Application of Complete Gradient Clustering Algorithm for analysis of wildlife spatial distribution

Abstract

Abstract Animal grouping is a very complex process that is triggered by a number of factors: habitat, social structure, season, predators' pressure etc. Proper identification of sites where natural concentrations of animals occur, is crucial for the explanation of such behaviour and for species and habitat management. Therefore, the selection of methods allowing to distinguish parts of a home range where a true tendency for aggregation takes place from areas where population density even if high, may be accidental, is so important. In this study, an innovative approach extending the possibilities provided by kernel density estimation was applied for analysis of species' distribution. The main purpose of our research was to identify areas of highest concentrations of animals and subsequently the main drivers for their habitat selection patterns. The data we used were collected within the home range of the European bison (wisent) Bison bonasus population in Bieszczady Mountains, a part of the Carpathians in the southeast of Poland. We considered both winter and vegetation seasons and used species presence data as determinant for selection of wisent habitats. The kernel density estimator of the data was first constructed to characterize spatial distribution. Next, by means of the Complete Gradient Clustering Algorithm, clusters corresponding to animal presence records were detected. Total numbers of observations belonging to main clusters were significantly higher than 50% (Herd 1: 65% vs. 61%, Herd 2: 75% vs. 93%). The results confirmed the natural tendency of wisents to aggregate. Moreover, the occurrences of concentration points (modes) were connected with the vicinity of feeding points or good conditions for grazing. Our results indicate that for both herds, the numbers of main clusters in winter were smaller than in the vegetative season (Herd 1: 4 vs. 5, Herd 2: 4 vs. 6). This is also justified because wisents merge into larger groups during winter around feeding points or in sites less exposed to wind or low temperatures. It is worth stressing that the algorithm does not require any assumptions concerning the data or the desired number and shape of clusters. This allows it to discover real structures within the dataset. In summary, we show how the commonly used nonparametric estimation technique can be adapted in a novel and effective approach for analyzing spatial processes in wildlife populations. The proposed methodology reveals powerful results for further investigation and refinements, especially when we consider an influence of environmental factors of various types.