POSTER SESSION 1
15:30–17:30, Monday, September 14
Poster Session | 1 | 2 | 3 | 4 | Schedule at a Glance
ABSTRACT 1243 | POSTER 45
GLOBAL DETECTION OF RED AND GREEN NOCTILUCA SCINTILLANS BLOOMS USING SENTINEL-2 IMAGES AND ADVANCED MACHINE LEARNING
Blooms of the marine dinoflagellate Noctiluca scintillans (400–1500 μm in diameter) typically discolor coastal waters red or green, depending on the pigmentation of their vacuoles, and can threaten marine ecosystems and fisheries by depleting dissolved oxygen, releasing ammonia, and disrupting planktonic food webs, leading to fish mortality, reduced recruitment, and altered trophic dynamics. Detection of red and green Noctiluca blooms from space remains challenging, mainly due to the characteristics of uneven biomass distribution and small-scale patchiness. Here, we developed a Noctiluca bloom detection model based on 10m-resolution Sentinel-2 multispectral imagery. We compiled a comprehensive dataset of citizen-reported bloom events and synchronized imagery across eight study regions worldwide, e.g., the North Sea, South China Sea, and Gulf of Mexico. To train the machine learning model, sample datasets of red and green Noctiluca blooms were generated by expert visual interpretation in pixel scale and spectral indices (e.g., band ratios, Maximum Chlorophyll Index) were used as input features. The performance of several ML algorithms (random forest, eXtreme Gradient Boosting, and local cascade ensemble) was compared. To decode model decisions, SHapley Additive exPlanation indices were employed to quantify the contribution of the top 10 spectral features to bloom pixel identification of Noctiluca blooms. This study provides a robust, transferable framework for operational monitoring of Noctiluca blooms at 10m resolution, and demonstrates the importance of spectral indices in the ML model, providing deeper understanding to the spatial and temporal distribution of Noctiluca blooms, supporting management of harmful algal blooms worldwide.
Xingda Chen*, MarSens, [email protected], https://orcid.org/0009-0008-2758-3310
Griet Neukermans, MarSens/UGent, [email protected], https://orcid.org/0000-0002-8258-3590
Shuisen Chen, Chinese Academy of Sciences, [email protected]
Poster Session | 1 | 2 | 3 | 4 | Schedule at a Glance
