Spatial autocorrelation potentially indicates the degree of changes in the predictive power of environmental factors for plant diversity. (January 2016)

• Record Type: Journal Article
• Authors: Kim, Daehyun; Shin, Young Ho
• Is Part Of: Ecological indicators. Volume 60(2016)
• Page Start: 1130
• Page End: 1141
• Publication Date: 2016-01
• Subjects: Spatial regression -- Moran's I -- Multi-scale approach -- Spatial eigenvector mapping -- Sindu coastal dunefield; Environmental monitoring -- Periodicals; Environmental management -- Periodicals; Environmental impact analysis -- Periodicals; Environmental risk assessment -- Periodicals; Sustainable development -- Periodicals; 333.71405
• DOI: 10.1016/j.ecolind.2015.09.021
• Languages: English
• ISSNs: 1470-160X

Abstract

Highlights: Dune plant diversity indices were modeled by soil and topographic attributes. In this regression, spatial filters were included as additional predictor variables. This inclusion led to decreases in the predictive power of most physical variables. Such decreases were especially obvious for spatially-autocorrelated variables. This was because the autocorrelated physical variables became redundant in the model. Abstract: In the literature of ecological indicators, more attention has yet to be paid to the potential effects of spatial autocorrelation (SAC) on the prediction of plant community structure. At a selected foredune ridge in a temperate coast of South Korea, this research developed two regression models: (1) a non-spatial ordinary least squares (OLS) in which the fine-scale (ca. 10 m) variability of diversity was predicted by soil and topographic parameters and (2) a spatial model in which spatial filters extracted by spatial eigenvector mapping were incorporated as additional predictors into the original OLS. After such incorporation, a reduction apparently occurred in the predictive power of the environmental variables, especially those with an inherently high amount of SAC. For example, Mg 2+ was the most significant predictor for species diversity in OLS, but it became insignificant in spatial regression. This indicates that, during the incorporation of SAC, the predictive importance of Mg 2+ was replaced by that of spatial filters. In other words, because the SAC of Mg 2+ was inherently strong (global Moran's I = 0.68, p < 0.001), this soil attribute became redundant when the spatial filters were added to the non-spatial OLS. These discussions corroborate the general idea of this paper that SAC potentially indicates the degree of shifts in the predictive power of environmental factors for plant diversity. In sum, we suggest that environmental variables, which are highly structured over space, should be the target of special attention and care in future modeling attempts aiming to predict the spatial patterns of plant species diversity in coastal dunes. This fine-scale approach can also be applied to macroecological studies along a variety of ecological systems, spanning latitudinal or disturbance gradients.