Advancing green tide monitoring: a spatiotemporal analysis in the Southern Yellow Sea using a novel deep learning and remote sensing fusion approach

Green tides caused by the macroalga Ulva prolifera are recurrent marine disasters in the Southern Yellow Sea (SYS), China. Accurate monitoring and spatiotemporal characterization are essential for effective mitigation. However, conventional remote sensing methods struggle to detect fine filamentous structures, and prevailing deep learning models suffer from spatial detail loss and scene-dependent variability that undermines multi-temporal consistency. To address these challenges, we propose UPS-Net-LDU, an improved semantic segmentation model that integrates a Hybrid Attention Gate (HAG) and an Edge-Enhanced Strip (EES) module into the U-Net architecture. Using 170 GF-1 WFV images from 2018 to 2025, we construct a multi-year distribution dataset covering the high-incidence period (May–August) of U. prolifera in the SYS. Quantitative validation shows that UPS-Net-LDU achieves an IoU of 73.44%, Precision of 82.93%, Recall of 86.52%, and an F1-score of 84.68%, outperforming several mainstream benchmarks. Ablation experiments confirm the individual contributions of the HAG and EES modules. Spatiotemporal evolution analysis reveals a persistent spatial pattern: source area in the Subei Shoal → northeastward transport corridor → convergence zone in the southeastern Shandong Peninsula, with notable interannual variability including an alternating pattern of high- and low-intensity years between consecutive years. Integrated analysis with sea surface temperature (SST) data indicates that SST warming within 15–24 °C during May–June facilitates algal expansion, whereas persistently high temperatures (>24 °C) during July–August suppress growth and promote decay, clarifying a dual-phase regulatory mechanism. This study offers a robust methodological tool for precise monitoring and delivers critical insights into the dual-phase thermal regulation of green tides, supporting early warning and management in the region.