Probabilistic machine learning-based phytoplankton abundance using hyperspectral remote sensing
Published in GIScience & Remote Sensing, 2025
Recommended citation: Kwon, D. H., Ahn, J. M., Pyo, J. C., Lee, J., Abbas, A., Park, S., Kim, K., Lee, H., & Cho, K. H. (2025). Probabilistic machine learning-based phytoplankton abundance using hyperspectral remote sensing. GIScience & Remote Sensing, 62(1), 2484864. https://doi.org/10.1080/15481603.2025.2484864
Korean water authorities rate an algal bloom by counting cells, while the aircraft that survey the water measure its colour, split into hundreds of narrow bands — so what is watched from above is not what the rules actually regulate. Seven years of airborne surveys flown between 2016 and 2022, paired with samples taken from the water at the same time, were used here to estimate cell counts directly from the imagery. The models recovered roughly 60% of the variation for one group of algae and 40% for another, narrowing the gap rather than closing it, and every estimate arrives with a statement of how much confidence to place in it. They also keep cyanobacteria, diatoms and green algae apart instead of lumping them together, and map each across whole water bodies, which is what makes it possible to single out the stretches that need attention.