Assessing bluegill (Lepomis macrochirus) habitat suitability using partial dependence function combined with classification approaches

Abstract

Classification approaches have been used to understand the habitat suitability of key species. Partial dependence function is an especially useful concept despite of a lack of studies that compare the results of the function against observations. Furthermore, there has been scant research investigating the relative performance of classification approaches for describing habitat suitability. Thus, we aim to assess the applicability of partial dependence function combined with classification approaches to describe habitat suitability of the bluegill Lepomis macrochirus, a riverine fish in the Kanto region of Japan. A total of 425 samples, along with eight environmental variables, were surveyed by the National Censuses on River Environments and were used throughout this research. Five classification approaches were combined with a partial dependence function to assess the habitat suitability of bluegill. The areas under the curves based on the training and test data were calculated 100 times using each of the five classification approaches. Additionally, partial dependence on individual and paired environmental variables was estimated using each combination of a five-classification approach and partial dependence function, and this dependence was plotted to determine the habitat suitability of bluegill. As a result, random forest approach demonstrated high predictive accuracy compared to other classification approaches. The combination of the partial dependence function and random forest described the peaks of habitat suitability observed in the field, both using individual and paired environmental variables. Moreover, habitat suitability based on pairs of environmental variables also indicated that bluegill changes their habitat throughout the year.