Sacred church forests as sources of wild pollinators for the surrounding smallholder agricultural farms in Lake Tana Basin, Ethiopia

Remnant natural forests act as a source of wild pollinators that are potentially relevant for crop pollination for sustaining the food production system of smallholder farms. In the Ethiopian highlands, sacred forests scattered within agricultural landscapes harbour wild pollinators. The contribution of wild pollinators to crop pollination services to surrounding smallholder agricultural farms is largely unknown. In this study, we empirically assessed the effects of Ethiopian church forest habitats on crop flower visitation by pollinators on smallholder farms in Lake Tana basin, Ethiopia. Crop flower visitation rates were recorded in 72 crop fields along distance gradients within a 1500 m buffer zone around 15 church forest patches. We constructed cross-validated generalized ad- ditive mixed models (GAMMs) to fit the crop flower visitation rates with fixed effect explanatory variables (distance from forest, forest patch size, forest age, forest functional diversity index, forest proximity index and crop type) that are known to potentially influence pollinator visitation in crop fields. Total flower visitation rates decreased with distance to the forest patches across an agricultural landscape. Visitation rates increased with increasing forest patch size, functional richness, proximity index, and age of church forest patches. Overall, we found that the rate of crop flower visitation varies with pollinator-dependent crop types and wild pollinator groups. An in-depth understanding of the forest biodiversity, cultural value, food security nexus enables the promotion of ongoing church forest stewardship in local and regional policies.


Introduction
Crop pollination is an essential ecosystem service for agricultural systems (Aizen et al., 2009;Klein et al., 2007;Lippert et al., 2021;Ollerton et al., 2011). The majority of the main global crop species, over 66%, benefit from animal pollination and contribute up to 30% of overall food production Kremen et al., 2002;Potts et al., 2010). Pollination-dependent crops are not only economically valuable (Gallai et al., 2016;Tibesigwa et al., 2019) but also include crops that provide essential micronutrients to increase overall nutrition diversity (Chaplin-Kramer et al., 2014;IPBES, 2016). Recent research has shown that wild bees are more efficient crop pollinators than managed bees (Blitzer et al., 2016;Garibaldi et al., 2013;Tibesigwa et al., 2019). Furthermore, a diverse community of wild pollinators from natural habitats provides more stable and adequate crop pollination compared with a single species (IPBES, 2016;Potts et al., 2010) for globally important crops such as oilseeds, pulses, fruits, vegetables, spices, and coffee .
Wild pollinators require natural habitats within agricultural landscapes that can provide floral resources and nesting habitats . Natural forest ecosystems are increasingly acknowledged in intensive agricultural areas for their ecological functions, which support essential ecosystem services contributing to local livelihoods and the well-being of the rural population (Ayanu et al., 2012;Costanza et al., 1997;Stave et al., 2017;Willemen et al., 2012). A unique type of natural forest, sacred forests, conserved by local communities through cultural or religious values, are found throughout the world (Avtzis et al., 2018;Dudley et al., 2009;Verschuuren et al., 2012). The remnant church forests in the Ethiopian highlands are examples of such sacred forest sites. Ethiopian church forests have been preserved for centuries by local communities due to their perceived cultural and religious significance (Cardelús et al., 2012;Wilson, 2016). These church forests appear as small 'islands' of green in the dry Ethiopian highlands landscape and are vital for global biodiversity preservation efforts (Aerts et al., 2008;Ermilov et al., 2012;Reynolds et al., 2017;Wilson, 2016). They provide important nesting habitats and foraging sites for wild insect pollinators such as solitary or honey bees, bumble bees, hoverflies or butterflies (Lowman, 2011;Wassie et al., 2010).
Wild pollinators from natural forest habitats move to nearby agricultural fields and provide pollination services in many cropping systems (Blitzer et al., 2012;Burns et al., 2009;Carvalheiro et al., 2010). The stability of crop flower visitation and crop production typically declines with the distance of crop fields from natural forest habitats (Carvalheiro et al., 2010;Decocq et al., 2016;Garibaldi et al., 2011). Consequently, pollination services provided by religious church forests have the potential to especially support the yield of adjacent pollinatordependent crops (Aizen et al., 2009;Garibaldi et al., 2011;Ricketts et al., 2008). However, the movement of pollinators to flowering crops over a landscape is influenced by the spatial configuration and characteristics of their forest habitats (Lonsdorf et al., 2009).
