現行與未來氣候下的台灣森林植物分布預測研究

Predicting the potential distributions of plant species and forests in Taiwan under present and future climates

林奐宇(Huan-Yu Lin)

指導教授 ： 胡哲明

國立臺灣大學/生命科學院/生態學與演化生物學研究所/博士(2020年)

https://doi.org/10.6342/NTU202000682

摘要

氣候與森林植物的分布緊密關連，而多項研究也證實，臺灣山地的帶狀植群分化、以及部分泛域植群類型的分布，均與氣候條件高度相關，尤其受到溫度與年間降水分配差異的影響。為瞭解臺灣森林植群與氣候的關係，進一步建立森林植群氣候棲位資訊，以及推估氣候變遷的可能影響，本論文整合臺灣既有植群調查及長期氣候資料，建立預測森林植物現生與未來分布的流程方法，並提出森林分布變遷之研究結果。茲就各章節研究成果摘述如下：

一、動態氣候降尺度模型與高解析氣候圖層之產製
氣候資料是生態研究的重要基礎，然而一般泛用氣候圖資之空間解析度以數公里至數十公里不等，難以反映山區起伏地形導致的氣溫與降水的劇烈變化。本章節利用臺灣氣候變遷推估與資訊平台（TCCIP）5公里網格氣候資料，經動態局部迴歸方法獲得區域範圍之海拔遞減率，作為內插校正參數，於R軟體設計一套自由尺度化之氣候降尺度模型，命名為clim.regression。經15處不同海拔氣象測站實測驗證，clim.regression推估月尺度氣候之平均絕對誤差為0.56°C（月均溫）、0.79°C（月均低溫）、0.80°C（月均高溫）及36.26mm（月累積降水），改善了TCCIP原始資料54.6–66.7%的誤差。Clim.regression共可針對歷史年度（1960–2009）及三個未來階段產製73種氣候因子，其自由尺度化、高準確度的特色，極適合在山地氣候與生態關係研究應用，亦是本論文進行後續章節研究之氣候資料來源。

二、以氣候為基礎的臺灣山地植群分布模擬與預測方法
相較於傳統航遙測影像判釋或現場調查方法，森林植群氣候的棲位模擬預測，是相對簡易而迅速獲得森林植群空間分布資訊的方法。本章節使用林務局國家植群多樣性調查計畫樣區資料，以及李靜峯等人建立之森林分類架構，經由clim.regression產生各森林類型之氣候幅度，再利用隨機森林方法，建立13種與氣候相關森林類型的棲位模型，並完成13種森林類型的現生分布預測。依據3817個樣區交叉驗證顯示，隨機森林對於現生植群分布預測之平均錯誤率為6.59%，森林分布預測結果與地形高度擬合，並反映出不同森林類型交會帶之過渡現象。本章節證實高解析度、高準確度的氣候資料，配合野外調查樣本及充分的機器學習訓練，可提供良好的植群現生分布資訊，作為預測未來變遷的基礎工具。

三、海拔梯度下，氣候變遷對於臺灣山地森林植群的影響程度為何？
目前全球森林與氣候變遷相關研究，主要集中在北半球中緯度，對熱帶森林的研究較少。以山地森林而言，普遍認為暖化可能導致森林植物的向上遷徙，對高海拔或山頂植群造成威脅與衝擊。臺灣山地森林跨越近4000公尺的海拔梯度，涵蓋熱帶至亞高山森林類型，利用前一章獲得之山地植群模擬預測方法，於本章節探討不同暖化情境下各森林類型的面積與海拔分布變化。所有暖化情境一致顯示，高海拔森林及中海拔雲霧林可能出現面積縮減，尤其以亞高山刺柏灌叢及臺灣水青岡落葉霧林首當其衝，將喪失大多數棲地（RCP 4.5）或瀕臨滅絕（RCP 8.5）。對於熱帶山地森林的預測結果則較為分歧，雖然熱帶森林的棲地面積在多數暖化情境下呈現逐步擴張的趨勢，然而在極端暖濕、或極端暖乾的狀況下，可能因水分有效性劇烈變化因素，導致熱帶山地霧林完全失去適存環境，熱帶季風林亦將出現顯著的棲地縮減，是不可忽視的氣候變遷威脅。綜上研究結果，本章節可推測出氣候變遷下的易危森林類型，並提出相對應的保育建議，作為後續監測及管理工作的參考。

四、被子植物雌雄異株物種的地理分布與及其生態相關因子之研究
本論文收集大量的植物分布紀錄與氣候資料，不僅限於氣候變遷研究應用，也是生態與生物地理分布研究的珍貴素材。本章節運用上述資料，以植物性別表現的空間分布型式為例進行研究，希能激發其他生態研究者的興趣與更多的參與。
雌雄異株是相對罕見的植物生殖模式，藉由不同雌雄個體異交產生種子延續後代。過去調查統計全球被子植物的雌雄異株種類約佔所有種類的6%，但許多例子指出不同區域的雌雄異株物種比例會有差異，地形起伏劇烈的海洋島嶼、熱帶森林、木本生活型、蟲媒授粉等現象，通常與較高的雌雄異株比例相關。臺灣的地理環境具有大陸至海洋的過渡特性，本研究發現，臺灣本島被子植物的雌雄異株整體比例約為8.2%，但從臺灣海峽至太平洋間，臺灣諸島被子植物相的雌雄異株比例呈現逐漸升高的梯度，呼應了Bawa氏提出的海洋島嶼較多雌雄異株物種的理論。此外，沿本島海拔梯度則發現，自然植群皆存在雌雄異株比例隨海拔升高而降低的現象，並在海拔2200公尺處存在顯著的遞減率轉折點；高海拔的雌雄異株比例劇烈陡降，推測可能與闊葉林與針葉林的交會轉換，以及高海拔授粉昆蟲相趨於單純化所致。但人工植群之雌雄異株比例則與海拔無顯著相關。搭配氣候因子分析，顯示雌雄異株植物常見於溫暖環境，兩性花植物則偏好於低溫的高海拔地區，雜性物種的出現則與氣候條件無顯著關聯。

