Annual mean biomass of Ulva prolifera in the Yellow Sea from 2007 to 2024 (V 1.0)

Recurrent blooms of Ulva prolifera (the dominant algal species responsible for green tides) in the Yellow Sea have become one of the most significant marine ecological hazards worldwide, posing persistent threats to coastal ecosystems, aquaculture, tourism, and regional economies. Since 2007, large-scale green tide events have occurred almost annually. Satellite remote sensing provides a feasible means for large-scale and long-term monitoring of Ulva prolifera. MODIS imagery, with its high temporal resolution and long observational continuity, has been widely used for this purpose. However, traditional index-based approaches (e.g., Floating Algae Index, FAI) rely on manually selected thresholds and are highly sensitive to observation conditions, limiting their robustness and scalability for long-term applications. To address these limitations, this dataset was generated using GTD-Net (Green Tide Detection Network), a knowledge-guided deep learning model specifically designed for Ulva prolifera detection under complex marine environments. GTD-Net integrates a heterogeneous, multi-dimensional training dataset with architectural enhancements to improve cross-scenario generalization. Based on GTD-Net extraction results, Ulva prolifera biomass was estimated using a physically based BPA-FAI relationship that accounts for both horizontal and vertical spectral mixing effects. This dataset provides a long-term record of annual mean biomass per unit area (BPA, kt km⁻²) of Ulva prolifera in the Yellow Sea from 2007 to 2024. It is intended to support studies on green tide dynamics, environmental drivers, climate variability, and ecosystem risk assessment.黄海海域反复发作的浒苔（Ulva prolifera，绿潮优势藻类）已成为全球最严重的海洋生态灾害之一，持续对沿岸生态系统、水产养殖业、旅游业及区域经济构成威胁。自2007年起，大规模绿潮事件几乎每年都会暴发。卫星遥感为浒苔的大规模、长期监测提供了可行技术手段。中分辨率成像光谱仪（MODIS）影像凭借高时间分辨率与长期观测连续性，已被广泛应用于该类监测工作。然而，传统基于指数的检测方法（例如漂浮藻类指数（Floating Algae Index，FAI））依赖人工选取阈值，且对观测条件极为敏感，这限制了其在长期应用中的鲁棒性与可扩展性。为解决上述局限，本数据集采用绿潮检测网络（Green Tide Detection Network，GTD-Net）生成——这是一种专为复杂海洋环境下浒苔检测设计的知识引导型深度学习模型。该模型整合了异构多维训练数据集，并通过架构优化提升了跨场景泛化能力。基于GTD-Net的提取结果，本研究采用基于物理机制的BPA-FAI关系估算浒苔生物量，该关系同时考虑了水平与垂直方向的光谱混合效应。本数据集提供了2007年至2024年黄海海域浒苔单位面积年均生物量（BPA，kt km⁻²）的长期序列记录，旨在为绿潮动力学、环境驱动因子、气候变异性及生态系统风险评估等相关研究提供支撑。