基于野外调查的居群分布信息和20个环境变量( 包括海拔及19 个气候变量), 采用最大熵模型(MaxEnt)和地理信息系统(ArcGIS)对中国特有种天山猪毛菜(Salsola junatovii Botsch.)的潜在分布区进行预测;并采用受试者工作特征曲线(ROC)和刀切法(jackknife test)分别检验MaxEnt 模型的精度和评估各环境变量在决定潜在分布区时的贡献量。结果显示:天山猪毛菜主要分布在中国新疆南部的12 个县(包括托克逊县、和硕县、和静县、焉耆县、拜城县、库车县、温宿县、乌什县、阿合奇县、阿图什县、乌恰县和阿克陶县),但在相关文献记载的曾有分布的阿克苏市、柯坪县和喀什市则本调查中没有发现。通过MaxEnt 模型预测,天山猪毛菜的潜在适生区主要集中在新疆南部(包括天山南坡和塔里木盆地西南缘)以及甘肃的西部和东部等,新疆西部地区有零星分布;在调查的15 个居群中,除居群14(位于乌恰县西部)外,其他14 个居群均位于潜在适生区内,其中的7 个居群位于非常适宜的潜在适生区内,表明利用MaxEnt 模型预测天山猪毛菜的潜在适生区效果较好。在20 个环境变量中,对预测结果贡献量位居前3 位的环境变量为最冷季度平均温度、最冷月最低温和最干季度平均温度,表明该种的分布与低温相关。此外,对天山猪毛菜潜在适生区与实际分布区差异的成因进行了分析。

Based on population distribution information and 20 environmental variables ( including altitude and 19 climatic variables) in field investigation, potential distribution areas of Chinese endemic species Salsola junatovii Botsch. were predicted by the maximum entropy model ( MaxEnt) and archaeological geographical information system (ArcGIS), MaxEnt model precision was tested by receiver operating characteristic curve ( ROC), and contribution of each environmental variable in deciding potential distribution areas was evaluated by jackknife test. The results show that S. junatovii is mainly distributed in 12 counties (including Toksun County, Hoxud County, Hejing County, Yanqi County, Baicheng County, Kuche County, Wensu County, Wushi County, Akqi County, Artux County, Wuqia County and Akto County) of the south of Xinjiang of China, but S. junatovii ever distributed and recorded in related literatures is not found in Akesu City, Keping County and Kashi City in this investigation. Based on prediction of MaxEnt model, the potential suitable distribution areas of S. junatovii is mainly located in the south of Xinjiang (including the south slope of Tianshan Mountain and the southwest rim of Tarim Basin) and the west and east of Gansu, and is also sporadically distributed in the west of Xinjiang. Among 15 populations investigated, except population 14 (locating in the west of Wuqia County), other 14 populations all are located in potential suitable distribution areas, in which 7 populations are located in very suitable potential distribution areas. It is indicated that the effect of MaxEnt model using to predicting potential suitable distribution areas of S. junatovii is better. Among 20 environmental variables, the environmental variables ranked the top three of contribution to prediction results are mean temperature of the coldest quarter, the minimum temperature of the coldest month and mean temperature of the driest quarter, which means that distribution of S. junatovii is related to low temperature. Besides, reason for difference between potential suitable distribution areas and actual distribution areas of S. junatovii is analyzed.