Institut National de Recherche en Informatique et en Automatique, Sophia Antipolis-Mediterranée, Research Team MERE, Unité Mixte de Recherche Systems Analysis and Biometrics, 2 place Pierre Viala, Montpellier, France.
Am Nat. 2010 Apr;175(4):E74-90. doi: 10.1086/650718.
Entropy maximization (EM, also known as MaxEnt) is a general inference procedure that originated in statistical mechanics. It has been applied recently to predict ecological patterns, such as species abundance distributions and species-area relationships. It is well known in physics that the EM result strongly depends on how elementary configurations are described. Here we argue that the same issue is also of crucial importance for EM applications in ecology. To illustrate this, we focus on the EM prediction of species-level spatial abundance distributions. We show that the EM outcome depends on (1) the choice of configuration set, (2) the way constraints are imposed, and (3) the scale on which the EM procedure is applied. By varying these choices in the EM model, we obtain a large range of EM predictions. Interestingly, they correspond to spatial abundance distributions that have been derived previously from mechanistic models. We argue that the appropriate choice of the EM model assumptions is nontrivial and can be determined only by comparison with empirical data.
最大熵原理（EM，也称为最大熵）是一种起源于统计力学的一般推理方法。它最近已被应用于预测生态模式，例如物种丰富度分布和物种面积关系。在物理学中，众所周知，EM 结果强烈取决于如何描述基本配置。在这里，我们认为对于生态学中的 EM 应用，同样的问题也至关重要。为了说明这一点，我们重点研究 EM 对物种水平空间丰度分布的预测。我们表明，EM 的结果取决于：（1）配置集的选择，（2）施加约束的方式，以及（3）EM 过程应用的范围。通过在 EM 模型中改变这些选择，我们得到了大量的 EM 预测。有趣的是，它们对应于先前从机械模型中得出的空间丰度分布。我们认为，EM 模型假设的适当选择并非微不足道，只能通过与经验数据进行比较来确定。
