In order to analyze the effect of natural biodiversity on grape losses, we performed exclusion experiments on the selected vines. In each vine three bunches were selected and randomly allocated to each of the three treatments: (1) exclusion of birds by installing a net with a broad mesh of 19 × 19 mm; (2) exclusion of birds and insects by installing a net with a fine mesh of 1.9 × 1.9 mm; and (3) control vines accessible to all species (no exclusion).

All nets were installed in early June, when the unripe fruits were already formed, but before they started to ripen. At this stage, the grapes were sufficiently robust to endure the net installation and were not yet attacked by insects or birds. The initial number of grapes per bunch was recorded, as well as the linear distance from each vine to the nearest vineyard edge.

By the end of August, all nets were removed, and the number of grapes lost, due to each pest type (birds, insects and fungi), was scored by visual inspection of the marks left on the grapes, according to field guides and the farmer's experience (Isaacs et al., 2003; Carisse et al., 2006; Hahn and Wold-burkness, 2008; Hoover et al., 2011; Mani et al., 2014).

Bird abundance in each vineyard was evaluated with four censuses per month in June, July and August 2016, the period of grape development. Censuses were performed between 07:00 and 10:00 at two sites with good visibility at opposite edges in each vineyard. Each census lasted 5 min, during which the horizontal distance to all birds seen or heard was estimated. Only birds within a radius of 50 m from the census point were used in the analyses. Bird species were divided into three groups – auxiliary, neutral and pest – according to Cramp and Perrins (1993) and our previous experience with the local bird diets (Cruz et al., 2013; Costa et al., 2014; da Silva et al., 2017). These categories were adjusted every month in order to reflect bird feeding habits in relation to the available resources. For instance, most species are largely insectivorous during the breeding season and consume large amount of fruits in the end of summer.

Insects in each vineyard were sampled with 10 × 25 cm yellow sticky traps (commercially available from Koppert Horiver), a broad sampling method particularly useful for flying insects (Thomson et al., 2004). The traps were sticky on both sides and were suspended from the lower wire that connects the different vines along straight lines. Five traps were placed in each vineyard approximately 30–50 m apart. Traps were operated once per month during the duration of the experiment (June–August); on each occasion, they were removed after 5 days and kept at 4–5 ∘C until insect identification. All insects were later extracted from the traps with a solvent, identified to the family level, sorted according to their morphotypes and counted. Only three families of microhymenoptera (Platygastridae, Diapriidae and Proctotrupidae) were grouped together due to their similar morphology. Like birds, all insects were divided into three functional groups regarding their main relation with agriculture crops, i.e., auxiliary, neutral and pests (Mani et al., 2014; Bostanian et al., 2015). This classification was temporally flexible (i.e., variable across months), in order to reflect the changing roles of insects in relation to their life cycle and food availability.

Grape losses were quantified in terms of the percentage of losses, in relation to the initial number of grapes present in each bunch (i.e., losses = N grapesfinal - N grapesinitial / N grapesinitial × 100).

To evaluate whether grape loss by birds and insects differed among grape caste and color, we used generalized linear mixed models (GLMMs) with nested effects, with color and caste nested within color as fixed effects, vineyard as a random factor and distance to the edge as a co-variable. We evaluated the proportion of grape loss due to insects and birds in the control treatment (no exclusion) with two GLMMs for each loss type: namely one to look for differences between the four castes and one to look for differences between grape color.

Significant differences between castes were explored with a general linear hypothesis followed by a Tukey multiple-comparison test. This test uses a Tukey multiple-comparison test to create confidence intervals for all pairwise differences between factor-level means, while controlling for the family error rate.

To evaluate the potential of natural biodiversity to control grape losses, we performed linear regressions between the percentage of fruit losses per bunch (response variable), with the abundance of each insect and bird functional guilds (pest, auxiliary and neutral) as explanatory variables.

All analyses were performed in R 3.05 (R Core team, 2016), using packages ggplot2, Rmisc, lmerTest (Kuznetsova et al., 2017) and multcomp.

Bird abundance – there were no differences in bird abundance across vineyards (F5.63=1.025; p=0.411; Fig. 2) or across months (F2.177=0.834; p=0.44; Fig. 2). In terms of functional diversity, most birds were neutral for the grapes (i.e., they were not pests or auxiliary agents). The abundance of neutral birds per census was higher in July than in the other two months (F2.171=7.59; p=<0.001; Fig. 2). The abundance of pest and auxiliary birds did not vary significantly throughout the season (F2.80=0.50, p=0.60, and F2.28=0.27, p=0.76, respectively) (Fig. 2).

