Search

Lichens associated with old forest are commonly assumed to be negatively affected by tree logging or natural disturbances. However, in this study performed in a spruce-dominated sylvopastoral landscape in the Swiss Jura Mountains, we found that genetic diversity of the epiphytic old-forest lichen Lobaria pulmonaria depends on the type of disturbance. We collected 923 thalli from 41 sampling plots of 1 ha corresponding to the categories stand-replacing disturbance (burnt), intensive logging (logged) and uneven-aged forestry (uneven-aged), and analysed the thalli at six mycobiont-specific microsatellite loci. We found evidence for multiple independent immigrations into demes located in burnt and logged areas. Using spatial autocorrelation methods, the spatial scale of the genetic structure caused by the clonal and recombinant component of genetic variation was determined. Spatial autocorrelation of genotype diversity was strong at short distances up to 50 m in logged demes, up to 100 m in uneven-aged demes, with the strongest autocorrelation up to 150 m for burnt demes. The spatial autocorrelation was predominantly attributed to clonal dispersal of vegetative propagules. After accounting for the clonal component, we did not find significant spatial autocorrelation in gene diversity. This pattern may indicate low dispersal ranges of clonal propagules, but random dispersal of sexual ascospores. Genetic diversity was highest in logged demes, and lowest in burnt demes. Our results suggest that genetic diversity of epiphytic lichen demes may not necessarily be impacted by stand-level disturbances for extended time periods.

In landscape genetics, model selection procedures based on Information Theoretic and Bayesian principles have been used with multiple regression on distance matrices (MRM) to test the relationship between multiple vectors of pairwise genetic, geographic, and environmental distance. Using Monte Carlo simulations, we examined the ability of model selection criteria based on Akaike’s information criterion (AIC), its small-sample correction (AICc), and the Bayesian information criterion (BIC) to reliably rank candidate models when applied with MRM while varying the sample size. The results showed a serious problem: all three criteria exhibit a systematic bias toward selecting unnecessarily complex models containing spurious random variables and erroneously suggest a high level of support for the incorrectly ranked best model. These problems effectively increased with increasing sample size. The failure of AIC, AICc, and BIC was likely driven by the inflated sample size and different sum-of-squares partitioned by MRM, and the resulting effect on delta values. Based on these findings, we strongly discourage the continued application of AIC, AICc, and BIC for model selection with MRM. [ABSTRACT FROM AUTHOR]

Spatial autocorrelation jeopardizes the validity of statistical inference, for example correlation and regression analysis. Restricted randomization methods can account for the effect of spatial autocorrelation in the observed data by building it into an empirical null model for hypothesis testing. This can be achieved, for example, based on conditional simulation, which fits a highly parameterized geostatistical model to the observed spatial structure, or, for data observed on a regular transect or grid, with Fourier spectral randomization methods that can flexibly model spatial structure at any scale. This study uses Moran eigenvector maps to extend spectral randomization to irregularly spaced samples., We present different algorithms to perform restricted randomization to suit different types of research questions: individual randomization of each variable, joint randomization of a group of variables while keeping within-group correlations fixed, and randomization with a fixed correlation between original data and randomized replicates (e.g. as input for simulation studies). The performance of the proposed Moran spectral randomization methods for regularly and irregularly spaced samples is assessed with correlation analysis of simulated data., Moran spectral randomization closely matched the spatial structure of original simulated data sets, with identical or nearly identical Moran's I values and power spectra, depending on the algorithm. In correlation analysis of two spatially autocorrelated variables, Moran spectral randomization produced correct type I error rates for stationary spatial data, even for very small and highly irregular samples, but was sensitive to linear trend. When one or both variables lacked spatial structure, Moran spectral randomization tests were more conservative than correlation t-tests., The proposed Moran spectral randomization method requires a minimum of parameterization and is able to address multivariate data with spatial structure at multiple scales, with the option of controlling levels of correlation with the original data. It can provide technically unlimited numbers of randomizations even for small samples while closely maintaining the spatial characteristics of uni- or multivariate data at all spatial scales. The method is applicable for correlation analysis of stationary, autocorrelated spatial or temporal series. Further research should assess whether the method can be extended to multiple regression analysis. [ABSTRACT FROM AUTHOR]

The linear regression model, with its numerous extensions including multivar-iate ordination, is fundamental to quantitative research in many disciplines. However, spatial or temporal structure in the data may invalidate the regression assumption of independent residuals. Spatial structure at any spatial scale can be modeled flexibly based on a set of uncorrelated component patterns (e.g., Moran's eigenvector maps, MEM) that is derived from the spatial relationships between sampling locations as defined in a spatial weight matrix. Spatial filtering thus addresses spatial autocorrelation in the residuals by adding such component patterns (spatial eigenvectors) as predictors to the regression model. However, space is not an ecologically meaningful predictor, and commonly used tests for selecting significant component patterns do not take into account the specific nature of these variables. This paper proposes "spatial component regression" (SCR) as a new way of integrating the linear regression model with Moran's eigenvector maps. In its unconditioned form, SCR decomposes the relationship between response and predictors by component patterns, whereas conditioned SCR provides an alternative method of spatial filtering, taking into account the statistical properties of component patterns in the design of statistical hypothesis tests. Application to the well-known multivariate mite data set illustrates how SCR may be used to condition for significant residual spatial structure and to identify additional predictors associated with residual spatial structure. Finally, I argue that all variance is spatially structured, hence spatial independence is best characterized by a lack of excess variance at any spatial scale, i.e., spatial white noise. [ABSTRACT FROM AUTHOR]