Management intensity affects insect pests and natural pest control on Arabica coffee in its native range

1. Agroforestry systems provide opportunities to reduce the trade-off between agricultural production and biodiversity, for example by enhancing a diverse community of species potentially acting as natural pest control agents. While management of agroforestry systems is intensifying across the globe, we lack insights into the impact of management intensity on pest levels and natural pest control, especially along broad management gradients and as compared with nat - ural forests. 2. We assessed the impact of management intensity on major insect pests (the coffee blotch miner, the serpentine leaf miner, the coffee leaf skeletonizer and


| INTRODUC TI ON
The species diversity and functional and structural complexity associated with agroforestry systems provides natural pest control in the form of suppression of pest levels by natural enemies without human intervention (Karp et al., 2013;Vandermeer & Perfecto, 2010). This is apparent, for example, when comparing agroforestry to the production of a single or few annual crops in open fields (Pumariño et al., 2015;Rosati et al., 2021). As such, agroforestry has the potential to both conserve biodiversity and provide an income to smallholder farmers (Cerda et al., 2017;Wei et al., 2018). However, agroforestry comes in many shapes and shades, and we lack insights into how management intensity affects pest levels and natural pest control in agroforestry systems, especially along broad management gradients (Pumariño et al., 2015;Schroth et al., 2000). A promising area to study the effect of management intensity is in the centre of origin of a crop, where broad management gradients exist that start with the crop growing in the shade of the natural forest and end with more intensively managed plantations (Zewdie et al., 2020). More intensive management of agroforestry systems such as coffee and cacao often decreases spatial heterogeneity and biodiversity (Geeraert et al., 2019;Steffan-Dewenter et al., 2007), and more natural and less intensively managed agroforestry systems are often characterized by lower pest levels, for example through a higher diversity and density of natural enemies (Clough et al., 2011). Studies on the effect of management intensity on pest levels and natural pest control along broad gradients may shed light on the sustainability of management intensification, and possibly identify thresholds where management intensification leads to abrupt changes in pest levels. From a management perspective, it is also important to know whether the patterns uncovered are consistent across seasons, and whether patterns are similar for different pest species.
Studies on management intensity in agroforestry have generally found higher pest levels and lower natural pest control in more intensively managed sites (Bianchi et al., 2006;Dalin et al., 2009).
For example, a study on coffee showed that damage by the coffee leaf miner Leucoptera coffeella decreased, and the richness and abundance of parasitoid wasps increased, with increasing forest cover in coffee plantations (Medeiros et al., 2019). Likewise, pest levels on cocoa increased with the intensity of management in cocoa agroforests, whereas natural predators increased with the density of native shade trees (Bisseleua et al., 2013). However, these management gradients are relatively short, and further insights may be gained from an investigation of the impact of management intensity on pest levels and natural pest control along a broader management gradient, which might range from the crop in its natural environment to a commercial plantation. The predominant pests on agroforestry crops frequently differ among years and among the different seasons (Teodoro et al., 2009), and from a management perspective it is thus important to know whether patterns found in one season are reflective of those in other seasons, and whether pest species show similar responses to management actions. Despite this, many studies have focused on single pest species and/or single seasons (Asfaw et al., 2019;Bagny Beilhe et al., 2020;Liebig et al., 2018). Landscape features, such as elevation and forest cover, may also be important in shaping pest levels. At the landscape scale, many coffee insect pests increase or decrease with elevation (Asfaw et al., 2019;Jonsson et al., 2015).
At the local scale, canopy cover, which is often lower in highly managed sites, has a strong impact on microclimate and may thereby affect organismal physiology and plant-insect-natural enemy interactions (Jamieson et al., 2012).
Management intensity can have a strong impact on natural pest control (Diehl et al., 2013;Martin et al., 2013;Rusch et al., 2010). For example, beneficial insects like parasitoids and predators are often more abundant in less managed production systems than in intensively managed systems Medeiros et al., 2019;Whitehouse et al., 2018). Several direct and indirect mechanisms may underlie this lower abundance of natural enemies of insects pests in more intensively managed sites, including the use of pesticides and scarcity of nectar plants, alternative hosts, overwintering sites and other shelters (Medeiros et al., 2019;Tscharntke et al., 2007).
However, few studies have examined the impact of natural enemies on pest insects when plants are growing in their native range (Chen et al., 2015). Overall, our findings highlight that ecological knowledge of natural pest control can be used to intensify production to improve the livelihood of smallholder farmers without jeopardizing natural pest control but only up to a certain point where it starts to deteriorate.

