應用廣義加法模式建構六種台灣針葉樹物種分布範圍與氣候因子之關係

Abstract

在全球氣候變遷的趨勢之下，未來台灣的氣候可能會變得更溫暖。由於植物的分布與氣候有相當重要的關連性，氣候的暖化可能會造成物種分布範圍的改變，甚至造成物種的滅絕。藉由氣候因子建立物種的分布模式，是評估氣候暖化對於植物分布範圍影響的作法。 本研究藉由統計上的廣義加法模式(generalized additive models, GAMs)，嘗試分析六個台灣山區優勢針葉樹物種，包括台灣扁柏(Chamaecyparis obtusa var. formosana )、台灣紅檜(Chamaecyparis formosensis)、台灣冷杉(Abies kawakamii)、台灣鐵杉(Tsuga chinensis)、台灣雲杉(Picea morrisonicola)、台灣二葉松(Pinus taiwanensis)，目前的分布範圍與目前氣候狀態之間的關連性，並建立物種的潛在分布模式。物種分布資料，來自於林務局的第三次台灣森林資源調查資料。以Thin-plate smoothing spline方法，將氣象觀測站長期記錄的月均溫和月總降雨量資料，平滑成在地理空間上連續分布的反應曲面。建立模式使用的一組氣溫和雨量解釋變數，彼此之間的相關係數皆低於0.9，可以避免模式產生多元共線性(multicollinearity)的問題。透過逐步選取的程序，從原本的解釋變數當中，找出對各別物種的分布範圍，最具有顯著性的解釋變數。 根據建立完成的六個物種分布模式，其中每一個模式都選入年均溫這項解釋變數，代表這六個物種的分布，皆受到年均溫的影響。年均溫的增加對於物種在一個地區出現的機率，有明顯的負面效應。相反地，沒有任何模式選入了年總降雨量，可能是因為台灣降雨量豐沛的原因。不過，季節性的降雨，對於某些物種的分布，仍然具有影響。 模式的預測結果，係藉由ROC曲線下面積(AUC)值與其他指標加以評估。評估結果顯示，台灣冷杉與台灣台灣鐵杉的預測結果最好，而台灣扁柏與台灣雲杉的預測結果較差，不過其結果仍然具有一定的準確性。

Under the predicted climate change scenarios, many of Taiwan’s endemic conifers may shift their distribution ranges in response. The first step in assessing the likely impacts of climate change on a species’ distribution is to develop a distribution model that relates the species’ current observed distribution to a set of possible climatic variables. In this study, a generalized additive modeling (GAM) approach was used to establish the distribution models for six dominant conifer species in the montane regions of Taiwan. The presence/absence data for each species were extracted from the inventory plot information of The Third Forest Resources and Land Use Inventory in Taiwan. Only plots with an elevation of at least 1000 m a.s.l. were included in this study. Climatic data were from available weather stations. A set of climate surfaces for mean monthly temperatures and precipitations was first developed by using thin-plate smoothing splines. The fitted climate surfaces were then used to estimate climatic parameters for each plot. A stepwise procedure was used to select the best climatic predictors for each species. Model performance was evaluated by area under the receiver-operating characteristic (AUC) curve, Kappa statistic, sensitivity, and specificity. The results showed that annual mean temperature was the only predictor included in all models, suggesting that the distributions of the six conifer species all were controlled by annual mean temperature. An increase in temperature would reduce the presence probability of the six species at a given site. In contrast, none of the models included annual mean precipitation, likely due to the abundant amount of precipitation in Taiwan. However, the influence of seasonal precipitation was important to some of the conifer species. Model performance criteria suggested the models for Taiwan fir and Taiwan hemlock had a good performance. Based on the AUC criterion, all the models reached the level of useful applications.