Overview
We compiled systematically sampled empirical taxa occurrence across the landscape, and inferentially assembled respective blue and green local food webs by combining these data with a metaweb approach. We quantified key properties of the inferred food webs, then analysed with GIS-derived environmental information how focal food-web metrics change along elevation and among different land-use types in blue versus green systems. Details are given below.
Assemble food webs using a metaweb approach
We applied a metaweb method to obtain the composition and structure of multiple local food webs across a landscape spatial scale10. A metaweb is an accumulation of all interactions (here, trophic relationships) among the focal taxa. In this study, we built our metaweb based on known trophic interactions derived from literature and published datasets, which themselves were all based on primary empirical natural history observations. We further complemented or refined the trophic interactions in the metaweb based on expert knowledge of primary observations that are not yet published or only accessible in grey literature. The expert knowledge covers authors and collaborators who have specific natural history knowledge on Central European plants, herbivorous insects, birds, fish, and aquatic invertebrates. Importantly, these observations were all based on empirical observations and/or unpublished data accumulated over considerable field research experience. The respective literature we referred, as well as the metaweb itself with information source of each trophic link (online repository), are provided in Supplementary Methods. By assuming that any interaction in the metaweb will realise if the interacting taxa co-occur, the metaweb approach allows an inference of local food webs if taxa occurrence is known. Such an assumption of fixed diets may lead to an over-estimation of the locally realised trophic links32, as it essentially ignores the possible intraspecific diet variation caused by resource availability61,62, predation risk63, temperature64, ontogenetic shift65, or other genetic and environmental sources66. Therefore, the food webs we inferred systematically using this method capture trophic relationships driven by community composition (species presence versus absence) but not the above-mentioned processes. Nonetheless, since the trophic interactions were based on empirical observations, the fixed diets can be seen as collapsing all intraspecific variations of diet-determining traits (or trait-matching) at species level, within which we know realisable interactions surely exist. This, together with co-occurrence as a pre-requisite, gives realistic boundaries for the potential interaction realisation, which is plausible and non-biased when applying to localised sites. With this approach, we were addressing a systematic comparison among potential local food webs between the blue and green systems and across the selected gradients. For sensitivity analyses considering the potential inaccuracy of the metaweb approach mentioned here, please see further below Food-web metrics and analyses and Supplementary Discussion.
We compiled taxa occurrence of four terrestrial and two aquatic broad taxonomic groups (“focal groups”) to assemble local green and blue communities, respectively and independently, based on the well-resolved data available. Each focal group referred to a distinct taxonomic group, and the within- and among-group trophic relationships captured most of the realised interactions. These focal groups were vascular plants, butterflies, grasshoppers, and birds in the green biome, and stream invertebrates and fishes in the blue biome. Notably, with “butterflies” we refer to their larval stage and accordingly their mostly-herbivorous trophic interactions throughout this study. Larval interactions were also the predominant interaction assessed for stream invertebrates (i.e., all interactions of stream invertebrates focussed on their aquatic stage, which is predominant larval). The occurrence data of these focal groups were compiled from highly standardised multiple-year empirical surveys of various authorities, all conducted by trained biologists with fixed protocols (Supplementary Methods). The information across sites should thus be representative and can be up-scaled to the landscape. The occurrences of plants, butterflies, birds, and stream invertebrates were from the Biodiversity Monitoring Switzerland programme (BDM Coordination Office67) managed by the Swiss Federal Office for the Environment (BAFU/FOEN). The occurrences of grasshoppers and fishes were from the Swiss database on faunistic records, info fauna (CSCF), where we further complemented fish occurrence from the data of Progetto Fiumi Project (Eawag). In terms of biological resolution, taxa were resolved to species level in most cases, while the plant and butterfly groups included some multi-species complexes. Insects of the order Ephemeroptera, Plecoptera, and Trichoptera were resolved to species, while all other stream invertebrates were resolved to family level. These were each treated as a node later in our food-web assembly, and referred to as “species”, as the species within such complexes and families mostly share the same trophic role. Spatially, the occurrence datasets adopted coordinates resolved to 1 × 1 km2. The species that were recorded in the same 1 × 1 km2 grid were considered to co-occurred. We took the co-occurring four/two focal groups to form local green/blue local communities, respectively. To obtain better co-occurrence across group-specific data from different sources (e.g., BDM and info fauna), we intentionally coarsened the grasshopper and fish occurrence to 5 × 5 km2 coordinates. This is arguably a biologically acceptable approximation considering the high mobility of these two groups. Also, we only included known stream-borne fishes and dropped pure lake-borne ones to match our stream-only invertebrate occurrence data. Across all 462 green and 465 blue communities we assembled, we covered 2016 plant, 191 butterfly, 109 grasshopper, 155 bird, 248 stream invertebrate, and 78 stream fish species. Unlike the knowledge of plant occurrence in green communities, we did not have detailed occurrence information of the basal components (e.g., primary producers) in blue ones. Therefore, we assumed three mega nodes—namely plant (including all alive or dead plant materials), plankton (including zooplankton, phytoplankton, and other algae), and detritus—as the basal nodes occurring in all blue communities, without further discrimination of identities or biology within. These adding to our focal groups thus cover major taxonomic groups as well as trophic roles from producers to top consumers in both blue and green systems.