K E Y W O R D S
agroforestry, elevation, Ethiopia, insect pests, management intensity, natural biocontrol, natural pest control, parasitism The native range of Arabica coffee is in the southwestern highlands of Ethiopia, Mount Marsabit in Kenya and the Boma plateau of Sudan (Anthony et al., 2002;Tesfaye et al., 2014). In southwestern Ethiopia, coffee production involves different management systems with highly different management intensities, including (i) forest coffee, where wild coffee is growing in the natural forest with little or no management, (ii) semi-forest coffee, where coffee is managed in a traditional way through filling gaps with seedlings, removal or slashing of competing plants and thinning of the canopy, (iii) semi-plantation coffee, where coffee is systematically planted and pruned, but still traditionally managed and (iv) plantation coffee, where coffee is managed using modern agronomic practices, more heavy pruning, the application of fertilizers, frequent weeding and herbicides (Hundera et al., 2013;Labouisse et al., 2008;Teketay, 1999). As these different production systems co-exist in a mosaic fashion in southwestern Ethiopia, this area provides a unique natural experimental setting in which to examine the impact of management intensity on pest levels and the potential for natural pest control. Understanding the impact of management practices and the natural environment on pest levels and natural pest control allows for the design and implementation of practices that maintain or enhance the survival, reproduction and efficacy of natural enemies (i.e. conservation biocontrol), which can contribute to the sustainability of coffee production in the landscape (Escobar-Ramírez et al., 2019).
The overarching aim was to identify the effect of management intensity, elevation and shade on pest levels during the wet and dry season, and the potential for natural pest control along a broad management gradient. To this aim, we conducted surveys of three major insect pests (the coffee blotch miner Leucoptera caffeina, the serpentine leaf miner Cryphiomystis aletreuta and the coffee leaf skeletonizer Leucoplema dohertyi) and damage by other free-feeding herbivores on Arabica coffee in its centre of origin in southwestern Ethiopia. We also assessed natural pest control by laboratory rearing of field-collected larvae of the coffee blotch miner, which is one of the most common insect pests in the region. To evaluate whether pest levels were regarded as problematic, we conducted interviews with the farmers. More specifically, we addressed the following questions: 1. What is the impact of management intensity, elevation and shade on the distribution of insect pests on Arabica coffee? Are patterns consistent between the wet and dry season? 2. What is the impact of management intensity on the parasitism rate and composition of the parasitoid community on the coffee blotch miner?
Based on a reduction of habitat quality for natural enemies, we expected all major coffee pests to increase, and parasitism to decrease, with coffee management intensity. Given previous reports on species-specific differences in life-histories and niches, we expected the pests to differ in their responses to season, elevation and shade.

