Sustainable landscape, soil and crop management practices enhance biodiversity and yield in conventional cereal systems

1. Input-driven, modern agriculture is commonly associated with large-scale threats to biodiversity, the disruption of ecosystem services and long-term risks to food security and human health. A switch to more sustainable yet highly productive farming practices seems unavoidable. However, an integrative evaluation of tar-geted management schemes at field and landscape scales is currently lacking. Furthermore, the often-disproportionate influence of soil conditions and agrochemicals on yields may mask the benefits of biodiversity-driven ecosystem services. 2. Here, we used a real-world ecosystem approach to identify sustainable management practices for enhanced functional biodiversity and yield on 28 temperate wheat fields. Using path analysis, we assessed direct and indirect links between soil, crop and landscape management with natural enemies and pests, as well as follow-on effects on yield quantity and quality. A paired-field design with a crossed insecticide-fertilizer experiment allowed us to control for the relative influence of soil characteristics and agrochemical inputs. We


| INTRODUC TI ON
Functional biodiversity generates essential ecosystem services to agriculture such as biological pest control, pollination and nutrient cycling. Yet, intensive agricultural practices threaten biodiversity and ecological functions thereby posing long-term risks to sustainable food production and human health Foley et al., 2005). A potential solution is the ecological intensification of farming, which allows farmers to enhance agricultural sustainability by managing biodiversity and yield-supporting ecosystem services via the reduction of intensive management practices Pywell et al., 2015). But which practices are suitable for this purpose and what is their relative effect on yield compared to conventional, intensive methods?
Productivity of the staple crop winter wheat Triticum aestivum (L.) directly depends on agrochemical inputs, as well as soil characteristics, crop management, herbivory by pests, weed pressure and infections with pathogenic fungi (Ray et al., 2012). Degradation of soil fertility (Edmeades, 2003), inappropriate sowing densities (Ozturk et al., 2006) and herbivory (Dedryver et al., 2010) increase the gap between potential and attained yield in intensive wheat farming systems. Accordingly, management for healthy soils and crop diversification enhances the fertility and nutritional balance of the soil and improves productivity (McLaughlin & Mineau, 1995;Ratnadass et al., 2012). Similarly, soil, crop and landscape management that favours beneficial biodiversity (e.g. predators) and their top-down control on herbivorous pests (hereafter 'pests') may indirectly improve yield. This was shown for fertile soils (Birkhofer et al., 2008;Tamburini et al., 2016), reduced pesticide applications (Jonsson et al., 2012;Krauss et al., 2011) and fields located in diverse, well-connected cropping systems and landscapes (Fahrig et al., 2015;Martin et al., 2019;. However, responses to landscape heterogeneity are often variable or taxon-specific (Karp et al., 2018), and management practices that enhance both pests and predators may not result in improved pest regulation and yield (Tscharntke et al., 2016). Soil, management and landscape factors may also interactively influence predators, pests and productivity (Lichtenberg et al., 2017). For instance, the benefits of high soil organic carbon content (SOC) and biological control for winter wheat yield across Europe were reduced by mineral fertilizer application (Gagic et al., 2017). These potential tradeoffs lower the anticipated benefits of management measures and farmers' trust in implementing research in real farming systems.
Implementation can further be hampered if drivers of biodiversity and productivity are difficult to manipulate. Even though natural habitat is important for biodiversity, the extent of cropland can rarely be reduced in very intensive cropping regions such as Germany. Other factors such as SOC content depend on the soil type, ploughing frequency and crop rotation. From a practitioner's perspective, it is advantageous to link biodiversity and yield to adaptable landscape structures (e.g. field size) or actual management practices (e.g. crop rotation diversity).
In this study and in contrast to previous research (Gagic et al., 2017;Garratt et al., 2018), we identify sustainable management practices that directly or indirectly reduce pests while maintaining beneficial biodiversity and high yields in conventional wheat systems. We concurrently consider variation of non-adaptable drivers (e.g. soil characteristics) and commonly used intensive farming practices (e.g. agrochemical application). Thereby, we aim to bridge the gap between science and practice by increasing our knowledge of local and landscape-scale management options that can achieve ecological intensification . We furthermore pre-empt criticism by practitioners regarding the transferability of results from small-scale, controlled experiments to real agroecosystems.
Based on previous research and ecological theory, we expected to find (a) positive effects of agrochemical usage, beneficial soil characteristics (e.g. high SOC) and reduced pests, weeds or diseases on wheat yield. However, we also hypothesized that (b) extensive crop management practices improving soil structure and fertility may directly increase productivity and add to ecological intensification. We further assumed that (c) soil, crop and landscape management that increases the spatial or temporal habitat and resource availability for specific arthropods may indirectly influence wheat productivity, either negatively (enhanced pest pressure) or positively (enhanced predator abundance). We investigated these potential links and their relative importance using structural equation modelling (SEM). As relationships between trophic levels (predators-pests-wheat) can be either negative (top-down processes, i.e. herbivory and pest control) or positive (resource-driven, bottom-up processes, i.e. host and prey availability) we assessed both options (Vidal & Murphy, 2018).

