Coherent structures are ubiquitous in spatiotemporal systems far from equilibrium. These structures provide concise descriptions of the system and its dynamics, and there is often interest in these structures themselves. We present a novel method for automated detection and labeling of coherent structures that can be applied universally to spatiotemporal systems with local dynamics. Adapted from its original development in strictly temporal systems, computational mechanics is a method of inferring a hidden-state model from data which extracts and quantifies structure in the data. Significantly, the structures discovered by computational mechanics are not always readily identifiable from the raw data. Computational mechanics has been successfully applied to identify known structures in cellular automata. Here, we demonstrate the method on two-dimensional DNS hydrodynamic models of vortex shedding. Ultimately, we hope to analyze large-scale climate data for automated detection of extreme weather systems and other (possibly hidden) climatological structures. Current capabilities of the Computational Mechanics in Python (CMPy) software package allows for data parallelization using Berkeley Lab’s Cori system, but significant scalability challenges remain to reach more realistic climate simulations. High-resolution simulations produce terabyte-size intermediates that require fully-distributed execution.