HabCluster: identifying the geographical boundary among intraspecific units using community detection algorithms in R

Abstract

Conservation management for a species generally rests on intraspecific units, while identification of their geographic boundaries is necessary for the implementation. Intraspecific units can be discriminated using population genetic methods, yet an analytical approach is still lacking for detecting their geographic boundaries. Here, based on landscape connectivity, we present a raster-based geographical boundary delineation method, habCluster, using community detection algorithms. Community detection is a technique in graph theory used to identify clusters of highly connected nodes within a network. We assume that the habitat raster cells with better connections tend to form a continuous habitat patch than the others, thus making the range of an intraspecific unit. The method was tested on the gray wolf (Canis lupus) habitat in Europe and the giant panda (Ailuropoda melanoleuca) habitat in China. The habitat suitability index (HSI) maps for gray wolves and giant pandas were evaluated using species distribution models. Each cell in the HSI raster is treated as a node and directly connected with its eight neighbor cells. The edge weight between nodes is the reciprocal of the relative distance between the centers of the nodes weighted by the average of their HSI values. We implement habCluster using the R programming language with the inline C++ code to speed up the computing. We found that the boundaries of the clusters delineated using habCluster could serve as a good indicator of habitat patches. In the giant panda case, the clusters match generally well with nature reserves. habCluster can provide a spatial analysis basis for conservation management plans such as monitoring, translocation and reintroduction, and population structure research.