Mixed effects of ecological intensification on natural pest control providers: a short-term study for biotic homogenization in winter wheat fields

Experimental area

Our study was conducted in the Heves Plain High Nature Value Area in North-Eastern Hungary. This region is one of the major target areas of agri-environmental schemes in Hungary (Kovács-Hostyánszki, Batáry & Báldi, 2011; Kovács-Hostyánszki & Báldi, 2012). The proportion of croplands in this region was 70%, the major crops being winter wheat, maize, as well as spring- and winter barley. Other major but less frequently sown crops were oilseed rape, sugar beet and sunflower. We selected 14 experimental winter wheat fields of 5–10 ha (geometric mean = 7.071 ha, S.D. = 3.53 ha). Seven of these had an adjacent, newly established (1 to 3 years old) set-aside field of sizes between 1.98 to 5.43 ha (geometric mean = 4.102 ha, S.D. = 1.65 ha), while the others, without any adjacent set-aside field, served as controls. The distance between the studied wheat fields was 1.54–10.84 km. The set-aside fields were sown by a seed-mixture, which included one leguminous (i.e., Medicago sativa) and two grass species (i.e., Lolium spp., Festuca spp.). During the maximum time of the 3-year set-aside management period, the use of agrochemicals was prohibited, and the fields were mown once a year in the second half of June (Kovács-Hostyánszki, Batáry & Báldi, 2011). The owners of these set-aside fields took part in the Hungarian Agri-Environmental Programme (HAEP-NHRDP 2007–13), in which the primary goal was to improve soil fertility and water retention, as well as to increase farmland biodiversity.

In 2014, we assigned an experimental plot of 45 × 20 m in each of the selected winter wheat field adjacent to the field margin (Fig. 1A). In the wheat fields with adjacent set-aside fields, the plot was at the edge towards the set-aside field. The farmers were asked to avoid applying fertilizers or insecticides within the experimental plots while maintaining all other conventional management practices. To assess the impact of fertilizer application and to avoid any interference between treatments, these areas were subdivided into two plots of 20 × 20 m, with a 5 m separation strip between them. One plot was treated with an NPK fertilizer at the usual rate of 95 kg N/ha in mid-April 2014, while the other plot received no fertilizer input (fertilizer control, Fig. 1A).

Soil characterization

We collected 15 cm soil cover to assess the soil organic content (SOC, a proxy for soil fertility) and pH. A total of 15 samples were taken per fertilizer control plots at the 10+ stage (BBCH scale, Zadoks, Chang & Konzak, 1974) of the wheat plants in mid-March and were stored at 4 °C. Prior to analysis, samples from each plot were sieved (at four mm mesh size) and blended. Subsequent soil analyses were conducted according to the Hungarian certifications (no. 08-0452 and 08-0206-2) and the method described by Mason (1983).

Landscape attributes

The landscape composition was calculated in a circle with 1,000 m-radius around each studied field. The proportion of arable fields, semi-natural habitats (semi-natural grasslands, semi-natural forests, tree lines, hedges and shrubs), urban areas and water bodies were measured using a GIS database (QGIS Development Team, 2018). We considered landscapes with >20% semi-natural habitats as structurally complex, while landscapes with <20% of such habitats were classified as simple (Batáry et al., 2011; Tscharntke, Batáry & Dormann, 2011). The presence/absence of set-aside fields adjacent to a study plot indirectly measures the spatial configuration (i.e., frequency of semi-natural habitats in the landscape) of the studied land use categories (Fahrig et al., 2011).

Arthropod and pathogen surveys

The abundance of aphids, their natural enemies and wheat pathogens was assessed by visual plant surveys in each plot along two parallel transects perpendicular to the margin, 5 m from each other at 5, 10 and 20 m from the edge of the wheat fields adjacent to the set-aside field, which were towards the habitat core. We randomly selected 17 wheat tillers per transect per distance (17 × 2 × 3 = 102 tillers per plot, Fig. 1B), and on each of these, we counted the number of grain aphids (mostly Sitobion avenae, Hemiptera: Aphidoidea), coccinellid adults and larvae (Coleoptera: Coccinellidae), syrphid larvae (Diptera: Syrphoidea), chrysopid larvae (Neuroptera: Chrysophidae), and the number of parasitized aphids (including mummies). We also visually assessed the major pathogens, including leaf spots (Septoria spp., Fusarium spp.), mildew and rust (Puccinia spp.) on the leaves. Plant surveys were conducted on three occasions in April (stem elongation stage; BBCH 35), May (head-forming stage; BBCH 50) and June (flowering stage; BBCH 60).

