2.1 Study area
The study was conducted in Nyankpala in the Tolon District of the Northern Region of Ghana. The area falls within the northern savanna ecological zone and has a unimodal rainfall pattern which occurs between April and October annually (GSS 2014).
Our seed removal experiment was performed in three different habitat types separated by a minimum distance of approximately 700 m: i.e., conventional farmland, parkland and woodland (Table 1). Ideally, conventional farmlands should contrast parklands as the latter seeks to promote ecological balance and the conservation of biodiversity by avoiding or limiting the use of agrochemicals as well as intensive tillage among other practices (Boffa 1999; Pantera et al. 2021; Sumberg and Giller 2022). However, in the study area, agrochemicals and other resources are usually applied similarly in both agriculture systems with the only notable difference being the presence or absence of woody plants. Even the abundance of woody plants in parklands are rapidly declining in the area due to unsustainable practices: e.g., intensive farming and fuelwood harvesting (Poudyal 2011). Studied cultivated lands consisted of smallholder farms that are owned by local community members while the woodland habitat, which is a secondary woodland, is owned and managed by the Savanna Agricultural Research Institute, Nyankpala.
Table 1
Description of study habitats included in the investigation. Estimates of total woody plant components are based on three 25 m × 25 m quadrats established in each habitat type.
Habitat | Description | Richness | Abundance |
Conventional farmland | Crop production on lands devoid of woody plants | 0 | 0 |
Agroforestry parkland | Crop production under scattered woody plants | 2 | 6 |
Woodland | An area of land dominated by woody plants with few grass or herb species | 5 | 109 |
In this study, tillage, planting, and application of agrochemicals (i.e., herbicides) on cultivated lands occurred earlier on the parkland compared to the conventional farmland at the time of field investigations (at least a weeks’ difference). Crops such as cowpea (Vigna unguiculata L.), maize (Zea mays L.), and pepper (Capsicum spp.) were commonly cultivated on the surveyed conventional farmland and parkland habitats. Woody plant abundance and richness was highest in the woodland habitat (i.e., Combretum sp., Detarium sp., Lannea sp., Terminalia sp., Vitellaria paradoxa) followed by the parkland habitat (i.e., Mangifera indica, V. paradoxa) (Table 1). Although woody plants were absent from the conventional farmland and its immediate periphery, it was surrounded by bushes and fallow lands.
2.2 Study species
The seeds of Faidherbia albida and Leucaena leucocephala species were used in this study because they are commonly associated with agroforestry practices in the northern savanna ecological zone of Ghana. F. albida seeds are approximately 6–8 mm long and 4–5 mm wide while L. leucocephala seeds are approximately 6–10 mm long and 4–6 mm wide (Sniezko and Stewart 1989; Lim 2012). Both seed species can be dispersed by insects, rodents, and wind due to their relatively small size (Yirgu and Tsega 2015; Sharma et al. 2022). Matured non-burst seeds of these species were collected from the study area in January 2022 and stored to prevent insect damage before field experiments.
2.3 Post-dispersal seed removal experiment
We used an experimental setup designed to either exclude or allow access to seeds by different taxa in the various habitat sites. To allow access to seeds by only invertebrates, we used a cylindrical metallic cage with a mesh size of 1.2 cm × 1.2 cm and referred to it as ‘closed treatment’ (cage dimension: height = 15 cm and diameter = 21 cm). However, we did not use an exclosure cage for the ‘open treatment’ because we wanted all seeds to be available to all seed predators (i.e., invertebrates and vertebrates, specifically rodents and birds).
In each habitat type, we established three transects that were separated by a minimum distance of 100 m for the deployment of seed removal treatments. Each transect had five seed stations that were 10 to 20 m apart with each station consisting of a pair of open and closed treatments that were between 1 and 3 m apart. For each treatment, ten seeds of each seed species were placed on a plastic plate for the removal experiment. All seed stations were monitored daily for seed removal data in the rainy season (period of the experiment: June 09–23, 2022). The experiment was repeated after days of rainfall because rain could lead to false seed removal, which is not attributable to fauna. Hence, complete seed removal data for the latter period of this investigation (June 16–23) was extracted and used for all analyses due to the absence of rainfall.
2.4 Statistical analysis
All statistical analyses were conducted in R programming software (R Core Team 2022). A generalized linear mixed effects model (GLMM) with a beta error distribution was used to test the effect of the following predictor variables on post-dispersal seed removal rate: habitat type, seed species and exclosure treatment. The dependent variable, seed removal rate (R), was defined as the proportion of seeds removed per species after 48 h using the equation \(R={N}_{removed}/{N}_{offered}\) (Meyer et al. 2017): i.e., where Nremoved is the number of seeds removed after 48 h and Noffered represents the total number of seeds per species placed at each seed station, which was 10. Additionally, we applied the transformation \(\left(R*\left(n-1\right)+0.5\right)/n\) where n is the sample size because R assumed extreme values of 0 and 1 (Smithson and Verkuilen 2006; Cribari-Neto and Zeileis 2010). Transects were included as a random effect to eliminate the effect of spatial pseudo-replication in the model. Due to the significance of interaction terms (i.e., habitat × Seed species) in the main model, we fitted separate models for each species to determine the effect of habitat and exclosure on post-dispersal seed removal (e.g., Rocha-Ortega et al., 2017). All GLMMs were performed using the package glmmTMB (Brooks et al. 2017).