Accurate characterization of salt caverns is a prerequisite for evaluating the feasibility of Compressed Air Energy Storage (CAES), yet high-noise urban environments pose challenges for conventional active-source geophysical methods. To address this, our study utilized Microtremor Array Measurements (MAMs) based on the Spatial Autocorrelation (SPAC) method to characterize deep salt caverns for a CAES project in Yunnan, China. The results reveal dissolution features and morphological asymmetry within the caverns: the upper sections are conical and brine-filled, whereas the lower sections are cylindrical, containing both brine and sediment. The identified caverns range from 45 to 190 m in height and 30–120 m in width, with an estimated total volume of approximately 4.6 × 106 m3. Additionally, localized wave velocity anomalies helped map connecting channels between four brine extraction wells, which were formed by differential dissolution and piping. By delineating cavern morphology and identifying fracture zones, joint networks, and hydraulic conduits that could potentially compromise sealing integrity, the MAMs approach provides critical geophysical insights for the safety evaluation and engineering design of CAES gas storage under complex geological conditions.