A Comparative Study On The Predictive Performance Of Ungulates Distribution Model In The Lesser Khingan Mountains

Species distribution is the key research field in animal ecology,wildlife conservation and nature conservation management.The study is based on the sample line survey data of 4 kinds of ungulates in the Lesser Khingan Mountains of Heilongjiang Province during the period from December 2018 to March 2019.Three species distribution model methods(generalized line model,GLM;generalized additive model,GAM;Random Forest,RF)were selected to respectively predict the distribution of ungulates.The AUC and TSS were used as the evaluation index to compare the predictive performance difference of different models in the same species distribution and the single model among different species distribution.The results are as follows:1)The distribution results of ungulate species show that,both roe deer and wild boar are widely distributed in the Lesser Khingan Mountains,and the distribution areas were similar,including the forest areas near the human settlements;the distribution areas of the red deer and moose are small,only in a few areas;2)When predicting the distribution of wild boar and red deer,GAM has the best prediction performance,and GLM performs better than RF.When predicting roe deer and moose,although the AUC and TSS of the three models are different,there is no statistical difference;3)The performance of three models to predict the distribution of red deer and moose was significantly better than that of roe deer and wild boar.However,when predicting the distribution of red deer and moose,there is no significant difference in prediction performance between the models;4)As a more flexible model method,GAM performs better in predicting the distribution of ungulates.However,for the distribution prediction of moose and red deer in rare and small distribution areas,the performance of RF is more stable and difficult to overfit.