Researchers at the Ulsan National Institute of Science and Technology (UNIST) have developed a clever AI-driven method to forecast how much CO2 is absorbed by terrestrial plants – the very same process that tackles around 30% of global emissions through photosynthesis. This new approach harnesses high-frequency data from the Himawari-8 satellite, generating a detailed, hourly picture of the gross primary production (GPP) that underpins our planet’s carbon uptake.
Led by Professor Jungho Im, the team’s model provides estimates at an impressive daily and hourly cadence, ideal for those who’ve ever wrestled with the limitations of polar-orbiting satellites that only capture a few snapshots per day. GPP, which measures the carbon fixed by plants during photosynthesis, is a key indicator for assessing ecosystem health and carbon sequestration. By using data collected every 10 minutes, the model significantly enhances prediction accuracy, all while integrating vital meteorological inputs like Aerosol Optical Depth (AOD). AOD, a measure of fine dust and particulate matter in the air, plays a major role in how sunlight filters through to the surface, affecting how efficiently plants can work their photosynthetic magic.
The researchers also made clever use of SHapley Additive exPlanations (SHAP) to clarify which factors are driving the model’s predictions. You’ll find that AOD exerts a noticeable influence during the early morning and late evening when the sun sits low in the sky, perfectly aligning with the idea that scattered light can make a significant difference to photosynthesis. According to Professor Im, the model’s ability to track carbon absorption across East Asia at a 2 km resolution around the clock offers a valuable tool for those involved in ecosystem studies and vegetation monitoring.
