Aerial-Ground Person Re-Identification (AGPReID) aims to match pedestrians from drastically different views. Part-based person re-identification methods have proven effective in capturing fine-grained details which are critical for better discrimination. However, applying part-based methods to AGPReID faces several challenges. First, the large view gap makes it difficult to properly align the feature of the parts. Second, view variation often causes some part features to be missing or invisible across views. To address these challenges, we propose a Context- and Visibility-Aware Part Learning (CVPL) framework which consists of Context-Aware Part Learning (CPL) and Visibility-Aware Part Learning (VPL). Specifically, for CPL, we observe that corresponding parts usually share similar context information. Therefore, we leverage such context to aid in the part feature extraction and alignment. Additionally, for VPL, we use the similarity values obtained during part extraction to estimate the visibility of each part, and further incorporate visibility into the training loss and part feature distance computation. Experiments on the CARGO and AG-ReID datasets demonstrate the effectiveness of our method.