利用相似度指標於出現與否地圖上偵測物種關聯性

Using similarity indices for investigating species association in presence-absence data

張雅梅(Ya-Mei Chang)；謝佩穎(Pei-Ying Hsieh)

核密度估計 ； Morisita指標 ； Horn重疊度指標 ； kernel density estimator ； Morisita index ； Horn overlap index

智慧科技與應用統計學報

22卷1期（2024 / 07 / 01）

26 - 41

繁體中文;英文

本篇研究主要利用四種相似度指標於PA地圖上偵測物種關聯性，其中Morisita指標與Horn重疊度指標原本只適用於物種豐度資料，利用核密度估計針對這兩個指標進行調整，使其適用於PA地圖上。於模擬研究中，在獨立的假設下，隨機生成兩個物種，建立各個指標的經驗分配，透過雙尾檢定去判別物種關聯性，並觀察各個指標的型一誤差及檢定力。從模擬的結果得知，出現率較低的資料中，指標較無法正確判別其關聯性；有用核密度估計修正後的指標容易把兩個具有關聯性的物種，判斷為不具關聯性。最後將本篇所提及的相似度指標應用於百靈鳥上，指標可以發現資料間是具有關聯性的。

In this research, we employ four similarity indices to explore species associations in presence-absence data. Two specific similarity indices, namely the Horn overlap index and the Morisita index, undergo modification through kernel density estimation to better suit the characteristics of presence-absence data. In our simulation study, we randomly generate two independent species to establish empirical distributions for each index. Subsequently, a two-sided test is employed to identify species associations, with a comparison of type I error and testing power for each index. The simulation results reveal that the indices struggle to accurately detect associations at low occurrence rates. Notably, indices modified by kernel density estimation tend to mistakenly identify two associated species as having no association. Finally, we apply these similarity indices to the Woodlark data.