This study focuses on applying geographic information system and a digital terrain model to generate a potential vegetation map for an ecosystem management area of the Liukuei Experimental Forest, southern Taiwan. The process included the following aspects. First, the digital terrain model was applied to produce 4 environmental factors (i.e., altitude, slope, aspect index, and whole light sky ratio). Then according to the 7 vegetation types of this study area published by Wang(1991), different discriminant functions were derived from the relationships between the above environmental factors and vegetation types. The results show that the generated discriminant functions for distinguishing 7 vegetation types are accepted at the 5% significance level, because the calculated Mahalanobis distance (316.81) was greater than the tabulated value (X20.005(24)=45.6) . Second, the probabilities of 7 vegetation types for each grid wee calculated based on those discriminant functions, and the potential vegetation map was then generated according to the vegetation type with the maximum probability. The result shows that the generated potential vegetation map coincides with the qualitative description from general vegetation studies. Third, in order to evaluate the accuracy, 34 field samples of floristic composition were chosen to compare with the results obtained from the discriminant functions. The results indicate a low accuracy (44%) and Kappa (0.29) value although the measurement of agreement is statistically significant. However, if the 7vegetation types are merged into 3 vegetation zones (i.e., only considering 1 factor of elevation in the discriminant function), the accuracy and the Kappa value are 74% and 0.55, respectively; the measurement of agreement is statistically significant as well. From the above results, we concluded that the discriminability of floristic composition with 3 vegetation zones is more effective than that with 7 vegetation types.
本研究應用地理資訊統和數值地形模型推算六龜試驗林北半部第1到16林班之海 拔高度、坡度、方位指數及全天光空城等4個環境因子，並引用Wang(1991)之植群分析結果 ，進行多組判別分析，求得7個植群型(含亞型)與4個環境因子間的判別函數。該組判別 函數之Mahalanobis D2等於316.81，大於X20.005(24)=45.6，顯示4個環境因子能有效判別7個 值群型。以數值地形模型推算之網格環境因子，遂點代人判別函數式，取最大機率之植群 型，產生六龜試驗林北半部l至16林班之潛在植群分佈圖。該植群分佈圖符合植群研究對各 植群型分佈環境之定性描述。野外進行種類組成之檢核點調查，共得34個有效樣本，野外調 查之植群型與判別式推測之植群型間同意度檢定的結果為顯著，但K值為0.29，屬低同意 度，實際推測正確率僅達44％。若依植群帶的劃分將各型合併，僅以海拔高度進行判別，判 別式在統計上可以被接受，且推測植群型與調查結果間的同意度檢定仍然顯著，而K值提高 為0.55，已屬中等同意度，實際推測正確的比率為74％ 。以本研究的結果而言，實際應用 地理資訊系統於預測六龜生態系經營區內的植群種類組成，以海拔高度區分楠榕、楠櫧及爍 林帶下層等三個植群帶的分類較為實用。
