Machine learning-assisted sea surface wind dataset (MLAWind)

The product is developed under the leadership of Weihao Guo, Rongwang Zhang, and Xin Wang from South China Sea Institute of Oceanology, Chinese Academy of Sciences. The product has a horizontal spatial resolution of 1°×1° and covers the period from 1950 to 2023 with a monthly average temporal resolution. Based on the Random Forest machine learning algorithm and the SHapley Additive exPlanations (SHAP) interpretable module, the product effectively integrates in-situ observations with satellite data. It achieves comparable accuracy to widely-used sea surface wind products, while significantly improving El Niño-Southern Oscillation (ENSO) forecast skill in boreal spring.本数据集由中国科学院南海海洋研究所的郭伟浩、张荣旺、王欣主导研发。该数据集空间分辨率为1°×1°，时间分辨率为月均尺度，覆盖时段为1950年至2023年。基于随机森林（Random Forest）机器学习算法与SHapley可加解释（SHAP）可解释模块，本数据集有效融合了原位观测数据与卫星数据，其精度可与主流海面风场产品相媲美，且显著提升了北半球春季厄尔尼诺-南方涛动（ENSO）的预报技巧。