Obtaining the best possible predictions of habitat selection for wintering Great Bustards in Cangzhou, Hebei Province with rapid machine learning analysis

Oct 25, 2020

Here, we present our results from two years of field data collection on habitat selection. We choose a machine learning model approach based on a rapid assessment methodology for the winter habitat of the Great Bustard. It is based on a spatial analysis of the best available environmental data, which were collected relatively quickly. These relatively new methods in ecology are based on an ensemble of decision trees and include algorithms such as TreeNet, Random Forest and CART used in parallel. In this study, we collected bustard droppings (presence only) from 48 locations between December 2011 and January 2012 and used the sites as training data. Droppings from 23 locations were collected in November 2012, and those sites were used as test data. We used eight environmental variables as predictor layers for the response variable of bustard presence/availability. We employed a Geographic Information System (ArcGIS 10.1 and Geospatial Modelling Environment) and Google Earth. Compared with the other three models, we found that predictions from Random Forest obtained a significant difference between presence and absence. According to this model, the three most important factors for wintering Great Bustards are distance to residential area, distance to water pools, and farmland area. Our model shows that wintering Great Bustards prefer locations that are over 400 m away from residential areas, within 900 m of water pools and on areas of farmland smaller than 0.5 km2. We think we can apply our analysis to Great Bustard management in our study area and the adjacent region and that this work sets a baseline for future research.


Submit a Comment

Your email address will not be published. Required fields are marked *

Pin It on Pinterest