Evidence for the importance of land use, site characteristics and vegetation composition for rooting in European Alps

Study sites

To obtain a cross-section of land-use types through the Eastern Alps (Fig. 2), rooting samples were taken from Tyrol (Austria) and from South Tyrol and northern Trentino (both in Italy), which include two climatic regions—the central European climatic region in the northern part and the sub-Mediterranean climatic region in the southern part of the research area⁴⁷. The average annual precipitation at the 13 study sites ranges from 400 to 2000 mm, with maximum rainfall observed from June to July⁴⁷. Mean annual temperature ranges from 0 °C to 9 °C. Additional climatic variability was added by sampling at elevations from 650 to 2680 m a.s.l. The bedrock in the research area is composed of calcareous sedimentary rock in the northern and southern regions and of crystalline rock in the main chain of the Alps, sometimes with superimposed calcareous isles: Stubai Valley (North Tyrol) is geologically dominated by silicate with transitions to limestone; Ötz Valley, Ziller Valley and Igls/Patsch (all North Tyrol), Passeier Valley, Mühlbach, Matsch, Ritten and Jenesien (South Tyrol) are geologically dominated by silicate; and Leutasch (North Tyrol), St. Vigil and Toblach (both South Tyrol) and Monte Bondone (near Trento) are geologically dominated by limestone. The pH of the topsoil (0–10 cm), which ranges from 3.7 to 7.8³², is determined by bedrock and land use⁴⁸. For more details on the study region, see Supplementary Appendix S1.

To be representative, the most widespread vegetation communities in the 13 study sites for all land-use types (arable land, intensively used hay meadow, lightly managed hay meadow, pasture, agriculturally unused grasslands, and forest) were analyzed (Supplementary Appendix S2). Overall, a total of 171 soil samples were taken, with 15 samples from arable land, 56 samples from intensively used hay meadows, 15 samples from extensively managed hay meadows, 16 samples from lightly stocked pastures, 32 samples from agriculturally unused grasslands, and 37 samples from forests. Meadows that are mown and fertilized with slurry and/or manure at least twice a year were defined as intensively used hay meadows. An extensively managed hay meadow was not fertilized and mown only once a year. Pastures were extensively grazed by cattle and/or sheep (annual average stocking intensity: 0.15–0.4 livestock units (LU) ha⁻¹ year⁻¹) but not mown. As arable land, we defined different crops typical for the region, especially maize and bread cereal crops, as well as vegetables and potatoes. Agriculturally unused grasslands included all grassland areas that were abandoned for at least five years or have never been used for agriculture, such as alpine grasslands. Finally, all permanent deciduous, coniferous or mixed forests were combined into the forest land-use type (thus, no energy forests).

Data collection and analysis

Vegetation and site variables depending on land-use types were used to explain the rooting parameters. As Fig. 1 shows, dependencies between explanatory variables and rooting parameters are not always strictly unidirectional. For example, vegetation composition influences rooting; however, rooting patterns can also influence vegetation composition. We considered as many different dependencies as possible in the applied methods and interpreted discovered statistically significant relationships as associations rather than causal (unidirectional) impacts.

Rooting parameters: root mass, root length and rooting depth

Overall, 171 rooting samples (Appendices S1 and S2) were taken between 1994 and 2017 in the field with core samplers of 6.8–7.7 cm diameter and a maximum core depth of 70 cm. Before coring, the vegetation was characterized with the standard phytosociological method of Braun-Blanquet⁴⁹ to directly connect rooting and vegetation characteristics. The size of the vegetation survey areas was determined by the minimal area of a plant community as the area with 90% or more of all plant species within this ecosystem. The survey area ranged between 1 m × 1 m in homogenous meadows and 10 m × 10 m in forests. Even though we chose the rooting survey areas to be homogeneous regarding vegetation composition, it was possible that the rooting measured in the soil cores was affected by species other than those located above the core area due to large heterogeneity within plant communities⁵⁰. Nevertheless, this error should be negligible.

As the data for this analysis were derived from a collection of rooting analyses from different research projects in the past 20 years using the same methodological approach, the number of samples per land-use type and per site was unbalanced (Supplementary Appendix S2). For example, some land-use types were represented only at one site (e.g., all agriculturally unused areas were at site I), while others were represented at three or even more than 10 sites. In addition, the number of samples within each land-use type was also unbalanced: 15 samples for arable land, 32 for agriculturally unused grasslands or 56 samples for intensively used hay meadows. The original data collection included the most common and important plant communities in the project areas except for arable land. Thus, the rooting of the most common crops (maize: n = 3; barley: 3; oat: 3; wheat: 3; and vegetables: 3) was analyzed near Innsbruck in an area specially selected for this purpose.

