Abrupt intensification and spatiotemporal dynamics of large-scale harmful algal blooms in China’s marginal seas mapped using multi-sensor satellite imagery and deep learning

• HABs mapped by MODIS, OLCI, and non-fluorescence VIIRS using a deep learning approach. • Multi-sensor integration captures more extensive HABs with lower outbreak frequency. • In 2023–2024, HABs increased by > 51% versus 2018–2022, especially in Bohai and South China Sea. • HABs in the Yellow Sea exhibit bimodal seasonal peaks in April and August. • Seasonal HABs in China shift spatially from southeast in spring to north in summer. Harmful algal blooms (HABs) pose increasing threats to marine ecosystems and coastal economies in marginal seas around China. While satellite observations have greatly improved our understanding of HABs distributions, most prior studies rely on single sensors, which are limited by temporal resolution and unable to capture the dynamic changes of short-lived blooms, and lack continuous assessments of recent trends and seasonal variability. This study generated a HABs record from 2018 to 2024 by integrating 74,882 images from Moderate Resolution Imaging Spectroradiometer (MODIS), Visible Infrared Imager Radiometer Suite (VIIRS), and Ocean and Land Color Instrument (OLCI). A Res-UNet deep learning model was developed to extract HABs across the Bohai Sea, Yellow Sea, East China Sea, and South China Sea. The model achieved high accuracy across all sensors, with an average F1-score of 0.85, and performed well even on VIIRS images lacking fluorescence bands. Validation against China’s marine disaster bulletins yielded 81.09% accuracy, and comparisons with the CIE-HABs dataset and high-spatial-resolution imagery (Sentinel-2 MSI, Landsat-8 OLI, and HY-1C CZI) further confirmed reliability. Compared with single sensor results, our multi-sensor dataset combined from six sensors identified more extensive bloom extent with lower outbreak frequency. Using our multi-sensor dataset, we detected a > 51% increase in HAB intensity during 2023–2024 relative to 2018–2022 mean levels, particularly in the Bohai and the South China Sea, coinciding with abnormal sea surface temperature increases. The Yellow Sea exhibited the strongest seasonality, with bimodal peaks in April and August, while summer-dominant patterns occurred in the Bohai Rim and the Pearl River Delta. Seasonal HABs in China’s marginal seas exhibited a northward shift, from the South China Sea in spring to the Bohai and the East China Sea in summer. Our findings highlight the value of multi-sensor integration and deep learning for HABs monitoring, offering new insights into bloom dynamics under anthropogenic and climate impacts.