Research on a systematic analysis of information on GPS array data

从1996年到1997年，我们分析了日本各地1000个站点的GPS数据。具体来说，在数据质量方面对数据集的详细调查限制了我们使用其中的614个数据点。我们在时间-空间域上进行贝叶斯平滑处理，使我们能够自动提取平均结构，并尝试基于得到的结果对平均结构进行建模。实际上，我们采用了基于贝叶斯方法的非参数平滑，并在每个时间和空间域交替进行残差分析。最后，我们将各站点GPS数据的变化分解为三个分量：时变分量、空变分量和其他分量。我们在国际会议上展示了这一成果，得到了与会人员的积极响应。

We analyzed GPS data at 1,000 stations covering all over Japan for two years frorn 1996 through 1997. Specifically, an detailed investigation on data sets in terrns of data quality limited us to use 614 data points among them. We performed a Bayesian smoothing procedure in time-spatial domain which allows us to extract the mean structure automatically, and tried to model the mean structure based on the result obtained. Actually, we applied the non-parametric smoothing based on the Bayesian approach and conducted the residual analysis alternatively in each of time- and spatial domains. Finally we decomposed the variations of GPS data for each station into three components : time-dependent, spatial-dependent, and other components. We presented the result at the international meeting, and obtain the positive response from the attendants.