Global Ocean Methane Seepage Flux Based on Submarine Observation of South China Sea Combine with Machine Learning

Data for article “Global Ocean Methane Seepage Flux Based on Submarine Observation of South China Sea Combine with Machine Learning” Marine methane seepage represents a critical component of the global carbon-water cycle, yet existing studies face persistent challenges: scarcity of long-term continuous monitoring data, difficulty in capturing the complex interactions of multiple environmental factors, and limited scalability from local observations to global estimates. This research addresses these gaps through three major innovations. First, we pioneer the application of machine learning with SHAP analysis in marine methane seepage prediction, achieving exceptional model accuracy (R² > 0.95) while ensuring model transparency and interpretability. Second, our study systematically identifies key environmental sensitivity factors controlling seepage dynamics from seafloor sedimentary layers to the seawater interface, revealing that pressure dominates high-intensity seepage events (61% contribution) with distinct tidal periodicity. Third, we provide robust quantitative estimates of methane flux at both regional (South China Sea: 0.964 ± 0.482 Tg CH₄/year) and global scales (11.465±0.382 ~ 49.613±1.394 Tg CH₄/year), global ocean methane seepage flux is significantly underestimated (beforehand 6-12 Tg CH₄/year).本数据集配套于论文《基于南海海底观测结合机器学习的全球海洋甲烷渗漏通量》。 海洋甲烷渗漏是全球碳-水循环的关键组成部分，但现有研究始终面临诸多持续性挑战：长期连续监测数据匮乏、难以捕捉多环境因子间的复杂相互作用，且从局地观测推演至全球估算的可扩展性受限。 本研究通过三项核心创新填补了上述研究空白。 其一，本研究率先将机器学习结合SHAP（SHapley Additive exPlanations）分析应用于海洋甲烷渗漏预测，模型精度优异（决定系数R²＞0.95），同时保障了模型的透明度与可解释性。 其二，本研究系统甄别了从海底沉积层至海水界面间控制渗漏动态的关键环境敏感因子，揭示出压强对高强度渗漏事件的贡献占比达61%，且其变化具有显著的潮汐周期性。 其三，本研究给出了可靠的甲烷通量定量估算结果：区域尺度上南海的甲烷渗漏通量为0.964±0.482 Tg CH₄/年，全球尺度上则为11.465±0.382 至49.613±1.394 Tg CH₄/年；当前全球海洋甲烷渗漏通量的认知值被显著低估（此前学界估算值仅为6~12 Tg CH₄/年）。