中文摘要：

本研究基于单质点拉格朗日集成轨迹模型（HYSPLIT），对珠三角地区2012年全年每3小时的后向轨迹进行了模拟。模拟采用了两套初始气象场来驱动HYSPLIT模型，其一为水平分辨率为0．5度的全球同化系统（GDAS）输出场，其二为水平分辨率为27km的气象研究与预报模型（WRF）的输出场。结果显示，当用于驱动HYSPLIT来模拟后向轨迹时，GDAS数据的可靠性依赖于季节和天气条件。GDAS和WRF输出场得到的HYS—PLIT后向轨迹的方向差异在冬季和春季珠三角盛行偏北及东北风场时为最大，而在秋季珠三角处在稳定系统控制下时为最小。大多数WRF输出场得到的后向轨迹的传输高度比GDAS得到的后向轨迹要高，这是由于更高的分辨率使得WRF输出场的边界层垂直运动更明显。在五种天气条件下，不同分辨率初始气象场得到的后向轨迹差别最大，分别为大陆高压，槽脊系统，鞍型场，热带气旋以及冷锋。案例分析的结果显示，绝大多数情况下，GDAS和WRF输出场得到的后向轨迹因传输高度不同导致了轨迹方向的差别，因此在模拟HYSPLIT后向轨迹时，初始气象场在垂直方向上的细节是一个尤为关键的要素。

英文摘要：

In this study, backward particle dispersion modeling is conducted using the Hybrid Single-Particle Lagrangian Integrated Trajectory （HYSPLIT） model for 2012 in the Pearl River Delta of China. The Global Data Assimilation System （GDAS） data with horizontal resolution of 0.5 degree and the output data from Weather Research and Forecast （WRF） model with horizontal resolution of 27 km are used to drive the HYSPLIT model, respectively. The results reveal that the applicability and reliability of GDAS data is dependent on seasons and synoptic conditions. The direction differences between GDAS and WRF output data generated trajectories are the largest in winter and spring when northerly and northeasterly winds prevail and the smallest in autumn when stable weather systems control the region. Most of trajectories generated from WRF output data transport to higher altitudes than those from GDAS data, due to the more convective boundary layer which the vertical velocity fields from WRF output data reflect. Five different kinds of synoptic conditions are identified which favor generation of large direction differences between trajectories, including continental highs, troughs and ridges, col pressure fields, tropical cyclones and cold fronts. Case studies suggest that in most situations, the large direction differences are mainly resuhed from the height differences between trajectories generated from GDAS and WRF output data, and vertical details of input data are more crucial than horizontal details in generating HYSPLIT trajectories.