Prediction of avian species composition from assemblage structure

• Main Author: Huang, Jing-Lun
• Other Authors: Eigenbrod, Felix
• Published: University of Southampton 2018
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.736765

This research focuses on how capitalising on community pattern, a character of ecological communities, could improve the predictability of community models, thus facilitating research in conservation. Patterns of communities not only depict phenomena but are also useful for predicting potential changes in species composition when patterns are governed by specific mechanisms. Most conventional prediction models do not take community pattern into consideration, despite the fact that incorporating community patterns into conventional models for predicting species richness and composition may enhance predictability. In this thesis, I assessed if incorporating two community patterns, nestedness and species co-occurrence, into conventional prediction models could improve the model predictability. Nestedness is a non-random species distribution pattern in which species in depauperate sites are contained in species-rich sites. Co-occurrence networks categorise species assemblages that reflect differential habitat requirements. I demonstrate that capitalising on nestedness provides a novel approach for improving the predictive power of species accumulation curves for species richness in unsampled areas. Specifically, while species richness is usually overestimated when the data are inputted in random order (the conventional approach), species richness is underestimated when the data are inputted in nested order. Taking an average of projected species richness of these two inputting orders dramatically lowers the prediction error rate, indicating that using nestedness in addition to random orders can greatly improve the predictive power of species distribution curves. I also show that network analysis can improve the ability to correctly classify site groups, which is the basis for calculating the indicator species value, by accurately reflecting similar ecological requirements of co-occurred species. Indicator species identified by network modularity, comparing to conventionally based on the k-means clustering method, can more successfully assign unsampled sites to the correct species groups and recognise representative species for the groups. These methods were tested using both British and Taiwanese bird assemblages. Both case studies supported the above conclusions, suggesting that the methods developed in this thesis have real promise for conservation applications. However, further work is required to assess whether these two novel pattern-based approaches are similarly applicable in other geographic regions or taxas.