Between 2014 and 2020, 74 flower strips were sown in the central Swiss plateau in the cantons of Bern, Freiburg, and Solothurn (mean altitude 548 m, range 413–981 m). The study area represents the typical agricultural landscape of the Swiss lowlands, characterized by a small-scale mosaic of intensively managed crop fields, meadows, and forest fragments in a comparably densely populated region (Figure 1). One to four different seed mixtures were sown in adjacent plots within one flower strip—hereafter called seed mixture plots (n = 196; see Supplementary Material 1 for illustration and seed composition of the flower strips). Flower strips were established between April 20th and May 25th by local farmers and left in place for a minimum of 100 days. No pesticides and fertilizers were used in the flower strips. All flower strips had a minimum width of 3 m and a maximum area of 0.5 ha. We used flower strip size (in hectares), flower coverage (in %), and flowering plant species richness (number of flowering plant species) as explanatory variables in the data analysis (see Table 1). Flower coverage and flowering plant species richness (only species with ≥ 0.1% coverage) was estimated in four representative squares (1 m²) in every seed mixture plot (simultaneously with the insect surveys—see below).

The surrounding landscape of every flower strip was described in a 1 km radius (Figure 1). Data on agricultural land-use provided by the cantonal agricultural offices allowed to calculate the share of agricultural and biodiversity promotion areas (BPA), BPA with high ecological quality, mean field size, and crop diversity for every landscape circle (see Supplementary Figure 2.1). BPA are extensively managed semi-natural habitats such as meadows, pastures, or hedges that are subsidized by the Swiss agri-environment scheme (Herzog et al., 2017). All farmers receiving direct payments have to manage at least 7% of their farmland as biodiversity promotion areas and today 98% of the Swiss farmland is managed under this scheme (Herzog et al., 2017). BPA with ecological quality, as defined by the Swiss Ordinance on Direct Payments, are BPA with high botanical and/or structural diversity and high ecological value (Herzog et al., 2017). A description of the different BPA types, management requirements and quality indicators can be found in the Supplementary Table 2.2. For crop diversity we used the Shannon–Weaver index (H) of the proportion of different crop types within each landscape circle (see Supplementary Table 2.1).

In addition to agricultural land use, we quantified forest area and total edge length for every landscape circle. Forest was defined as the summed area of densely and loosely stocked forests and smaller woody elements covered with trees and bushes (in hectares). Edge length was defined as the total length of the perimeter of forests, woody elements, and hedges (in km) and served as proxy for landscape heterogeneity (Martin et al., 2019). The data used to calculate these variables was obtained through the swissTLM3D dataset (Version 1.8 provided by the Federal Office of Topography Swisstopo) and verified using orthophotos from 2013 to 2017 (Bundesamt für Landestopografie Swisstopo, 2021). All spatial analyses were conducted in ArcGIS (Version 10.8.1).

Insects were collected repeatedly within each seed mixture plot in intervals of 9–12 days after the start of flower blooming (between May and July). The number of surveys varied among flower strips with a minimum of two and a maximum of seven surveys but was equal for all seed mixture plots within a flower strip. All insects were collected with a sweep-net with a diameter of 40 cm. For each survey, the sweep net was swept 20 times on a standardized straight line at walking pace. Insect surveys were only done on days with favorable weather conditions (temperatures between 20 and 35°C and no rain or strong winds). Captured insects were frozen at −18°C and subsequently counted and categorized into the following groups: honeybees (Apis spp.), bumblebees (Bombus spp.), other wild bees, hoverflies (Syrphidae), parasitic wasps (Ichneumonidae), soldier beetles (Cantharidae), rove beetles (Staphylinidae), ladybirds (Coccinellidae), green lacewings (Chrysopidae), ants (Formicidae), other flies (Diptera), and all others (e.g., Heteroptera and Orthoptera). Larvae and imagines were summed within each group. In addition, all the wild bees from 80 seed mixture plots in 36 flower strips from the years 2015 to 2017 were identified to species level.

From these counts the following indicators for insect abundance and diversity were calculated: (i) Total insect abundance was defined as the mean number of all individuals, (ii) pollinator abundance as the mean number of honeybees, bumblebees, other wild bees, and hoverflies, (iii) natural enemy abundance as the mean number of parasitic wasps, hoverflies, green lacewings, and ladybirds, (iv) wild bee abundance as the mean number of bumblebees and other wild bees, (v) wild bee richness as the number of wild bee species, and (vi) wild bee diversity as the Shannon-Wiener diversity index of wild bees. The mean numbers were always calculated as the sum of all individuals divided by the number of surveys per seed mixture plot (see Table 2).

To assess the impact of landscape characteristics and flower strip traits on insect abundance and diversity, we fitted linear mixed-effects models with total insect abundance, pollinator abundance, natural enemy abundance, wild bee abundance, as well as wild bee richness, and diversity as response variables. All explanatory variables described above (see also Table 1) were included as fixed effects in every model. To account for dependencies between observations, flower strip id and year were included in the models as crossed random effects. Pairwise correlations between explanatory variables were assessed using Pearson’s correlation coefficient (r_s) and agricultural area was excluded from the models due to a strong negative correlation (r_s > −0.60) with forest area. All linear mixed-effects model were fitted using the following formula:

lmer [y ∼ flower strip size + flower coverage + flowering plant species richness + BPA area + BPA area with quality + forest area + edge length + (1| Flower strip ID) + (1| Year)].