In the laboratory, the soil cores were split into the O-horizon (if present) and mineral soil layers of various thicknesses (0–3 cm, 3–8 cm, 8–13 cm, 13–23 cm, 23–38 cm, 38–53 cm, and > 53 cm). Root extraction was performed manually with the roots cleared of soil in sieving cascades under running water⁵¹. Afterwards, the roots were sorted into three size categories¹⁸: (1) very fine roots (diameter between 0 and 1 mm); (2) fine roots (diameter between 1 and 5 mm); and (3) coarse roots (diameter between 5 and 20 mm). Roots of woody species with a diameter larger than 20 mm were not taken into account, as the distribution and diameter of coarse roots (especially trees) in the soil vary greatly spatially; therefore, a single survey cannot be representative of the rooting of an ecosystem^50,52. The reason for this classification was due to the different functions of the classes. Very fine roots have a dominant role in the uptake of water and nutrients and may be the main source of stabilized carbon input to soil¹. Fine roots are mainly responsible for the transport, anchoring and storage of carbohydrates and are also able to take up water. Coarse roots are important for water transfer and the stabilization of plants. To account for the different specific root lengths (SRLs) of very fine and coarse roots from herbaceous and woody species²⁹, we classified the single samples according to the cover of herbaceous and woody species from the phytosociological surveys into pure grassland samples, mixed grassland samples (dominance of woody species: < 50%) and dwarf shrub-rich or tree-rich samples (dominance of woody species: > 50%)¹⁸. The conversion of root mass to rooting length was carried out using previously published Eqs. ¹⁹ (Table 1). Finally, the maximum depth (RD_90%), above which 90% of the total root mass was found, was calculated for each root sample using the equation:

where RM_90% is 90% of the total root mass (kg m^-2) and m_max is the maximum slope of the saturation curve. In the same way, the depths above which 50% (RD_50%) and 95% (RD_95%) of the total root mass occurred were calculated. In forests and in dwarf shrub-rich communities, the rooting depths and distributions could be biased by the fact that the sampling depth was very shallow, which could lead to underestimating the 50%, 90% and 95% rooting depths⁵³. In grassland ecosystems, croplands and in dwarf shrub rich communities, however, the 70 cm sampling depth is sufficient because most roots are within the top 30 cm¹⁸.

Environmental variables

For every root sample, we collected 79 potential impact variables on rooting, including 19 site variables, six land-use variables and 53 vegetation variables (see Table 2 and Appendices S1, S3 and S4).

Vegetation variables

In total, 53 vegetation variables were collected and divided a priori into four groups (Table 2, Supplementary Appendix S3). Variables included in the richness group were ‘number of plant species’, ‘number of taxonomic groups’ and ‘functional types’ (after³⁸). All variables that displayed information on the mean species cover, plant cover variance or dominance of species, the Shannon–Wiener and Evenness indices (both after⁵⁴) and the total vegetation cover were allocated to the community composition group. We calculated the Shannon–Wiener and Evenness indices⁵⁴ for species composition, functional types and functional traits.

The cover of functional types group included variables that provide information on the abundance, dominance and composition of single plant functional types (see Supplementary Appendix S3). Finally, the community-level trait group (see Supplementary Appendix S3) contained leaf, plant height and root traits (effect traits in sensu⁵⁵) used to assess the relative effects of aboveground and root trait turnover at the community level. They were calculated for each sample using trait values taken from the literature and the measured abundance of each species within the single community (i.e., community weighted mean⁵⁶). We used mean root density and main rooting depth for the single species^57,58,59. The rooting density of the species was classified into sparse, medium dense, dense, and very dense roots⁵⁹. The mean leaf size and plant height of the species (sources:^60,61; http://www.floraweb.de/; own observations) were classified according to the following thresholds. Plant height was divided into small (mean plant height < 20 cm), medium (20–40 cm), large (40–90 cm) and very large (> 90 cm) species. Leaf size was classified as small (mean leaf area < 10 cm²), medium (10–70 cm²) and large-leaved (> 70 cm²) species. In accordance with other authors^62,63, most plant species showed clear allometric allocation trends between leaves, stems and root biomass for different groups of plant species. In particular, a trend towards a decreased root mass fraction with plant size was detected.

Site characteristics

Important meteorological parameters were measured at eight study sites at a distance of < 150 m from the rooting samples using different microclimate stations. For the five sites without site-specific climate measurements, we used data from standard meteorological stations of the regional or national Weather Service departments at a distance of < 5 km and at approximately the same elevation.

For detailed soil characterization, soil profiles were investigated directly at the sample site or at a representative site with the same land-use type and identical plant communities in the immediate vicinity (< 50 m distance). Undisturbed soil samples of 250 cm³, as well as disturbed soil samples (~ 2 kg), were collected from the main root horizon for subsequent physical and chemical analyses. The disturbed soil samples were air dried and passed through a 2 mm sieve. For all soil samples, we analyzed pore size distribution, soil bulk density, soil particle density, total soil porosity, soil texture, and soil organic C and pH (for technical details and exact descriptions of the employed methods, see Supplementary Appendix S13).

Furthermore, we calculated mean Ellenberg’s indicator values (EIV⁶⁴) for temperature (T), moisture (F), soil reaction (R) and soil productivity or fertility (N) for all study sites. EIVs are bioindicators for ecological requirements in Central Europe of a single plant species in a competitive relationship. We weighted the species-specific EIVs with the cover of each species from the vegetation relevées after Braun-Blanquet⁴⁹ to create a community-weighted mean for each site.

