SPOT衛星影像應用於德基水庫集水區歷年土地利用變遷之分析

遙測技術爲獲得集水區上地利用情形之最迅速方法。本文遂利用此技術，以SPOT衛星影像並配合歷年現地土地利用調查資料分析德基水庫集水區92年至96年之土地利用及土地變遷偵測之情形，以供德基水庫防災及治理所需。於土地利用分類方面，本文採用階段式分類法，第一階段以光譜特徵區分大類別，第二階段加入紋理特徵輔助分類，將農地與裸露地於大類別中取得再進行小類別之分類。由分類結果顯示光普特徵配合紋理特徵進行分類，可反應出較佳之分類結果。於變遷偵測方面，利用主成分轉換並以第一主成分作爲判釋的特徵，以判定前期分類之結果是否於後期發生變動。本文應用條件機率的概念建立信賴區間，提供信心水準作爲變遷判釋的依據。此法最大之特點爲改進傳統利用試誤法或經驗法決定變遷門檻值過於主觀之缺點，且賦予變遷門槛值統計意義，提升變遷偵測準確度。由歷年土地利用分類結果配合德基水庫水位與含水量資訊，顯示水體及裸露地面積的增加與地颱風或暴雨事件發生有關。

關鍵字

衛星遙測；土地利用；變遷偵測

並列摘要

Remote sensing images and technologies have been widely applied to environmental monitoring, especially in landuse/landcover classification and change detection. In this paper we present a two-stage Bayesian classification approach which is capable of achieving high classification accuracy. In the first stage, spectral features were used for classification of major landcover classes including water body, forest, grassland, agriculture, and bare land. In second stage, textural features were used for more detailed classification of hare land and agriculture. The bare land class was further classified into two subclasses-bare soil and built-up. The agriculture land was classified into orchard, vegetation garden, and tea plantation subclasses. For change detection, a hypothesis-test-based multispectral algorithm which involved principal component analysis and conditional probability distribution was adopted for landcover change detection in the Techi reservoir watershed. Under the null hypothesis of no change, the 95% confidence intervals of individual classes could be established. The results showed that the proposed change detection method can achieve high accuracies. It also showed increase in water body and bare land was related to heavy rains and typhoon events.

並列關鍵字

Satellite remote sensing ； Landuse/landcover ； Change detection

國際替代計量

SPOT衛星影像應用於德基水庫集水區歷年土地利用變遷之分析

全文下載

主題瀏覽