提高生态位模型时间转移能力的方法研究

针对经典生态位模型模拟物种潜在分布时,模型转移能力较低,影响模拟准确性的问题,以MaxEnt模型为例,研究基于Spearman’s rho相关性系数的环境变量筛选,基于方差膨胀因子VIF的环境变量筛选和调节"Regularization Multiplier"模型参数,对模型时间转移能力("预测系数"|r_f|和"追溯系数"|r_b|)的影响。通过对比分析,提出了提高生态位时间转移能力的方法。利用美国阿肯色流域鲤鱼2个时期的分布数据与环境数据,验证了通过降低模型维度和复杂程度提高模型时间转移能力的有效性。结果表明:增大模型参数"Regularization Mutiplier"值对转移能力的提高最显著,但值过高,会使物种适生概率与环境变量的反应曲线过于平滑而导致信息丢失;而增大"Regularization Multiplier"值和VIF方法相结合,可以使得模型|r_f|和|r_b|均大于0.85。建议在保证模型自适应的基础上,将变量相关性分析与模型参数调节相结合,以提高模型的转移能力。本研究结果能够为生态位模型精度的验证和时间转移能力的提高提供借鉴经验。

Study of methods to improve the temporal transferability of niche model

Low model transferability affects the accuracy of the predicted potential species distribution.Taking MaxEnt model as an example,we studied three methods to improve the temporal transferability by reducing the model dimension and complexity:Environment variables selection based on the Spearman''s rho correlation coefficient r ,and selection based on the variance inflation factor VIF ,and adjustment of the parameter of "Regularization Multiplier" model.The Arkansas River fish distribution and environmental data at two periods were used for the verification of the methods to improve the "forecasting" and "backcasting" transferability of the model.The results showed that:The improvement of "Regularization Mutiplier" value enhanced the model''s transferability significantly;However,the high "Regularization Mutiplier" value caused the smoother response curves of the relationship between the species distribution probability and environment variable and resulted in information loss.Combination of the parameter "Regularization Multiplier" and the variable selection based on VIF caused the greater (>0.8) value of r_f and r_b.We proposed to combine variable correlation analysis with model parameter adjustment to improve the transferability of the model.This research can provide reference for niche model accuracy validation and improvement of temporal transferability.