To assess if the differing number of surveys had a significant effect on our response variables, we fitted an additional linear-mixed effects model for every response variable including number of surveys as (only) fixed effect and flower strip id and year as crossed random effects. As the number of surveys had no significant effect on any response variable, it was subsequently not included in the main models. We also tested for quadratic effects of BPA area and BPA area with ecological quality, yet as there were no significant relationships, quadratic effects were not included in the final models.

Models were fitted using the R package lme4 (Bates et al., 2015) and model summaries calculated using the package lmerTest (Kuznetsova et al., 2020). Normality and homogeneity of the residuals were visually checked by using Tukey-Ascombe and QQ-plots. Spatial autocorrelation in the model residuals was checked with a Morans-I test [package DHARMa; (Hartig and Lohse, 2021)] and visually using bubble plots [package sp; (Pebesma et al., 2021)] but no remaining autocorrelation was detected (p > 0.05). Marginal effect plots were done using package sjplot (Lüdecke et al., 2021). The package vegan (Oksanen et al., 2021) was used to calculate the Shannon diversity indices. All statistical analyses were conducted in R Version 3.6.2 (R Core Team, 2021).

The area of BPA in the surrounding landscape was negatively correlated with all biodiversity indicators, except for natural enemy abundance. This was unexpected, as we assumed that BPA in the landscape would increase insect abundance and diversity in flower strips. Previous research has mostly shown positive effects of semi-natural habitats on the abundance and diversity of ecosystem service providers at the landscape scale (Batáry et al., 2015; Holland et al., 2017; Zingg et al., 2019). Grassland habitats, especially extensively managed (low input) meadows which represented the most abundant BPA in our study, often harbor high pollinator species richness and abundances, as they provide mating, foraging and nesting sites (Pfiffner et al., 2018). In our case, however, a high proportion of BPA in the landscape apparently diluted insect abundance in flower strips. We suppose that in our case observed insect densities in flower strips more strongly reflect the distribution of individuals among alternative floral resources in the surrounding landscape than the total abundance in a given area (Haenke et al., 2009; Carvell et al., 2011). As indicated by the intermediate landscape-complexity hypothesis (Tscharntke et al., 2012), relative insect abundances in flower strips may be highest in landscapes with intermediate proportions of semi-natural farmland habitats and when moving from moderate to high landscape complexity, dilution effects may increasingly outperform facilitation. Indeed, most of our study landscapes would categorize as rather complex on an international scale, as they harbor on average 20% (±14) natural areas (i.e., forests) and an additional 5.4% (±1.9) semi-natural farmland habitats (i.e., BPA).

Irrespective of this result, it is the overall abundance of service providers in the landscape that determines the strength of provided ecosystem services. To support large and effective populations of service providers on the long-term, a network of diverse semi-natural habitats is indispensable (Sutter et al., 2018). Thus, landscape level greening, i.e., a high share of biodiversity promotion areas is crucial to maintain ecosystem services in the long term. Particularly annual flower strips are only temporary available habitats that can bypass resource scarcity but are plowed in autumn (Ramseier et al., 2016). Annual flower strips are therefore not sufficient to ensure a safe overwintering of arthropods and need complementing perennial habitats at spill-over distance (Ganser et al., 2019).

On the other hand, overall insect abundance in flower strips increased with the area of forest in the landscapes. Natural habitats are known to promote overall biodiversity and a share of at least 20% is often needed to maintain high species richness in agricultural landscapes (Tscharntke et al., 2012; Zingg et al., 2018). Nevertheless, none of the functional groups was correlated with forest area. This suggests that forests facilitate insect species and groups that are attracted by resources offered in flower strips but are not considered as pollinators or natural enemies of crop pests. Although, forest edges can potentially act as a source of pollinators, the effect may further depend up on microclimatic conditions (Bailey et al., 2014), the vegetation (Roberts et al., 2017), as well as the specific nesting behavior of a given species (Odanaka and Rehan, 2020). As highlighted by Ewers and Didham in their comprehensive review on habitat fragmentation and edge effects (2006), different invertebrate species are likely to vary strongly in their response to habitat edges and the composite of often contrasting responses may result in little to no effects on overall abundance and species richness observed.