Church forests are shrinking and decreasing in density, with noticeable declines in biodiversity due to livestock grazing within them, wood extraction, and encroaching adjacent agricultural practices (Bongers et al., 2006;Reynolds et al., 2017;Teka et al., 2013). The decline of wild pollinators raise concerns about declining food production stability (Potts et al., 2016) and threaten household food security in Ethiopia. Wild insect pollinators are negatively affected by the loss of natural habitats that provide nesting and floral resources (Pereira et al., 2012;Ricketts et al., 2008). This, in turn, means that crop pollination services provided by church forests to the surrounding smallholder agricultural fields are jeopardized, resulting in lower pollinatordependent crop yields, reduced local livelihood options, and food insecurity. To reverse this trend, understanding the linkage between church forests and nearby agricultural farms through crop pollination services is one approach to support conservation and improve the livelihood of rural people. This could help change the perception of rural farmers who perceive wild pollination services as an infinite gift of nature and promote the preservation of church forest ecosystems adjacent to smallholder crop fields.
Earlier studies have assessed the vegetation diversity and contribution of the church to biodiversity conservation within the boundary of church forests (Aerts et al., 2016;Wassie et al., 2010;Woods et al., 2020). The contribution of wild pollinators for crop pollination services to surrounding smallholder agricultural farms is unknown. In this study, we used field data, remote sensing imagery, and a spatially explicit model of central place foraging to quantitatively assess and map crop pollination services in smallholder farms as provided by church forestbased wild pollinators in the Lake Tana basin in 2020. Our objectives were to (i) spatially characterize sacred church forest patches and pollinator-dependent croplands; (ii) assess church forest floral resources, nesting sites, and wild pollinator guilds; and (iii) model the provision of pollination services using indicators for flower visitation on four types of crop fields varied along distance gradients from church forests, forest patch size, forest floral functional diversity index, forest age, forest proximity index and crop types.

Description of the study area
We conducted our study in the sacred church forests of the Ethiopian highlands and surrounding pollinator-dependent crop fields in the Lake Tana basin of the Blue Nile, located in the Amhara Region of Ethiopia (36 • 44 ′ 40 ′′ to 38 • 13 ′ 13 ′′ N and 10 • 56 ′ 17 ′′ to 12 • 44 ′ 55 ′′ E) (Fig. 1). The study area covers approximately 15,000 km 2 , including Lake Tana with a surface area of 3000 km 2 . The elevation ranges from 1782 to 4109 m above sea level. The study area has a tropical highland monsoon climate with an average annual temperature of 19.6 • C, with a maximum of 30 • C and a minimum of 14 • C. Rainfall is strongly seasonal, with a dry season between the end of October and mid-June and a pronounced rainy season between June and September. Accordingly, mean annual rainfall ranges between 800 and 2000 mm (Conway and Schipper, 2011;Mohamed et al., 2005).
The landscape of the study area is dominated by intensively cultivated farmland and crops, with interspersed patches of church forests, woodland, shrubland, built-up areas, wetlands, water bodies, and Fig. 1. Map of the study area and experimental design. Location of the study area in the north-western Ethiopian highlands (left-hand side) and land cover map including distribution of the five different sub-landscapes of the Lake Tana basin (right-hand side). Elevation map was created using ALOS PALSAR 12.5 m resolution data (https://search.asf.alaska.edu) and the land cover map was produced by the authors using PlanetScope satellite imageries with a resolution of 3 m (http s://www.planet.com), see section 2.3. grasslands. The livelihood of local people depends on smallholder agriculture in the Lake Tana basin. The smallholder farms surrounding the church forests grow a mix of perennials and seasonal crops (mainly cereals (67-74% of crop area), pulses (12-20%), oilseeds (4-6%), vegetables, fruit crops, and coffee) (CSA, 2020) to minimize the potential risks of crop failure and secure their livelihood requirements. Food security and nutritional diversity in the area are low (Karlberg et al., 2015).

Methodological approach
The methodological approach (Fig. 2) was developed based on the evidence that church forest habitats provide pollination service to surrounding pollinator-dependent crop fields. Lonsdorf et al. (2009) presented a spatially explicit model to evaluate the relative abundance of bees in landscapes by assuming bees diffuse out from their nest and use foraging habitats indiscriminately with respect to foraging returns. This model assumes homogeneity in landscape and does not account for behavioural mechanisms by which bees can select appropriate foraging resource sites. The more recent model of Olsson and Bolin (2014) built a habitat use model from central place foraging theory, demonstrating how to predict what patches foragers should use in a specific landscape. The model, which is general for any central forager, shows that for any given patch quality, there will be a maximum distance that any forager would be willing to travel.
