Results

Comparison with no pest control

Overall, relative to farms without any pest control method, biocontrol interventions had a strong negative effect on pest abundance and crop damage, which were reduced by 55% (95% Confidence intervals (C.I.) = -64.62 to -44.21, p< 0.0001) and 60% (C.I. = -71.44 to -45.38, p< 0.0001), respectively (Fig. 1). Crops subject to biocontrol exhibited a 62% increase in yield (C.I. = 38.58 to 91.57, p< 0.0001). However, we found no significant overall effect of biocontrol on natural enemy abundance (-19%, C.I. = 38.58 to 91.57) (Fig. 1). There was substantial heterogeneity for all outcome measures, suggesting unexplained variation (Pest abundance,I2 = 54,98%; Crop damage, I2 = 51.35; Yield, I2= 69,20%, Natural enemy abundance, I2 = 92.35) (Fig.1). Hence, we used meta-regression to elucidate the effect of potential moderators.

Factors affecting biocontrol effectives

Biocontrol intervention technique

Overall, the most tested biocontrol approaches were botanical pesticides (n = 244), followed by intercropping (n = 163) and push-pull (n = 46), followed by both field margins (n = 38) and augmentation/introduction (n = 38). We found that crop yield was significantly affected by the nature of the biocontrol intervention (p=0.0001), with botanical pesticides and push-pull increasing yield by 92% (C.I. = 50.48 to 147) and 80% (C.I. = 52.13 to 114), respectively (Fig. 2c). In contrast, the specific biocontrol technique adopted had no significant effect on pest abundance (p=0.21), crop damage (p=0.30), or contrasting effects on natural enemy abundance (p=0.35).

Crop type

Across all outcome measures, the impact of biocontrol was measured predominantly in cereal crops (n = 457), followed by pulses (n = 155), vegetables (n = 207), fruits (n = 28) and fibres (n = 43). Biocontrol had an overall significant negative effect on pest abundance across all crop types, with cereal pests showing a 61% reduction (C.I. = -77.84to -34.66), followed by vegetable pests with a 54% reduction (C.I. = -63.47 to -35.31) (Fig.3a). Pest abundance in pulses and fruits showed a 52% and 39% decrease in pests respectively (pulses: C.I. = -71.80 to -20.29; fruits: C.I. = -62.79 to -2.59) (Fig.3a).
We found that biocontrol had a strong negative effect on crop damage in all crop types tested (cereal: 60%, C.I. = -71.37 to -45.01; vegetables: 46%, C.I. = -62.05 to -24.55; pulses: 44%, C.I. = -60.72 to -20.29; fruits: 38%, C.I. = -58.69 to -7.27) (Fig3b). Yield was positively affected by biocontrol, but this varied according to crop type; yields in vegetables increased by 57% (C.I. = 16.08 to 135) and pulses by 61% (C.I. = 5.91 to 145), while cereals and fibres showed an increase of 36% and 29% respectively (cereal: C.I. = 18.13 to 58.75; fibres: C.I. = 25.55 to 33.26) (Fig.3c). The specific crop type in which biocontrol interventions were tested did not influence the abundance of natural enemies (NEA, p = 0.06, Fig.3d).

Target pest taxon

Biocontrol interventions had a significant negative effect on the abundance of all pest taxa (p<0.0001), with lepidopteran pests showing the greatest decline (-63%, C.I. =-73.47 to -49.30) (Fig.4a). The crop damage of all taxa was strongly negatively affected by biocontrol interventions (p=0.012), with damage caused by Blattodea showing a 79% reduction (C.I. = -95.45 to -7.49) with biocontrol implementation (Fig.4b). We found that exposure to biocontrol interventions had a significant positive effect on yield where Coleoptera, Lepidoptera and Blattodea were the targeted pests (Fig.4c, Coleoptera: 157%, C.I.= 11.28 to 316; Lepidoptera: 65%, C.I. = 32.49 to 106.51; Blattodea 51%, C.I. = 17.62 to 94.37). There was no detectable effect of pest taxon on NEA response to biocontrol (p = 0.60; Fig.4d).

Comparison of research and farmers’ fields

Across all outcome measures, effect sizes did not differ significantly between farming types. In terms of cropping systems, the size of the negative effect of biocontrol on pest abundance was marginally higher in smallholder farms (66%, C.I.: -78.14 to -47.21) than in research farms (48%, C.I.= -59.62 to -33.64) (Fig.5a). Crop damage showed a similar pattern, where reduction in small holder farms (-69%, C.I.: -81.83 to -47.87) marginally exceeded that of research farms (45%, C.I.= -55.53 to -34.04) (Fig. 5b). With regards to yield, the proportional increase was almost equal in the two cropping types (small farm: 59%, research farm 67%). in neither case was NEA affected by biocontrol interventions.

Comparison with synthetic pesticides

The effectiveness of biocontrol interventions compared to synthetic pesticides was measured mostly for botanical pesticides (n = 339), followed by intercropping (n = 26) and augmentation/introduction (n = 23). We found no studies comparing the effect of field margins or push-pull with pesticides on their ability to control crop pests.
Although biocontrol interventions showed marginally greater pest abundance and damage, and reduced yield compared to synthetic pesticides, we found no significant difference between the two treatments (Fig.6, pest abundance: 23%, C.I. = -10.30 to 69.04; crop damage: 87%, -2.06 to 246; yield: -7%, C.I. -24.04 to 11.48). NEA: 43%, C.I. = 5.26 to 116.62). Conversely, the abundance of natural enemies was significantly greater following biocontrol implementation compared to the application of synthetic pesticides (43%, C.I. = 5.26 to 116.62) (Fig.6).

Landscape composition

Our search yielded seven studies that explored the effect of landscape composition on biocontrol delivered to crops in SSA. Four studies showed a positive effect of proximity to natural habitat, or proportion of natural habitat within a given buffer, on natural enemy activity (i.e., parasitism and predation) (Henri et al. 2015; Milligan et al. 2016; Kebede et al. 2018; Soti et al. 2019). Only three studies explored the interactive effects of landscape complexity and farm management on pest control effectiveness (Tsafack et al.2013; Midega et al. 2014; Kebede et al. 2019). All studies found an interactive effect of management and landscape composition, though the low sample size did not allow for quantitative analysis here.