使用大熊猫数据评估Biomod2和MaxEnt分布预测模型的表现

物种分布模型是物种研究和保护者常用的工具.不同模型的预测结果可能相差很大,对研究者选择模型造成一定的难度.本研究使用大熊猫的实际分布数据评估了两种常见物种分布模型Biomod2和最大熵模型(MaxEnt)的表现,运用ROC曲线下面积(area under the curve,AUC)、真实技巧统计值(true skill statistics,TSS)、KAPPA统计量3种指标综合评估了两种模型预测结果的准确度.结果表明: 当使用的物种分布数据和模拟重复次数足够多的时候,两者都能够给出相当准确的预测.相对于MaxEnt,Biomod2的预测准确度更高,尤其是在物种分布点稀少的情况下.然而,Biomod2使用难度较大,运行时间较长,数据处理能力有限.研究者应基于对预测结果的误差要求来选择模型.在误差要求明确且两个模型都能满足误差要求时,建议使用MaxEnt,否则应优先考虑使用Biomod2.

Evaluating the performance of species distribution models Biomod2 and MaxEnt using the giant panda distribution data

Species distribution models (SDMs) are widely used by researchers and conservationists. Results of prediction from different models vary significantly, which makes users feel difficult in selecting models. In this study, we evaluated the performance of two commonly used SDMs, the Biomod2 and Maximum Entropy (MaxEnt), with real presence/absence data of giant panda, and used three indicators, i.e., area under the ROC curve (AUC), true skill statistics (TSS), and Cohen’s Kappa, to evaluate the accuracy of the two model predictions. The results showed that both models could produce accurate predictions with adequate occurrence inputs and simulation repeats. Comparedto MaxEnt, Biomod2 made more accurate prediction, especially when occurrence inputs were few. However, Biomod2 was more difficult to be applied, required longer running time, and had less data processing capability. To choose the right models, users should refer to the error requirements of their objectives. MaxEnt should be considered if the error requirement was clear and both models could achieve, otherwise, we recommend the use of Biomod2 as much as possible.