南仁山欖仁溪森林動態樣區中樹種個體死亡模型之建立

Modeling individual tree mortality for forest species of Lanjenshi forest dynamics plot, southern Taiwan

黃思博(Ssu-Po Huang)

指導教授 ： 謝長富

國立臺灣大學/生命科學院/生態學與演化生物學研究所/碩士(2007年)

https://doi.org/10.6342/NTU.2007.02716

摘要

本論文以南仁山欖仁溪森林動態樣區三次調查的資料，利用邏輯回歸分析的方法，建立樣區內樹種個體死亡預測模型，並以實際調查的資料來驗證模型的預測能力。
以1991∼1997年間兩次調查的資料，計算個體與其他個體間競爭關係及個體本身植株大小等相關變數，共25種變數來建立9197模型，另外，以1991∼2005年三次調查的資料，除計算前述之變數外，亦增加生長或死亡等相關變數，共35種變數來建立9105模型。將各樹種的資料用邏輯回歸分析後，能夠成功建立9197模型的有41種，包括冠層樹種24種和次冠層以下樹種17種，而能夠成功建立9105模型的有47種，包括冠層樹種26種和次冠層以下樹種21種。兩種模型中有近半數的樹種預測死亡率和胸高直徑（DBH）的倒數有關，這些樹種的小徑級植株死亡率很高，但到了大徑級時的死亡率就很低，除此之外，沒有特定的變數普遍影響各樹種的預測死亡率，各個樹種的邏輯回歸死亡模型都不相同。
利用隨機亂數作為模擬個體死亡的閥值，再比較各植株實際死亡情形與預測的死亡情形，9197模型中預測成功率達80﹪以上者有15種，成功率介於70∼80﹪者有14種，而9105模型中預測成功率達80﹪者有8種，成功率介於70∼80﹪者有17種，9197模型和9105模型對於優勢樹種的預測成功率差異不大，但對於整體實際死亡率來說，有使用生長、死亡相關變數的9105模型比9197模型預測來得準確，大部分樹種各徑級植株的預測死亡率和實際死亡率有相同的趨勢，且實際死亡率和預測死亡率相似，顯示這些樹種的死亡預測模型有良好的預測能力，而預測能力差的樹種大部分是植株數量太少或者實際死亡率大於20﹪，要改善這些模型的預測能力可能要增加取樣數量或者選用更好的閥值。

關鍵字

欖仁溪森林動態樣區 ； 森林動態 ； 死亡 ； 邏輯回歸 ； 模型

並列摘要

This study was used logistic regression function for modeling individual trees mortality. The data came from three times remeasurements of the Lanjenshi forest dynamics plot（Lanjenshi FDP）. The prediction ability of these mortality models were validated using independent data that was different from the modeling data. Two kinds of models were developed using two different periods. The 9197 models were using the data among 1991-1997 years and the 9105 models were using the data among 1991-2005 years. There were 41 species successfully developed the 9197 models that were using 25 variables include tree size, density and individual competition. There were 47 species successfully developed the 9105 models that were using 35 variables include tree growth and death variables besides tree size, density and individual competition. For many species, we found the hyperbolic transformation of diameter (DBH-1) to be highly significant in predicting the high mortality rates for small diameter trees and decreasing mortality rates for larger diameters. To compare the predictive and actual mortality was used the random number as the threshold to simulate trees mortality. The overall predictive success rates of 15 species are over 80﹪in 9197 models and 8 species are over 80﹪in 9105 models. The overall predictive success rates of 9197 and 9105 models are no different in dominate species. But prediction ability of overall mortality of 9105 models which were used tree growth and death variables in dominate species batter than 9197 models.

並列關鍵字

Lanjenshi FDP ； forest dynamics ； logistic regression ； mortality ； model

參考文獻

葉定宏，2006，南仁山欖仁溪樣區木本植物社會15年期動態。國立台灣大學生態學與演化生物學研究所碩士論文。

Sun, I.F. 1993. The species composition and forest structure of a subtropical rain forest at southern Taiwan. Ph.D. dissertation. University of California, Berkeley.

Avila, O. B., and H. E. Burkhart. 1992. Modeling survival of loblolly pine trees in thinned and unthinned plantations. Canadian Journal of Forest Research 22:1878-1882.

Bigler, C., and H. Bugmann. 2003. Growth-dependent tree mortality models based on tree rings. Canadian Journal of Forest Research 33:210-221.

Bravo-Oviedo, A., H. Sterba, M. del Rio, and F. Bravo. 2006. Competition-induced mortality for Mediterranean Pinus pinaster Ait. and P. sylvestris L. Forest Ecology and Management 222:88-98.