Earth Observation Data for Monitoring Economic Activities
Start of implementation: 2021
Technology type: Big data, artificial intelligence, geographic information system mapping, remote sensing
Technology service provider: Earthlab AI Systems
ITD’s Partner ADB Department: Economic Research and Development Impact Department (formerly known as the Economic Research and Regional Cooperation Department)
In line with ADB’s Operational Priorities:
· Fostering regional cooperation and integration
o Global and regional trade and investment opportunities expanded
o Regional public goods increased and diversified
· Strengthening governance and institutional capacity
o Strengthened public management and financial stability
Various developments in the past years, including additional investments in satellite capabilities, advancements in algorithms and data processing tools, and increasing data accessibility, have led to the increasing popularity of using earth observation data to generate insights. COVID-19 further accelerated its use as more governments and organizations are exploring alternative data sources given that the pandemic affected the collection and sharing of official statistics.
Earthlab AI Systems, which was the team selected in the “Earth Observation Data Challenge,” explored whether it was possible to use different sources of data collected by satellite missions for correlating and predicting economic activities. The original intent was to have an integrated dashboard for monitoring multiple economic indicators across different countries, but it was later decided that the initiative would instead focus on investigating whether there were correlations between proxy indicators using earth observation data and economic activities in three countries: Georgia, the Philippines, and the Republic of Korea.
Besides remote sensing and geographic information system mapping, the team used artificial intelligence, specifically machine learning, and big data to make sense of the earth observation images taken by satellites. The amount of earth observation data collected by satellites is huge. Making sense of the information can be challenging given the extent and capacity of humans to interpret data. Machine learning can be used to detect patterns and similarities in large amounts of data.
Four approaches were used:
1. Ships in ports. The presence of ships in ports was used as a proxy for trade volume from international shipping. Synthetic aperture radar images from the European Space Agency Copernicus Mission Sentinel 1 were analyzed since radar satellite data are not affected by weather conditions such as clouds. Areas that showed highly significant positive correlations between the number of ships in their ports and economic indicators included the Port of Davao and the Port of Gwangyang. The Port of Busan, meanwhile, showed a negative correlation between the average monthly number of ships in the port and exports as an economic indicator. Other factors could have affected the presence of ships in ports. For instance, the movement restrictions in the first half of 2020 saw an increase in the number of ships in the Port of Manila because of quarantine restrictions.
2. Nitrogen dioxide concentrations. Nitrogen dioxide is emitted when fuel is burned, and as such is associated with human activity. The European Space Agency’s Sentinel-5p L2 measures nitrogen dioxide concentrations globally, alongside other variables. There was also a clear seasonal effect on the nitrogen dioxide concentrations across all three countries. While all calculated correlation coefficients were positive and some were statistically significant, there were inconclusive results on whether there was a significant correlation between nitrogen dioxide measurements and economic indicators. This may have been due to various reasons, including the complexity of the relationship between nitrogen dioxide and different economic indicators and meteorological conditions (such as precipitation and wind). It is also possible that the time series of 3 years was too short.
3. Containers in ports. The number of containers in ports was also studied as a possible proxy indicator for trade volume. Images from PlanetScope, which takes images of the land surface of the earth every day using approximately 130 satellites, were analyzed. The small sample size may have contributed to inconclusive results. There was a strong but not significant negative correlation between the number of containers and the import and export activity. This may have been affected by the port’s space utilization. For instance, containers may have piled up when trade slowed down.
4. Nighttime lights. Nighttime lights can provide insights on areas that had limited available data on socioeconomic indicators. The working assumption in this initiative was that there was a correlation between nighttime light and industrialization in a given area. However, there were inconclusive results. Further research may be needed to determine the relationship between economic activities and nighttime light. It is possible that the available instruments at the time the satellite images were collected were not sensitive enough to capture the subtle differences in nighttime light. Additional calibration, cleaning, and processing of the data may be needed as the available images did not seem to show significant differences between the average pixel radiance per month. There may also have been other variables that influenced the relationship between economic activity and nighttime light.
This initiative showed a lot of potential but time, resources, and technical support from experts are needed to explore these approaches further. Nonetheless, these results may be used as a starting point to pursue additional economic analyses. Similar studies should consider unique geographical and socioeconomic contexts. A clear understanding of the tools and technologies is also important to ensure correct interpretation and analysis. Care must be taken in analyzing and forming conclusions considering that some of these approaches used data that covered limited time periods. Data that cover longer periods can provide a better understanding of the relationships of different variables.