In addition, also the share of BPA with ecological quality tend to promote overall insect and pollinator abundance. This stands in contrast to overall BPA area that had a negative effect but is more in line with our hypotheses. We assume, that the fulfilment of ecological quality requirements led to a pronounced improvement of landscape-scale habitat quality, capable to foster insect populations on the large scale and the long term. There are several reasons, why BPA with quality may represent an overall improvement of habitat quality for multiple insect groups: First, grassland BPA with high ecological quality harbor a higher botanical diversity, known to be a major driver of pollinator abundance and richness (Albrecht et al., 2007). Second, in grassland BPA with quality, the use of hay conditioners, knowing to tremendously reduce insect abundance is not allowed (Frick and Fluri, 2001). Third, in woody BPAs with high ecological quality (e.g., hedges) additional structural elements such as piles of branches or stones and nest boxes for wild bees are offered. BPA with high ecological quality have thus previously shown to enhance different components of biodiversity at the landscape level and may thus represent an important key to foster pollinators and insects overall (Meichtry-Stier et al., 2014; Zingg et al., 2019). BPA with lower ecological quality, on the other hand, may not be sufficient to provide sufficient key resources for many insect groups.

Both, pollinator, and wild bee abundance increased with flower coverage inside the flower strips. This corroborates our hypotheses and shows that not only landscape, but also the flower strip configuration and ecological quality is decisive. Flower strips thus need to contain high flower density to support high pollinator numbers (Krimmer et al., 2019). Yet, no effect on species richness or diversity was observed. Species-specific reactions to flower strip traits have often been documented (Wood et al., 2015) and many wild bee species show a distinct response depending on floral and nesting resources (Pfiffner et al., 2019). Apparently, the annual flower strips investigated here attracted mainly generalistic wild bees (98.7% of the observed wild bee individuals were polylecthic species) and, as a consequence, the wild bee communities observed were dominated by few, but abundant species (68% of all individuals belonged to only five species). Indeed, the flower strips investigated here were often dominated by few flowering plants such as Phacelia or Buckwheat that seem to be mainly attractive to these few generalists. In this regard, it is not very surprising that no effect of flower coverage on species richness or diversity was observed. However, also flowering plant species richness had no effect on wild bee richness and diversity. This can be explained by the fact that the presence of key plants is more important for wild bee diversity than flower coverage or plant species diversity per se (Sutter et al., 2018; Warzecha et al., 2018). Although our seed mixtures (Supplementary Table 1.1) contained some of the key plants mentioned by Sutter et al. (2018) and Warzecha et al. (2018), such as Phacelia tanacetifolia, Anthemis tinctoria, and Origanum vulgare, P. tanacetifolia was the only key plant that properly emerged. Including additional key plants and improving their emergence may therefore be important to support species diversity and enhance the stability and resilience of ecosystem services (Warzecha et al., 2018).

From the parameters related to configurational complexity, neither field size, nor crop diversity or edge length had a significant effect on insect abundance or diversity. This contrasts our expectations based on the findings that pollinators and natural enemies and their services increase with configurational landscape complexity (Krimmer et al., 2019; Martin et al., 2019; Albrecht et al., 2020). A diverse cropping matrix with small fields can provide a temporal continuity of food, nesting sites and shelter and promote biodiversity, including pollinators and natural enemies (Bianchi et al., 2006; Hass et al., 2018; Sirami et al., 2019). A lack of effect of configurational complexity measures can be explained in three ways: First, all of our flower strips were established in rather heterogeneous agricultural landscapes with comparably small fields (mean ± SD: 1.0 ha ± 0.2), high crop diversity (mean ± SD: 15 ± 1.4 crop types according to Supplementary Table 2.1) and many patches of non-crop areas. Hereby the contrast, or gradient, for configurational complexity may have been insufficient to detect any ecologically relevant effect and corroborates earlier results from the Swiss lowland (Zingg et al., 2018). Second, field margins in the study area are often small to non-existent or of low ecological quality, as they are impacted by agrochemical drift from mostly intensively managed fields. As a result, the small field size in the Swiss lowlands does not necessarily correlate with more habitat for insects, and unproductive structural complexity in marginal areas may be low compared to nearby regions in neighboring countries (Tschumi et al., 2020). Finally, as highlighted above, the variable and often diverging responses of individual species to habitat edges and the complex effects of habitat edges on species interactions may simply have blurred an ecological signal in our functional groups (Ewers and Didham, 2006).

Remarkably and in contrast to our expectations, none of the selected landscape characteristics and flower strip traits had a significant effect on natural enemy abundance (with the exception of a negative trend of flower coverage). A recent meta-analysis corroborates these findings by showing that flower strip traits and landscape context had little effect on biological control (Albrecht et al., 2020). One explanation might be that seed mixtures designed to promote pollination are not optimal for the promotion of other ecosystem services such as pest control (Windsor et al., 2021). Although in our study various seed mixtures were used, at least some of them where specifically designed to attract pollinators. Hereby, and in contrast to pollinators, natural enemies strongly depend on the distribution of their prey or hosts and cannot reproduce with floral resources alone. Another explanation for the lack of effects might be that the group of natural enemies consists of several groups with highly specialized and diverging life-histories. The communities of natural enemies in our study were mainly dominated by the abundance of hoverflies and parasitic wasps (see Supplementary Material 3). While hoverflies are rather generalist predators, parasitic wasps are highly specialized on few or single host species. Although the occurrence of both groups is driven by the availability of resources for larvae and adults (Moquet et al., 2018), they often have very differing requirements. Lumping various natural enemy groups together might have resulted in too much variance in their responses to landscape characteristics and flower strip traits to detect any biological signal.