Application and comparative performance of network modularity algorithms to ecological communities classification

Network modularity (community structure) is a wellstudied large-scale connectivity pattern in networks [1,2], with several detection algorithms described in the literature (for a review, see [2]). A comparative study of module assignment accuracy by the different algorithms using benchmark networks showed that algorithm performance varies according to network size and the level of intermodule mixing [3]. A comparative evaluation of different community detection algorithms using real ecological data is however lacking so far. Here we test the applicability of network modularity algorithms as a method to classify plant species communities. Community ecologists seek to understand the processes underlying organism and environment interaction dynamics of diversity, abundance, and composition of species in communities [4]. Vegetation science focuses on the ecology and composition of plant communities [5]. A basic task of the vegetation ecologist is to characterize, identify and distinguish different vegetation units that comprise plant species with similar habitat preferences. A common traditional approach is the making of relevés [6], which comprise a catalog of all plant species that occur in a vegetation plot together with their respective degree of coverage (i.e., frequency). Plant communities are ascertained by sorting the relevés in vegetation tables according to the occurrence of diagnostic species. Large and complex vegetation tables can however become error-prone and do not provide a concise overview of the whole data. A potentially more critical limitation is that the method demands an a priori knowledge about the respective diagnostic species whose identity can be a matter of debate. Diagnostic species include those particular species whose occurrence in the relevés may serve as an important telltale for the plant community classification. These include the character species, whose occurrence is typical to specific plant communities, and the differential species, whose occurrence can be used to distinguish related plant communities, but are not limited to a single community. Computer based methods including network applications have been used in Abstract