Insect abundance – insect abundance differed among vineyards (F5.333=3.165; p=0.008; Fig. 3); the vineyard with the most insects was Boas Quintas (16.2 individuals per trap), and the vineyard with the least insects was Messias (3.0 insects per trap). There was a small, non-significant trend for insect abundance to increase over the season (F2.1093=1.38; p=0.252) (Fig. 3). In terms of functional diversity, most insects were pests, and only a few insects were neutral. The abundance of insect pests was much higher than that of the other two functional groups in the three months (F2.1091=17.835; p=<0.0001).

Overall, grape losses were low or negligible at most vineyards (mean = 0.59 %). The vineyard with the highest percentage of fruit losses was, by far, Messias (7.50 %), where the percentage of fruit losses by fungi reached 7.00 %, (Fig. 3). All other vineyards had very few grape losses, particularly those of São Lourenço and Aveleda, where there were no documented losses due to birds or fungi, and only 1.00 % of the fruits were lost due to insect activity (Fig. 4). The overall mean grape losses across all vineyards were 2.01 % due to fungi, 0.08 % due to birds and 0.24 % due to insects.

Regarding the differences between grape castes, Chardonnay had the highest proportion of insect losses (0.43 %), and Touriga Nacional the lowest (0.08 %) (Fig. 5a). White grapes (Arinto and Chardonnay) had slightly more losses due to insects (0.70 %) than red grapes (Baga and Touriga Nacional) (0.50 %), though these differences were not significant (F1.119=0.98; p=0.337).

The nested GLMM showed that grape loss by insects in the control bunches differed only among castes (F2.119=9.74; p=<0.001), with Chardonnay and Baga showing slightly higher losses than both Arinto and Touriga Nacional (Fig. 5a, Table 1a). In relation to birds, grape loss differed marginally among castes (F2.117=3.02; p=0.05) and significantly between caste colors (F1.119=8.38; p=0.004), with a preference for red castes (Fig. 5b, Table 1b).

Similarly, the vulnerability to fungi differed among castes (F3.365=67.06; p=0.003), with Baga being the caste with the greatest losses due to fungi, and Touriga Nacional the caste with the fewest losses (Fig. 5c, Table 1c).

Grape losses by insects were positively correlated with insect abundance (rs=0.89; p=0.03, Fig. 6a), but not with bird abundance (rs=0.54; p=0.2), abundance of insect pests (rs=0.09; p=0.92) or auxiliary insects (rs=-0.2; p=0.71). Grape losses by birds were correlated neither with overall abundance of birds (rs=-0.46; p=0.35) or insects (Fig. 6b; rs=0.02; p=1), nor with the abundance of pest birds (rs=0.66; p=0.18) or auxiliary birds (rs=0.02; p=1). Finally, the abundance of auxiliary birds was not significantly correlated with insect abundance (rs=0.26; p=0.41).

The proportion of grape losses due to insects, birds and fungi was independent from the distance to the vineyard edge (rs=0.08, p=0.38; rs=-0.12, p=0.20; and rs=-0.06, p=0.50, respectively).

We observed a very low proportion of grape losses by birds and insects, but other studies report a higher percentage of grape loss attributed to these pests. In an experiment in New Zealand, Kross et al. (2012) registered 3.5 % of grapes damaged by birds, and Bournier (1976) calculated that crop losses in California can reach up to 15 %, chiefly due to Platynota stultana (Lepidoptera: Tortricidae) (Bournier, 1976; Kross et al., 2012).

The fact that the percentage of grape losses attributed to fungi was much higher than those attributed to birds and insects was largely influenced by the vineyards of Caves Messias. Such high vulnerability of these vineyards to fungi could be due to specific microclimatic conditions (e.g., high relative humidity).

Similarly to previous studies, we registered a significant influence of caste color on grape losses due to birds. The preference for the red color may be explained by its attractiveness to birds (Whitney, 2005; Gagetti et al., 2016). The fact that insects did not show a preference for the red color in our study goes against previous studies (Takahara and Takahashi, 2016) and may be explained by the high effectiveness of the chemical treatments in our study region.

The biochemical composition of grapes is another factor that influences the preference of grapes by birds and insects; indeed, our results show a significant preference for Chardonnay and Baga grape castes over Arinto and Touriga Nacional by insects and Baga by birds. The chemical and tactile characteristics of the different castes can affect their vulnerability to several bird and insect pests (Varandas et al., 2004; Bellí et al., 2007; Saxton et al., 2009), but unfortunately we did not quantify grape composition in the present study.

Few studies evaluated grape losses by birds; however, the few studies that tried to quantify this problem suggest that changes are relatively small, namely lower than 5 % in South Africa (Dignon, 2013) and lower than 9 % in North America (Anderson et al., 2013).

