The impact of land use on non-native species incidence and number in local assemblages worldwide

While the regional distribution of non-native species is increasingly well documented for some taxa, global analyses of non-native species in local assemblages are still missing. Here, we use a worldwide collection of assemblages from five taxa - ants, birds, mammals, spiders and vascular plants - to assess whether the incidence, frequency and proportions of naturalised non-native species depend on type and intensity of land use. In plants, assemblages of primary vegetation are least invaded. In the other taxa, primary vegetation is among the least invaded land-use types, but one or several other types have equally low levels of occurrence, frequency and proportions of non-native species. High land use intensity is associated with higher non-native incidence and frequency in primary vegetation, while intensity effects are inconsistent for other land-use types. These findings highlight the potential dual role of unused primary vegetation in preserving native biodiversity and in conferring resistance against biological invasions.


Supplementary Figures
. Spatial cross-validation of models with respect to spatial biases from uneven sampling across biomes using LU-type as the only fixed effect. Supplementary Figure 7. Spatial cross-validation of model results with respect to the location of the assemblage on an island or on the mainland. Supplementary Figure 8. Spatial cross-validation of models with respect to the location of the assemblage on an island or on the mainland using LU-type as the only fixed effect. Table title  Supplementary Table 1. The classification of LU-types andintensity levels in our analysis. Supplementary Table 2. Sources of regional non-native species distribution for the taxa analysed Supplementary Table 3. Invalid or unresolved species in PREDCITS assemblages that were removed from analyses Supplementary Table 4. Percentages of local assemblages with non-natives across all taxa and for each taxon separately. Supplementary Table 5. Odds ratios of non-native incidence in local assemblages in response to the interaction of LU-type and LU-intensity.

Supplementary Tables
Supplementary Table 6. Odds ratios of non-nativeincidence in local assemblages in response to LU-type. Supplementary Table 7. Comparison of models of non-native incidence, number and proportion that do or do not include the assignment of assemblage locations to one of the 14 biomes distinguished in Supplementary References 13 into the random effects structure. Supplementary Table 8. Cross-validation of robustness of coefficient estimates in the models of non-native incidence, number and proportion computed by excluding assemblages from one biome in turn. Supplementary Table 9. Comparison of model results for non-native incidence, number and proportions without and with including the size of the sampling area of assemblages as additional fixed-effects predictor Supplementary Table 10. Comparison of models of non-native incidence, number and proportion that either account or do not account for the location of the assemblage on an island or on the mainland Supplementary Table 11. Test of robustness of coefficient estimates in the models of nonnative incidence, number and proportion computed by excluding assemblages from t islands. Supplementary Table 12. Number of non-native species in local assemblages in response to the interaction of LU-type and LU-intensity. Supplementary Table 13. Number of non-native species in local assemblages in response to LU-type. Supplementary Table 14. Proportion of non-native species in local assemblages in response to the interaction of LU-type and LU-intensity. Supplementary Table 15. Proportion of non-native species among all species in local assemblages in response to LU-type. Supplementary Table 16. Number of all assemblages and of those with at least one nonnative species, given separately for the LU-types.
Supplementary Figure 1. Geographical distribution of local assemblages with at least one non-native species for the taxon. A, ants; b, birds; c, mammals; d; spider; e, vascular plants.
The colors indicate distribution of assemblages with at least one non-native species (blue points) and the assemblage with no non-native species (black points); respectively. Silhouette illustrations for the taxa are from PhyloPic (http://phylopic.org), contributed by various authors under public domain license.
Supplementary Figure 2. Odds ratio of non-native incidence in local assemblages in response to LU-type (a) and LU-intensity (b) separately. The logistic generalized linear mixed effects model with LU-type (a) and with LU-intensity (b) were used separately (n=11,713, 9869; respectively). Odds ratios (the means and standard errors) were back-transformed pairwise contrasts to the reference level 'primary vegetation' and 'minimal use'; respectively. The asterisks in the figure indicate significant differences (p values: **<0.01 and ***<0.001; respectively).
Supplementary Figure 3. Number of non-native species for assemblages with at least one nonnative species in response to LU-type (left panel) and LU-intensity (right panel) separately. Numbers of non-native species were back-transformed from a generalized linear mixed effects model (GLMM) with a compact letter display of all pairwise comparisons of estimated marginal means (n=2450, 2314; respectively). The means denoted by a different letter indicate significant differences (p<0.05). Data are presented as mean values and standard errors.
Supplementary Figure 4. Non-native proportions of local assemblages in response to changes in LU-type (left panel) and LU-intensity (right panel) separately. Non-native proportions were back-transformed from a generalized linear mixed effects model (GLMM) and a compact letter display of all pairwise comparisons of estimated marginal means (n=2,450, 2,314; respectively). Means denoted by a different letter are significantly different (p<0.05). Data are presented as mean values and standard errors.
Supplementary Figure 5. Spatial cross-validation of the full model results with respect to spatial biases from uneven sampling across biomes. Models were re-run by excluding all assemblages from one biome in turn. a, models of non-native incidence (n=11,693); b, non-native species number (n=2,314) and c, non-native species proportions in local assemblages (n=2,314). Symbols show the mean coefficient values across all models, grey bars their 95% confidence intervals and coloured bars the mean standard errors. All models used LU-type, LU-intensity and their interaction as fixed effects and study-site blocks nested in study sites (SSB/SS) as random factors. Models of non-native incidence were run 14 times, corresponding to the 14 biomes in Supplementary References 13 . Models of non-native species number and proportions were run 11 times because some biomes had too low numbers of assemblages. Coefficient estimates of statistical models are back-transformed pairwise contrasts to the reference level 'primary vegetation under Minimal use' in each model. The 'na' in (a) indicates that urban assemblages under intense use could not be included due to lack of variation in the response (all assemblages had alien species).
Supplementary Figure 6. Spatial cross-validation of models with respect to spatial biases from uneven sampling across biomes using LU-type as the only fixed effect. a-f, models of nonnative incidence across all taxa (n=11,713) and each taxon separately (n=407, 3,978, 1,114, 762 and 4,453; respectively); g,h, models of non-native species number and i, j models of nonnative species proportions across all taxa (n=2,450) and for each taxon separately (n=113, 457, 292, 182 and 1,406; respectively). Symbols show the mean coefficient values across all models, grey bars their 95% confidence intervals and coloured bars the mean standard errors. The values are back-transformed pairwise contrasts to the reference level 'primary vegetation'.The NA in (d, e, f) indicates that these LU-types were removed from the analysis due to low numbers of the assemblages (see Methods for details). In case of non-native species number, mammals were not analyzed due to low variance in species number per assemblage. Spiders were not analyzed due to low sample size. Silhouette illustrations for the taxa are from PhyloPic (http://phylopic.org), contributed by various authors under public domain license. Figure 7. Spatial cross-validation of model results with respect to the location of the assemblage on an island or on the mainland. We re-run two full models, one including all assemblages and the other one all but those from islands. a, models of non-native incidence (n=11,693); b, non-native species number (n=2,314) and c, non-native species proportions (n=2,314). Symbols show the mean coefficient values across the two models, coloured bars indicate the mean standard errors. All models used LU-type, LU-intensity and their interaction as fixed effects and study-site blocks nested in study sites (SSB/SS) as random factors. Coefficient estimates of statistical models are back-transformed pairwise contrasts to the reference level 'primary vegetation under Minimal use' in each model. The 'NA' in (a) indicates that urban assemblages under intense use could not be included due to lack of variation in the response (all assemblages had alien species).