| Study system
The study was conducted in Gomma and Gera districts of Jimma zone in southwestern Ethiopia (7°37′-7°56′N and 36°13′-36°39′E; Figure 1a). The area receives an annual rainfall in the range of 1480 to 2150 mm, with the main rainy season between June and September. Mean daily minimum and maximum temperatures are 12°C and 28°C, respectively. The elevation ranges from 1400 to 2500 m a.s.l. The study area is a mosaic landscape with remnants of natural forests and shade coffee plantations but also open areas with annual agriculture. The region is the centre of origin of Arabica coffee Coffea arabica L., and coffee can be found as an understory shrub in the unmanaged natural forests as well as in intensively managed commercial plantations. Improved varieties with resistance to coffee berry disease Colletotrichum kahawae are sometimes used by farmers and are common in plantations, but these improved varieties are not known to exhibit cross-resistance (or cross-susceptibility) against insect pests (Labouisse et al., 2008). Coffee is a major cash crop, and provides the main source of income for many of the smallholder coffee farmers in the region. Arabica coffee is attacked by a diverse group of insect species (Le Pelley, 1973). Three lepidopteran species specialized on coffee are very common in our study area. The coffee blotch miner Leucoptera caffeina (Washbourn) [family Lyonetiidae] oviposits its eggs in rows of 1 to 13 eggs (Notley, 1956), and the larvae create a distinct blotch mine while feeding gregariously in the upper side of the leaf (Figure 1b). The serpentine leaf miner Cryphiomystis aletreuta (Meyrick) [family Gracillaridae] also mines the upper side of the leaf, but leaves a distinctly different imprint as it feeds in a snake-like manner ( Figure 1c). The coffee leaf skeletonizer Leucoplema dohertyi (Warren) [family Epiplemidae] has a free-feeding larva that feeds on the underside of the leaf, and its damage is very distinct as it leaves behind the upper epidermis and veins ( Figure 1d). Finally, there are several other free-feeding insect species that feed on Arabica coffee, including several lepidopterans, slugs and snails ( Figure 1e; Abedeta et al., 2015;Crowe, 2004). The coffee berry borer, the main insect pest on coffee berries in coffee growing regions in the world, is not common in this landscape (Mendesil et al., 2004).
In our study area, one of the most problematic insect pests on Arabica coffee is the coffee blotch miner L. caffeina. The main natural enemies of the coffee blotch miner are parasitoids (Mendesil et al., 2011), with only occasional predation on mines by ants and birds. In other countries where the related leaf miner species L. coffeella is the main challenge, predatory wasps, ants, birds and bats are reported to predate on L. coffeella (Androcioli et al., 2018;De la Mora et al., 2008;Fernandes et al., 2010;Librán-Embid et al., 2017). While the parasitoid community of the coffee blotch miner has been described in detail from plantations in other African countries (Godfray & Hassell, 1989;Notley, 1956), few studies have been carried out in Ethiopia (Mendesil et al., 2011), and none across a broader gradient of coffee management, to assess the potential for natural pest control of the coffee blotch miner in the center of the crop's origin in southwestern Ethiopia.
The blotch miner is native from eastern Africa, is specialized on plants in the genus Coffea and a few other plant species, and has a particularly high density on Arabica coffee (Notley, 1956;Washbourn, 1940). While the exact origin of the leaf miner and its past distribution are unknown, there is no scientific record or living memory that the species has invaded the study area, and the species has thus likely coevolved for hundreds, if not thousands, of years with both coffee plants and parasitoids in the study area.
Permits for the field study in Gomma and Gera were granted by Jimma zone Coffee and Tea Authority (No. BS/8075/8-96/83). No ethical approval was required.

| Study area, insect survey and site descriptors
A total of sixty sites distributed across the four different coffee management systems were selected from the Gomma and Gera districts, with a minimum distance between sites of 0.5 km (Figure 1a; see Zewdie et al., 2020 for detailed information on site selection).
Following Gole et al. (2002), Labouisse et al. (2008) and Hundera et al. (2013), sites were classified into forest, semi-forest, semiplantation and plantation coffee systems. Forest sites had no or little management, semi-forest sites were traditionally managed through gap-filling with seedlings, removal or slashing of competing plants and thinning of the canopy, semi-plantation sites were traditionally managed but planting and pruning was more systematic, and plantation sites were managed following modern agronomic practices with heavy pruning, application of fertilizers, frequent weeding and application of herbicides, though no insecticides are used ( Figure S2; Gole et al., 2002;Labouisse et al., 2008;Hundera et al., 2013;Zewdie et al., 2020). Within each site, a 30 × 30 m plot was established, and 16 coffee shrubs were selected on the intersections of the 10 m gridlines ( Figure S3). A survey of the three insect pests and damage by free-feeding herbivores was conducted in both the wet (July-August 2018) and dry season (January-March 2019). For this, we randomly selected two branches (as based on being representative for the typical size and structure of branches on the coffee shrub) from each of the 16 shrubs in each plot. On these two branches, we counted the total number of leaves, as well as the number of leaves damaged, separately for the coffee blotch miner, serpentine leaf miner, coffee leaf skeletonizer and damage by other free-feeding herbivores. Values were pooled across branches and shrubs to obtain site-level averages for analyses. We do note that pest-induced leaf drop, which is known from, for example the coffee leaf miner C. coffeella (Dantas et al., 2021;Waller et al., 2007), might result in an underestimation of pest and herbivory levels.
F I G U R E 1 Overview of the study location and system. The inset in (a) shows the location of the study area (rectangle) in Jimma zone in southwestern Ethiopia. The sixty sites are plotted on the map of the study area, where shapes and colours of the symbols indicate the coffee production system in a site. The white and green background colours represent open and forested areas, respectively. The photos below the map depict characteristic damage by the three major coffee pests and other free-feeding herbivores: (b) a leaf mine of the coffee blotch miner Leucoptera caffeina (with a parasitoid nearby), (c) a leaf mine of the serpentine leaf miner Cryphiomystis aletreuta, (d) the coffee leaf skeletonizer Leucoplema dohertyi and (e) damage by other free-feeding herbivores.
For each site, elevation was recorded by GPS, and canopy cover was estimated from five pictures taken from above the coffee shrubs using ImageJ software v. 1.50i (Schneider et al., 2012; see Zewdie et al., 2020 for details).