There are more studies evaluating grape losses due to insects, although most of them focus on the effects of a single insect pest (Hoffman and Dennehy, 1987; Moschos, 2005), while others focus on losses in general, like a study in Brazil which registered 4 % of grape losses in the whole country (Oliveira et al., 2014). Other studies analyzed the losses caused by Lobesia botrana in grapes, and authors registered 5.7 % losses in some years (Hoffman and Dennehy, 1987). In our study, insect and birds losses (1.42 % and 0.08, respectively) were slightly lower than those reported in previous studies, which is normal considering the multiple factors affecting grape losses, such as year, localization, age of the vineyard and castes.

In summary, there is some evidence that insects select certain grape varieties due to their physiological and morphological characteristics – such as skin thickness, color and nutrient content (Galvan et al., 2008) – and this needs to be taken into account in order to manage vineyards more effectively, for example by improving the conditions for the reproduction of insectivorous birds in areas with more vulnerable castes. Other evidence is the preference of birds by red caste colors and a non-preference for Arinto. This can be correlated because Arinto is a white caste and therefore not visually appealing to birds. But, as the percentage of losses caused by birds was low, we cannot draw great conclusions about the non-preference for Arinto.

In an effort to reduce costs and the environmental footprint without jeopardizing productivity, alternate strategies have been developed focusing on the economic level threshold for pesticide application and habitat management, i.e., integrated pest management, integrated production and biological production strategies (Perini and Susi, 2004; Mccracken, 2011).

In contrast with our expectations, we could not detect any measurable effect of natural biodiversity on pest suppression. We showed that vineyards with more insects had more losses, as most insects sampled in the vineyards were pests (74.5 %). Interestingly, the proportion of auxiliary insects was also very low, likely due to the use of pesticides, which is likely to hinder their potential role as biocontrol agents. Such a negative relationship between insecticide toxicity and the abundance of biocontrol agents (spiders, lacewings, carabids and parasitoids) has also been observed by Thomson and Hoffmann (2006) in Australia. It should be noted that the sticky traps that we used in our study were not appropriate to sample spiders, which might also be relevant as biocontrol agents in our study area.

Our bird census revealed that the abundance of bird pests and auxiliary birds was similar. Bird feeding behavior changes throughout the season, namely by consuming mostly insects early in the season, which corresponds to the breeding season, when insects are crucial dietary items for their offspring (Herrmann and Anderson, 2007). The breeding season of most bird species in the study area corresponded to the first months of our experiment (June–July), while later on (August–September) they began feeding on grapes which had ripened. In our dataset, vineyards with a higher density of bird pests had considerably more grape losses; however, this relationship was not statistically significant due to the high heterogeneity of the dataset, the overall low effect of birds and the small number of vineyards sampled. Such a low impact of frugivorous birds may be a common trait of the vineyards of Bairrada region, which are highly embedded within a complex landscape matrix that provide shelter, breeding sites and alternative feeding areas for many birds (Pithon et al., 2016). This might contrast with the damages documented in extensive vineyards in more simplified landscapes, such as those in California and Alentejo (southern Portugal), where large flocks of birds such as starlings (Sturnus sp.) can have considerably higher impacts (Stevenson and Virgo, 1971; Curtis et al., 1994).

We did not find any effect of the distance to the edge of the vineyard on the proportion of grape losses. However, most losses were recorded in the first 100 m from the vineyard edge. In previous studies, grape losses caused by birds tended to decline with increasing distances from the edge (Somers and Morris, 2002). Most avian species only visit the vineyards occasionally for feeding, and they rapidly look for cover in the edge habitats (Pithon et al., 2016). Accordingly, we can expect that the vines closest to the edge are visited more often by frugivorous birds (Stevenson and Virgo, 1971; Somers and Morris, 2002; Saxton et al., 2004); thus more losses are predicted in this area. Another factor that may affect grape losses is type of edge, which might influence the abundance and diversity of frugivorous and insectivorous birds, thus potentially affecting grape losses. Unfortunately, for the reasons discussed above, we could not detect such effects in the present study.

Likewise, grape losses caused by insects tended to decline with increasing distances from the edge, although not significantly (Hoffman and Dennehy, 1987). Grape losses are presumably dependent on the ratio of auxiliary and pest insects, which might both be inflated near the edge due to the greater habitat complexity, thus canceling any positive or negative effect on grape losses (Nicholls et al., 2001; Williamson and Johnson, 2005; Sciarretta and Trematerra, 2014; Steel et al., 2017).

This study constitutes an important first step to evaluate the potential role of natural biodiversity on grape productivity in Portugal. Nevertheless, we failed to detect an appreciable effect of biodiversity on grape losses due to insects, fungi or birds, mostly because these losses were already strongly limited by intensive pesticide spraying, which is common practice in the region. Further studies comparing alternative management actions, such as integrated production and biological production strategies, will likely shed new light on the real potential of natural communities to suppress pest outbreaks, thus providing a valuable test to the biotic resistance hypothesis.