Supplementary
Supplementary Figure 8. Spatial cross-validation of models with respect to the location of the assemblage on an island or on the mainland using LU-type as the only fixed effect. We re-run two full models, one including all assemblages, and one all but those from islands. a-f, models of non-native incidence across all taxa (n=11,713) and each taxon separately (n=407, 3,978, 1,114, 762 and 4,453; respectively); g,h, models of non-native species number and i, j models of non-native species proportions across all taxa (n=2,450) and for each taxon separately (n=113, 457, 292, 182 and 1,406; respectively). Symbols show the mean coefficient values across the two models, coloured bars indicate the mean standard errors. The values are back-transformed pairwise contrasts to the reference level 'primary vegetation'.The 'NA' in (d, e, f) indicates that these LU-types were removed from the analysis due to low numbers of the assemblages (see Methods for details). In case of nonnative species number, mammals were not analyzed due to low variance in species numbers per assemblage. Spiders were not analyzed due to the model convergence by excluding the assemblages from island. Silhouette illustrations for the taxa are from PhyloPic (http://phylopic.org), contributed by various authors under public domain license. Includes synonyms such as "ancient woodlands", "old-growth forests" or "natural grasslands" Any disturbances identified are minor.
One or more disturbances of moderate intensity.
One or more disturbances that are severe enough to markedly change the natural ecosystems. Primary sites in urban areas are intensely used.

Secondary
The   . 'NA's indicate that the respective factor levels could not be included in the analysis due to low assemblage number. For non-native species number, mammals were not analyzed due to low variance in species number of assemblages. Spiders were not analyzed due to low sample size (see the details in Methods).     Table 11. Test of robustness of coefficient estimates in the models of nonnative incidence, number and proportion computed by excluding assemblages from islands. The values of coefficient estimates are back-transformed pairwise contrasts to the reference level 'primary vegetation under Minimal use' in the full model and to the reference level 'primary vegetation' in theLU-type only model. Shown are mean estimates, mean standard errors of coefficient estimates across the two models without assemblages from islands and with the assemblages from islands. 'NA's indicate that the respective factor levels could not be considered in the models due to low sample size, lack of response variation (nonnative species number in case of mammals), or problems with (spiders for models without assemblages from islands (see the details in Methods)).