To monitor the activity density (a proxy for abundance from pitfall trap data, hereafter referred as ‘abundance’) of the surface-active natural enemies, four pitfall traps were installed along the transects (Fig. 1B) in each experimental (i.e,. fertilized and control) plot and operated for two weeks in late May–early June. The traps consisted of plastic cups of ∼114mm diameter, each of which filled with approximately 250 ml of a 50% propylene glycol solution saturated with NaCl and a drop of odorless detergent to reduce the surface tension. A green plastic roof protected the traps from litter and rain. The pitfall traps were emptied after 14 days, the catch was sorted in the laboratory, and the abundance of adult ground beetles and cursorial spiders (Lycosidae) was calculated.

Estimating yield

In June, all experimental plots were hand-harvested. Two sub-samples of 0.58 × 0.58 m area were taken at each distance (5, 10 and 20 m), resulting in a harvest sample of 1 m² in total. Harvested ears were transported in cotton bags and dried at 24 °C for 48 h in a climate chamber under 10% RH. After threshing, we measured the total fresh grain mass per replicate batch, as well as fresh and dry mass of sub-samples in order to calculate the total grain yield (g/m²) (standardized to a moisture content of 14%); the grain and the ear mass were used for further analyses.

Statistical analyses

We used linear mixed models to test the responses of pests (aphid abundance), pathogens (prevalence of leaf spots expressed as percentage), natural enemies (abundance of ground beetles and spiders) and winter wheat yields (estimated as mass of wheat- ear and grain) to soil properties, nitrogen use, landscape complexity and configuration. To achieve a normal distribution of the residuals, all response variables were square-root transformed. We created sets of single-argument models on each response variable, to avoid collinearity and maximize the predictive power of explanatory variables, only one explanatory variable was considered in each model (Burnham & Anderson, 2002). As for explanatory variables, we included soil properties including pH (continuous variable) and SOC (continuous variable), nitrogen treatment (categorical variable with two levels: fertilizer applied/not applied), presence of adjacent set-aside field (categorical variable with two levels: with/without bordering set-aside field), or landscape complexity (categorical variable with two levels: simple/complex). In all models, the farm ID (nominal variable) was used as a random effect. For these aforementioned models we used the function ‘lme’ from the ‘nlme’ package (Pinheiro et al., 2017). Afterwards, each set of models was tested to select the best ones based on the Akaike Information Criterion (AICc) corrected for small samples sizes (Burnham & Anderson, 2002) using the ‘model.sel’ function from the ‘MuMIn’ package (Bartoń, 2016). When ΔAICc >2, the ‘best approximating’ model was selected as the most parsimonious explanation (see Table 1 for details). When more than one model had ΔAICc <2, we applied a model averaging approach for the uncertainty in model selection and obtained robust parameter estimates (Grueber et al., 2011). During model averaging, we built all possible models with the given explanatory variables and parameter estimates of the best models (ΔAICc <2), which were fully averaged with the models’ AICc weights. We used the ‘dredge’ and ‘model.avg’ functions from the ‘MuMIn’ package for model averaging. All the analyses were compiled in R 3.4.1 (R Core Team, 2019). The datasets collected on different sampling occasions were pooled for analyses. For all analyses, the alpha was set to 0.05.

Pests and pathogens

The presence of adjacent set-aside fields significantly affected aphid abundance. There were significantly more aphids in winter wheat fields without adjacent set-aside fields than in those with set-aside fields (Table 2, Fig. 2A). Neither soil properties, nor nitrogen use or landscape complexity affected aphid abundance. We did not find any significant effects of within- and between-field factors on the prevalence of aphids. The prevalence of leaf spots was higher in complex than in simple landscapes (Fig. 2B).

Natural enemies

Our results did not indicate significant impact on the abundance of the two studied groups of ground-dwelling predators by any studied factors (Table 2, Fig. 3). However, the model selection procedure revealed that individual models, including pH and fertilizer treatments, were the two most parsimonious ones to describe the changes in carabid abundance, while those featuring landscape complexity and the presence of adjacent set-aside fields, as well as SOC and fertilizer treatments, were the best models for describing the changes in spider abundance (Table 1). In addition, soil pH had a marginally positive effect on the abundance of ground beetles, suggesting that, on average, slightly more beetles were present in the soil with neutral pH.

Yield estimations

Ear and grain mass were significantly higher in fields without adjacent set-aside fields (Table 2, Figs. 2C, 2D) but no other variable had an impact on yield.

