Twitter, as the most popular social media platform, has made a great revolution in producing real-time user-generated data. This research aims to propose a method to extract the latent spatial pattern from geotagged tweets. We take both spatial autocorrelation and spatial heterogeneity into account while revealing the underlying pattern from geotagged tweets. Moreover, the textual similarity is considered to extract spatial-textual clusters. The method was implemented and tested during hurricane Dorian on the east coast of the U.S. The results proved the superiority of the proposed method against Moran’s Index and VDBSCAN algorithms in extracting clusters with various densities.