| Study design
A total of fourteen pairs of conventional winter wheat fields (28 fields) were selected in 2014 near Würzburg/Germany (49°47′N, 9°57′E) along a landscape heterogeneity gradient defined by the % semi-natural habitat cover within a 1-km radius around fields (Figure 1a, detailed design in Appendix S1 in Supporting Information).
Fields were paired according to contrasting soil organic carbon content SOC (Figure 1b). To avoid correlations of SOC with other variables, we took care to (a) pair fields with similar pH, soil texture and field margin quality; (b) include pairs with different management practices (e.g. till or no-till, mineral vs. organic fertilizer input); and (c) minimize correlations between semi-natural habitat and other configurational and compositional landscape variables used in the analysis (Appendix S1). Average distance (mean ± SD) within and between field pairs was 1.07 ± 0.9 and 11.2 ± 3.8 km.
In each field, a randomized, fully crossed experiment with pyrethroid insecticide and mineral fertilizer was implemented in four adjacent treatment plots (12 × 14 m each, total 4 × 28 = 112 plots, Figure 1b). For both treatments, the type and amount of substance and the timing of application followed regional practices and recommendations ( Figure 1c; Appendix S1). Fertilizer (total 190 kg/ha) was applied at three growth stages and compared well to fertilization levels of surrounding conventional wheat fields (mean ± SD = 215.7 ± 42.8 kg/ha, range 120-339 kg/ha, farmer survey). Insecticide (active substance of beta-cyfluthrin, 300 ml/ha) was sprayed once after the first of three biodiversity surveys ( Figure 1c) to test for temporal effects of insecticide application on pests and predators.
In contrast to the integrated pest management framework of the Bavarian Regional Agricultural Institute, our study design required standardized insecticide application on all fields, even if pest thresholds were not crossed.