Landscape influence on pests and pathogens

Agricultural landscapes are not static and may influence disease dynamics not only through its structure but also through its own dynamics (Plantegenest, Le May & Fabre, 2007). We found that the landscape configuration and complexity were key factors in the distribution of pests and pathogens, but the distribution of pathogenic fungi did not support the hypothesis that a higher proportion of semi-natural habitats in the landscape can mitigate the negative effects of local (intensive) field management practices on the within-field abundance of natural enemies, since complex landscapes did not restrict the distribution of pathogens. Although a previous study (Pfender et al., 2006) using a complex air pollution models (CALPUFF) demonstrated that the spores of pathogenic fungi had high deposition within a radius of 1–2 km around their source, the complexity of the landscape could actually lead to a locally higher infection rate due to a decreased air transfer distance of about 0.4 km (Plantegenest, Le May & Fabre, 2007). This finding was similar to our results on the prevalence of leaf spots, which was higher in the plots within the complex landscape. The decreased infection in simple agricultural landscapes could also be explained by the more effective chemical treatments (Plantegenest, Le May & Fabre, 2007; Gagic et al., 2017). We also demonstrated that the studied ground-dwelling predators seemed to be unaffected by the landscape constrains, leaving our hypothesis (i) unsupported. With regard to carabid beetles, in contrast to a previous study (Cole et al., 2002) we found that local species diversity in agricultural fields did not differ significantly when compared to semi-natural areas, owing to the high turnover of the typical agrobiont species. Our local spider assemblages were enhanced in the patches surrounded by a larger percentage of non-crop habitats, agreeing with earlier results by Clough et al. (2005) and Hendrickx et al. (2007). Additionally, the local communities in landscapes consisting of small and disconnected patches were characterized by a species composition with low beta-diversity. This homogenization caused by agricultural intensification may suggest that local assemblages are becoming more unsaturated, most probably because of the loss of specialist –and typically less competitive –species with low dispersal ability, such as many ground-dwelling predators (Clough et al., 2005; Hendrickx et al., 2007). In addition, highly mobile species could mask the effect of spatial heterogeneity between habitats, as demonstrated on orthopterans by Marini et al. (2011).

The beta diversity of natural enemies is masked by their mobility

We observed that less grain aphids were on plots that had adjacent set-aside habitats, while the natural enemies seemed unaffected, thus our hypothesis (ii) was just partially supported. This could be explained by the spillover of natural enemies into crop fields, resulting in better pest control (Woodcock et al., 2016). Any increase in aphid population growth, as well as any increase in aphid suppression with an increase in landscape complexity, could be explained by the higher availability of alternative resources and overwintering habitats in the semi-natural habitats around crop fields, benefiting both pests and their natural enemies (Martin et al., 2015; Karp et al., 2018). Additionally, the low abundance of aphids in the plots next to set-aside fields suggested that EI is also characterized by the contribution of other natural enemy groups to pest control (e.g., parasitoid wasps; Martin et al., 2015) or the isolation effect by the adjacent non-crop areas (Dainese et al., 2019; Karp et al., 2018).

We observed no effect of the presence of an adjacent set-aside field on the abundance of natural enemies. The available evidence (Hendrickx et al., 2007) suggests that the more specialized species (even natural enemies) abandon the isolated habitat patches that, consequently, contain only a few generalist species of high dispersal ability. The lack of response in two groups of natural enemies at the within-field scale might suggest that the intensive, asymmetric species flows (or spillovers) between the fields during the ripening phase of winter wheat masked eventual differences in habitat utilization between croplands and the semi-natural habitats (sensu lato Marini et al., 2011).

Yields: conditions within intensive fields are connected to local land-use traditions

Although we demonstrated no effect of fertilization on the abundance of pests, pathogens or their natural enemies, we found that the grain yield was higher in plots without adjacent set-aside than with them. In addition, the yield remained unaffected by soil pH or SOC. These results partially supported our hypothesis (iii), pointing to below- and above-ground ESs. At the below ground, the unaffected yield by soil parameters might be caused by the increased nitrogen loss rate in calcareous soils and the less efficient N utilization in the soil with intensive fertilizer application (Ju et al., 2009). A good quality topsoil with high SOC levels and optimal pH supports a healthy soil fauna (Scheu, 2001). A previous study by Pettersson, Tjallingii & Hardie (2007) revealed that higher abundance of grain aphids in fields with high SOC could be explained by their sensitivity to any changes in plant quality. Above ground, the observed lack of effect of fertilizer use on natural enemies, such as spiders and carabids, might imply that high SOC levels and optimal pH could support the soil fauna, which provides sufficient food for these groups to successfully resist management related disturbances. This may lie behind the marginally positive effect of pH on carabid abundance.

The fact that we did not detect any direct effect of the fertilizer on crop yield might lead to the conclusion that high soil fertility itself was more important than the external inputs. The lack of a relationship between the yield and any soil parameter such as pH and SOC could also be explained by the fact that while the best soils were farmed by intensive methods, the owners used their less fertile land as set-aside fields. We also found high yields in fields without adjacent set-aside fields. This could be explained by between-site variability in soil properties, which were independent of the proportion of arable land in the landscape, and being masked by intensive management practices (Williams & Hedlund, 2013; Bartomeus, Gagic & Bommarco, 2015; Gagic et al., 2017). Crop yield is influenced by a combination of biotic and abiotic factors. A better understanding of the interactions between above- and below-ground processes and ESs might ensure high crop yields, but this also requires the investigation of how they co-vary within and between crop fields (Gagic et al., 2017; Tamburini et al., 2016).