An Approach to Discovering Spatial–Temporal Patterns in Geographical Processes

Abstract

Spatial data mining focuses on searching rules of the geographical statement, the structures of distribution and the spatial patterns of phenomena. However, many methods ignore the temporal information, thus, limited results describing the statement of spatial phenomena. This paper focuses on developing a mining method which directly detects spatial temporal association rules hidden in the geographical process. Through such approach, geographical process can be extracted as a particle which exists in spatial temporal-attribute dimensions. By setting customized fixed-window, geographical process in one time interval is organized as a record with attribute value and spatial orientation change. Spatial temporal association rules can be found in geographic process mining table.

$$ \left[ {{\hbox{TimeInterva}}{{\hbox{l}}_i},{\hbox{MovingDirectio}}{{\hbox{n}}_m},{\hbox{P}}} \right]{ } \Rightarrow \left[ {{\hbox{TimeInterva}}{{\hbox{l}}_j},{\hbox{MovingDirectio}}{{\hbox{n}}_n},{\hbox{Q}}} \right] $$

To verify this mining approach, it is applied on AVHRR MCSST thermal data for extracting Indo-Pacific warm pools frequent movement patterns. The raw data provided by PO.DAAC, whose time spans of 20years from 1981 to 2000 with 7days time particle, has been used to mining spatial temporal association rules. In the experiment, we extract warm pool within 30N 30S, 100E 140W and use 28C as temperature threshold. After which Warm Pool s geographical process table is established so as to describe the variation of warm pool in spatial temporal-attribute dimension. In the mining process, 18 spatialtemporal process frequent models can be found by setting minimal support threshold at 10% and confidence threshold at 60%. The result shows such a methodology can mine complicated spatial temporal rules in realistic data. At the same time, the mining result of warm pool s frequent movement patterns may provide reference for oceanographers.