We conceptualize the pollination modelling frameworks proposed by Lonsdorf et al. (2009) and Olsson and Bolin (2014) to model crop pollination service (see Fig. 2). In the fragmented landscapes of the Lake Tana basin, we assume that wild pollinators are central-place foragers originating from scattered church forest patches. During the summer crop growing season, wild pollinators nesting in church forests diffuse into available flowering crop fields within agricultural landscapes. All bees are central-place foragers, including honey bees, bumble bees, and solitary bees that use the mass-flowering crops in the surrounding agricultural matrix as a foraging source (Cresswell et al., 2000).
Our model was built by: (i) considering wild central place foragers in church forests in a heterogeneous landscape; (ii) integrating behavioural response of pollinators towards different crops; (iii) considering forest age and forest functional diversity index and (iv) including wild pollinators visitation rates across four distance buffers from church forest habitats for use in crop pollination models. For modelling the pollination services, relevant factors that influence the decay in pollination service with distance, include indicators for the biophysical traits of church forests and pollination-benefiting crop fields that influence crop flower visitation rate (Table 1). A detailed study of floral resources and nesting habitats inside church forests revealed habitat functional diversity to which such forest patches may support wild pollinators providing pollination services. Forest proximity to crop fields within a wild pollinator's flight range (1500 m) was incorporated in the model to account for pollinator home-range overlap.

Spatial characteristics of church forests and croplands
The spatial configuration and characteristics of forest patches may influence the flower visitation of wild pollinators from forest habitats to the surrounding crop fields. To spatially describe the sacred church forest patches and pollinator-dependent croplands, we made a land cover map of the Lake Tana basin. The land cover map was based on PlanetScope satellite imagery for 2019 and 2020 provided by Planet Labs Inc. (www.planet.com) with 3 m spatial resolution and four spectral bands, blue (455-515 nm), green (500-590 nm), red (590-670 nm) and near-infrared (780-860 nm) (Planet Team, 2017). To account for the spatial and classification errors, we averaged the two years (2019 and 2020) maps to produce the representative land cover of the basin. Eight land cover classes were identified: cultivated land, church forest, woodland, wetland, shrubland, grassland, water, and built-up. Reference data for the eight land cover classes were collected from in-situ field survey, aerial photographs (2015 acquired with a ground sample distance of 25 cm) and imagery available in Google Earth for 2020 (spatial resolution: 0.65 m/pixel) (http://earth.google.com). In total, 260 polygons of reference data were collected for image classification. We made a supervised classification of land cover types with Orfeo ToolBox random forest classifier (Grizonnet et al., 2017) in QGIS, version 3.16.4 (www.qgis.com), where 70% of the reference polygons were used for classification purposes and the remaining were used for accuracy assessment. The overall accuracy was derived from the error matrices based on collected ground-truth data (Congalton, 1991). The overall classification accuracy for the 2020 forest and cropland classes  from the PlanetScope image was 93.5%, and the kappa coefficient was 0.91, indicating a high degree of accuracy.
We extracted the sacred church forest patches (>0.5 ha) across the Lake Tana basin from the land cover map by using point shapefiles of church locations (Wassie et al., 2010). We identified 1058 patches of church forests covering a total area of 10,630 ha, with forest patch sizes ranging from 0.5 to 234.5 ha. We selected 15 church forest patches for this study in five sub-landscapes ( Fig. 1). This selection was based on different forest patch sizes, different levels of patch isolation and the availability of different pollinator-dependent crops around them. In general, all sampled church forests were surrounded by the variegated landscape of smallholder agricultural fields.
To assess crop pollination services provided by the church forest patches, we delineated 1500 m wide buffers around each church forest patch, based on an assumed maximum foraging distance for wild insect pollinators originating from the church forest habitats. For field data collection purposes, a detailed map of pollinator-dependent crop types grown near the 15 church forest patches (1500 m radius) was digitized from the same PlanetScope satellite images for the year 2020 assisted by ground survey data.
