| 摘要: | 植物生長空間分布與氣候及環境息息相關，評估森林資源現狀、分布及健康狀態是當今關注的重點。在全球氣候變遷下，台灣自然災害頻率與強度的增加影響到植群及生態系的變化，植群空間分布會因應各種現象而有所改變。本研究主要目的是探討曾文水庫集水區各種植群在地形、氣候、衛星影像光譜上的特徵，希望建立氣象與環境因子與植群分布之關聯性。衛星影像記錄著多時段的地表資訊，不同植物就有不同的波譜特性，森林是由植群所組成，植群是一個區域中生活在同環境下的植物集合體，並與環境有某種程度的相關性；在地理區域中，植群類型可以當作生態系類型的代表，每個森林群落都有它特殊的生長條件，了解這些資訊有助於幫助集水區對於植群現況的推估及輔助監測森林資源。本研究經過文獻探討影響植群的相關因子，分別使用地形因子、氣象因子、光譜特徵並與各種類植群製作圖表來解釋相關情形，結果說明僅以單一因子難以區分或說明所有植群的空間特徵分布。因此嘗試使用多種資料進行植群分類，包含最大概似法及決策樹分類。結果顯示僅以最大概似法監督式分類結果整體精度為58.5%，說明光譜資訊不足以滿足於植群分類。而使用決策樹法將地形、植生指數、氣象因素分類結果，經過事前剪枝及事後修剪，以CART分類效果較佳正確率為76.9%，分類結果也清楚表示分類規則及因子重要性。因子重要性從高到低分別為海拔高度、全年溫差、夏季溫度、全年雨量。未來研究可以先透過最大概似法分類後結合決策樹的方法，以高度、氣候因子預測及回推出植群空間變化，以了解集水區內植群長期的生長時序變化。
The spatial distribution of plants is related to the climate and the environment. To assess the current situation, distribution and health status of forest resources are important issues in forest managements. Because of climate change, the increase of the natural disaster frequency and the change of the intensity of the natural disasters in Taiwan affect the changes of the vegetation and the ecosystem. This study explores the relation between environment factors and spectral characteristics of various plant populations in Tseng-Wen reservoir watershed.
In the geographical area, the type of vegetation can be used as a representative of the ecosystem type; so we must first understand the plant population status and climate status of forest land, each forest community has its special growth conditions. Understand the status is helpful to assess the condition of the vegetation and to monitor forest resources. First, the study investigated the literature to explore the relevant factors of vegetation. Using topographical factors, meteorological factors, spectral characteristics and making statistical charts with various types of vegetation. The results show that it is difficult to distinguish or explain the spatial distribution of all the vegetation only by a single factor. Therefore, the study use a variety of data for vegetation classification, including the Maximum Likelihood Classifier Supervised and Decision Tree. The results show that the kappa value is 0.41, which indicates that the spectral information is not enough to meet the classification of the vegetation when only use the maximum likelihood classifier supervised. Using the decision tree method with terrain, the index of the vegetation and the meteorological factors. In results, the correct rate of CART classification is 76.9%. The classification results also clearly indicate the classification rules and the importance of the factors which importance from high to low is terrain, annual temperature difference, summer temperature, rainfall. In future, vegetation classification can be followed by the maximum likelihood classifier method and then combined with the decision tree practice. And use terrain, climatic factors to predict and trace the space changes of vegetation, to understand the long-term sequence changes of the vegetation in the catchment area. |