| Direct and indirect drivers of yield
We considered a total of 34 explanatory variables as potential direct or indirect drivers of wheat yield in our system. These can be separated into soil, crop management, landscape and biotic factors and were sampled at different spatial scales (Table 1; Appendix S1). The assessed soil characteristics likely to influence plant growth, yield, but also ground-dwelling predators (e.g. by providing suitable habitat conditions, Birkhofer et al., 2008) included SOC as proxy for soil organic matter, topsoil type and pH. We derived crop management data from the experimental set-up (fertilizer and insecticide treatment) and from farmer surveys performed in autumn 2014. The survey covered recent crop management in 2013/2014 and long-term management (mean ± SD data availability 11.25 ± 0.43 years), focussing on practices known or assumed to (in)directly affect pests, predators and yield. For instance, frequent no-till and organic fertilization should increase productivity but also ground-dwelling predators, as habitat is less disturbed and offers abundant alternative prey (Birkhofer et al., 2008). Similarly, we selected landscape heterogeneity variables in 1 km radius around study fields (ArcGIS v. 10.4, ESRI) including configuration and composition variables relevant for pests and their natural enemies. For example, landscapes with abundant semi-natural habitat cover or high crop diversity may enhance predators and pest regulation, thereby increasing productivity .
The set of biotic variables that may directly (herbivory, disease) or indirectly (predation) determine yields comprised the abundance and activity density of pests and predators in our treatment plots.
Responses to soil characteristics, management and landscape aspects and effects on productivity may be taxon-specific Redlich, Martin, Wende, et al., 2018). We therefore divided arthropods into eight functional groups (four pest and four predator groups, Table 1). Within the pests, we differentiated between two main aphid species (Hemiptera: Sternorrhyncha), the English grain aphid Sitobion avenae (Fabricius), which preferentially occurs and feeds on the wheat grain thereby reducing grain yield, and the rose-grain aphid Metopolophium dirhodum (Walker), which feeds on stem and leaves and should influence plant biomass. web-building spiders [Araneae]). While the first group is likely to benefit from large-scale landscape heterogeneity, the latter is thought to respond to field management. Abundances of all four pest and the two leaf-dwelling predator groups were counted on 50-100 tillers during three transect surveys (two after insecticide application, Figure 1c). Abundances per survey were then standardized to 50 tillers and summed across surveys to estimate pest pressure and pest control potential throughout the growing  Figure 1c).

| Estimation of wheat yield
Compensation effects in response to pest or disease pressure are common in crops and can be identified by assessing different yield components. In every treatment plot, we measured three aspects of wheat productivity at harvest (Table 1; Figure 1c). Yield quantity was estimated as grain weight ('grain yield', t/ha) and plant biomass (t/ha).
Both measures were extrapolated from four 0.25 m −2 subsamples.
Thousand kernel weight (TKW, g) acted as proxy for yield quality. Yield components were treated as final response variables in our analysis (Table 1), but were also tested for correlations among each other.
The final, simplified SEM showed good fit (D separation test, Fisher's C = 356.2, p = 0.215, Shipley, 2009) and conditional model fit (R 2 ) of individual path models ranged from 37% to 74% of variance explained (Figure 2). We assessed the relative importance of explanatory variables included in the final SEM using standardised path coefficients scaled by mean and variance (Schielzeth, 2010).
Additional mixed effects models tested for temporal insecticide effects on arthropods that may be masked in the SEM due to pooling of abundances across surveys. We fitted separate models for all pests and both groups of leaf-dwelling predators (no temporal sampling for soil-dwellers was conducted).

| Direct effects on yield
Our average (±SE) grain yield (8.5 ± 0.3 t/ha) and TKW (44.2 ± 0.4 g) on fertilized plots compared well to mean yields in 2014 on the remaining area of our wheat fields (8.1 ± 0.3 t/ha, farmer surveys), in Germany (grain yield 8.6 t/ha, TKW ~ 44 g) and the neighbouring France, Europe's largest producer of wheat (grain yield 7.46 t/ha, TKW ~ 47 g). Plant biomass and grain yield were strongly correlated and increased with fertilizer application by 26% and 30% respectively (Figures 2 and 3a,d). Grain yield (not plant biomass) was significantly lower on soils with high sand content compared to loamy soil (−36%, Figures 2 and 3b,e). We found no direct links between yield and insecticide application, pH or SOC. The pest S. avenae had the strongest effect on yield quantity, with reductions being greater for biomass than grain yield (Figures 2 and 3c, Contrary to expectations, we did not observe any direct relationships between extensive crop management and yield quantity (grain yield and plant biomass). TKW, however, was larger on fields with high crop rotation diversity or no-till soil preparation (Figure 2; Figure S1B,C).  showed positive responses to sowing date and was more prominent on fields sown later in the year (Figure 2; Figure S2E).

