Untangling human development and natural gradients: Implications of underlying correlation structure for linking landscapes and riverine ecosystems

Increasingly, ecologists seek to identify and quantify relationships between landscape gradients and aquatic ecosystems. Considerable statistical challenges emerge in this effort, some of which are attributable to multicollinearity between human development and landscape gradients. In this paper, we measure the covariation between human development—such as agriculture and urbanization – and natural landscape gradients – such as valley form, climate and geology. With a dataset of wade-able streams from coastal Oregon (USA), we use linear regression to quantify covariation between human activities and landscape gradients. We show that the correlation between human development and natural landscape gradients varies dramatically with the scale of observation. Similarly, we show how the correlation varies by region, even within a scale of interest. We then use a simulation experiment to demonstrate how this inherent covariation can hinder statistical efforts to identify mechanistic links between landscape gradients and features of aquatic ecosystems. We illustrate the negative consequences of the underlying correlation structure for statistical efforts: infl ated goodness-of-fi t metrics and infl ated error terms on key coeffi cients that may undermine model building. We conclude by discussing the current best statistical practices for dealing with multicollinearity as well as the limitations of existing statistical tools.