We obtained information on the crop types of the Lake Tana basin growing in the main seasons of the year 2019 and 2020 from the regional agricultural office and Ethiopian Statistical Authority (CSA) for the study area (ANRS BoA, 2020; CSA, 2020). Out of 53 cultivated crop types, we identified 37 pollinator-dependent crops (see supplementary information 1 appendix A). Then, four pollinator-dependent crop types grown in the study area were therefore selected with the largest production area and economic importance (income). A crop is considered pollinator-dependent if pollination is necessary to maximize fruits or seeds used for human consumption. Crops were classified according to their dependence on pollinators based on the extent of the reduction in crop production. For this study, we obtained information on the dependency of perennial and annual crops on insect pollinators from Aizen et al. (2009), Gallai et al. (2009) and Klein et al. (2007). These authors defined pollinator dependence classes as: (i) none (production does not increase with pollination), (ii) 0-10% production reduction (little), (iii) 10-40% reduction (modest), (iv) 40-90% reduction (high) and (v) >90% reduction without pollinators (essential). In this study, pollinator visitation rates were recorded on mango (Mangifera indica) (high), coffee (coffee arabica) (modest), horse bean (Vicia faba) (modest), and field pea (Pisum sativum) crop fields, all crop with a high pollinator dependence. In agricultural landscapes, the level of pollination services provided to crop fields (focal patches) frequently depends on the proximity of the farms to one or more forest patches. The magnitude of crop flowervisiting rates across crop fields is influenced by the number of forest patches within a foraging range of pollinators. (Fig. 3). For each crop field, we calculated the forest proximity index (Moilanen and Nieminen, 2002;Radford and Bennett, 2004;Winfree et al., 2005) by considering all surrounding church forest patches within a radius of 1500 m, about the maximum foraging distance of central place pollinators Marzinzig et al., 2018;Osborne et al., 2008). This index is well suited for highly fragmented landscapes that measure forest proximity relative to target crop fields. The proximity index was computed by using crop fields and forest patches within the threshold distance based on the partial sampling function of FRAGSTATS version 4.2 (Fig. 3).

Church forest functional diversity
We collected information on floral resources, nesting sites, and wild pollinator guild in the field to describe the role of church forests for pollination. The floral abundance of tree species was used to represent the functional diversity of church forests based on the values of speciesbased functional traits: floral tree height (mean height, m), DBH (cm), nesting pollinator guilds and flowering periods (flowering and peak flowering in months).
We obtained floral resource data from 69 vegetation plots established within 15 sacred church forest patches from April to October 2020. The area of the sample church forest patches ranges from 7 ha to 234.5 ha. For the 15 church forest patches, linear transects were laid systematically starting from the southwest outer edge of the forest in each patch area. Along the study transects, a total of 69 sampling plots of Spatial data of crop field sampling plots and church forest patches extracted from the land cover data. (Mcgarigal et al., 2012;Radford and Bennett, 2004) Crop types (attractiveness of crops) Crop types grown by smallholder farmers attract wild pollinators differently.
Four major crop types: Mango, coffee, horse bean and field pea were identified and included (CSA, 2020). (Fahrig et al., 2011;Raderschall et al., 2021) 20 m × 20 m were laid at 100 m intervals for small patch areas (≤30 ha) and 200 m intervals for large patch areas (>30 ha). All floral species surveyed were identified to the species level using the Flora of Ethiopia and Eritrea books (Edwards et al., 2000(Edwards et al., , 1995Hedberg et al., 2003;Hedberg and Edwards, 1989). The flowering months of floral resources of church forest habitats were identified based on (Admassu et al., 2014;Fichtl and Adi, 1994). The flowering period of most of the floral trees started late in October of the year, reached a peak during the dry season, and ended in June. The phenological variation between church forest floral resources and crops in smallholder agricultural fields were examined to understand the dispersal of wild insect pollinators from their nesting church forest habitats.
Based on the church forest traits, we calculated the forest functional diversity at the church forest patch-level using the functional richness (FRic) metric (Mason et al., 2005) for each of the 15 church forest patches. We calculated the forest FRic value of each tree species weighted by its abundance and then clustered by each church forest patch. The average value of FRic was calculated for each church forest patch (Eq. (1)). All the forest inventory data were standardized before calculating the functional richness index to avoid scale effects (Casanoves et al., 2011). The forest functional richness index analysis was carried out in R version 4.0.2 (R Core Team, 2020), using the 'FD' (Laliberté and Legendre, 2010) and 'vegan' (Oksanen et al., 2019) packages. Gower's distance calculation method (Gower, 1971) was used to build a distance matrix of the functional space.
where FR ic = the functional richness of plant traits (floral tree height, diameter at breast height, floral tree abundance, flowering period) c in community i, SF ci = the niche space filled by the species within the community, and R c = the absolute range of the character.

Crop flower visitation rate
The crop flower visitation rate as an indicator for pollination service was assessed by counting pollinator visitation in crop fields along a distance gradient from church forest patches in the study area. For this, eighteen (18) radial transects of length 1500 m were established starting from the edge of the nearest church forest habitat towards the adjacent crop fields. Wild bees and bumble bees regularly forage within 1500 m of their natural nesting habitats Marzinzig et al., 2018;Osborne et al., 2008;Steffan-Dewenter and Kuhn, 2003), and crop visitation at greater distances from unmanaged pollinators is likely to be negligible (Garibaldi et al., 2011;Ricketts et al., 2008;Serna-Chavez et al., 2014). Except for five transects on two of the largest church forest patches surrounded by different crops, each church forest had one transect. We expected that the effects of church forest habitats on pollination ecosystem service would decay with distance (i.e., greater crop flower visitation would occur at small distances to church forests). Four crop field sampling plots with 2 m × 2 m ground area at each transect were established at distance buffers 200 m, 500 m, 1000 m, and 1500 m. In total, 72 sampling plots were laid at the distance buffers to record crop flower visitation rates. Due to the similarity of most agricultural fields, these crop field plots were adequate. The distance from church forest habitats to crop field sampling plots was measured using a handheld GPS unit.
