Landslides can persist within the landscape for years to decades. Multi-temporal landslide inventories have been used to provide insight into how overall landslide hazard can evolve through time, often following a disturbance event such as an earthquake. However, these inventories can rarely be used to infer how landslide hazard evolves on an individual landslide scale because they (1) only map new landslides or use the previous epoch to map existing landslides through time, or because (2) they map landslides independently through time. In both cases, it is difficult or impossible to cross-reference between epochs for datasets consisting of several thousand landslides, as is typical following a large earthquake. In this study, we present a novel semi-automated method to cluster related landslide polygons across different epochs based on the geometric and topographic properties of individual polygons. We apply this method to a published multi-temporal inventory for the 2015 Gorkha Earthquake in Nepal, containing over 200 000 independently mapped landslide polygons between 2014 and 2020. Using this method, we automatically link the trajectory of 48% of landslides within the inventory through time, thus reducing the amount of time spent manually digitizing. By linking individual landslides through time, we can explore how individual landslides respond to an earthquake in time and space. We use our results to infer how long existing landslides, in the form of clusters, persist in the five years following the Gorkha earthquake, and thus, for the first time, assess how the hazard posed by both new and existing landslides changes. We also discuss the challenges when generating manually-mapped multi-temporal landslide inventories and provide suggestions for future applications.