摘要
我们介绍一种估计驯鹿(Rangifer tarandus)的空间分布的方法。描述野生动物物种分布的数据,其内在的空间自相关性包含可以用来提高野外调查效率的信息。我们的数据包括驯鹿粪便小球数、卫星图像和一个数字化地形模型。基于这些综合数据,我们采用普通逻辑回归、自动逻辑回归和吉布斯抽样器预测驯鹿的空间分布。通过训练数据集,对每种方法所采取的不同抽样设计结果进行比较。结果表明,在预测精度(质量)降低15%的情况下,样方数量可减少75%。我们的结果表明,就精度而言,吉布斯抽样器要优于逻辑回归,但结果取决于区域光谱的均质性,以及高度曲线相对位置的取样设计。当对驯鹿的分布进行更精细的建模时(<1km),我们的结果证明结合空间信息进行预测是必要的。
We describe an approach to estimation of the spatial distribution of reindeer (Rangifer tarandus).Spatial autocorrelation,inherent to the data describing the distribution of wildlife species,contains information that can be utilized to improve the effciency of field inventories.Our data included reindeer fecal pellet counts, satellite imagery and a digital terrain model.We applied ordinary logistic regression, autologistic regression,and the Gibbs sampler to predict spatial distribution of reindeer based on the combined data.A training set was used to compare the outcome for different field sampling designs for each method.Results suggested the possibility to reduce the number of plots by up to 75% with a 15% reduction in prediction accuracy (quality).We also showed that the Gibbs sampler outperformed, in terms of accuracy,the logistic regression.The outcome,however,was dependent on the spectral homogeneity of the area and on the relative position of the sampling design to the elevation curves.Our results justify the incorporation of spatial information when modeling the distribution of reindeer at finer scales (< 1 km).
