About 30% of the carbon dioxide emitted worldwide is absorbed by plants through photosynthesis. However, it has been difficult to accurately determine how much plants absorb by time of day. Recently, domestic researchers developed artificial intelligence (AI) technology that can precisely analyze this process on an hourly basis. This technology is expected to be a significant aid in responding to climate change and establishing carbon neutrality policies.
Professor Lim Jung-ho's research team at Ulsan National Institute of Science and Technology (UNIST) announced on the 1st that they developed a model that predicts the amount of carbon dioxide absorbed by plants on an hourly basis by training AI with ultraviolet and meteorological information from satellites. The research findings were published in the international academic journal in the field of remote sensing, "Remote Sensing of Environment," on the same day.
Photosynthesis is the fundamental natural process by which plants absorb carbon dioxide using sunlight and produce organic matter. The amount of carbon absorbed in this process is quantified as "gross primary production (GPP)." This is a key indicator showing the carbon storage capacity of ecosystems.
Previous GPP predictions relied on polar orbit satellites that could only observe a few times a day. As a result, there were limitations in accurately reflecting how factors such as sunlight, clouds, and fine dust that change over time affect photosynthesis. In particular, aerosols, including fine dust, absorb or scatter sunlight and directly or indirectly influence photosynthesis, but no technology existed to precisely track that influence by time of day. To overcome these limitations, Professor Lim's research team developed a new predictive model combining AI and high-frequency satellite data.
The research team created a model that can predict GPP on an hourly basis by training AI with data sent every 10 minutes from Japan's Himawari-8 geostationary satellite. This enables much more detailed analysis of the photosynthetic responses of plants than before.
The model uses not only weather information but also "aerosol optical thickness (AOD)" to indicate how much aerosols in the atmosphere absorb or scatter sunlight. AOD is a satellite observation index that indirectly demonstrates the concentration of particulate matter like fine dust, altering the intensity and properties of sunlight which influences photosynthesis.
The research team employed explainable AI techniques (SHAP) to verify what information AI used for its predictions. The results showed that during times when sunlight is weak, like in the morning and evening, AOD had the greatest influence on photosynthesis predictions. This reflects the characteristic that as the solar angle decreases, the proportion of scattered light increases, which sensitively alters the photosynthetic reaction.
Professor Lim said, "We can estimate carbon absorption responses over a day for the East Asia region at a spatial resolution of 2 km," adding, "This could be utilized in various fields, from analyzing carbon flow in ecosystems to monitoring vegetation responses and carbon modeling based on light environments."
References
Remote Sensing of Environment (2025), DOI: https://doi.org/10.1016/j.rse.2025.114735