Localized Spatial Association: A Case Study for Understanding Vegetation Successions in a Typical Grassland Ecosystem

Abstract

Spatial heterogeneity exists widely in geographical space. We here suggest that spatial association mining should consider spatial variations when designing knowledge-discovering models and applying the models in geographical studies. A Quadtree-based framework was proposed to mine localized spatial associations. Unlike many other approaches, the novelty of this Quadtree-based frame is its suitability in finding strong spatial association rules that are valid in smaller areal patches or “hot spot” area rather than the whole region, and therefore provides domain experts an insight to explore further the associations among different geographical phenomena. The principle underlying the Quadtree-based algorithm is that it employs a Quadtree data structure to explore the multi-level nodes (each representing a patch at that level), recursively. This recursive process is used to check whether the explored nodes satisfy predefined criteria (including support and confidence threshold for association rules, and event density and minimum area for each node), denoting a strong association. Practical application in an ecology study proved that, compared to traditional global association mining, the proposed model and algorithm for mining localized association rules are more meaningful under spatially heterogeneous environment.