Exploring marine abnormal association pattern from long-term multiple remote sensing images is a key issue in the context of global change. As traditional spatiotemporal analysis has great challenge to obtain this pattern, we propose a raster-oriented mining framework to address such issue, which consists of three components, i.e. data pretreatment component, mining component and visualization component. Data pretreatment component is to extract the abnormal information and discretize the variations into quantitative levels, and to construct the mining transaction table with space, time, marine elements and their variation types.
Mining component is to design an efficient mining algorithm to find the frequent patterns (candidate patterns) and identify the meaningful patterns. Visualization component is to visualize the discovered spatiotemporal association patterns with a group of thematic maps by a recursive strategy from (k-1)-dimension to k dimension. Finally, a case study over Pacific Ocean was shown to demonstrate the effectiveness and the efficiency of the proposed mining framework.