Weekly green tide mapping in the Yellow Sea with deep learning: Integrating optical and synthetic aperture radar ocean imagery

Abstract

Since 2008, the Yellow Sea has experienced the world's largest-scale marine disaster, the green tide, marked by the rapid proliferation and accumulation of large floating algae. Leveraging advanced artificial intelligence (AI) models, namely AlgaeNet and GANet, this study comprehensively extracted and analyzed green tide occurrences using optical Moderate Resolution Imaging Spectroradiometer (MODIS) images and microwave Sentinel-1 synthetic aperture radar (SAR) images. However, due to cloud and rain interference and the varying observation frequencies of the two types of satellites, the daily green tide coverage time series throughout the entire life cycle often contain large gaps and missing frames, resulting in discontinuity and limiting their use. Therefore, this study presents a continuous and seamless weekly average green tide coverage dataset with a resolution of 500 m, by integrating highly precise daily optical and SAR data for each week during the green tide breakout. The uncertainty assessment shows that this weekly product conforms to the life pattern of green tide outbreaks and exhibits parabolic-curve-like characteristics, with a low uncertainty (R2=0.89 and RMSE=275 km2). This weekly dataset offers reliable long-term data spanning 15 years, facilitating research in forecasting, climate change analysis, numerical simulation, and disaster prevention planning in the Yellow Sea. The dataset is accessible through the Oceanographic Data Center, Chinese Academy of Sciences (CASODC), along with comprehensive reuse instructions provided at 10.12157/IOCAS.20240410.002 (Gao et al., 2024).