Rockfalls generate seismic signals of great utility to detect and monitor rockfall activity. Event locations can be estimated based on arrival times, amplitudes or polarization of the seismic signal. However, strong surface topography variations can significantly influence seismic wave propagation and hence compromise the localization. Here we specifically use the imprint of topography on the seismic signal to better constrain the source location. Seismic impulse responses are predicted using Spectral Element based simulation of 3D wave propagation in realistic geological medium. Subsequently, rockfalls are localized by minimizing the misfit between simulated and observed inter-station energy ratios. We test the method on rockfalls at Dolomieu crater, Piton de la Fournaise volcano, La Réunion. Both single boulder impacts and distributed granular flows are successfully localized. The full rockfall trajectories is resolved, analyzing the signals in sliding time windows. Results from the highest frequency band (here 13-17 Hz) yield the best spatial resolution, making it possible to distinguish detachment positions less than 100 m apart. Taking into account surface topography allows to employ all signal components, enhancing localization especially when few stations are available. Limitations and noise robustness of the localization method are assessed using synthetic signals. Precise representation of the topography is shown to control the localization resolution, which is not significantly affected by the assumed impact direction or a globally faster seismic velocity model. We propose that the method can improve the monitoring of rockfall activity in real time once a simulated database for the region of interest is created.