| Indirect effects on yield via herbivory
At the landscape scale, only two factors influenced pest populations ( Figure 2). Sitobion avenae was less numerous in landscapes with large fields (Figure S2A), while Oulema adults-and to a lesser extent their offspring (indirect effect)-were enhanced by semi-natural habitat cover ( Figure S2F,G). These correlations were rather weak though, as were the negative follow-on effects for yield quantity and quality respectively.
Of those, all leaf-dwelling predators responded positively to prey density, particularly to abundances of S. avenae (active flyers) and Oulema larvae (passive flyers; Figure 2; Figure S4B,C). Accordingly, insecticide-driven reductions of pest Oulema larvae indirectly affected passive flyers (indirect effect = −0.36). We did not, however, observe direct effects of insecticide on leaf-dwelling predators, neither using pooled abundances (SEM) nor distinct survey rounds (additional analyses). Temporal effects of insecticide application on soil-dwelling predators could not be tested, as pitfall traps were only opened once after insecticide application ( Figure 1c; Table 1).
Instead, reduced soil disturbance, i.e. no-till as opposed to deep ploughing, increased ground-dwelling spiders and carabids by 32% and 49% respectively (Figures 2 and 4a,d). Furthermore, predatory carabids had higher activity densities in fields with frequent organic fertilizer application and late sowing of winter wheat (Figure 4e; Figure S4F). Densities of ground-hunting spiders increased with SOC content yet declined with the addition of plant residues (Figure 4b; Figure S4E). Leaf-dwelling predators were not affected by soil management, but crop rotation diversity enhanced the abundance of passive flyers (Figures 2 and 4f). Surprisingly, passive flyers were not affected by any other crop management factor.

Activity density
In contrast to our predictions, landscape aspects at the scale assessed here played a minor role for predator abundances, with two exceptions. The activity density of ground-hunting spiders increased in landscapes with small crop fields (Figures 2 and 4c), yet decreased with high edge density ( Figure S4D). Active (but not passive) flyers were less abundant in landscapes with high edge density ( Figure S4A), which confirmed that active flyers respond more strongly to landscape than passive flyers. Still, we did not observe the expected positive response of active flyers to landscape heterogeneity.
We only observed positive relationships between predator and pest abundances, an indication for the prevalence of bottom-up, resource-driven processes in our system. Consequently, no beneficial indirect effects of predator abundances on yield were found.

| D ISCUSS I ON
Input-driven, modern agriculture is commonly considered as basis for high crop production but also linked to ongoing health and environmental problems. Together with growing food demands, this calls for more sustainable farming systems able to maintain current levels of productivity. Ecological intensification (i.e. utilizing ecosystem services) aims to achieve this goal. In this study, we identify potential pathways for ecological intensification by investigating multifactorial direct and indirect drivers of yield in a real-world cereal agroecosystem. Although agrochemical inputs had the relatively strongest influence on yield quantity, our results indicate that reduced pesticide application may increase yield quality without causing additional pest pressure. We also demonstrate that biodiversity-enhancing management options such as no-till soil preparation and high crop rotation diversity can enhance natural enemies.

| Reduction of agrochemical use
In our study, ambiguous effects of fertilizer and insecticide highlight alternative options for maintaining yield quality and quantity under conditions of reduced agrochemical inputs. Even though grain yield and biomass were lower in non-fertilized plots, so was pest pressure ('plant vigour hypothesis', (Price, 1991)). This enhanced TKW, an important indicator of seeding vigour and milling quality of wheat.
While insecticide decreased Oulema larvae, S. avenae as the most abundant and yield-damaging pest was not effectively controlled.
Instead, S. avenae numbers quickly rebounded after short-term insecticide-driven reductions (Krauss et al., 2011). Accordingly, a reduction of fertilizer and insecticide in this system may benefit yield quality without significantly increasing pest pressure by S. avenae.
An interesting finding, considering that insecticide and fertilizer treatments were applied following regional practices. However, although we used the typical application time, amount and substance, our experimental set-up required standardized insecticide applications independent of recommended thresholds defined by integrated pest management frameworks. On real farms, unnecessary, precautionary spraying may cause severe environmental damage and death of non-target species such as natural enemies. Even though we did not observe any negative effects of insecticide on leaf-dwelling predators, this could also be due to the specific type of insecticide used (pyrethroid) or rapid resettlement of relatively mobile web-building spiders and ladybirds from adjacent unsprayed plots.