Crop pollinator visitation, number of flowers, and wild pollinator species were recorded during two flowering seasons from late July to late October in 2019 and 2020 when most crop species were at their peak flowering period. Observations on crop field plots were made under favourable weather conditions (sunny and no precipitation), between 10:00 and 15:00. Following the standardized methodological approach used by Klein et al. (2003) and Ricketts (2004), we conducted a transect walk starting from the edge of the church forest to record crop flower visitation for each plot. We performed simultaneous observations by two ecologists at each field plot to count flower visitation for 15 min. Visitations to mango and coffee flowers were determined on five randomly selected branches at each plant (Carvalheiro et al., 2010), and counting of flower visitation was assisted by using recorded videos and photographs. Since the flowers of mango and coffee are clustered, the recorded video facilitates counting flower visits by watching the pollinator movement in between-flowers per inflorescence (i.e., bumble bees) and pollinator swapped between flowers (i.e., honey bees, wild bees, and hoverflies). The mean crop flower visitation in each cluster of flowers (inflorescence branch) was divided by the total number of flowers, then multiplied by the average number of flowers estimated in

Modelling pollination in church forest landscapes
We constructed a generalized additive mixed models (GAMMs) function in the package 'mgcv' (Wood, 2017) using R version 4.0.2 (R Core Team, 2020) to assess the response of crop flower visitation rate as a function of distance gradients from the 15 church forest habitat patches. The crop flower visitation rate was fitted with the fixed effect explanatory variables (Table 1). The fixed effect variables of church forest patches include: forest patch size, forest floral functional diversity index, and forest age. The variables of pollinator-dependent crop fields include: distance from nearest church forest habitat and crop types. All explanatory variables were included in our model and the transect was considered as a random effect. We used the random factor 'transect' into the explanatory variables to avoid pseudo replication.
GAMMs can uncover systematic patterns of variation in crop flower visitation along with the distance from remnant forest patches. We chose GAMMs instead of generalized linear models because we had no a priori assumption that the relationship between visitation and variables would be linear. Our visitation rate observations along each transect are considered as a dependent to assess the service distance decay. By specifying the predictors inside the model, GAMMs can also allow the form of the smooths to change by distance lags along each transect. We used the cubic regression spline function denoted by 'cr' as smoother for each explanatory variable in the model. Cubic regression splines have a cubic spline basis defined by a modest-sized set of knots as k = 4 (distance lags) spread evenly through the variables. They are penalized by the conventional integrated square second derivative cubic spline penalty (Wood, 2017). We included the crop types as a covariate for our model due to different crop flower attractiveness to wild pollinators. Explanatory variables such as forest patch area and forest age were log 10 -transformed to avoid scale effects and meet the assumptions of normal distribution and homogeneity of variances in the analysis. This is to deal with skewed data, scale and measurement unit differences among explanatory variables.
A 10-fold cross-validation expressed as generalized cross-validation (GCV) score was used for evaluating the GAMMs, using the crop flower visitation observation from original data. This was repeated such that each observation in the sample was used once as the validation data to investigate the model stability and predictability. Each visitation observation was used in the model development. The best-fitted GAMMs for the crop flower visitation rate was chosen based on the lowest GCV value.

Spatial characteristics of church forests and pollinator crops
The forest patches and surrounding pollinator crops are characterized below.
Church forests: A total of 1058 patches of church forests were extracted from the land cover map based on church building shapefiles in the entire study area, covering a total area of 10,600 ha. Church forest patches cover about 1% of the study area ranged from 0.5 to 234.5 ha with a mean (±SD) patch size of 28.5 (±55.4) ha. Small church forest patches were the most abundant type of remnant habitat in the study area. Church forest patches were separated from one another by an average distance of 1.75 km ± 0.97 SD. The density of church forest patches distributed over the southern and eastern areas of the Lake Tana basin and had the highest number of patches, largest patch size, and shortest inter-patch distance. There were relatively fewer and smaller forest patches in the north-western part of the study area. Small forest patches were more isolated than larger patches.
