identifying regions highly vulnerable to climate change and disasters, providing valuable guidance for disaster relief efforts.

To illustrate the technology’s capabilities, the researchers examined changes in North Korea before and after United Nations sanctions were imposed on the country. By applying the model to satellite images from 2016 and 2019, three key trends in North Korea’s economic development emerged. First, economic growth became increasingly concentrated in major cities like Pyongyang, exacerbating urban-rural disparities. Second, satellite imagery revealed substantial changes in areas designated for tourism and economic development, including new construction and meaningful alterations. Third, traditional industrial and export development zones experienced relatively minor changes.

Meeyoung Cha, a data scientist on the research team, emphasized the significance of this interdisciplinary effort, stating, “This is an important interdisciplinary effort to address global challenges like poverty. We plan to apply our AI algorithm to other international issues, such as monitoring carbon emissions, disaster damage detection, and the impact of climate change.”

Jihee Kim, an economist involved in the research, highlighted how this approach could enable detailed examinations of economic conditions in developing countries at a low cost, ultimately reducing data disparities between developed and developing nations. She stressed that this is crucial because many public policies require accurate economic measurements to achieve their goals, whether they pertain to growth, equality, or sustainability.

The research team has made the source code publicly available on GitHub and intends to further refine the technology, applying it to newly updated satellite imagery annually. The results of this groundbreaking study, authored by Ph.D. candidates Donghyun Ahn at KAIST and Jeasurk Yang at NUS, were published in Nature Communications under the title “A Human-Machine Collaborative Approach Measures Economic Development Using Satellite Imagery.”