Taking the above-assembled local communities then drawing trophic links among species (nodes) according to the metaweb yielded the local food webs (illustrated in Fig. 1), representatively covering the whole Swiss area. Notably, although our understanding of trophic interactions indeed encompassed some links across the blue and green taxa (e.g., between piscivorous birds and fishes), our occurrence datasets did not present sufficient spatial grids where these taxa co-occur. We, therefore, did not include such links, nor assembled blue-green interconnected food webs, but the blue and green food webs separately instead (but see Supplementary Discussion). Also, we dropped isolated nodes, i.e., basal nodes without any co-occurring consumer and consumer nodes without any co-occurring resource, from the inferred food webs. These could possibly be passing-by species that were recorded but had no trophic interaction locally, or those that interact with non-focal taxa whose occurrence information was unknown to us. We thus had to exclude them to focus on evidence-supported occurrences and trophic interactions. Nonetheless, across all cases, isolated nodes were rather rare (averaged less than 3% of species occurred in either blue or green communities).
Environmental data
We acquired environmental data across all of Switzerland (42,000 km2) on a 1 × 1 km2 grid basis (i.e., values are averaged over the grid) from GIS databases, with which we mapped environmental conditions to the grids where we assembled food webs. These included: topographical information from DHM25 (Swisstopo, FOT), land-cover information from CLC (EEA), and climate information (averaged over the decade of 2005–2015) from CHELSA. Among environmental variables, elevation and temperature are essentially highly correlated. In this study, we took elevation as the focal environmental gradient throughout, as after accounting for the main effects of elevation on temperature, the residual temperature was not a good predictor of the food-web metrics we looked at (see next section, and Supplementary Table 4). In other words, by analysing along the elevation gradient, we already captured most of the temperature influences on food webs. Based on the labels provided by the GIS databases, we categorised the originally detailed land cover into the five major land-use types that we used in this study, namely forest, scrubland, open space, farmland, and urban area. Forest includes broad-leaved, coniferous, and mixed forests. Scrub includes bushy and herbaceous vegetation, heathlands, and natural grasslands. Open space encompasses sparsely vegetated areas, such as dunes, bare rocks, glaciers and perpetual snow. Farmland include any form of arable, pastures, and agro-forestry areas. Finally, urban area is where artificial constructions and infrastructure prevail. As each grid could contain multiple land-use types, we then defined the dominant land-use type of the grid as any of the five above that occupied more than 50% of the grid’s area. Analyses separated by land-use types with subsetted food webs (land-use-specific analyses) were based on the grids’ dominant land-use type. There were a few grids where the dominant land-use type did not belong to the focal major five, e.g., wetlands or water bodies, and a few where no single land-use type covered more than 50% of the area. Food webs of these grids were still included in the overall analyses but excluded from any land-use-specific analyses (as revealed in the difference in sample sizes between all versus land-use type subsetted food webs in Fig. 2; analyses details below).
Food-web metrics and analyses
We quantified five metrics as the measures of the food webs’ structural and ecological properties. For the fundamental structure of the food webs, the number of nodes (“No. Nodes”) reflects the size of the web, meanwhile represents local species richness (though the few isolated nodes were excluded as above-mentioned). Connectance is the proportion of realised links among all potential ones (thus bounded 0–1), reflecting how connected the web is. We also derived holistic topological measures, namely nestedness and modularity. Nestedness of a food web, on the one hand, describes the tendency that some nodes’ narrower diets being subsets of other’s broader diets. We adopted a recently developed UNODF index68 (bounded 0–1) that is especially suitable for quantifying such a feature in our unipartite food webs. On the other hand, modularity (bounded 0–1 with our index) reflects the tendency of a food web to form modules, where nodes are highly connected within but only loosely connected between. Nestedness and modularity are two commonly investigated structures in ecological networks and have been considered relevant to species feeding ecology24 and the stability of the system69. Finally, we measured the level of consumers’ diet niche overlap of the food webs (Horn’s index70, bounded 0–1), which essentially depends on the arrangement of trophic relationships (thus the structure of the webs), and could have strong ecological implications as niche partitioning has been recognised to be a key mechanism that drives species coexistence71,72. We selected these fundamental and holistic properties as they are potentially more relevant to the processes that may have shaped food webs across a landscape scale (e.g., community assembly), in comparison to some node- or link-centric properties. Also, addressing similar metrics as in the literature13,69 would facilitate potential cross-study comparison or validation.
