基于物种分布模型对未来气候变化下云南松毛虫在四川省适生区的预测

为明确云南松毛虫Dendrolimus houi于未来气候变化下在四川省的分布情况，运用最大熵（maximum entropy，MaxEnt）模型中的刀切法和Pearson相关系数分析法对未来气候数据（2050年和2070年）、林地因子及人为因子进行权重划分，筛选出影响云南松毛虫潜在分布的重要且相关系数较低的环境因子，并结合openModeller中的人工神经网络（artificial neural network，ANN）模型、生物气候（BioClim）模型、气候空间（climate space，CS）模型、气候信封（envelope score，ES）模型、基于统计概率和规则集的遗传算法（genetic alorithm for rule-set production，GARP）模型、MaxEnt模型和支持向量机（support vector machine，SVM）模型对云南松毛虫未来的适生区进行预测，利用AUC（area under curve）评价模型的精确度。结果发现，最暖季降水量、人类足迹指数、最冷季降水量和海拔对云南松毛虫在未来气候条件下的潜在分布有较强影响，在2050年贡献率分别为27.2%、16.0%、2.0%和4.9%，在2070年贡献率分别为20.6%、16.8%、9.7%和4.9%。对比7种模型的预测结果，发现SVM模型在2050年和2070年对云南松毛虫适生区预测的AUC为0.93，预测精确度最高，具有较高的可信度；该模型预测结果显示，从2050年至2070年云南松毛虫在四川省的总适生面积增加了4 269.8 km²，其中，中、低适生区面积共减少了17 185.8 km²，高适生区面积增加了21 455.6 km²。

Projections of Yunnan pine moth Dendrolimus houi in Sichuan Province under future climate change based on species distribution model

In order to understand the dynamic distribution of Yunnan pine moth Dendrolimus houi in Sichuan Province under the future climate, the Jackknife method in maximum entropy (MaxEnt) model and Pearson correlation coefficient analysis was used to divide the future climate data (2050 and 2070), forest land factors and human factors into weights, screen out the important environmental factors influencing the potential distribution of D. houi and with low correlation coefficients, combined with the seven species distribution models in openModeller, including artificial neural network (ANN), Biocliom, climate space (CS), envelope score (ES), MaxEnt, genetic alorithm for rule-set production (GARP), support vector machine (SVM), predict the future suitable areas of D. houi and use AUC values to evaluate the accuracy of the model. The results showed that precipitation in the warmest season, human footprint index, precipitation in the coldest season and altitude had strong influences on the potential distribution of Yunnan pine moth under future climate conditions. Their contribution rates were 27.2%, 16.0%, 2.0% and 4.9% in 2050, and 20.6%, 16.8%, 9.7% and 4.9% in 2070, respectively. Among the seven species distribution models, the AUC value predicted by SVM model for 2050 and 2070 was 0.93, which had the highest accuracy and strong credibility. From 2050 to 2070, the overall suitable areas may increase by 4 269.8 km²; the moderately and slightly suitable areas may decrease by 17 185.8 km², and the highly suitable areas may increase by 21 455.6 km².