Abstract The performance of different forcing datasets for trajectory prediction in the South China Sea and the sensitivity of prediction accuracy from selected datasets and model-based methods have been studied using Lagrangian models. Seven global ocean forecast or reanalysis forcing datasets including three current forcing datasets (CMEMS, CMEMS-R, and GlobCurrent), three wind forcing datasets (NCEP, ERA5, and IFREMER) and one waves forcing dataset (MFWAM) were selected to advect the virtual drifters for modeling the drift trajectories of dummies and SVP drifters in four experimental regions. We use an integrated analysis framework to conduct several holistic prediction experiments, evaluate quantitatively the accuracy of trajectory prediction using statistical indicators and FSLE analysis, and to further estimate the uncertainty factors from data and model levels. The experimental results show that the CMEMS-R can meet the needs of trajectory prediction of dummies within 48 hours in Region 1, but neither CMEMS nor CMEMS-R can better follow the complex ocean current pattern in Region 2, which leads to the poor results of the 24-hour trajectory prediction driven by CMEMS or CMEMS-R. In the deep water regions (Region 3-4), the performance of the GlobCurrent derived from satellite-observations for modeling the trajectories of SVP drifters is comparable with the CMEMS in general, and shows obvious advantages in Region 3, even though CMEMS has a finer spatial and temporal resolution. The two wind forcing datasets, ERA5 and IFREMER, show roughly similar performance in the trajectory prediction in Region 1. Compared with the significant difference between current forcing datasets, the three wind forcing datasets have relatively consistent temporal and spatial variations in the other three experimental regions. In four experimental regions, the separation distance between the simulated and actual trajectories increases in a uniform linear mode with mean separation velocity. In general, the selection of current forcing fields has more influence on the trajectory prediction of the three types of floating objects than other factors. The ocean current pattern represented by related current forcing datasets is the key to this essential influence.

中文翻译：

---

使用拉格朗日模型对南海多源强迫数据集进行漂移轨迹预测的评估

摘要 使用拉格朗日模型研究了不同强迫数据集对南海轨迹预测的性能以及所选数据集和基于模型的方法的预测精度的敏感性。选择了七个全球海洋预测或再分析强迫数据集，包括三个海流强迫数据集（CMEMS、CMEMS-R 和 GlobCurrent）、三个风力强迫数据集（NCEP、ERA5 和 IFREMER）和一个波浪强迫数据集（MFWAM）来平流虚拟用于模拟四个实验区域中假人和 SVP 漂移器的漂移轨迹的漂移器。我们使用综合分析框架进行多次整体预测实验，使用统计指标和 FSLE 分析定量评估轨迹预测的准确性，并从数据和模型层面进一步估计不确定因素。实验结果表明，CMEMS-R能够满足1区48小时内假人轨迹预测的需要，但CMEMS和CMEMS-R都不能更好地跟踪2区复杂的洋流模式，导致结果较差由 CMEMS 或 CMEMS-R 驱动的 24 小时轨迹预测。在深水区（区域 3-4），卫星观测所得的 GlobCurrent 的性能与 CMEMS 总体上与 CMEMS 相当，并且在区域 3 中显示出明显的优势，尽管 CMEMS 具有更精细的空间和时间分辨率。两个风力强迫数据集 ERA5 和 IFREMER 在区域 1 的轨迹预测中显示出大致相似的性能。与当前强迫数据集之间的显着差异相比，三个风力强迫数据集在其他三个实验区域具有相对一致的时空变化。在四个实验区域中，模拟轨迹和实际轨迹之间的分离距离以均匀线性模式随着平均分离速度增加。总的来说，当前强迫场的选择对三种漂浮物的轨迹预测的影响比其他因素更大。相关海流强迫数据集所代表的洋流模式是这种重要影响的关键。模拟轨迹和实际轨迹之间的分离距离随着平均分离速度以均匀线性模式增加。总的来说，当前强迫场的选择对三种漂浮物的轨迹预测的影响比其他因素更大。相关海流强迫数据集所代表的洋流模式是这种重要影响的关键。模拟轨迹和实际轨迹之间的分离距离随着平均分离速度以均匀线性模式增加。总的来说，当前强迫场的选择对三种漂浮物的轨迹预测的影响比其他因素更大。相关海流强迫数据集所代表的洋流模式是这种重要影响的关键。