To first gain a glimpse of the structure of the blue and green food webs, we performed a principal component analysis (PCA; Fig. 3a) on the inferred food webs (n = 462 and 465 in green and blue, respectively) taking the four structural metrics (number of nodes, connectance, nestedness, and modularity) as the explaining variables of blue versus green system types. We then confirmed that system type, elevation, and land-use type were all important predictors of food-web metrics (whereas the residual temperature after accounting elevation effects was not) by conducting general linear model analyses, taking the former as interactive predictors while the latter response variables (Supplementary Tables 3, 4). To check how elevation influences food-web properties in blue and green systems separately, and how food-web properties depend on each other, we ran a series of piecewise structural equation modelling (SEM)73 analyses on inferred food webs (Fig. 3b, c) whose dominant land use can be defined (n = 421 and 430 in green and blue, respectively). This was also conducted on subsetted webs of each of the five major land-use types (Supplementary Figs. 1 and 2). The SEM relationships were derived from linear mixed model analyses with dominant land-use type as a random effect (assumption tests see Supplementary Figs. 12–17). The SEM structure of direct effects was set according to the literature13,69 and is illustrated in Fig. 3b. In short, this structure tests the dependencies from elevation (an environmental predictor) to food-web metrics (ecological responses). The further dependencies among food-web metrics themselves were assigned with the principle of pointing from relative lower-level properties to higher-level ones. That is, from number of nodes (purely determined by nodes) to connectance (determined by numbers of nodes and links), further to nestedness and modularity (holistic topologies, determined further by the arrangement of links), then to diet niche overlap (ecological functional outcome).
Finally, to check and visualise the exact changing patterns of food webs, we applied generalised additive models (GAMs) to reveal the relationships between food-web metrics and the whole-ranged elevation (Figs. 4 and 5), as well as a particular comparison between food webs in forests and farmlands below 1500 m a.s.l. (Supplementary Fig. 5), as this elevation segment covered most of the sites belonged to these two land-use types. We also performed a series of linear models (LMs) and least-squared slope comparisons based on land-use-specific subsets of food webs (Figs. 4 and 5; Supplementary Figs. 3 and 4), to investigate whether food-web elevational patterns are different among land-use types (assumption tests see Supplementary Tables 5 and 6). In the GAMs analyses, specifically, we simulated two sets of randomised webs, i.e., “keep-group” and “fully”, as the null models to compare with the inferred ones74. Both randomisations generated ten independently simulated webs from each input inferred local food web, keeping the same number of nodes and connectance as of the latter. On the one hand, the keep-group randomisation shuffled trophic links from an input local web but only allowed them to realised fulfilling some pre-set within- and among-group relationships. That is, in green communities, birds can feed on all groups, grasshoppers on any groups but birds, while butterflies only on plants; in blue communities, fishes can feed on all groups, while invertebrates on themselves and the basal resources. These pre-set group-wide relationships captured the majority of realistic trophic interactions compiled in our metaweb. On the other hand, the fully randomised webs shuffled trophic links disregarding the biological identity of nodes. The GAMs of nestedness, modularity, and niche overlap illustrated the patterns of these randomised webs (Fig. 5). Comparing among the three types of webs, the patterns exhibited already by fully randomised webs should be those contributed by variations in web size and connectance, while the difference between keep-group and fully randomised webs by the focal-group composition of local communities, and the difference between inferred and keep-group randomised webs further by the realistic species-specific diets. In addition, we also applied the same GAMs and LMs approach to analyse node richness, as well as both realised and potential diet generality (vulnerability for plants) of each focal group (Supplementary Figs. 6–11). These analyses provided hints about the changes in community composition and species diet breadths along elevation and among land-use types, which helped explain the detected food-web responses in mechanistic ways.
In addition, to check if our findings were shaped or strongly influenced by the potential inaccuracy of using the metaweb, we repeated the above PCA, SEM, and GAM analyses as a series of sensitivity analyses. We generated food webs based on our locally inferred ones (i.e., the observations) but with random 10% link removal. This procedure mimics the effect of potential intraspecific diet variation (mentioned earlier) so that some trophic interactions in the metaweb do not realise locally. Overall, these analyses with link removal showed that our conclusions are qualitatively and quantitatively highly robust, and only very minorly affected by the such potential inaccuracy of metawebs, which is also in accordance to other food-web studies (see e.g., Pearse & Altermatt 201575). All details and outcomes of these additional analyses are given in Supplementary discussion.
All metric quantification and analyses were performed under R version 4.0.3 (R Core Team76). All applied packages and functions were described in Supplementary Methods, while the R scripts performing these tasks can be accessed at the online repository provided.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com