| Rearing of parasitoids
To assess the impact of management intensity on the potential for natural pest control of the coffee blotch miner, we selected four sites from each of the four management systems (forest, semiforest, semi-plantation and plantation). In the wet season 2018, we collected 100 leaves that were mined by the coffee blotch miner from each site. These infested leaves were collected randomly from the subset of leaves with medium to large mines and with an unbroken mine surface. As densities of the blotch miner were very low in the forest system, we could not collect enough infested leaves for a meaningful comparison, and forest sites were thus excluded from the analyses. Leaves were stored in plastic boxes (25 × 16 × 6 cm) with two sheets of toilet paper to absorb moisture and prevent fungal growth. Boxes were stored in the laboratory away from direct sunlight, where they were kept for 35 days, when host and parasitoid emergence ended. For each site, we then calculated the proportion of emerged parasitoids divided by the summed number of emerged adult blotch miner moths and parasitoids, which is henceforth referred to as the parasitism rate. The emerged parasitoids were classified into the families Braconidae, Encyrtidae, Eulophidae and Ichneumonidae based on morphological characteristics (e.g. Askew, 1968;Ubaidillah, 2006).

| Interview on perceived problems with different insect pests
To link the empirical pest levels with the perception of these pest levels by farmers, we interviewed the farmers in the forest, semiforest and semi-plantation systems, and the sub-managers in the plantation system. To avoid leading questions and an influence of differences in education or knowledge among farmers and submanagers, we asked two open questions during a larger interview on coffee diseases, and phrased it as 'Did any pest or pathogen change in abundance over the last three decades, and if so, has it become more problematic?' and 'What are the main challenges and threats you faced during your coffee production in recent years?' We then noted which type of insects that were perceived as a problem. All interviews were conducted by the same two persons (Biruk Ayalew and Dinkissa Beche).

| Statistical analyses
Statistical analyses were conducted using R v 3.6.3. Mixed-effects models were fitted using the functions lmer and glmer in the lme4 package (Bates et al., 2015), and logit models using function glm in the base package (R Core Team, 2021). We evaluated model fit using diagnostics in the sjPlot and DHaRma packages (Lüdecke, 2020), tested for significance using the Anova function in the package caR (Fox & Weisberg, 2019), and compared coffee production systems using the function emmeans in the package emmeans (Lenth, 2020).
To assess the impact of season, management system, elevation and shade on pest levels, we modelled the average number of infested leaves at the site level (the coffee blotch miner, serpentine leaf miner, coffee leaf skeletonizer and damage by other free-feeding herbivores) as functions of season, management system, elevation and shade using linear mixed models with Gaussian distribution and identity link function. Variance inflation factor was <5 in all models, indicating that the models had no problems with multi-collinearity (James et al., 2013). To account for sampling the same plots during the wet and dry season, we included the random effect plot identity. As the effect of management system, elevation and shade may differ between the two seasons, we included the two-way interactions between season and all other variables. As patterns differed between the seasons (Table S1), we also ran season-specific models using a linear model with identity link function in the base R package. We scaled all predictor variables to zero mean and unit variance to obtain standardized regression coefficients. We used the function moran.randtest from the package aDesPatial to test for spatial autocorrelation in the residuals (Diniz-Filho et al., 2003). Spatial autocorrelation was not significant (p < 0.05), with the exception of a weak pattern of spatial autocorrelation for the serpentine leaf miner in the dry season (p = 0.03).
To assess the impact of coffee production system on the parasitism rate of the coffee blotch miner, we modelled the proportion of parasitoids out of the total sum of emerged hosts and parasitoids as a function of management system using a generalized linear mixedeffects model with a binomial distribution and logit link, where we accounted for overdispersion by including the random effect plot identity. We did not include shade and elevation due to the lower number of replicates in this analysis. To test whether the composition of the parasitoid community was affected by coffee production system, we modelled the composition of the parasitoid community as a function of coffee production system using the function adonis2 in the package vegan (Oksanen et al., 2019). We complemented these analyses at the level of individual parasitoid families using the models described for parasitism rate above, with site-level counts as a response variable and a Poisson link.
To assess differences in the problems perceived by farmers with the different insect pests between the four production systems, we modelled the probability of a farmer reporting problems with the coffee blotch miner, serpentine leaf miner, coffee leaf skeletonizer and damage by other free-feeding herbivores as functions of coffee production system using generalized linear models with a binomial distribution and logit link. We conducted a similar model for the proportion of farmers that reported a problem with at least one of the four damage types.

