Using Satellite Data and Machine Learning to Track Cambodia’s GDP in Near Real Time

The IMF study shows that satellite data such as nighttime lights, air pollution, and vegetation health, when combined with machine-learning models, can reliably estimate Cambodia’s GDP growth in near real time, overcoming delays in official statistics. It finds that adding satellite indicators improves forecasting accuracy by over 20 percent and provides valuable regional insights for faster, better-informed economic policymaking.

Prepared by economists at the International Monetary Fund (IMF) in collaboration with data providers such as NASA, the Food and Agriculture Organization (FAO), and climate research institutions, the paper tackles a familiar policy problem in developing economies: how to understand what is happening in the economy when official data arrive too late. Cambodia’s GDP is published only once a year and with long delays, leaving policymakers without timely signals during fast-changing conditions. The paper shows how satellite data, collected continuously from space, and modern machine-learning techniques can help estimate Cambodia’s GDP growth in near real time, offering a practical solution to this information gap.

Why Traditional Data Are Not Enough

Cambodia has made progress in improving its national statistics, but important gaps remain. Annual GDP figures are not sufficient for real-time decision-making, and many high-frequency indicators, such as trade, credit, prices, or tourism, also come with reporting lags. As a result, policymakers often have to rely on partial or outdated information. The paper argues that this challenge is not unique to Cambodia and that alternative data sources are needed to complement official statistics. Satellite data stand out because they are available for almost all countries, updated frequently, and measured consistently over time and across regions.

What Satellite Indicators Reveal

The study focuses on several satellite-based indicators that reflect real economic activity. Nighttime lights capture the brightness of cities and towns at night and are linked to electricity use, urbanization, tourism, and industrial production. Nitrogen dioxide emissions reflect pollution from factories, power plants, and vehicles, making them a useful proxy for industrial and transport activity. Vegetation-related indicators measure plant health and water stress, providing insight into agricultural output, which remains central to Cambodia’s economy. Rainfall data add information about climate conditions that affect food production. The paper finds that changes in these satellite indicators are positively correlated with Cambodia’s GDP growth, suggesting that what satellites observe from space closely mirrors what is happening on the ground.

Turning Data into GDP Nowcasts

To combine satellite data with traditional economic indicators, the authors use a machine-learning method known as a random forest model. This approach is well-suited to handling many variables and capturing complex, non-linear relationships. Because Cambodia does not publish quarterly GDP, the authors first construct a quarterly GDP series by interpolating annual data using the growth patterns of major trading partners, weighted by Cambodia’s export shares. The model is then trained using both satellite indicators and standard macroeconomic variables such as trade, inflation, credit, exchange rates, and tourism arrivals. The data are split into past observations used for training and more recent observations used for testing, ensuring that the model’s performance reflects its ability to predict unseen data. As new satellite data become available, the GDP nowcast can be updated monthly.

What the Results Show

The results are encouraging. The model closely tracks historical GDP movements and produces realistic estimates for recent quarters. For early 2025, it estimates year-on-year GDP growth of around 5.7 percent in the first quarter and 6.7 percent in the second quarter. Importantly, adding satellite indicators improves the accuracy of GDP estimates by more than 20 percent compared with models that rely only on traditional data. Among satellite variables, nitrogen dioxide emissions are particularly influential over the long term, highlighting the role of industrial activity, while nighttime lights become more important in recent years, reflecting urban growth, rising electricity use, and tourism patterns. Vegetation indicators contribute steadily, with larger effects during periods of agricultural stress, underscoring agriculture’s ongoing importance.

Why This Matters for Policy

Beyond national averages, satellite data provide valuable regional insights. The paper shows that nighttime lights differ significantly across Cambodian provinces, revealing stronger activity in urban centers like Phnom Penh and coastal areas, and weaker signals in tourism-dependent or remote regions. Vegetation indicators similarly highlight differences in agricultural conditions across the country. These spatial details allow policymakers to identify localized slowdowns or strengths and design more targeted responses. Overall, the paper concludes that satellite data should not replace official statistics, but they can greatly strengthen economic monitoring where traditional data are limited. For Cambodia and many similar economies, looking at the economy from space offers a powerful new tool for timely, informed policymaking.