运用最大熵模型和随机森林模型对东北红松分布的模拟

Predicting Potential Distribution of Pinus koraiensis in Northeast China by MaxEnt Model and Random Forest Model

利用东北三省气候、土壤和地形等数据和红松分布样点,分别在MaxEnt软件和R语言中输出建模精度、因子重要性和分布范围,对比最大熵模型和随机森林模型对预测物种潜在分布结果的差异。结果表明:(1)最大熵模型训练数据AUC为0.927,检测数据AUC为0.865,随机森林模型的AUC为0.902,两个模型精度接近;(2)两个模型的输出结果有所差异,最大熵模型结果显示影响红松潜在分布的主要因子顺序是气候、地形和土壤,随机森林模型显示的结果是土壤、气候和地形;(3)两个模型模拟的红松潜在分布结果有着很大的重合度,主要集中于东北地区的中东部,说明东北地区的中东部最适合红松的生长。最大熵模型和随机森林模型精度相似,但是输出结果具有一定差异,原因是输入的训练样本不同,在构建物种分布模型时,需要考虑输入样本的合理性,分析样本对预测物种分布造成的影响。

The experiment was conducted to compare the differences between the MaxEnt and random forest models for predicting the potential distribution of species.Using the data of climate,soil and topography and the distribution of Pinus koraiensis in the three northeastern provinces,the modeling accuracy,factor importance and distribution range were output in MaxEnt software and R language,respectively.MaxEnt training data AUC value is 0.927,detection data AUC value is 0.865,random forest model AUC is 0.902,and the accuracy of the two models is close.The output results of the two models are different,the main factors affecting the potential distribution of P.koraiensis are climate,topography and soil,and the results of random forest models are soil,climate and topography.The potential distribution results of P.koraiensis simulated by the two models have a large degree of coincidence,mainly concentrated in the central and eastern parts of the Northeast,indicating that the central and eastern parts of the Northeast are most suitable for the growth of P.koraiensis.MaxEnt and random forest models have similar precision,but the output results have some differences.The reason is that the input training samples are different.When constructing the species distribution model,it is necessary to consider the rationality of the input samples and analyze the possible influences brought by the samples to the species distribution.