| RE SULTS
We inspected pests and herbivory on 1499 ± 295 (mean ± SD) leaves per site. The percentage of infested leaves varied among the species, with 8.2% of the leaves attacked by the coffee blotch miner, 3.8% by the serpentine leaf miner, 38.9% by the coffee leaf skeletonizer and 27.9% by other free-feeding herbivores. The pests also showed remarkable differences in their seasonal dynamics. The coffee blotch miner was three times more common in the wet season (12.0 vs. 4.4%, respectively), whereas the serpentine leaf miner was seven times more common in the dry season (0.9 vs. 6.7%). The coffee leaf skeletonizer had very similar incidence in the wet and dry season (38.0% and 39.7%, respectively), and the percentage of leaves damaged by free-feeding herbivores decreased from 30.6% in the wet season to 25.2% in the dry season.
Management intensity had a major impact on pest levels in the wet season, whereas it did not leave a detectable imprint on pest levels in the dry season ( Figure 2, Table 1 and Table S1). In the wet season, infestation by the coffee blotch miner, coffee leaf skeletonizer and other free-feeding herbivores was similar in the forest, semi-forest and semi-plantation systems, whereas incidence was significantly higher in the plantation system than in the semiforest and semi-plantation systems for all three species, and higher in the plantation system than in the forest system for the coffee blotch miner (Figure 2A,C,D). A similar but weaker trend was apparent for the serpentine leaf miner ( Figure 2B). In the wet season, all pests decreased with elevation, whereas in the dry season only infestation by the coffee leaf skeletonizer decreased with elevation ( Figure 3, Table 1). In the wet season, infestation by the coffee leaf skeletonizer was higher with increasing shade, whereas no other pests were affected by shade in either the wet or dry season ( Figure S1, Table 1).

The number of farmers or sub-managers with perceived problems
with insect pests increased with management intensity (Figure 5;

F I G U R E 2
The impact of management system on (A) coffee blotch miner Leucoptera caffeina, (B) serpentine leaf miner Cryphiomystis aletreuta, (C) coffee leaf skeletonizer Leucoplema dohertyi and (D) damage by other free-feeding herbivores in the wet and dry season. Shown are box plots based on the percentage of infested leaves at the site level, separately for the wet and dry season. Letters above the box plots are based on Tukey-adjusted pairwise differences among management systems for the wet season (management system had no significant effect on pest levels in the dry season). Tables S4 and S5). Farmers did not perceive any problems with insect pests in the forest system and only few farmers reported problems with the coffee blotch miner and/or coffee leaf skeletonizer in the semi-forest and semi-plantation systems, whereas more than half of the sub-managers in the plantations reported problems with one of these two pests ( Figure 5). None of the interviewees mentioned problems with the serpentine leaf miner or damage by free-feeding herbivores.