Pollinator crops: Based on the land cover map, 986,000 ha of the Lake Tana basin is under crop production. The total aggregated cropland area within the 1500 m pollinator foraging buffer zones of all church forests was 702,800 ha (76.4%), of which 196,800 ha (28.0%) include croplands that could benefit from wild pollinators (>0% pollinatordependent crops). The area of croplands calculated around the sampling church forest patches at different buffer distances is presented in Fig. 4 and Table 2. The forest proximity index to target crop fields was calculated considering neighbouring forest patches (1500 m radii) varied from 0.96 to 9.21.

Church forest functional richness index
A total of 4590 individual floral trees belonging to 121 tree species and 61 families were recorded along the 18 transects in 15 church forests. 67% of the tree species surveyed in church forests provide floral resources and nesting habitat for wild pollinators outside of the crop flowering period when floral resources are limited. The five most species-rich families of floral resources were Fabaceae (16 species), Rubiaceae (12), Euphorbiaceae (6), Rutaceae (4), and Oleaceae (4). The sacred forest species composition is old-growth, with Juniperus procera, Olea europaea, Celtis Africana, Prunus Africana, Maytenus arbutifolia, and Combratum mollie being the predominant pollen source vegetation of the study area. We identified a strong seasonal variation of tree flowering periods. The flowering periods of floral tree species varied from three to twelve months (Fig. 5). For all floral species, flowering began at the end of the rainy season (October) and peaked in the dry season (from November to May). In the dry season, these floral resources are an  (14) were revealed to be sources of high quantities of nectar or pollen (supplementary information 2). The remaining floral species were medium nectar or pollen sources for wild pollinators.
As a result, the mean forest floral functional richness index in the 15 church forests was 4.61 (range: 0.49-11.31). Forest floral functional richness differs among church forest patch size and age for all forest patches (Table 3).

Table 2
Key characteristics of the 15 church forests surveyed in this study including forest patch area and area of pollinator-dependent croplands at four different radii per transect around each church forest patch. The ID numbers are indicated on the maps in Fig. 3.

Modelling pollination service distance decay in afromontane church forests
We recorded 2872 pollinator flower visits (38.5% honey bees, 36% solitary bees, 10% bumble bees, and 15% hoverflies) during the peak flowering of each crop species under study. Bumble bees followed by hoverflies highly visited mango flowers. In this study, the mean visitation rate was 0.46 ± 0.024 pollinator visits per crop flower head per 15minutes observation period. The pollinator visitation rate depended on the crop type ( Fig. 6; supplementary information 1 appendix C). Fruit crops had higher wild pollinator visitation rates than pulses.
We investigated the effects of forest patch size, forest age, functional richness index, crop field distance from nearest forest patch, crop type and forest proximity on the crop flower visitation rate along distance gradients from sacred church forest habitats in Lake Tana basin, Ethiopia. The GAMMs results (Table 4; Fig. 7) revealed that the effects of church forest habitats on pollination ecosystem service decay with distance (i.e., greater crop flower visitation rate occur at small distances to church forests). Table 4 describes two components, the parametric (unsmoothed) coefficients and smoothing terms. The effective degrees of freedom (edf), states how 'wiggly' the fitted line is. For an edf of 1, the explanatory variables were estimated to have a linear relationship to the response, and higher values indicate the more complex fit. Flower visitation rates were significantly different along distance gradients in different crop-type fields due to consistent differences in wild pollinators' preference for different crop flowers. Mango flowers had the highest floral visitation rate (p < 0.001) followed by coffee flowers (p < 0.001) ( Table 4). Horse bean (p < 0.001) and field pea (p = 0.001) crops had comparable pollinator visitation rates along distance gradients from church forest patches. There is clear evidence of visitation rate variation on different crop types along the same distance gradient from church forest habitats since wild pollinators had different forage preferences (Fig. 7a). Among the four crops studied, coffee and mango fields attracted more wild insect pollinators than other crops.
The generalized cross-validation (GCV) scores of the models were taken as an estimate of the mean square prediction error based on a 10fold cross-validation technique. We performed an ANOVA and found that the model fitted with a lowest GCV value of 0.02 had an R 2 of 0.88. Table 4 reveals that the size of the church forest patches and the distance from forest patches had a significant influence on crop pollination service to adjacent crop fields. Crop flower visitation rate differed significantly across church forest patch size gradients (p = 0.025).
