Canonical correlation analysis (CCA) has been applied in data-driven distributed process monitoring. However, conventional CCA monitoring heavily relies on data covariance that can be easily corrupted by outliers and is sensitive to data collinearity. This paper proposes a neighborhood stable CA (NSCA)-based local-global modeling approach to achieve efficient robust monitoring of multiunit processes with possible outliers and data collinearity. First, the training data of a local unit are decomposed into three parts, namely, low-rank dominant, sparse gross error, and random noise parts by considering the correlation constraints from neighboring units. Second, stable dominant subspaces that capture dominant variations of a local unit and neighboring units are constructed. The relationship of the local unit and neighboring units is characterized by performing CA within the two dominant stable subspaces. Finally, local-global monitoring statistics are constructed based on which the process status and fault characteristics can be identified. The effectiveness of the proposed NSCA-based local-global monitoring is verified through experimental studies on a numerical example and a lab-scale distillation process.