Natural and semi-natural habitats provide important resources for many plant and animal species. The destruction of such habitats might reduce biodiversity and disturb ecosystem functions. Sub-Saharan Africa suffers particularly under the destruction and deterioration of ecosystems due to demographic pressure. In this study, we analysed ecosystem functions in a small remnant of East African coastal forest and the surrounding anthropogenic landscape. We measured pollination activity, predation, and seed dispersal. For each study plot, we also recorded local habitat conditions, which may also influence ecosystem functions. We found no significant difference between the natural forest and anthropogenic agro-environments for seed dispersal nor pollination. Insect predation showed highest rates inside the forest and decreasing rates in the open agro-environments. Local environmental conditions strongly affected ecosystem functions. For example, flower availability showed positive effects on pollination activity, and the availability of herbs on the ground positively influenced the level of predation. Rather homogenous ecosystems such as eucalyptus plantations and arable land showed lowest levels of ecosystem functions. Here, planting of undergrowth such as herbs and flowers may improve ecosystem functions. Our findings underline that natural forest as well as heterogeneous agro-environments provide a large variety of ecosystem functions, which strongly depend on site-specific microhabitat conditions.

Our study area is located in southern Kenya, 20km north of Mombasa, at an elevation of about 40m a.s.l. The climate of this region is hot and humid year-round (with mean temperatures of 25.5 °C and average rainfall 1100 mm/a). The climate is characterised by strong seasonality, with two rainy seasons lasting approximately from end-March to July (average seasonal rainfall 631 mm) and from October to December (average rainfall 287 mm) (Jaetzold et al., 2010). The region shows medium to weak agricultural potential as characterised by the agro-ecological zonation of Coastal Lowlands with a medium to long cropping season, intermediate rains and a weak short second rainy season. The soils consist mainly of luvisols. The beginning and end of the rains vary strongly among years and especially the second rainy season is unreliable (Jaetzold et al., 2010).  The landscape consists of a mosaic of different ecosystems, including natural- and semi-natural and anthropogenic ecosystems (Appendix A: Fig. A1a). In the centre of our study area there is the sacred Kaya Kambe forest, a small 75ha remnant of East African coastal forest (Fungomeli et al. 2020, Habel et al. 2023). Such Kaya forests (i.e. Mijikenda sacred forests) are found along the Indian Ocean coastline of southern Kenya (Robertson & Luke 1993, Githitho 2003, Matiku 2005, Shepeard-Walwyn 2014, Fungomeli et al. 2020), and represent small and geographically isolated forest patches (mean size 120.4ha, 75% smaller than 150 ha) (see Nyamweru et al. 2008, Shepeard-Walwyn 2014, Fungomeli et al. 2020). Despite its small habitat size and geographic isolation, these forest patches still provide suitable habitats for a large variety of typical and rare forest species (Luke & Githitho 2003, Wijtten et al. 2011, Habel et al. 2023). Human activities such as the removal of deadwood, selective tree logging, hunting and grazing of cows and goats have led to severe disturbances and subsequently to reduced vegetation regeneration with negative effects on biodiversity (Kibet 2011).** **Management and the protection of Kaya Kambe forest is controlled by the elders of the local community (Habel et al. 2023). Shifts in cultural beliefs and an increasing demand for land and forest products poses a severe threat to Kaya Kambe forest (Luke and Githitho 2003). The other habitat types are settlement area (Kambe village), orchards (with mango and banana trees mainly), arable land (mainly maize, sorghum, cowpeas), eucalyptus tree plantation, and shrubland (Luke and Githitho 2003). The study area is confined by a lead mine in the north and a road in the south.

Study set-up

We applied the Rapid Ecosystem Functions Assessment according to Meyer et al (2017). This low-tech method is easy to use and allows standardized collection of data on ecosystem functions (Meyer et al. 2015). We assessed the following proxies, all relevant to conduct successful agricultural farming: pollination activity, predation, and seed dispersal. These proxies are of high relevance for the production of food crops. Predation and thus regulation of pests may increase yields in agro-environments (Meyer et al. 2015). Pollination activity is a key ecosystem service for the production of seeds and thus yield stability (Isaacs et al. 2017, Bishop et al. 2022). Seed dispersal is another mutualistic interaction pivotal for effective regeneration of ecosystems due to animal-mediated seed dispersal (Meyer et al. 2015).

We measured these three proxies of ecosystem functions in natural habitats (the forest interior, forest edge), semi-natural habitats (orchards, shrubs), and anthropogenic habitats (plantations, agricultural fields, and settlements). Data collection was performed from 26th of February to 15th of March on 10x10m study plots, which were set in all habitat types. These study plots were arranged along habitat gradients from the forest interior and the forest edge (200m inside the forest, at the forest margin 8m from forest edge, and at the forest edge) into the semi-natural and anthropogenic habitats (8m distant from forest edge 16m, 32m, 64m and 128m distant from forest edge). It was not possible to have all transects at equal distance, but distances were in all case sufficient to minimize spatial autocorrelation. In part of the analyses we subsume the plots near the forest edge (-8m, 0m 8m) into a forest margin category. In total, ecosystem functions were measured along 17 such gradients (Fig. A1a). We added further study plots in a eucalyptus plantation (Fig. A1a).

Pollination activity was measured using coloured pan trap (white, blue, and yellow), all three mounted on a wooden stick about one meter above ground (Fig. A1b). These colours were chosen, as they are known to have highest sampling efficiency across a wide range of pollinators (Habel and Ulrich 2021). Pan traps were filled with saturated salt solution and a drop of detergent to reduce the surface tension and activated during the morning for 8 hours. Afterward, the collected material was taken from the field. We pooled counts of Hymenoptera, Diptera, Coleoptera, and Lepidoptera to estimate pollinator abundances. For measuring predation, we used green artificial caterpillars made out of plasticine. These 4cm long caterpillars were pinned on earth-coloured paper at the ground and were exposed in the field for 24h to measure predator activity during the day and night. In total, we established 10 artificial caterpillars per study plot. We considered bite marks on the dummies from arthropods, birds, small mammals, and slugs (see Meyer et al. 2017). We did not record bite mark frequency because one predator could cause multiple bite marks on the same dummy. Dummies, which vanished or got damaged by humans were not evaluated. Seed dispersal was measured by placing three seed plates (wooden plates with 25 small pits) at each plot and counted the removed seeds after 24 hours (Fig. A1b). We used peeled sunflower seeds. We did not consider the taxa that removed the seeds. All raw data are contained in Appendix B.

Environmental characteristics

We recorded the following habitat characteristics: Herb-, shrub-, tree-, canopy- and litter coverage (categorized into 0-25%, 25-50%, 50-75% and 75-100%). Furthermore, we assessed the average litter height (in cm) and the presence of dead wood (no, little, medium, much), as well as nectar sources (0, <25, <50, <75, and <100 blossoms). For comparability, these percentage and abundances classes were coded as 0, 1, 2, 3. Raw data are provided in Appendix B.