临近空间通常指距离地球表面20~200km高度的地球空间，是国家安全空间战略新高地。临近空间大气除了随高度、地理纬度、经度和季节变化的气候特性以外，还存在其他多种时间、空间尺度上的复杂变化。由于临近空间探测的手段多样，相应的探测仪器和其探测原理也互不相同，所得到的测量结果刻画的变化特征也存在多时空尺度特性。针对临近空间多尺度数据同化的需求，根据具体的数据资料，研究开发相应的尺度分离方法，是亟待解决的问题。本论文采用多尺度子空间变换（MWT）和经验模态分解（EMD）两种分离方法对临近空间再分析大气资料MERRA-2数据进行了尺度分离研究。结果表明，EMD适用于对数据的初步分离，得到大致尺度范围；MWT适合更高精度的需求；两种方法都能达到多种尺度的分离效果。研究结果可促进临近空间多尺度数据同化技术的发展，以提高模型预报的准确性。

Near space typically refers to the region of Earth's atmosphere located 20 to 200 km above the surface, which is a new strategic frontier for national security in space. The atmospheric conditions in near space not only vary with altitude, geographic latitude, longitude, and season, but also exhibit complex changes across various temporal and spatial scales. Due to the diversity of near space exploration methods, the corresponding detection instruments and their principles vary, resulting in measurements that reflect multi-temporal and spatial characteristics. Addressing the need for multiscale data assimilation in near space, developing appropriate scale separation methods based on specific data is an urgent issue. This thesis employs two separation techniques, Multiscale Window Transform (MWT) and Empirical Mode Decomposition (EMD), to conduct scale separation studies on the MERRA-2 reanalysis atmospheric data of near space. The results indicate that EMD is suitable for preliminary data separation, providing a rough scale range; MWT is more suited for higher precision requirements; both methods achieve separation effects across multiple scales. These findings can promote the development of multiscale data assimilation techniques in near space, thereby enhancing the accuracy of model forecasts.