Relatively larger church forest patches had higher pollinator populations that provide a significant contribution to crop pollination. The church forest patch area and age significantly correlated with floral species' functional richness and diversity. Flower visitation rates on surrounding crop fields are positively related to church forests' floral functional diversity index. Church forest age was positively associated with the crop flower visitation rate. Among the explanatory variables, forest age had a large effect on crop flower visitation rate (p = 0.015) along distance gradients from forest habitats. The forest proximity index was related to wild pollinators' home range overlap in surrounding church forest patches (Fig. 7c). Flower visitation rates on crop field sampling plots differed significantly with different values of forest proximity index to focal crop fields within the 1500 m threshold distance (p < 0.001). The relationships between crop flower rate and   Table 4, does not allow for interpreting the shape of the regression line. Therefore, visualization is important to interpret our nonlinear fitted regression models. In Fig. 7 we see that pollinator-dependent crop fields found in-between lots of forest patches nearby may benefit from pollinators nested in those proximal church forest habitats. This reflects the probability of pollination service connection provided by proximal church forest habitats at a given distance. We observed that the pollination service decline over greater distances was mitigated by the contribution of central place foragers in different church forest patches within their foraging distance. Crop fields located close to forest patches within a threshold distance of 1500 m benefited from greater visitation rates.

Discussion
A growing body of literature has demonstrated the contributions of natural habitats for harbouring wild pollinators to provide pollination services to adjacent agricultural farms (Garibaldi et al., 2011;Potts et al., 2010;Tibesigwa et al., 2019). Simultaneously, as the human footprint expands, the amount of natural forest, free of substantial human activities, is rapidly declining, which degrades many forestdependent ecosystem services (Potts et al., 2010;Watson et al., 2018). The focus on the culturally important sacred forests in this study offered new empirical insights into the role of pollination services to nearby smallholder crop fields in space and time, at scales important to land management decisions.
Our research adds to previous studies (Bodin et al., 2006;Fahrig et al., 2011;Ricketts, 2004) by examining a large number of landscape characteristics combined with biodiversity data. We measured crop flower visitation rates across the landscape and built GAMMs to estimate the effects of distance from forest patches, patch size, patch proximity, patch age, and patch functional diversity on this pollination service. We observed the strongest effects of distance from forest patches and forest patch variables on crop visitation rates.
Our model assumed that sacred forests are the primary islands for wild pollinators in degraded and intensively cultivated landscapes. GAMMs can show the pollination service decay functions in complex structures of ecological data (Polansky and Robbins, 2013;Wood, 2017). In our study, observations of the visitation rates were not considered independent since the values along each transect depend on distance lags from forest patches. There is an implicit assumption that the dependency structure can be accounted for by nested records that considered a unique combination of visitation rates on the four knots (distance lags) from the complete set of observations. The estimated degrees of freedom for each explanatory variable in our model result is greater than zero, suggesting that the variables enhance model fit.
Church forests exist as 'islands' within agricultural fields, and due to their small size and influences from neighbouring land uses, functional isolation (also known as patch proximity) may be especially essential. Our study examined the contributions of the sacred forests, the last natural ecosystems in degraded landscapes of our study region, in maintaining the wild pollinator population. These remnant forests are isolated in the landscape. Therefore, we treated them as the only habitat location for wild pollinators, from which they disperse. This conceptualization is comparable to the island biogeography theory (MacArthur and Wilson, 1967), which emphasizes the distance from the dispersal target habitats and island variables.
As expected, the overall flower visitation rate on neighbouring crop fields declined with distance from forest patches, a pattern consistent with previous studies (Aizen et al., 2009;Garibaldi et al., 2011;Klein et al., 2007;Ricketts et al., 2008). Our results also show that for some crop fields located at 1500 m from a forest patch, a slight increase in visitation is observed. We expect this was due to pollinators living in other nearby forest patches. Crop visitation rates were also positively linked with church forest size, as large forests have relatively higher floral tree species abundance for supporting wild pollinators. Finally, visitation rates vary among crop types because not all insect pollinator groups are similarly attracted by crop flowers . For example, among the crops studied, mango flowers were the most visited highest pollinator dependence (Carvalheiro et al., 2010).
Forest functional diversity was also a significant predictor of crop visitation rates, highlighting the importance of big, mature floral trees with extended flowering periods for pollinator persistence. Field surveys by Lowman (2011) support the findings that larger floral trees in church forests are used as a foraging resource and provide critical habitat for pollinators. Older forests are likely composed of assemblages of flowering tree species that preserve pollinators that have long since disappearing from the surrounding area. It is supposed that old-aged forest patches have higher vegetation structures and higher levels of biodiversity (Martin et al., 2013;Santos et al., 2008). Among all variables, the weaker effect of forest age on visitation rate may have occurred because most of the sacred forests in our study are old-aged to support wild pollinators . It might be because pollination services peaked in forest areas older than 100 years, with the exception of varying capacities to offer other ecosystem services.
