For this study we previously evaluated the impact of several frequently used non-crop plant species in habitat management; Fagopyrum esculentum (Polygonaceae), Borago officinalis P (Boraginaceae), Lobularia maritima (Brassicaceae), Coriandrum sativum (Apiaceae), Calendula officinalis (Asteraceae), and Vicia faba (Fabaceae) on the resource availability for parasitoids (floral and extra floral resources), quantifying insect visits, phenology and ease of use in two agricultural landscapes of central Chile. For this pilot study we selected the two most attractive and easy to use flower strips which were F. esculentum and V. faba (unpublished data). In this study we focused on the main Aphidiinae parasitoids and the main aphids attacking brassica farms as well as their most abundant predators in central Chile. The central valley of Chile is characterized by a temperate Mediterranean climate, with dry summers and mild, rainy winters (Sarricolea et al. 2017). Temperatures vary from 25 to 35°C in spring-summer (September–March) and between 3 to 13°C in winter, with precipitations ranging from 22 to 130 mm during spring and from 300 to 900 mm in winter (June–August) (Del Pozo et al.; Montes et al. 2011). The natural vegetation is characterized by sclerophyllous forest in the coastal mountain range and foothill of the Andes, thorn scrub (Acacia caven) mainly in the eastern part with Nothofagus forests mostly close to the Andes. Between these two mountain ranges most agriculture is carried out, however patches of natural and semi-natural elements persist throughout (Zhao et al. 2016). However, in the last 40 years field crop coverage has begun to decrease almost by half, with notable decreases observed in cereal and legume crops (Del Pozo et al. 2024). The coverage of export-oriented orchards and vineyards has increased from 194,947 ha to 492,587 ha (del Pozo et al, 2024), as well as forest plantations (with a consequently 12.7 – 27.0% reduction per 10 years of native forests).
A total of 20 farms were first selected in a landscape gradient (see Table 1 in study system), during spring to late spring of 2023. In half of these farms flower strips were established in autumn and again in spring, allowing the same level landscape context for both management treatments.
Landscape attributes and site selection.
In order to compare landscapes attributes of farms with and without flower strips, the compositional complexity of the landscapes that surrounded 1 km diameter buffers was determined. At each buffer for each farm, using satellite images with the aid of software QGIS (Qgis.org, 2024) all elements and percentages of cover using Chile map of landscape covers (Zhao et al. 2016; Hernández et al. 2016) were identified using supervised classification methods. All gathered information was also complemented by field visits. The following cover was obtanied: AWV: Areas without vegetation; ST: Urban settlements; NF: native forests; BW: water; AS: Arborescent scrub; PL: Forest plantations; GR: Agricultural grassland; CP: Crop fields; TBF: Total buffer; BSH: natural habitat; PBSH: proportion natural habitat (NF+AS); PBW: Proportion water (BW); % SNH: proportion of semi natural area (PBSH+PBW). The percentage of semi natural area (% SNH) was determined as the sum of the following covers: natural forest, natural shrubs and water bodies. Pastures ( GR) were not includes in the % of SNH as in the study area managed exotic pastures did not present significant floral resources. All selected farms were at least 1 ha in size, with the greatest experimental units not surpassing 2 continuous ha of brassicae crops. During the winter of 2023 historical rainfalls occurred in central Chile (five to ten fold the average), which meant that many farmers lost their farms to river flooding (Garreaud, 2023). Due to these we were left with six farms with flower strips and eight farms without flowers. However, the mean % of SNH between the two groups was not significantly different with a mean ± SD of 3.62 ± 3.36 and 4.93± 2.77 for flower and non-flower treatments respectively. The percentage of arable crops in the buffer was highly correlated (t = -55.631, df = 1238, p-value < 2.2e-16; cor -0.845) with the % SNH, therefore farms with higher SNH had lower % of arable crops.
Abundance of aphids and parasitism rates.
Within the selected farms, 40 plants were inspected in a random pattern during two consecutive days (allowing for a total of 80 sampling points per farm), and for each of these the total number of aphids was counted on 10 leaves on the underside. The number of parasitoid mummies was also registered. A subsample per farm was collected and taken back to the laboratory to estimate species composition (aphids and parasitoids). All species were determined using keys from Blackman & Eastop, (2000; 2007), Starý (1995) and Tomanović et al. (2014).
Abundance of predators.
For each group of farms, we determined the abundance of predators from the sampling of abundance of aphids taking into account these organisms differences in their stalking/hunting behaviors as well as habitat preferences. Most abundant predators were collected manually, as well as with the help of an entomological aspirator through a standardized inspection, with a 5 minute sampling effort per plant. Ten pitfall traps per farm were also randomly placed among brassicas within the same crop row during one week. The organisms were then collected and preserved in 70 % alcohol for further identification.
Statistical analyses
All analyses were carried out using R version 4.3.2. (R Core Team 2021). Data was first explored with measures of central tendency for means and standard error and standard deviation using the tidyverse package (Wickham et al. 2019). To understand the effect of the management treatment (main explanatory variable) in our study on predator abundance, aphid abundance and parasitism rates, we used generalized linear mixed models with the lme4 package (Bates et al. 2015) assuming a poisson distribution for the count variables and binomial for the proportions (mummification rates). Random variables “farm” and nested “plant per farm" were included in the first full exploratory models. Only plant per farm effects were retained in all models as other random variables did not incorporate significant variability. We assumed a random intercept for the random factors, as exploratory analysis indicated only variation at the intercept per farm of the management effect. Landscape complexity measured as % SNH was used as a continuous predictor within all models. Model simplification from full models resulted in the removal of the interaction between % SNH and treatment, as it did not significantly decrease the amount of residual deviance of the model, nor increased significant predictive power (as were all non-significant interactions in the full models). Model overdispersion was analyzed through the aods3 package (Lesnoff and Lancelot, 2022) to reduce unexplained variability within the models. As models under a poisson distribution were overdispersed, the final models assumed a negative binomial distribution of the response variables, which fixed the overdispersion in all selected models. The resulting models were compared with a subsequent type II Anova analysis to select the best model. The proportion of mummies per colony sample was analyzed considering a binomial distribution with a logit link function and no overdispersion was detected.