空間資料庫知識探索之研究─以集集大地震引致之山崩為例

Abstract

隨著資訊科技的進步,資料的收集紛紛邁入自動化與電腦化，造成空間資料的迅速累積，衛星遙測影像、GPS的資料收集、行動通訊裝備以及各種與位置相關的交易(Transaction)，提供我們大量具有空間參考的資料，坐擁如此龐大的資料，其中包含了許多寶貴的資訊與知識，如何從這些資料中提煉出有價值的知識，是當前面臨的一大課題。目前越來越受重視的資料探勘技術可以協助決策者從大量資料中找出有價值的知識，但是大部分的應用仍只限於屬性資料的分析，對於空間資料的處理分析少有著墨，而地理資訊系統雖然具有強大的空間資料處理能力，但缺乏對於屬性資料的進階處理。 有鑒於台灣地處地震帶，部分地區每年均要歷經多次的地震以及周期性的大地震，往往帶來嚴重的山崩以及土石鬆動的現象，若再歷經豪雨後，將帶來嚴重的土石流，造成生命財產的重大損失。民國88年9月21日於台灣中部地區發生芮氏規模7.3的地震，此次強震造成2000多人死亡，難以估計的財產損失，以及數量龐大的坡地災害。而在此次強震中亦獲得大量的坡地破壞資料及地震記錄，可供專家學者進行地震對山崩影響的研究，得以對地震引致山崩的行為有進一步的認識。 本研究嘗試以九二一大地震所累積的大批地理資料，結合地理資訊系統空間資料分析與處理的功能，建立資料倉儲作為資料探勘的基礎，分別以預測型資料探勘技術與描述型資料探技術，對地震山崩空間資料庫作全面性的知識探索。其中預測型資料探勘技術包含有類神經網路模式、決策樹模式、案例式概念學習及貝氏分類器等多個具互補性之資料探勘技術，分別探討各模式效能並評估其整合預測模式的建立。而描述型資料探勘則包含有線上分析處理(OLAP)、關聯法則及等級相關分析，用以找出引致地震山崩的推論法則及關聯樣式(association pattern)。所使用的資料素材則包含了向量式以及網格式資料，藉以探討各種資料探勘技術在空間資料庫知識發掘的適用性以及其整合研究，獲得良好的結果。期望透過此研究了解 引發地震山崩的機制，建立山崩災害機率的潛感圖以及知識庫和模式庫，並且探討各種資料探勘技術在空間資料庫知識探索上之應用性，提供防災決策支援上的參考。

With the progress of information science and technology, the data collection march toward the automation and computerization. The fast accumulation of the spatial data, such as satellite image, GPS data recorder, mobile communication equipment and various kinds of location-based transaction, offer a large number of geo-referenced data. There include a lot of valuable information and knowledge among these data. How to refine out valuable knowledge from these data is a great subject faced at present. Data mining can help the policymaker to find out valuable knowledge from a large amount of data, but most application are still only limited to the analysis of the attribute data. It is difficult to deal with the spatial data. And though geographical information system is powerful in analyzing spatial data, but it lack the advanced ability to deal with sophisticated attribute data analysis. Because Taiwan is located in the earthquake zone, some areas go through a lot of earthquakes and periodic heavy earthquakes every year and cause serious landslide. If after going through the torrential rain again, will bring serious debris flow and cause great losses of the lives and properties. An earthquake of magnitude 7.3 on Richter scale occurred in the middle region of Taiwan on September 21, 1999. This earthquake caused more than two thousand people died, severe property loss, and a large number of landslides. A large number of landslide data and earthquake strong motion records were obtained for the experts and scholars to carry on the research of landslide influence of the earthquake. This research collects data of landslides triggered by Chi-Chi earthquake, and with the powerful data-processing function and spatial analysis ability of Geographic Information System (GIS), Data Mining modeling, the basic data of research region, and Chi-Chi earthquake strong motion records to establish the landslide database and data warehouse. A new strategy, which combines several models based on different philosophy, not only reduce the uncertainty of predictive modeling, but also improve the accuracy. In our study, a Decision Tree, Artificial Neural Network, Bayes Classfier, and Exemplar-based Concept Learning were individually applied to a spatial data warehouse. The result of each model and two kinds of modeling-integration methods, including horizontal integration and vertical integration, were then evaluated. Furthermore, the spatial association patterns are typically not encoded in database, but are rather embedded within the spatial framework of the geo-referenced data. The analysis of the association pattern between the occurrence of Chi-Chi earthquake-induced landslide and background environmental characteristics is used as a case study to demonstrate the potential of spatial data mining techniques, like OLAP, association rule mining and Spearman rank correlation. With the analysis results, we derived a suspecious potential map and build the knowledge base and model base. Verification proofed the result to be good. So the analysis methods mentioned by this research are suitable for the risk assessment of landslide hazard triggered by earthquake and can be used as the tool for disaster mitigation decision support.