| DISCUSS ION
To investigate pest levels and natural pest control along a broad gradient of management intensity, we focused on Arabica coffee in its center of origin. Levels of the three major pests in the study area and damage by other free-feeding herbivores were highest in the plantation system, whereas they were lower and relatively similar in the forest, semi-forest and semi-plantation systems. The insect pests TA B L E 1 The impact of management system, elevation and shade on the coffee blotch miner Leucoptera caffeina, the serpentine leaf miner Cryphiomystis aletreuta, the coffee leaf skeletonizer Leucoplema dohertyi, and free-feeding herbivory, separately for the wet and dry season. Shown are standardized regression coefficients (SC), F-values (with numerator and denominator degrees of freedom between brackets in subscript) and p-values from linear mixed models. Significant p-values (p > 0.05) are shown in bold. Altitude (m a.s.l) Free feeding herbivory (%) (d) differed in seasonal dynamics, declined with elevation, and were not or weakly influenced by shade levels. The higher pest levels in the plantation were matched by a lower parasitism rate and a distinct parasitoid community. Together, these findings suggest that intensification of management does not gradually increase pest levels and erode natural pest control-rather, we found no effect of management intensity when moving from coffee growing in the natural forest to semi-forest and semi-plantation systems, but there was an abrupt increase in pest levels and decrease in natural pest control in the plantations. Such insights are important to guide decisionmaking in landscapes such as southwest Ethiopia where management intensity is rapidly increasing, but at the same time there are strong incentives to increase sustainability.

