地理視覺化探索分析是一種利用地理空間的互動展示來探索多維度時空資料的方法，自組織映射圖網路的視覺化技術是目前最有效的方法，但仍存在部分問題，如資料的分類方法以及類神經網路的大小需依賴研究者主觀判斷。本研究提出「先分解再合併」的方法，整合群聚演算法逐步合併自組織映射圖網路神經元，達到動態調整分類數量的方式，配合其他視覺化及空間地圖的展示，來探索資料隱含的關係與空間特徵。本研究使用臺北市2000年早晚尖峰時間車流量資料進行分析，發現未經探索的原始資料在相關係數以及簡單迴歸的判定係數上無法找出資料的關連（r＝0.1682 to 0.1941, R^2＝0.0283 to 0.0377）。經由地理視覺化探索分析後，可在資料當中找到兩個主要的集合。此兩集合不但在相關係數及判定係數上均有顯著的改善（r＝0.4843 to 0.7500, R^2＝0.2345 to 0.5626），同時在空間上也可區分為兩個不同的分區，顯示地理視覺化探索分析對於確認地理資料的空間特性有所助益。研究結果顯示本研究提出的方法容易觀測到空間或是資料的關連特徵，未來可應用本研究的方法分析其他的時空資料。
Geovisualization is a method to explore spatial knowledge hidden in multidimensional geographic and temporal data via interactions with map and graph. The visualization of Self-Organizing Map (SOM) is one of the most effective methods but still has issues of what size the network should be. This study proposed a novel method called ”Divide and Regroup”, to integrate clustering analysis and SOM neurons interactively and dynamically for finding hidden data relations and spatial patterns. We chose rush-hour traffic flow data of Taipei City in the year 2000 as a case to demonstrate the effectiveness of this approach. The correlation coefficients and coefficients of determination of the unclassified data were low (r=0.1682 to 0.1941, R^2=0.0283 to 0.0377). Two major groups of traffic flow data were recognized using the Geovisualization approach. The correlation coefficients and coefficients of determination of the classified data have improved (r=0.4843 to 0.7500, R^2=0.2345 to 0.5626). Furthermore, these two groups showed different spatial patterns indicating that the Geovisualization approach is useful for identifying spatial characteristics hidden in geographic data. The results demonstrated the effectiveness of the novel method for Geovisualization.
