請用此 Handle URI 來引用此文件: http://hdl.handle.net/11455/49755
標題: 小樣本下的種類數估計
Small Sample Estimation of Species Richness
作者: 沈宗荏
黃文瀚
關鍵字: 數學類, 統計學
基礎研究
摘要: Estimating the number of species (or species richness) of an assemblage is recognized as a classic, basic and important issue with respect to biologists/ecologists. However, a plethora of estimators had been developed in the literature. In practice, a sample proportion adopted in any field study is hardly larger than 1% due to the survey cost.A promising estimator which combines the generalized jackknife procedure with the linear regression approach is included in this project. In order to meet the practical aspect, the new method is supposed to perform well when the sampling proportion is small. In some preliminary tests using two forest data sets, the proposed method outperforms some existing approaches and qualifies in the practical use, but its standard error is inflated with the estimation of the response variable with respect to the linear regression model. The bootstrap method is promising to make the estimation of the chosen response variable stable. That would therefore follow a new, stable, reliable richness estimator.A simulation study would then be conducted to compare the performances of the conventional estimators as well as the developed one. The two forest data sets are employed to be the sampling populations varying with the quadart size. It would be expected that real datasets should be more realistic than artificial ones. The data sets are briefly described as follows. The Barro Colorado Island (BCI) plot: 50 ha (1000500 m), locating in Barro Colorado Island, Panama. Six censuses have so far been surveyed. The second dataset is the Pasoh plot located in Pasoh Forest Reserve, Malaysia. The census carried out during 1985 to 1987 was used in this study.
在一固定群落中如何透過抽樣資料估計種類數,長久以來一直是生態上一個古典並且重要的課題。雖然文獻上已經發展許多的估計方法,但是對生態學家來說方法似乎過多,所以生態學家很難選擇一個適合他們的方法。實際上這是一個實務的問題,因為大多數的田野調查(field study)實際抽樣比例都不會超過1%,因此如何在低抽樣比例下提供一個合理的方法便是本計畫書的主要目的。本計畫主要結合廣義摺刀估計與線性迴歸模式進而提出一新的估計方法。因應實務上的需要,新估計方法強調在低抽樣比例的應用,利用實際資料所做的初步測試,新估計方法的表現比傳統所使用的估計方法來的好,特別是低抽樣比例下。但是也遭遇新的問題,即是新估計方法變異較大,因為在線性迴歸模式中的反應變數並不是很穩定,因此考慮利用拔靴法來穩定反應變數,初步的模擬試驗也看到它的效果,確實能降低變異,因此新提出的估計方法也會較為穩定。本計畫利用電腦模擬比較現有的一些估計方法,以及新提出之方法。抽樣母體是兩筆真實的森林數據,全部做過普查,因此相當珍貴,再者,它們是真實資料,因此比較能反映出母體的真實情況。一筆來自巴拿馬 (Panama) 的動態樣區, 另一筆則來自馬來西亞(Malaysia),面積皆是50公頃。由於是普查資料,因此可以任意切割抽樣區塊的大小,因此也可以研究區塊大小的選定如何影響估計方法的表現,進而提出一抽樣的準則。
URI: http://hdl.handle.net/11455/49755
其他識別: NSC98-2118-M005-001
文章連結: http://grbsearch.stpi.narl.org.tw/GRB/result.jsp?id=1877995&plan_no=NSC98-2118-M005-001&plan_year=98&projkey=PA9807-1354&target=plan&highStr=*&check=0&pnchDesc=%E5%B0%8F%E6%A8%A3%E6%9C%AC%E4%B8%8B%E7%9A%84%E7%A8%AE%E9%A1%9E%E6%95%B8%E4%BC%B0%E8%A8%88
顯示於類別:應用數學系所

文件中的檔案:
沒有與此文件相關的檔案。


在 DSpace 系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。