Spatio-temporal Mining of Solar-Terrestrial Satellite Observational Data for Distributed Data System

Rie Honda, Kochi University, Japan

abstract: A large amount of physical data on the Earth's magnetosphere has been obtained by spacecrafts and accumulated on the database. These data consists of a various type of spatio-temporal data such as magnetic field, electric field, particle information, which are sampled at different time intervals by individual sensors. Generally, each type of data has been stored at a geographically distant site and thus researchers had difficulties in conducting integrated analysis by utilizing multiple attributes effectively.

Recently, Murata et al. has developed Solar-Terrestrial data Analysis and Reference System (STARS): the distributed database system that connects these data sites on the internet by using Web service. STARS provides a seamless workbench for researchers, thus it facilitates the advanced analysis such as data mining.

In this study, we examined implementation of data mining as the global process on this STARS workbench. We developed the process which re-samples multiple attribute data at the specific time points over the internet by utilizing STARS. Clustering of observational data by Kohonen's self-organizing map was conducted with the retrieved data, aiming to create discriminator of the magnetic domain (e.g., magnetosheath, tail lobe) where the spacecraft exists. We further examined the construction of statistical model of these domains in the Solar-Earth fixed space, based on the clustering result.


Keywords: Data mining, clustering, SOM, Solar-Terrestrial, magnetosphere