Combining occurrence and abundance distribution models for the conservation of the Great Bustard.

Chunrong Mi,Falk Huettmann,Yumin Guo,Rui Sun

doi:10.7717/peerj.4160

Abstract

Species distribution models (SDMs) have become important and essential tools in conservation and management. However, SDMs built with count data, referred to as species abundance models (SAMs), are still less commonly used to date, but increasingly receiving attention. Species occurrence and abundance do not frequently display similar patterns, and often they are not even well correlated. Therefore, only using information based on SDMs or SAMs leads to an insufficient or misleading conservation efforts. How to combine information from SDMs and SAMs and how to apply the combined information to achieve unified conservation remains a challenge. In this study, we introduce and propose a priority protection index (PI). The PI combines the prediction results of the occurrence and abundance models. As a case study, we used the best-available presence and count records for an endangered farmland species, the Great Bustard (Otis tarda dybowskii), in Bohai Bay, China. We then applied the Random Forest algorithm (Salford Systems Ltd. Implementation) with eleven predictor variables to forecast the spatial occurrence as well as the abundance distribution. The results show that the occurrence model had a decent performance (ROC: 0.77) and the abundance model had a RMSE of 26.54. It is noteworthy that environmental variables influenced bustard occurrence and abundance differently. The area of farmland, and the distance to residential areas were the top important variables influencing bustard occurrence. While the distance to national roads and to expressways were the most important influencing abundance. In addition, the occurrence and abundance models displayed different spatial distribution patterns. The regions with a high index of occurrence were concentrated in the south-central part of the study area; and the abundance distribution showed high populations occurrence in the central and northwestern parts of the study area. However, combining occurrence and abundance indices to produce a priority protection index (PI) to be used for conservation could guide the protection of the areas with high occurrence and high abundance (e.g., in Strategic Conservation Planning). Due to the widespread use of SDMs and the easy subsequent employment of SAMs, these findings have a wide relevance and applicability than just those only based on SDMs or SAMs. We promote and strongly encourage researchers to further test, apply and update the priority protection index (PI) elsewhere to explore the generality of these findings and methods that are now readily available.

Highlights

The knowledge of species occurrence and abundance distribution provides fundamental information for conservation biology (VanDerWal et al, 2009; Drew, Wiersma & Huettmann, 2011; Primack, 2012; Johnston et al, 2015)
We found that the area of farmland, the distance to residential areas, to ditches and to expressways were the top four most important variables that influenced bustard occurrence
These results indicate that a high protection index (PI) is located in the center, north and northeast regions of the study area, and they indicate a sporadic and fragmented distribution, which could represent a priority protection site if a conservation decision was to be made

Summary

Introduction

The knowledge of species occurrence and abundance distribution provides fundamental information for conservation biology (VanDerWal et al, 2009; Drew, Wiersma & Huettmann, 2011; Primack, 2012; Johnston et al, 2015). Understanding how environmental factors are related to species occurrence and abundance distribution and how they are explicit in time and space are priorities in current biodiversity conservation (Drew, Wiersma & Huettmann, 2011; Martín et al, 2012). Species distribution models (SDMs) are empirical ecological models that relate species observations to environmental predictors (Guisan & Zimmermann, 2000); usually this process is done using machine learning algorithms (Drew, Wiersma & Huettmann, 2011, see Mi et al, 2017 for an application). Increasing attention has been paid to these problems in recent years (e.g., Yen, Huettmann & Cooke, 2004; Martín et al, 2012; Howard et al, 2015; Ashcroft et al, 2017; Fox et al, 2017)

Methods

Results

Conclusion