样本量对MaxEnt模型预测物种分布精度和稳定性的影响  

Effects of Sample Sizes on Accuracy and Stability of Maximum Entropy Model in Predicting Species Distribution

作　　者：陈新美[1] 雷渊才[1] 张雄清[1] 贾宏炎[2] 

机构地区：[1]中国林业科学研究院资源信息研究所,北京100091 [2]中国林业科学研究院热带林业实验中心,凭祥532600

出　　处：《林业科学》2012年第1期53-59,共7页Scientia Silvae Sinicae

基　　金：国家自然科学基金项目(31170588);科技部社会公益研究专项(2005DIB5J142)

摘　　要：以实际调查的4个物种的34个不同样本量(5,6,8,10,15,20,25,30,40,50,60,70,80,90,100,120,150,180,200,220,250,300,350,400,450,500,550,600,650,700,800,900,1000,1200)为例,模拟计算分析不同的样本量对MaxEnt物种分布模型的精度和稳定性的影响。结果表明:总体上来看,样本量的大小对MaxEnt模型预测物种空间分布的精度影响不大,在样本量较小时,精度不稳定,随着样本量的增大(训练数据在样本量50左右,检验数据在样本量120左右),MaxEnt模型的预测精度越来越稳定。Prediction of species distribution and its changes play more and more important roles in the fields of ecological protection and application as well as global climate changes. It is impracticable to survey species distribution in large area, especially rare species. Considering that very few species distribution data have been accumulated, employ species distribution model fitting technique is highly necessary in the process of predicting species distribution. Sampling size has an important influence on expense of actual survey and accuracy of model prediction. In terms of accuracy of species distribution model and expense of forest survey, it is necessary to investigate the least sampling size when species distribution models reach the most accuracy. Thirty-four different sampling sizes（5, 6, 8, 10, 15, 20, 25, 30, 40, 50, 60, 70, 80, 90, 100, 120, 150, 180, 200, 220, 250, 300, 350, 400, 450, 500, 550, 600, 650, 700, 800, 900, 1 000 and 1 200） of four species were used to simulate the influence of different sample sizes on the precision and stability of MaxEnt species distribution model. The results showed that sampling sizes had no obvious influence on MaxEnt. The accuracy of MaxEnt was unstable when sampling size was small, but as sampling size was increasing（sampling size of training data was about 50, test data was about 120）, the accuracy was more stable.

关 键 词：样本量 最大熵物种分布模型 AUC 预测精度 标准差 

分 类 号：S757[农业科学—森林经理学]