| Enhanced soil fertility and structure
Although nitrogen tends to have the largest influence on productivity, soils with high carbon content (SOC) often augment yield (Gagic et al., 2017;Garratt et al., 2018). In our study region that was not the case, which could be explained by very low average SOC levels in Germany (mean ± SE: 1.39 ± 0.12%). However, we observed a positive influence of different soil management practices on yield and grounddwelling predators. The enhanced condition and nutritional balance of diversely cropped, no-till soils provides optimal growing conditions for high quality crops, as illustrated here by TKW (McLaughlin & Mineau, 1995;Ratnadass et al., 2012). Furthermore, high soil organic matter content favours saprophagous insects such as springtails, which are important alternative prey for soil-dwelling predators.
In contrast, ploughing alters physical characteristics of the soil and speeds decomposition of soil organic matter, causing unfavourable conditions for below-ground prey. Soil-dwelling predators therefore benefit from the combined effects of soil conservation practices (notill, high crop diversity, addition of organic fertilizer, Bai et al., 2018).

| Ecological crop management
In addition to enhancing soil fertility and yield quality, diverse crop rotations are characterised by temporal resource diversity and reduced chemical inputs (McLaughlin & Mineau, 1995). This may explain the greater abundance of passive flyers (wind-dispersed spiders) on these fields. In addition, timely (as opposed to late) sowing of wheat allows for earlier crop maturation and can reduce pest pressure (Acreman & Dixon, 1985), as seen for M. dirhodum.
Although numbers of predatory carabids were also lower on early sown fields, this response was less pronounced and could potentially be counteracted by no-till soil preparation.

| Diversifying the landscape
In contrast to soil and crop management, the role of landscape heterogeneity was small in our system. This could relate to limited coverage of the land use intensity gradients due to other constraints in site selection (e.g. contrasting SOC conditions) or the choice of spatial scale. The scale of landscape effects differs depending on the life history and movement capability of taxa (Chaplin-Kramer et al., 2011;Martin et al., 2016). In a previous multi-scale landscape study , we highlighted the advantages of landscape-level crop diversity at small and large scales, both for natural enemy abundances (<250 and >2,000 m scale) and biological control (<500 m scale). In other landscape studies, landscape composition or configuration at 1,000-m scales were the main drivers of biodiversity and ecosystem services (Bosem Baillod et al., 2017;Rusch et al., 2013;Tamburini et al., 2016). To avoid issues with multiple testing, we restricted the analysis of landscape effects in this study a priori to a well-supported intermediate spatial scale. While we may have missed some small or large-scale landscape effects, we nevertheless observed relevant relationships, with ground-dispersing spiders responding positively to small field sizes, but not to increased edge density. Predators in landscapes with small fields benefit from easy access to field boundary habitat and on-field prey (Fahrig et al., 2015).
Yet a high density of crop-non crop boundaries may hinder their dispersal into fields if these structures act as barriers or provide competing resources (Ratnadass et al., 2012). Last, our monitoring efforts did not cover larger aerial predators like Vespid wasps that may have responded more readily to changes on the landscape scale.
In contrast to previous studies, pest densities did not decrease with landscape heterogeneity (Bianchi et al., 2006;Bosem Baillod et al., 2017) or host crop cover (Schneider et al., 2015). Instead, Oulema and S. avenae were more abundant in landscapes with large amounts of semi-natural habitat and smaller fields. Additional habitats and easy access to fields may therefore favour those pests in conventional agriculture. This calls for caution against proclaiming semi-natural habitat as a panacea of ecological intensification.
Alternatively, heterogeneous landscapes may experience lower landscape-wide spraying intensities with less negative effects on pests, a possible correlation we could not test.

