Breeding range shift of the red-crowned crane (Grus japonensis) under climate change

Earth System), which accounted for more climate change processes and included dynamic vegetation [44]. The spatial resolution of both current and future climate data is 5�5 arc-min- utes. The model output from HadGEM2-ES is average on 30 years centered at 2030s (2020– 2049), 2050s (2040–2069), 2070s (2060–2089), and 2080s (2070–2099) before downscaling. We used Maximum Entropy (MaxEnt) [35], which is one of the most popular and powerful modeling technique for predicting distribution of species [45]. For parameter settings, we used default parameter in MaxEnt. We used 10-fold cross validation to assess model performance. Data are divided randomly into 10 subsets with equal sample size. Then nine subsets of the data were used to train model, and the remaining one subset was used to test model perfor- mance. We used area under the receiver operating characteristic (ROC) curve (AUC) to assess our model [46]. The AUC value ranges from 0 to 1. Higher AUC value means greater discrimi- nant ability of bioclimatic niche model. Model evaluation was performed on the average AUC of the 10 cross validation runs. For each cross validation run, we chose the threshold which maximize the sum of sensitivity and specificity to convert the predicted suitability binary map (presence and absence). This threshold is equivalent to that from maximizing true skill statistic (TSS), which is recommended as a good measure for the performance of ecological niche model when predictions are binary maps [47]. We summed together the 10 binary maps from

10 fold cross validation models. We then set the areas where the grid value larger than 5 (i.e. areas where at least 6 models predicted the presence of this species) as the presence area for the continental population of the red-crowned crane. Variable importance measures the relative importance of environmental variables in deter- mining the distribution of species in a niche model. Two measurements of variable importance can be calculated in MaxEnt software: percent contribution and permutation importance [37]. Percent contribution depends on the particular path to get to the optimal solution and a differ- ent algorithm could result in different percent contribution values, therefore we did not use it in this study. Permutation importance measure depends only on the final MaxEnt model [37]. The importance for each variable is determined by randomly permuting the values of that vari- able in the training dataset and measuring the resulting decrease in training AUC, and normal- ized to percentages. A larger decrease indicates that that variable is more important in the model. The ranking of relative importance can help us find out which variables are more important in affecting the distribution of the red-crowned crane, and assessing the potential effects of changing of those variables under climate change on the breeding range of the red- crowned crane. To assess the effects of human activities in breeding range of the continental population of the red-crowned crane, we used human influence index as an indicator of intensity of human

activities. The human influence index was calculated from nine global data layers including human population pressure (population density), human land use and infrastructure (built-up areas, nighttime lights, land use/land cover), and human access (coastlines, roads, railroads, navigable rivers). The dataset is produced by the Wildlife Conservation Society (WCS) and the Columbia University Center for International Earth Science Information Network (CIESIN) (available at http://sedac.ciesin.columbia.edu/data/set/wildareas-v2-human-influence-index- geographic). The value of human influence index ranges from 0 to 64, higher value means heavier human activities in this raster cell. Results The average test AUC for the 10 fold cross validation runs is 0.943, and the standard deviation is 0.004. The high AUC value means our models can predict the distribution of the breeding range of the continental population of the red-crowned crane very well. The low standard PLOS ONE Breeding range shift of the red-crowned crane (Grus japonensis) under climate change PLOS ONE | https://doi.org/10.1371/journal.pone.0229984 March 12, 2020 5 / 16