Assessment of alternative approaches for bioclimatic modeling with special emphasis on the Mahalanobis distance
Assessment of alternative approaches for bioclimatic modeling with special emphasis on the Mahalanobis distance
复制标题
DOI：
10.1016/s0304-3800(02)00327-7
复制
发表时间：
2003-02-01
影响因子：
3.1
通讯作者：
Kadmon, R
中科院分区：
文献类型：
--
作者：
Farber, O;Kadmon, R
我们将马氏距离的概念引入到生物气候模拟中。具体地说，我们认为，由马氏距离定义的气候包络比标准的直线包络(例如，由BIOCLIM产生的那些)对物种分布的预测更准确。我们的假设基于三个基本原理：(1)由马氏距离产生的气候包络是倾斜的，因此可以处理气候变量之间的相关和相互作用；(2)马氏包络是椭圆形的，因此RID，更好地反映了生态位理论所表达的中心趋势原理；(3)马氏预测是基于整个数据而不是最外部的观测数据，因此对异常值不那么敏感。我们使用192种木本植物在以色列的分布数据来检验我们的假设。基于四个准确度指标(敏感度、特异度、总体准确度和Kappa统计量)的验证测试支持了我们的假设，并表明马氏模型产生的预测比相应的直线模型产生的预测要准确得多。另外的模拟实验表明，马氏模式的优越性不能与它们的椭圆形有关，也不与它们处理气候变量之间的相关性的能力有关。因此，我们的结论是，马氏模型的主要优势来自于这样一个事实，即它们的气候包络是使用所有观测数据定义的，而不是建立在最外面的观测数据基础上的直线包络。(C)2002 Elsevier Science B.V.保留所有权利。
We introduce the concept of the Mahalanobis distance to bioclimatic modeling. Specifically, we argue that climatic envelopes defined by the Mahalanobis distance produce more accurate predictions of species distribution than standard rectilinear envelopes (e.g. those produced by BIOCLIM). We base our hypothesis on three rationales: (1) the climatic envelope generated by the Mahalanobis distance is oblique, and therefore, may cope with correlations and interactions among the climatic variables; (2) the Mahalanobis envelope is elliptic, a rid therefore, better reflects the principle of central tendency as expressed by niche theory; (3) Mahalanobian predictions are based on the whole data rather than on the outermost observations, and are therefore, less sensitive to outliers. We test our hypothesis using data on the distribution of 192 species of woody plants in Israel. Validation tests based on four measures of accuracy (sensitivity, specificity, overall accuracy and the Kappa statistic) support our hypothesis, and suggest that Mahalanobis models produce predictions that are significantly more accurate than those produced by corresponding rectilinear models. Additional simulation experiments demonstrate that the superiority of Mahalanobian models cannot be related to their elliptic-shape, of their ability to cope with correlations among the climatic variables. Accordingly, our conclusion is that the prime advantage of Mahalanobian models originates from the fact that their climatic envelopes are defined using all the observations, as opposed to rectilinear envelopes that are founded on the outermost observations. (C) 2002 Elsevier Science B.V. All rights reserved.