Mining geographic episode association patterns of abnormal events in global earth science data

Abnormal events in earth science have great influence on both the natural environment and the human society. Finding association patterns among these events has great significance. Because data in earth science has characteristics of mass, high dimension, spatial autocorrelation and time delay, existing mining technologies cannot be directly used on it. We propose a RSNN (range-based searching nearest neighbors) spatial clustering algorithm to reduce the data size and autocorrelation. Based on the clustered data, we propose a GEAM (geographic episode association pattern mining) algorithm which can deal with events time lags and find interesting patterns with specific constraints, to mine the association patterns. We carried out experiments on global climate datasets and found many interesting association patterns. Some of the patterns are coincident with known knowledge in climate science, which indicates the correctness and feasibilities of our methods, and the others are unknown to us before, which will give new information to this research field.