Adaptive detection of statistically significant regional spatial co-location patterns

Abstract

Regional spatial co-location patterns refer to subsets of spatial features that often co-occur in close geographical proximity in certain localities of space. Discovering regional spatial co-location patterns is still very challenging because it is difficult to specify appropriate thresholds for prevalence measures without prior knowledge and to detect natural localities of regional spatial co-location patterns automatically. On that account, an adaptive method is proposed in this study. First, a non-parametric significance test is constructed to evaluate the prevalence of spatial co-location patterns. Then, an adaptive pattern clustering approach is developed to detect hotspots of each candidate regional spatial co-location pattern. Finally, all statistically significant regional spatial co-location patterns and their localities are detected by iteratively expanding these hotspots. Comparisons between this adaptive method and two state-of-the-art methods are carried out with both simulated and ecological datasets (i.e. a wetland species dataset in northeast China). Experiments show that the proposed adaptive method allows detecting regional spatial co-location patterns effectively and with less prior knowledge than the state-of-the-art methods.