Wet season
The insect pests differed in their seasonal dynamics. The coffee blotch miner and free-feeding herbivory were more common in the wet season than in the dry season, whereas the serpentine leaf miner showed the reverse pattern. In contrast, the coffee leaf skeletonizer had very similar infestation levels in the two seasons. While seasonal patterns of the serpentine miner, skeletonizer and free-feeding herbivory have not been described previously, our finding of higher densities of the coffee blotch miner in the wet season matches previous findings (Ababulgu, 2010;Abedeta et al., 2015). These patterns also match previous reports by Lomelí-Flores et al. (2010) of higher abundances of the coffee leaf miner Leucoptera coffeella during the wet season in Mexico. In contrast, in Brazil, the world's largest producer of coffee, the incidence of this leaf miner species reaches its peak in the dry season (Pereira et al., 2007). The effect of management intensity on the three insect pests and damage by free-feeding herbivores was strong during the wet season, but undetectable during the dry season. While we do not know why patterns were more pronounced in the wet season, this finding is important from an applied perspective, as it emphasizes that we should not rely on data collected in a single season to devise management plans. In the wet season, pest levels were relatively similar in the forest, semi-forest and semi-plantation systems, whereas densities in the more intensively managed plantation were higher. This was confirmed by interviews, which revealed that the pests were perceived as unimportant in the natural forest and only rarely caused problems in the semi-forest and semi-plantation systems, but reached outbreak densities in the plantation system. While these findings are congruent, we do note we cannot rule out that some of the differences in the interview responses are due to differences between the sub-managers of plantations and smallholder farmers in their level of education and related awareness of the pests. Our findings are in line with previous studies that reported that insect pests were more problematic in plantations than in semi-forest and garden production systems (Abebe, 1987). In particular, we found that leaf miner densities were very low in the natural forest systems compared to the more disturbed sites. This is in line with Mendesil et al. (2011) and could be related to the environmental requirement of the leaf miners. In our study, all species decreased with elevation in the wet season. While the decrease in F I G U R E 5 Perceived problems by farmers and sub-managers with coffee blotch miner, coffee leaf skeletonizer, or at least one of them. The number of sites per management system are shown between brackets on the x-axis. None of the farmers reported problems with the serpentine leaf miner or damage by free-feeding herbivores. the abundance of other coffee pests such as antestia bugs and coffee berry borer with elevation has been previously reported (Asfaw et al., 2019;Babin et al., 2018), no previous study has highlighted that the effect of elevation varies with the season.
Several environmental factors may explain the difference in pest levels between the natural forest, semi-forest and semi-plantation systems on the one hand, and the plantations on the other hand. While it is often highlighted that the lower shade levels in more intensively managed production systems may cause higher pest levels Teodoro et al., 2009), our findings showed that pest species were mostly unaffected by shade levels on Arabica coffee in their native range. When comparing our findings to studies in other parts of the world, it is important to realize that while coffee is growing along a broad gradient of management intensity in southwestern Ethiopia, shade levels are high across the full gradient, and coffee grown in full sun is near to non-existent. It might then be that the effect of shade is non-linear, with no effect in the gradient from low to high canopy cover in our study, but a strong effect of shade when directly comparing shade and sun coffee. Besides environmental factors, the pattern of higher pest levels in intensively managed productions systems may be explained by bottom-up and top-down effects. From a bottom-up perspective, the most striking change along the management gradient is the frequent use of cultivars in the plantation system, whereas coffee grown in the natural forests and the smallholder farms is largely reflective of the wild genetic reservoir of Arabica coffee. While these cultivars were selected for resistance to coffee berry disease caused by Colletotrichum kahawae, and are not known (nor believed) to be more susceptible to insect pests, the genetic uniformity in the plantation system may favour the adaptation of insects to the crop.
However, a recent study in our region did not find an effect of genetic diversity on coffee diseases .
From a top-down perspective, management intensification may reduce the diversity and abundance of natural enemies, for example through the reduction of vegetation complexity, which may result in a scarcity of nectar plants, alternative hosts, overwintering sites and other shelters (Medeiros et al., 2019;Vandermeer & Perfecto, 2010) or through the direct effects of use of pesticides on natural enemies (e.g. Schmidt-Jeffris et al., 2022). Matching this expectation, we found that the parasitism rate was lower in the plantations than in the less intensively managed systems, which supports the hypothesis that top-down effects could explain the higher densities of insect pests in the plantations. Lower levels of natural pest control in more intensively managed sites has also been suggested in other coffee and cocoa agroforestry systems, as based on lower densities and diversity of natural enemies like parasitoid wasps, spiders and ants in the more intensively managed sites (Bisseleua et al., 2013;Kone et al., 2014;Sperber et al., 2004). Our characterization of the parasitoid community revealed that the coffee blotch miner was mostly attacked by parasitoids from the hymenopteran families Encyrtidae, Eulophidae and Braconidae. The dominance of these families reflects the findings of classic investigations on the parasitoids of coffee blotch miners in Kenya and other parts of Africa (Kerrich, 1969;Notley, 1956), as well as a more recent characterization of the parasitoid community at coffee research stations in Ethiopia (Mendesil et al., 2011). Interestingly, the impact of management intensity on the parasitoid community differed among parasitoid families, with a notable drop in the relative abundance of eulophid parasitoids in the plantation system. This might suggest that some parasitoid families are more sensitive to management intensity than others, but unfortunately there are no other studies available to validate this hypothesis. One interesting direction might be to explore spatial factors that might affect natural enemies, such as the network structure of coffee farms or the distance to the natural forest (van Nouhuys & Hanski, 2005). Another interesting research direction would be to investigate whether the patterns detected for parasitoids can be generalized to other natural pest control agents, such as predatory wasps, ants, birds and bats (Hoehn et al., 2009;Karp et al., 2013;Philpott & Armbrecht, 2006).
We found that pest levels and natural pest control were similar in the natural forest and farms with low management intensity, whereas pest levels reached outbreak densities, and parasitism was reduced, in the plantation sites. This highlights that natural enemies may reduce pest pressures by insects below an economic threshold in the forest and less intensively managed farms, whereas economic losses might occur in the plantation system. To design management actions that can retain or enhance natural pest control in high production systems, we need taxonomic studies that describe the parasitoid species morphologically and molecularly, and ecological studies that investigate the life cycles and habitat requirements of the different parasitoid species. Ecologically informed management may then conserve natural pest control even during the process of management intensification, which gives the dual benefit of sustainable production of coffee and an improved livelihood for smallholder farmers. Conserving natural pest control and associated parasitoid diversity is even more important for southwestern Ethiopia, which harbours both the wild genetic reservoir of coffee and the coevolved natural enemies of many coffee pests. Tack led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication. We thank Moritz Stüber for help with producing the map, and the coffee owners for allowing us to work in their coffee plots.

CO N FLI C T O F I NTE R E S T S TATE M E NT
None of the authors have a conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are available via the Dryad Digital Repository https://doi. org/10.5061/dryad.h9w0v t4nw (Shimales et al., 2023).

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information can be found online in the Supporting Information section at the end of this article.     Table S5. Pairwise comparisons for the probability that a farmer perceives problems with the coffee blotch miner, skeletonizer or any pest between the semi-forest, semi-plantation and commercial plantation production systems. Shown are the production systems compared, estimates, standard errors (SE), z-ratios and p-values.
Comparisons with the forest system are not included due to complete separation (i.e. 100% "failures"), in which case logit (