An Automatic Method for Discovering Significant Regional Spatial Colocation Patterns

Discovery of regional spatial colocation patterns facilities understanding of the spatial dependency of different spatial features at the regional scale. However, two challenges remain:①appropriate thresholds for prevalence measures are difficult to specify without prior knowledge; and ②natural localities of regional spatial colocation patterns with different densities and shapes can hardly be automatically detected. On that account, an automatic method for discovering significant regional spatial colocation patterns is proposed in this paper. First, a nonparametric statistical model is developed to test for significance of spatial colocation patterns. Then, an adaptive spatial clustering method is modified to detect hot spots of each candidate regional spatial colocation pattern that is not identified as a statistically significant spatial colocation pattern at the global scale. At last, all hot spots are iteratively expanded until no larger statistically significant localities can be detected. Comparison between this automatic method and an existing method is carried out with both simulated and ecological datasets. Experiments show that the regional spatial colocation patterns can be effectively detected with less subjectivity and prior knowledge by this automatic method.