Effectiveness of augmentative biological control depends on landscape context

Biological pest control by natural enemies is an important component of sustainable crop production. Among biological control approaches, natural enemy augmentation is an effective alternative when naturally occurring enemies are not sufficiently abundant or effective. However, it remains unknown whether the effectiveness of augmentative biocontrol varies along gradients of landscape composition, and how the interactions with resident enemies may modulate the collective impact on pest suppression. By combining field and lab experiments, we evaluated how landscape composition influenced the effectiveness of predator augmentation, and the consequences on pest abundance, plant damage, and crop biomass. We show for the first time that the effectiveness of predator augmentation is landscape-dependent. In complex landscapes, with less cropland area, predator augmentation increased predation rates, reduced pest abundance and plant damage, and increased crop biomass. By contrast, predator releases in simple landscapes had a negative effect on pest control, increasing plant damage and reducing crop biomass. Results from the lab experiment further suggested that landscape simplification can lead to greater interference among predators, causing a decrease in predator foraging efficiency. Our results indicate that landscape composition influence the effectiveness of augmentative biocontrol by modulating interactions between the introduced predators and the local enemy community.

. Model selection for landscape effects and potential interactions with predator releases on lepidopteran larval abundance, plant damage, crop biomass, predation rates, and natural enemy abundance. The overall best model (most parsimonious) and competing models are presented. The overall best models are bolded. The AICc values, AICc difference (ΔAICc) and conditional coefficient of determination 1 (R 2 ) are given for each model. Models were selected based on second order Akaike Information Criterion (AICc). All final models were tested for spatial autocorrelation in the residuals using the mantel test from the package ade4 (Dray & Dufour 2007).

Effects of augmentative predator releases on final crop yield
For the purposes of this paper, we draw distinction between marketable crop yield and crop biomass. Marketable crop yield of cabbage is a result of both the harvested head weight and the cosmetic injury to the head. Crop biomass, on the other hand, is an indicator of plant productivity, which is significantly correlated with head weight (Pearson's r = -0.58, P = 0.005).
but does not account for the quality component. In fact, although high levels of lepidopteran defoliation consistently increase feeding injury of cabbage plants (i.e., plant damage was significantly correlated with the mean abundance of lepidopteran larvae, Pearson's r = 0.33, P = 0.002), they may not always affect harvested head weight due to the ability of brassica crops to tolerate relatively high levels of defoliation without significantly affecting final weight (Burkness et al. 2005, Liu et al., 2004. For this reason, we evaluated the potential effects of augmentative biocontrol on final crop yield using both quantity (i.e., marketable cabbage head weight) and quality (i.e., cabbage head damage) measures ( Supplementary Fig. S1).
Supplementary Fig. S1. The effect of augmentative releases of predators on final crop yield based on (a) quality (i.e., cabbage head damage) and (b) quantity measures (i.e., cabbage head weight) in landscapes of varying complexity. Predicted responses for the control (solid lines) and augmentative release (dashed lines) treatments are calculated from the set of best supported linear mixed-effects models (lme4). Effects of the interactions between treatment and landscape complexity were significant (P < 0.05) for crop yield quality, but not for crop yield quantity. In the top Figures (a and b) every point represents the mean treatment value in a given experimental plot. The bottom figs. (c and d) are effect sizes (mean ± 95 % CI) for crop yield quality (c) and crop yield quantity (d) based on the difference in the marginal means between plots with and without predator releases across the landscape complexity gradient. A positive effect size indicates that the mean of the predator plots is larger than the mean of control plots, while a negative effect size indicates a higher control mean. Pairwise comparisons were individually calculated at even intervals across the landscape complexity gradient. Asterisks denote effect sizes that are significantly different from zero (P < 0.05). Summary statistics of the LMER models used to estimate marginal means and confidence intervals are available in Supplementary  Table S3.
Supplementary Table S3. Statistical models for the effects of landscape composition and potential interactions with predator releases on both final crop yield quality (i.e., cabbage head damage) and quantity (i.e., cabbage head weight). Statistical models were used to estimate mean and 95% CI of effect sizes for landscape effects and potential interactions with predator releases (Supplementary Fig. S1.). Dashed lines represent interaction terms not included in the final models because they were not significant (P > 0.05). Boldface text indicates significant relationships (P < 0.05). ---------------1. The statistical significance of fixed effects and interaction terms were estimated using mixed-effect models (lmer) interpreted with a Satterthwaite approximation (Kuznetsova et al. 2017, Luke 2017.