Floral abundance and richness and their peak blooming periods are likely to influence pollinators' habitat use throughout the year. Among the afromontane forest characteristics, floral trees bloom during the dry season (from September to June). Therefore, during the crop growing summer season (July to August), the main foraging resources nearby are the mass-flowering crop fields. As a result, pollinators from forest habitats are likely to spillover to crop fields in large numbers. Previous research has investigated the importance of forests with more tree species for multiple ecosystem services (Gamfeldt et al., 2013;Hector and Bagchi, 2007). Our study, which takes into consideration essential floral tree traits, reveals continuous positive relationships between floral tree functional diversity index and pollination services. It also emphasizes the need of maintaining a floral tree diversity in order to provide a future potential for pollination service supply. As a result, our study expands on previous research (Garibaldi et al., 2013;Klein et al., 2007;Kremen et al., 2007), none of which takes into account forest functional diversity. Our study revealed high forest functional diversity may help to offset pollinator population declines throughout the dry season and, as a result, contribute to pollination services delivery during the summer. Our results strongly support many floral species and longer flowering seasons inside church forests provide continuous floral resources over a more extended period. This happens due to the temporal mismatch of the flowering period of floral trees and mass-flowering crops.
The pollination service impacts human wellbeing in the area. Survey data of 2019 to 2020 for the Central Statistical Agency of Ethiopia shows that on average, smallholder farmers grew pollinator-dependent crops on 22.4 percent of the total cultivated area in the study area (CSA, 2020). The mean estimates the annual economic value for the four studies crops in Lake Tana basin was about US$160.5 million for the year 2019 and 2020. Crop vulnerability for these four crops is estimated to range between 10% and 40% of lost yield in the absence of pollinators (Aizen et al., 2009;Klein et al., 2007).
Biodiversity hotspots located in natural ecosystems are found within human-modified landscapes all around the world (Jacobson et al., 2019;Sloan et al., 2014). Lessons learned from our study might be applied to areas where forest fragments remain intact. As a result, the findings of this study have implications for a wide range of remaining forest patches. Details are likely to vary by location; for example, the drier study area may differ from the other moist parts of southern Ethiopia, where wild pollinators are only restricted to natural habitats. In practice, our findings suggest that sacred forests are beneficial for ecosystem services that support smallholder livelihoods. While other degraded landscapes are affected by the loss of ecological function and resilience (Fischer et al., 2021;Karp et al., 2012;Tscharntke et al., 2008), loss of crop diversity (Khoury et al., 2014), and increasing disconnection of rural people from nature (Ives et al., 2017). In addition, quantitative evidence on pollination services supplied by wild pollinators outside of their native forest habitat can help support the sustainable use, conservation and restoration of remaining natural forest patches. Despite efforts, many of the causes of pollinator decline have remained unchanged over space and time: land-use change, habitat fragmentation, increased wood extraction, livestock grazing and invasive alien species, along with climate change, could have an impact on forest biodiversity and disrupt ecosystem processes and crop pollination services (Cardelús et al., 2019;Scull et al., 2017). Restoration efforts could include connecting the remnant forests with vegetation corridors creating vegetation and stonewall buffers around the patches, and replanting deforested areas, and establishing new patches in the landscape (Aerts et al., 2016;del Río-Mena et al., 2021;Woods et al., 2017).

Conclusions
Sacred natural forests are an important part of degraded landscapes that provide various ecosystem services, including crop pollination to rural people in the Lake Tana basin. These forest habitats provide foraging and nesting resources for wild pollinators throughout the year. Wild pollinators that reside in their natural habitats go to neighbouring crop fields to pollinate them. In our study, we found that: 1) crop flower visitation rates in smallholder crop fields decline at varying distances from adjacent forest remnants across an agricultural landscape; 2) distance from forest patch, patch size, patch isolation, patch functional diversity and patch age were significantly correlated with crop pollination services; 3) sacred forests (regardless of size, age, or functional variety) can influence flower visitation rates in surrounding crop fields; however, this effect is mitigated by forest patch proximity across the landscape; 4) flower visitation rates on different crops vary according to their pollinator dependency and distance from forest patches in space and time. Different wild pollinator groups (i.e., wild bees, bumble bees, solitary bees, and hoverflies) may respond differently to different crop types. As per our findings, re-emphasizing the conservation and management of remaining natural forests will improve our ability to optimize pollination services. This can help improve the nutrition, food security and income of rural farmers while simultaneously achieving biodiversity and ecosystem conservation of the SDGs.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.