Identifying the sources of sub-synchronous oscillations (SSOs) in power systems with increasing shares of inverter-based resources (IBRs) is challenging when representative models and detailed controller information are unavailable. This paper presents a purely data-driven method for identifying dominant oscillation contributors using time-synchronised phasor measurement unit (PMU) data. A short-time Fourier transform (STFT)-based detector separates events into stages based on persistent changes in the dominant oscillatory frequency. Within each stage, an EDMD analysis window is automatically selected, after which Extended Dynamic Mode Decomposition (EDMD), grounded in Koopman operator theory, is applied to estimate oscillatory modes and quantify participation at individual PMU locations.

The method is evaluated using two real IBR-induced SSO events in Australia and Denmark, featuring an approximately stationary 21 Hz oscillation and an evolving oscillation around 2.8 Hz, respectively. The dominant contributing locations are consistent with independent post-event investigations by AEMO and Energinet. Comparisons with measured spectral magnitudes show that locations with the largest oscillation magnitudes do not necessarily have the highest EDMD participation factors. Sensitivity analyses provide evidence that the method remains informative under incomplete measurement coverage. These findings support practical post-event oscillation source investigation and future extension towards real-time implementation.