Computational protocol: Relative Importance of Current and Past Landscape Structure and Local Habitat Conditions for Plant Species Richness in Dry Grassland-Like Forest Openings

Similar protocols

Protocol publication

[…] The statistical analyses were performed using S-plus 4.6 and Statistica 7.0 . The dependent variable, the number of species, was normally distributed, and so no transformation of the data was needed.To reduce the number of independent PDSI and slope variables, we used a pair-wise correlation matrix (Statistica 7.0 ) and selected the least correlated ones for use in subsequent analyses (for PDSI – median value for December and June; for slope – median and maximum value). To assess the relationship between all variables and their categories, we also calculated a pair-wise correlation matrix () in Statistica 7.0 .To take into account the linear spatial trends (spatial auto-correlation) of parameters describing localities across the study area, we used variation partitioning according to Borcard et al. and Legendre and Legendre , which enabled us to separate the pure spatial component from the pure environmental component and their shared contribution. To identify complex spatial trends, seven derived geographical variables were constructed by including all quadratic and cubic combinations of x and y as suggested by Borcard et al. : x, y, x2, xy, y2, x3, x2y, xy2, y3. We used the values selected in the stepwise regression (x, y, xy) as covariates in all subsequent tests to remove the effect of spatial position of the localities (spatial auto-correlation) because we wanted to study only the pure effect of the environment and not of the spatial component.The tests of the effects of all the independent variables () on species richness were performed in three steps. First, we tested the effect of each variable on species richness separately without any covariates, which represents the fraction of variation explained by non-spatial environmental variation and spatially structured environmental variation together (e.g., shared contribution of environmental and spatial variation). Second, we used the geographical coordinates of the center of each locality (those selected in the stepwise regression) as covariates and tested the effect of each independent variable separately, which represents the fraction of variation explained by non-spatial environmental variation , . Finally, to obtain the pure effect of each variable after removing spatially structured environmental variation and any other shared variation with all other parameters, we used geographical coordinates and all significant factors from the second analyses as covariates (according to and ). We used an analysis of variance (ANOVA) with type III sum of squares (S-plus 4.6) to identify a real effect of particular factors without the effect of all other factors. In the case of categorical variables with more than one category (substrate) and variables with multiple levels (slope, PDSI), we analyzed these categories or levels together using the difference between the models with and without all of the categories or levels. The means, medians and ranges of each predictor and dependent variable per locality are presented in .For assessing the species-area relationship, we compared the fit provided by two alternative functions, a logarithmic function and a power function, using SPSS . The test revealed that the logarithmic function explained a higher amount of variation in our data (R2 = 0.36) than the power function (R2 = 0.22). We thus decided to use the logarithmic function in the following analyses.Alternatively, it is also possible to test the effect of studied factors on species richness using a multimodel inference analysis based on AIC , which would alleviate the problem of testing many partially correlated variables using p-values. The comparison of models based on AIC yielded extremely similar results to our first approach, and thus, only the results based on the p-values are presented.To assess the relative importance of the three groups of factors (current landscape structure, historical landscape structure and habitat conditions) for species richness, we analyzed the effect of each of these three groups of factors alone. We also tested the effect of each of these groups of factors after using the other two groups of factors as covariates. Because each group contained too many independent variables, we chose only those factors that were significant when testing their independent effect on species richness using only coordinates as covariates. We used variance partitioning (according to ) to calculate the proportion of variance explained by each group of factors. We expressed the portion of explained variance as a relative part of the total explained variance. […]

Pipeline specifications

Software tools Statistica, SPSS
Application Miscellaneous
Diseases Pulmonary Fibrosis, Xerostomia