應用遙測與地文資料以共用鄰域法進行陳有蘭溪上游集水區莫拉克颱風新增崩塌地環境敏感度分類之研究

A Shared Near Neighbours Approach to the Landslide Classification Induced by Typhoon Morakot Using Remote Sensing and Geological Data at Upstream of the Chen-Yu-Lan River Watershed

衛強(Chiang Wei) ； 葉惠中(Hui-Chung Yeh) ； 陳彥璋(Yen-Chang Chen) ； 曹崇銘(Tsung-Ming Tsao)

《國立臺灣大學生物資源暨農學院實驗林研究報告》 28卷4期 (2014/12) Pp. 217-237

https://doi.org/10.6542/EFNTU.2014.28(4).1

摘要

本研究以遙測及地文資料應用共用鄰域法（Shared Near Neighbours, SNN）分析臺灣中部臺大實驗林轄區內陳有蘭溪上游集水區於莫拉克颱風後所增加崩塌地環境敏感度分類之研究。經福衛二號衛星影像前後期影像判釋結果，增加之崩塌地面積達624.5公頃，佔臺大實驗林（以下簡稱本處）轄區面積1.9%；以面積1公頃及22°（40%）坡度門檻值進行篩選，較均勻之崩塌地共有51處，其中位於陳有蘭溪上游本處第25至42林班者計有48處；以衛星光譜影像分析結果與航照圖比對相當吻合，顯示以多光譜遙測衛星影像進行長期或突發性災害評估之優越性與重要性。本研究以水土保持技術規範所使用之簡確評估法評估崩塌地之環境敏感度分級為依據，以共用鄰域法對地文因子進行環境敏感度分析，研究結果顯示共用鄰域法分類之整體正確率為60.4%，與以傳統最大概似法或貝氏分類法分類之整體正確率僅為39.6%有明顯之提昇。後續研究仍需進一步檢討相關崩塌地其他地文參數、人為影響因素及水文因子之必要性以提高分類精度。

關鍵字

遙測與地文資料 ； 共用鄰域法 ； 崩塌地 ； 環境敏感度分類

並列摘要

This study using remote sensing data analyzed the environmental sensibility of the upstream landslide areas of the Chen-Yu-Lan watershed in the National Taiwan University Experimental Forest (NTUEF) after the Typhoon Morakot in 2009 by the Shared Near Neighbors(SNN) technique. There were a total of 624.5 ha landslide areas delineated by FORMOSAT-II imagery, which accounted for 1.9% of total area. 51 landslide sites based on the threshold of 22° slope and 1 hectare area were located using satellite imagery together with aerial photo and GIS related coverage. Among these 51 landslide areas, 48 were located in the upstream of the Chen-Yu-Lan watershed. Comparing with aerial photo after the Typhoon Morakot, landslide sites were correctly identified and indicating superiority and importance of multispectral satellite imagery for monitoring long-term event. Furthermore, the Simplicity Method was used to extract geological data for classifying degree of environmental risk for each landslide. These geological data was then set as the variables for classification using SNN approach compared with the classification done by Simplicity Method. The result showed that the precision of classification using SNN is up to 60.4%. Compared with the accuracy of 39.5% obtained from traditional Maximum Likelihood Classifier or Bayesian Classifier, the precision has significantly increased. More geological data, anthropogenic influence and hydrological factors may be necessary for clarifying landside area. Other clustering methods combined with SNN are needed to achieve more accurate results.

並列關鍵字

remote sensing and geological data ； Shared Near Neighbours (SNN) ； landslide ； classification of environmental sensibility