關鍵字

氣候 ； 氣候變遷衝擊 ； 生態棲位模型 ； 山地森林植群 ； 隨機森林 ； 臺灣

並列摘要

Climate plays a vital role in shaping the distribution of forest and plant species. Some authors had reported that climate, especially the temperature and seasonal partitioning of rainfall, is significantly correlated with the altitudinal zonation of mountain forests and also to some of the distribution of azonal vegetation. This dissertation incorporated vegetation survey and historical climate data to reconstruct climatic niches for mountain forests of Taiwan, and project their distributional changes under future climatic scenarios. The critical outcomes of each chapter summarized as below:

1. A dynamic downscaling approach to generate scale-free regional climate data.
Climate variables, particularly temperature and precipitation, are the most well-known key factors related to vegetation zonation. However, to obtain climate data adequately for the requirement of ecological studies is challenging due to the difficulty of data integration and the complexity of downscaling, especially for mountainous regions. In this section, a synthetic approach combining bilinear interpolation and dynamic local regression was conducted to develop a scale-free climate downscaling model in R environment, namely clim.regression. Based on the original 5km x 5km gridded climate surface from TCCIP, clim.regression can generate 73 climatic variable estimates specific to the user-defined points of interest for historical (1960–2009) and future periods (2016–2035, 2046–2065 and 2081–2100), which reduced prediction error by 54.6–66.7% relative to the original gridded climate data for temperatures. The result is adapted to the uses of ecological researches and is the source of climate data of this dissertation.

2. Climate-based approach for modeling the distribution of montane forest vegetation.
A climate-based ecological niche model may provide an effective alternative to the traditional approach for assessing limitations, thresholds, and the potential distribution of forests. In this section, a machine-learning method based on scale-free climate variable estimates and classified vegetation plots was applied to develop niche models for the 13 climate-related forest types in Taiwan, and to generate a fine-scale predicted vegetation map. The result supported that the machine-learning approach is sufficient to handle a large number of variables and to provide accurate predictions, which has the potentials in projecting the distributional changes of forests under different climate change scenarios.

3. How much does climate change alter the distribution of forests across a great altitudinal gradient?
Taiwan is a high-mountain island with substantial altitudinal variations and diverse forest types driven by climate. In this study, we used the scale-free climate variable estimates and an established machine-learning approach to project the distributional changes of 13 climate-related mountain forest types under selected global warming scenarios. The results demonstrated a consistent trend of the drastic habitat contractions of subalpine Juniperus woodland and the deciduous Fagus broadleaved forests. It also revealed that tropical montane cloud forest and tropical winter monsoon forest might be highly vulnerable under the extreme warm-humid or warm-dry climatic conditions because of the sever change of water availability. For mitigating the risk of climate change to the vulnerable forest types, adapted conservation strategies were suggested according to the environmental characteristic of each forest type.

4. Geographical distribution of dioecy and its ecological correlates based on fine-scaled species distribution data.
A great deal of species occurrence and climate data is valuable in different ecological researches. An example is shown here to demonstrate such use in elucidating the complicate distribution patterns of plant sexual systems. In this section, I used species occurrence and historical climate data in exploring the geographical distribution of sexual expression systems of the flowering plants in Taiwan.
It was reported that the incidence of dioecy varied among local floras and suggested inclining to tropical and oceanic environments. We found the average incidence of dioecy in the flora of Taiwan to be 8.2%, but it exhibits geographical variations from islets in the Taiwan Strait to the Pacific Ocean. An apparent two-step decreasing pattern of dioecy percentages with elevation was also found, which shows a distinct transition at the altitude of 2200m. The overall analysis indicated that spatial variations of dioecy were associated with eco-correlates of land cover, elevation, woodiness, species richness, and mean annual temperature. Results of this section partially support Bawa’s hypothesis of a higher incidence of dioecy on oceanic islands, and consistent with Baker and Cox’s observations of more prosperous dioecious species on high-mountain islands in the tropics and subtropics.

並列關鍵字

Climate ； Climate change impacts ； Ecological niche modeling ； Mountain forest vegetation ； Random forest ； Taiwan

參考文獻

Chapter 1.

Beniston, M. 2003. Climatic change in mountain regions: A review of possible impacts. Climatic Change 59: 5–31.

Davis, M. B. and R. G. Shaw. 2001. Range shifts and adaptive responses to quaternary climate change. Science 292: 673–679.

IPCC. 2014. Summary for policymakers. In: Climate Change 2014: Impacts, adaptation, and vulnerability. Part A: Global and sectoral aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 1–32.

Lee, P.-F., K.-Y. Lue and S.-H. Wu. 2006. Predictive distribution of Hynobiid salamanders in Taiwan. Zoological Studies 45: 244–254.