國立政治大學統計學系
學 術 演 講
主講人：邱春火副教授(國立臺灣大學農藝學系)學 術 演 講
題 目：A Mixture-Model Approach to Robust Species Richness Estimation under
Detection Heterogeneity
時 間：民國115年3月9日 (星期一) 下午1：30
地 點：國立政治大學逸仙樓050101教室
摘 要：
Accurate estimation of species richness remains a central challenge in ecology, particularly under heterogeneous detection probabilities and when data are collected using multiple sampling schemes. Traditional estimators, including maximum likelihood and jackknife methods, often exhibit instability or fail to satisfy essential consistency criteria as sampling effort increases. In this study, I develop and evaluate an improved species richness estimator that addresses these limitations across three major data types: individual-based abundance data, sample-based incidence data, and integrated datasets composed of heterogeneous sampling formats.
The proposed estimator is derived from a mixture-model framework using a moment-based approach and can also be interpreted as a bias-corrected extension of Chao’s lower bound via the Good–Turing frequency formula. Extensive simulation studies were conducted under homogeneous and heterogeneous assemblage models with varying coefficients of variation and sampling efforts. Performance was evaluated using bias, root-mean-square error (RMSE), variance, and confidence interval coverage.
The new estimator consistently satisfies the essential criterion that bias and RMSE decrease with increasing sample size. Compared with Chao1 and Chao2, it exhibits lower bias and lower RMSE under heterogeneous detection, while maintaining more accurate confidence interval coverage. The results show that the proposed estimator provides a more reliable and stable framework for richness estimation across heterogeneous and integrated sampling designs, thereby offering practical advantages for large-scale biodiversity assessment and ecological monitoring.