Land-use types

Past and present management practices at the study sites were recorded by interviewing landowners. All landowners were asked to specify the actual type (arable land, hay meadow, pasture, agriculturally unused grasslands, or forest) and intensity of use (intensively or extensively managed) or the number of years since abandonment.

Key species

The idiosyncratic effects of particular species must be accounted for⁶⁵. These are effects that cannot be satisfactorily explained by the weighted mean or distribution of functional composition or traits but rather by the abundance of particular plant species^12,27. Therefore, these key species are not specific to a functional type of plant; they may belong to grasses, forbs, shrubs or trees. For this work, we could not or did not want to define key species a priori, but we considered all species with a presence > 1% and a mean plant cover > 1% as potential key species. In sum, we identified 29 potential key species. Our aim was to determine whether one of these potential key species, in addition to the vegetation and site variables, has additional explanatory power for the rooting parameters and thus actually becomes a key species.

Statistical analyses

As the number of vegetation variables (Supplementary Appendix S3) and site variables (Supplementary Appendix S4) were too many with respect to our sample size (n = 171), we used two principal component analyses (PCAs) to summarize these two sets of variables into components. For each PCA, components were uncorrelated with each other, but variables were highly correlated within the corresponding component. Therefore, each component comprised the common information of the variables correlated with this component. Variables not correlating with any components were included in the further analyses as original variables. In this way, the number of variables used in the subsequent regression analysis could be significantly reduced.

In order to operationalize and test hypothesis 1, separate regressions for each rooting parameter were computed. The estimated standardized regression coefficient with its standard error represents whether the component/variable has a statistically significant partial correlation with the corresponding rooting parameter. To allow for different associations dependent on the land-use type, a separate coefficient for each land-use type was applied (see Fig. 1). The remaining indirect effects of site variables on rooting using vegetation components/variables (Fig. 1, dashed arrows) were not considered (cf. Supplementary Appendix S13). To summarize the associations of the independent components/variables on each rooting characteristic, we counted whether a components/variables was statistically significant within the different rooting characteristics (root mass, root length, rooting depth). The ratio of significant estimates of the vegetation composition components/variables to the number of all estimates of these components/variables was used as a measure of vegetation composition importance. The same procedure was applied to the site characteristics. To provide a measure of the importance of the land-use types, we calculated the ratio of the number of times the estimates for each components/variables were different with respect to land-use type within a rooting characteristic versus all possibilities. Investigating these importance measures provides an answer to the relevance of the vegetation composition compared to site characteristics and land use. To elaborate the importance of the vegetation composition in more detail (hypothesis 2), the ratio of significant estimates of each vegetation group to the number of all estimates of this vegetation group was used as a measure of its importance.

The regressions underlying the computation of the importance measure (used for hypothesis 1 and 2) required many coefficients (number of components/variables multiplied by the number of land-use types). Unfortunately, we did not have enough samples for each land-use type compared to the number of used components/variables. Therefore, we chose a two-step procedure. For the land-use forest types, intensively used hay meadows and agriculturally unused grasslands, the sample sizes sufficed, and the regression with all components/variables was computed. From these calculations, we obtained the statistically significant components/variables (inductive part of the two-step procedure).

Only the significant components/variables from the inductive part were used in the second step for the remaining land-use types (arable land, lightly used hay meadows and pastures). The components/variables with a p-value < 0.15 were considered in the second step to investigate potential relationships with rooting parameters for the remaining land-use types. The second step was therefore an explorative approach but nevertheless gave a first insight into possible important components/variables facing a restrictive sample size.

For all regression analyses, we checked the model assumptions. No severe multicollinearity between the explanatory components/variables was present (variance inflation factor (VIF) < 4). Using residual diagnostics residuals were symmetrically distributed and showed no severe heteroscedasticity. Nevertheless, we used bootstrapped standard errors in our analyses (bootstrap samples 1000). Cook’s distance was checked for possible outliers. We examined in detail the samples with Cook’s distance values that were substantially larger than the rest and looked at the relative changes in the estimates when dropping these samples⁶⁶. Very few samples (< 1.8% per model) had to be excluded from the regression analysis, as the relative changes of the estimates (> 25%) indicated highly influential single observations. For all models, the F-tests indicated the necessity of allowing different level and slope coefficients with respect to land-use types (p < 0.05, for details, cf. Supplementary Appendix S5).

To check hypothesis 3 key species for rooting parameters were identified. A PCA was computed with the abundances of all potential key species (presence > 1% and mean plant cover > 1%). We investigated whether all these species were summarized into PCA components, i.e., into species groups with similar habitat requirements. Species not included in any component were treated as their own component (however, in our study, all species were included in a component). The multiple correlation coefficient (R²) of each component with the vegetation and site components/variables was computed. A high R² denotes that the information of the key species is covered by the vegetation and site variables.

All technical details and further detailed descriptions of the methods can be found in Supplementary Appendix S13. Statistical analyses were conducted with Stata/MP 13.1 for Windows.

Source: Ecology - nature.com