Spatial Autocorrelation and Autoregressive Models in Ecology

Abstract

Recognition and analysis of spatial autocorrelation has defined a new paradigm in ecology. Attention to spatial pattern can lead to insights that would have been otherwise overlooked, while ignoring space may lead to false conclusions about ecological relationships. We used Gaussian spatial autoregressive models, fit with widely available software, to examine breeding habitat relationships for three common Neotropical migrant songbirds in the southern Appalachian Mountains of North Carolina and Tennessee, USA. In preliminary models that ignored space, the abundance of all three species was correlated with both local‐ and landscape‐scale habitat variables. These models were then modified to account for broadscale spatial trend (via trend surface analysis) and fine‐scale autocorrelation (via an autoregressive spatial covariance matrix). Residuals from ordinary least squares regression models were autocorrelated, indicating that the assumption of independent errors was violated. In contrast, residuals from autoregressive models showed little spatial pattern, suggesting that these models were appropriate. The magnitude of habitat effects tended to decrease, and the relative importance of different habitat variables shifted when we incorporated broadscale and then fine‐scale space into the analysis. The degree to which habitat effects changed when space was added to the models was roughly correlated with the amount of spatial structure in the habitat variables. Spatial pattern in the residuals from ordinary least squares models may result from failure to include or adequately measure autocorrelated habitat variables. In addition, contagious processes, such as conspecific attraction, may generate spatial patterns in species abundance that cannot be explained by habitat models. For our study species, spatial patterns in the ordinary least squares residuals suggest that a scale of 500–1000 m would be appropriate for investigating possible contagious processes.

Predicting tree distributions in an East African biodiversity hotspot: model selection, data bias and envelope uncertainty

“…This might simply reflect the high variability in predictor sets chosen by selection and the associated omission of different autocorrelates (Lichstein et al, 2002). It may also be the case that no single range-specific autocovariate could carry sufficient information to identify the true scales at which aggregation occurs (van Teeffelen and Ovaskainen, 2007).…”

Palaeo-precipitation is a major determinant of palm species richness patterns across Madagascar: a tropical biodiversity hotspot

“…Spatial effects are here incorporated directly into model residual structure. As a result, the overall residuals will be by definition spatially autocorrelated, but they can be decomposed into a spatial component and an error component [54]. The latter was again checked for spatial autocorrelation by computing Moran's I correlograms.…”

The shape of mammalian phylogeny: patterns, processes and scales

“…Following Lichstein et al [82] and Dormann et al [83], we tested for the presence of spatial autocorrelation in the regression residuals from our initial, ordinary least-squares (OLS) model with Moran's I correlograms, and accounted for the spatial autocorrelation with spatial autoregression (SAR, simultaneous autoregressive model). We used an SAR error model, which models the autoregressive process in the error term and has been recommended as a reliable spatial method [84].…”