优化后最大熵模型在模拟驼鹿适宜栖息地分布中的应用  

Optimized MAXENT Model in Simulating Distribution of Suitable Habitat of Moose

作　　者：于沿泽[1] 张明海[1] 杜海荣 李倩[1] 张立博[2] 穆文静 Yu Yanze;Zhang Minhai;Du Hairong;Li Qian;Zhang Libo;Mu Wenjing(Northeast Forestry University, Harbin 150040, P. R. China;Chinese Research Academy of Environmental Sciences;Jilin Forestry Survey and Design Institute)

机构地区：[1]东北林业大学,哈尔滨150040 [2]中国环境科学研究院 [3]吉林省林业勘察设计研究院

出　　处：《东北林业大学学报》2019年第10期81-84,95,共5页Journal of Northeast Forestry University

基　　金：黑龙江省青年基金项目(QC2014C027)

摘　　要：以大兴安岭根河地区驼鹿分布数据为例,综合考虑样本选择偏差、独立训练集与验证集划分、模型复杂度,模拟了冬季驼鹿适宜栖息地分布。结果表明:最大熵模型默认设置下模型存在严重的过拟合,通过小样本赤池信息准则(AICc)选择的最优特征类型与正则化乘数组合,训练集受试者操作曲线下面积(AUC)虽有下降,但测试集AUC显著提高,模型过拟合情况得到显著改善;使用AICc最小模型拟合驼鹿分布发现,冬季驼鹿主要分布在远离人为干扰(城镇)、坡度低、靠近河流、林间道路、沼泽灌丛的地方;根河林业局驼鹿冬季适宜栖息地面积为1371hm2,真实统计技能(TSS)值为0.54。最大熵模型的应用,应根据不同的物种出现数据特点及研究目的,选择合适的参数设置。We took moose distribution data in GenHe, Great Khingan as an example, comprehensively considered sample selection bias, independent partitioning training data and test data and model complexity, then we modeled the habitat suitability of moose in winter. Under the default setting of MAXENT, there is a serious over-fitting of the model, however, model with combination of the optimal feature type and regularization multiplier selected by Akaike information criterion corrected for small samples (AICc) significantly improve condition of over-fitting. Although AUC for training data reduced, area under receiver operating curve (AUC) for test data increased significantly. Using the model with the smallest AICc model moose distribution, we found that moose usually tended to select places far from anthropogenic disturbance, with low slope, near river, near forestry road and bush in winter. The suitable habitat for moose in GenHe is 1371 ha, and true skill statistic (TSS) of classification is 0.54. Therefore, the application of MAXENT should select appropriate settings and processing methods according to study purpose and characteristic of presence data.

关 键 词：最大熵模型 驼鹿 栖息地分布 

分 类 号：S718.6[农业科学—林学] S864.1