Premium
Evaluating the performance of neutrality tests of a local community using a niche‐structured simulation model
Author(s) -
Takeuchi Yayoi,
Innan Hideki
Publication year - 2015
Publication title -
oikos
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.672
H-Index - 179
eISSN - 1600-0706
pISSN - 0030-1299
DOI - 10.1111/oik.01703
Subject(s) - niche , neutral theory of molecular evolution , overfitting , ecology , biology , abundance (ecology) , neutrality , ecological niche , statistics , evolutionary biology , computer science , mathematics , machine learning , genetics , philosophy , epistemology , habitat , artificial neural network , gene
Understanding the processes that underlie species diversity and abundance in a community is a fundamental issue in community ecology. While the species abundance distributions (SADs) of various natural communities may be well explained by Hubbell's neutral model, it has been repeatedly pointed out that Hubbell's SAD‐fitting approach lacks the ability to detect the effects of non‐neutral factors such as niche differentiation; however, our understanding of its quantitative effect is limited. Herein, we conducted extensive simulations to quantitatively evaluate the performance of the SAD‐fitting method and other recently developed tests. For simulations, we developed a niche model that incorporates the random stochastic demography of individuals and the nonrandom replacements of those individuals, i.e. niche differentiation. It therefore allows us to explore situations with various degrees of niche differentiation. We found that niche differentiation has strong effects on SADs and the number of species in the community under this model. We then examined the performance of these neutrality tests, including Hubbell's SAD‐fitting method, using extensive simulations. It was demonstrated that all these tests have relatively poor performance except for the cases with very strong niche structure, which is in accordance with previous studies. This is likely because two important parameters in Hubbell's model are usually unknown and are commonly estimated from the data to be tested. To demonstrate this point, we showed that the precise estimation of the two parameters substantially improved the performance of these neutrality tests, indicating that poor performance can be owed to overfitting Hubbell's neutral model with unrealistic parameters. Our results therefore emphasize the importance of accurate parameter estimation, which should be obtained from data independent of the local community to be tested.