AI-revealed long-term spatio-temporal characteristics of yellow sea green tides

Read full text Summarise with VIRAT
Authors: Xinliang Pan, Longxiao Zheng, Xiaofeng Li, Jiangtao Li, Mengqiu Wang, Yanfang Xiao, Rongjie Liu, Keunyong Kim, Song Gao, Xiaomin Ye, Maohua Guo, Ying Xu
+8 more Lie Sun, Ning Wang, Xiaoqing Cai, Joo‐Hyung Ryu, Jue Huang, Yuhan Zhang, Huanhong Ji, Tingwei Cui
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Amid accelerating global environmental change over recent decades, floating macroalgae blooms have expanded worldwide. Notably, the annual Ulva prolifera bloom in the Yellow Sea has emerged as the world's largest green tide disaster, significantly impacting the marine environment and economy. However, the complex ocean–atmosphere environment presents challenges for high-accuracy automatic remote sensing detection. Consequently, a long-term, high-resolution daily dataset remains absent, which limits understanding of spatiotemporal characteristics and hinders disaster prevention. In this study, we designed a high-accuracy automated model, Comprehensive Algae Net (CANet), for green tide detection in satellite images using deep learning. The model enables high-accuracy green tide detection in cloud edges and turbid waters, with a Kappa coefficient of 0.875, and outperforms traditional models such as U-Net and FCN. Based on CANet, we constructed a daily 250-m-resolution multivariate dataset (2007–2024) capturing unmixed area and biomass dynamics. Key findings indicate that over the past 18 years, green tides have cumulatively affected 23% (70,914 km²) of the South Yellow Sea, spanning 526 km east–west and 682 km north–south, with an average annual duration of 82 ± 14 days. Despite interannual variations, the disaster scale can be predicted from the initial area (R² = 0.66, p < 0.05). The model and dataset provide a valuable basis for mechanism analysis and improved operational monitoring.

Fields of Study & Subjects

Marine and coastal plant biology Remote-Sensing Image Classification Seaweed-derived Bioactive Compounds

Publication details

Journal / source
Figshare
Publication date
25 March 2026
Language
EN