An event-based model for discovering frequent spatio-temporal patterns: application to the study of tropical cyclones

Discovering frequent spatio-temporal patterns from dynamic phenomena is challenging due to continuous changes affecting both spatial and semantic dimensions. Existing frequent pattern mining methods often simplify complex phenomena into time-stamped point events regardless of their lifetime and spatio-temporal coverage. This article introduces an event-based data model that extracts events from continuous geographical data. A hierarchical framework models the lifecycles of events using raster time-series datasets. Events are then mapped to Discrete Spatio-Temporal Evolution Patterns (D-STEP) and evaluated using complementary local spatial parameters that continuously characterize their movement, density, coverage, and intensity over their lifetime at the local spatial level. Evaluated through a case study of tropical cyclones in the Western North Pacific Ocean, this model shows promising potential in identifying frequent spatio-temporal patterns and trends. The model preserves the primary data and can be generalized to other Earth-related phenomena.