Combining Landform Thematic Layer and Object-Based Image Analysis to Map the Surface Features of Mountainous Flood Plain and Surrounding Areas

2009年8月的莫拉克颱風事件，對台灣南部地區帶來累積雨量超過2,000毫米的極端降雨事件，造成高屏溪流域受損嚴重，縱觀災害區域，河道兩側坡地多處側蝕崩塌，流路變遷造成了橫跨的道路與橋樑中斷，沿岸高灘地及河階地聚落嚴重水患以及堤岸毀損，支流土砂所形成的沖積扇更阻斷了主河道形成堰塞湖。這些災情都突顯出對河道變遷及其周遭區域細部地形地物瞭解的重要性。 遙測影像判釋技術可在短時間內有效且快速的獲取大面積的空間資訊，近年來衛星影像的解析度及波段數大幅改善，使得其在天然災害防治的應用更加實務化，應用於探討河道周圍的地物特徵是十分適合而值得發展的技術。本研究採用物件式的影像分析軟體Definiens Developer 7.0，對荖濃溪流域寶來與六龜地區的災後QuickBird多波段高解析度衛星影像進行判釋。物件式的影像分析以均質化的影像物件區塊為分析單元，突顯了區塊的形狀、紋理以及區塊間的相互關係等特性，透過一連串合理的分類條件以及流程，將能進行類似人工判釋的快速半自動化地物判釋。除了衛星影像的光譜資訊外，運用數值地形模型DTM來進行空間地形分析，產製出特定的地貌主題圖層，能更進一步輔助判釋，協助區分影像光譜中容易混淆的類別如崩塌地與河床，顯著的改善判釋成果之正確性，並能夠定量地劃分出具有地形意涵的地貌區位範圍，如河階地及灘地。 本研究提供了一個定量、快速、廣域且相對自動化的氾濫原地物分析流程，將氾濫原分類出主流路、次流路、沙洲、高灘地與較低位河階、較高位河階、沖積扇、周邊崩塌地及鄰近的人工構造物等八種地物類別，協助我們瞭解河川地形的演育發展以及人與自然的互動。判釋成果將可作為近河的人工構造物之安全評估、災害防治與救災規劃以及土地利用規劃的參考。The Typhoon Morakot on August 2009 brought more than 2,000 mm of cumulative rainfall in southern Taiwan, the extreme rainfall event caused serious damage to the Kaoping River basin. The losses were mostly blamed on the landslides along sides of the river, and shifting of the watercourse even led to the failure of roads and bridges, as well as flooding and levees damage happened around the villages on alluvial plain and river terraces. Alluvial fans resulted from debris flow of stream feeders blocked the main watercourse and debris dam was even formed and collapsed. These disasters have highlighted the importance of mapping the watercourse alteration, surface features of flood plain area and artificial structures soon after the catastrophic typhoon event for natural hazard mitigation. Interpretation of remote sensing images is an efficient approach to acquire spatial information for vast areas, therefore making it suitable for the interpretation of terrain and surface features near the vast flood plain areas in a short term. The object-oriented image analysis program (Definiens Developer 7.0) and multi-band high resolution satellite images (QuickBird) were utilized to interpret the flood plain features from Liouguei to Baolai of the Kaoping River basin after Typhoon Morakot. Object-based image analysis split an image into meaningful homogenized segments for obtaining information such as shapes, textures, area and the mutual relationships of segments. After segmentation, we develop classification criteria to classify the image segments semi-automatically. Specific “landform thematic layers” produced by digital terrain models (DTM) are also employed along with the above process. These layers are especially helpful in differentiating some confusing categories in the spectrum analysis with improved accuracy, such as landslide and riverbed. The river terraces and alluvial plain, which are with significant meaning in terrain, can also be defined qualitatively. Finally, a fast, quantitative, large-scale and semi-automated image interpretation process is proposed. The result including river channel, vegetation, sandbar, alluvial plain, river terrace, alluvial fan, landslide, and the nearby artificial structures of mountainous flood plain and surrounding areas. It can be used as references for safety evaluation of riverside engineering structures, disaster prevention and mitigation, and even future land-use planning