| Compensation potential
Our study shows that wheat can compensate for unfavourable soil conditions, weed competition and crop damage caused by pathogenic fungi or herbivory via changes to other yield components (Freeze & Bacon, 1990). For instance, crop damage by the plant pathogens rust and Fusarium (maximum infestation rate 58% and 34%, respectively), and feeding damage by Oulema larvae were very obvious on some fields. With the exception of TKW, however, our data did not confirm fears of participating farmers that this would lead to significant yield losses. In addition, yield reductions due to weed cover and S. avenae infestation had a greater negative influence on plant biomass than on grain yield. At the levels reached in our system, controlling these factors by agrochemical means is therefore less relevant than expected for final crop productivity.

| Linking biodiversity and pest control
Soil management and landscape aspects have the potential to enhance biodiversity, pest control services and consequently yield (Rusch et al., 2013;Tamburini et al., 2015). However, the negative relationships of the underlying top-down processes pest control and herbivory could not, or only partly, be confirmed. Instead, pests and predators primarily related to host and prey availability respectively. There are several explanations for this finding: (a) benefits of pest control are masked by effects of bottom-up resource availability and positive density dependence (Pywell et al., 2015;Vidal & Murphy, 2018). In this case, the only way to see benefits for biological pest control would have involved a predation experiment (e.g. predator exclusion). (b) Rather than natural enemy abundances, other trait-based functional biodiversity measures may have been a better indicator for pest control efficiency and yield (Gagic et al., 2015). (c) The importance of specific predator groups varies across the season. High initial densities of soil-dwelling predators at the beginning of the growing season are imperative for reducing initial pest infestations, while top-down control by this predator group may be lower or non-significant later on, as observed here (Birkhofer et al., 2008). (d) The recorded generalist predators are likely to have a more significant indirect effect on yield in other cropping systems that suffer from herbivory by chewing insects. In those cases, the management options identified here offer numerous possibilities for supporting pest control via ecological intensification.

| SYNTHE S IS AND MANAG EMENT IMPLIC ATIONS
We combined a landscape gradient with a factorial field-scale experiment to identify practicable management options that can sustain biodiversity and winter wheat yield. Strengthening observed links between natural enemies and landscape (mean field size) or local management aspects (intensity of soil management, addition of soil organic matter, crop rotation diversity) may restore the pest control potential in this and other farming systems. These biodiversityenhancing practices, together with expected benefits of agrochemical reductions and improved soil quality, provide important tools for ecological intensification. Accordingly, currently weak links between multiple ecosystem services and yield could be fostered by future adaptive soil, crop and landscape management and reduced agrochemical inputs. Considering the large-scale threats of conventional agriculture to biodiversity, the environment and human health, anticipated yet limited yield losses are an acceptable price relative to the economic and environmental benefits of a turnaround to more sustainable agriculture. In fact, we may even find that the strength of observed trophic relationships increases once legacy effects of long-term pesticide use fade away, further reducing potential yield gaps. Last, although this study is a snapshot of our system in time and results are closely linked to the climate in 2014, the observed effects of soil, management and landscape factors are likely to persist under different climatic conditions and may even be exaggerated due to climate change. These biodiversity-enhancing measures may strengthen the resilience of cropping systems to future global change drivers such as new pests and diseases or extreme weather events.

ACK N OWLED G EM ENTS
We thank the farmers for kindly permitting us to work on their land, Open access funding enabled and organized by Projekt DEAL.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available via the Dryad Digital Repository https://doi.org/ 10.5061/dryad.bk3j9 kd9m (Redlich et al., 2020).