云底高度的时空特征对分析云系演变有重要作用，其多源信息的融合分析方法尚不完备。当前广泛采用的逐步订正方案，其核心思想是寻求多源信息的最大一致性。其难点在于：一致性规则难以获得；阈值不确定性大。这也是获得高时空分辨率的准确云底高度场的瓶颈。本研究拟采用随机森林算法（RF）将数值模式预报、卫星辐射数据和多种云的地面自动观测结合，建立一种数据驱动（兼顾物理机制）的融合分析算法，进而改进经典三维云分析模式LAPS中的云底高度融合精度。该方法灵活、快速，对云底高度进行分析融合的同时，也利用显式学习规则和变量重要性度量来研究三维环境场对云底高度的作用机理，探索多源信息的一致性关系。本研究是三维云分析模式向人工智能转化的创新性探索，为三维云场的精确预报和应用提供依据和支持。

The temporal and spatial characteristics of cloud base height have an important effect on the evolution of cloud system, and the fusion analysis method of multi-source information is not complete yet. The core idea of current broadly adopted step-by-step program is to seek the greatest consistency of multi-source information. The difficulty is that the consistency rules are difficult to obtain and the setting of threshold is very uncertain. This is also the bottleneck of obtaining an accurate cloud base height field with high temporal resolution. In this study, we use random forests algorithm (RF) to combine numerical model forecasting, satellite radiation data and ground observation of various clouds to establish a data-driven ( with physical mechanism) fusion analysis algorithm, and then improve the classical three-dimensional cloud analysis model LAPS to get high degree of fusion accuracy. The method is flexible and fast. The physical mechanism of the three-dimensional environment field to cloud base height, and the relationships among multi-source information are also studied by using the explicit learning rules and the variable importance measure. This study is an innovative exploration of 3D cloud analysis model to artificial intelligence transformation, which provides the basis and support for the accurate prediction and application of 3D cloud field.