Study Site
The study acoustically surveyed two fragments of Transkei coastal scarp forest (Defined in Mucina and Rutherford 2006) and six former crop fields near the mouth of the Qora River, south-east of the town of Willowvale (32.26 °S; 28.50 °E) in the communal land tenure Transkei are of the Eastern Cape Province, South Africa (Fig. 1). Two large intact forests, Manubi, and Dwesa, surveyed by Moir et al. (2020a) were compared against our sites. The area lies within the Maputaland-Pondoland-Albany biodiversity hotspot, a centre of high endemism and species diversity, which is a priority area identified by Strassburg et al. (2020) in terms of carbon sequestration and reversing the extinction debt. The greater Willowvale region is characterized by a mosaic of Transkei coastal scarp forest, thornveld, dune thicket, and grassland landscapes (von Maltitz et al., 2003; Mucina and Rutherford, 2006; Berliner, 2011). Here, despite 91% of households falling below the poverty line, field cover declined from 12.5% in 1961 to 2.7% in 2009, with a peak in field abandonment during the transition to democracy in the early 1990s (Shackleton et al., 2013).
Figure 1: Study site locations within the Willowvale area, Eastern Cape, South Africa during 1942 and 2022, pre- and post- agricultural cropland farming respectively. 1960s fields: sites 2 & 3, 1970s fields: sites 4 & 5, 1980 sites: 6 & 7, and forest fragment sites: 8 & 9.
Njwaxu and Shackleton (2019) had previously identified and described 43 former fields and seven forest fragments in the broader Willowvale area. Each former field was assigned a decade in which it was abandoned and plots of 10 m x 20 m were set in the middle of the former field to record plant species (Njwaxu and Shackleton, 2019). We selected six of these sites for our study, two each abandoned during the 1960s, 1970s, and 1980s respectively for bat recording. The fields are located within a matrix of dispersed rural villages and homesteads, sand roads, and forest fragments (Shackleton et al., 2013; Njwaxu and Shackleton, 2019). While abandoned in terms of cropping these fields remain utilised by the local community for informal harvesting of non-timber forest products and cattle grazing (Njwaxu and Shackleton 2019). In addition, two fragments of Transkei Scarp forest, embedded within the agricultural landscape, were monitored for comparison (Figs. 2: G and H).
Figure 2: General habitat structure of eight sites for insectivorous bat passive acoustic monitoring. Former fields: A, B, (abandoned in 1960s); C, D, (abandoned in 1970s); E, F, (abandoned in 1980s); G, H, (Forest fragments).
Permitting and Ethical clearance
Sampling was conducted under a permit issued by the Economic Development Environmental Affairs and Tourism Department of the Eastern Cape Province (HO/RSH/11/2023). Ethical clearance was obtained through the Stellenbosch University Animal Care and Use Committee (ACU-2023028082). Landowner permission was obtained from the local Chief (LOA202309051) and headman (LOA20230905 2).
Survey methods
Four Wildlife Acoustics Song Meter SM4 (SM4BAT) and four Song Meter SM2+ (SM2BAT) full spectrum ultrasonic monitors were deployed across eight sites (two sites per treatment and two forest fragments), over two separate sampling periods in 2023 (Online Resource 1: Table S1); one in the austral autumn (March 24 — April 3) where sites were surveyed for seven nights (56 detector nights) and the second in the austral spring (5 — 10 September) where sites were surveyed for five nights (40 detector nights). Spring sampling was restricted to five nights due to inclement weather conditions. Six to seven nights (36 — 42 detector nights) is deemed sufficient to reach bat species sampling saturation for this region (Moir et al., 2020a). Sampling was conducted under minimal lunar illumination (Appel et al., 2016) and acoustic sampling nights where precipitation exceeded 3 mm were discarded during analysis (following Fischer et al., 2009), as bats reduce their activity during periods of heavy rain (Appel et al., 2019).
To account for potential microphone sensitivity discrepancies, owing to using two different bat detector models, monitors were switched to the alternate model on the fourth and third night for autumn and spring, respectively. Furthermore, microphone sensitivity (SM4BAT vs SM2BAT) was assessed with no significant effects detectable of the two different microphones used. Each monitor was set to trigger at a minimum of 12 kHz, with 384 kHz sample rate and 12dB gain in full spectrum (see Online Resource 1: Table S2). The monitors were scheduled to commence recording a half hour prior to sunset and cease a half hour post sunrise (approximately 14 recording hours), following Moir et al. (2020a) and as stipulated by Kunz and Parsons (2009) to coincide with peak insect activity (Fenton, 1980). Microphones were positioned towards flyways and under as little vegetation overhang as possible to mitigate call attenuation and capturing call echoes (Weller and Zabel, 2002) and placed 1.5 — 2 m above the ground. Sampling sites, where possible, were separated by a minimum distance of 250 — 450 m to mitigate simultaneously capturing calls from the same individual on separate detectors (Moir et al., 2020a; Law et al., 2015). However, as ultrasonic microphones have a range of roughly 30 m, the likelihood of capturing the same bat simultaneously is improbable (Law et al., 2015).
Fire disturbance and the invasive shrub Lantana camara were noted as present or absent within each site during autumn and spring for subsequent analysis. Evidence of these two factors were simply recorded as present or not.
Acoustic Analyses
Acoustic data were processed in Wildlife Acoustics Kaleidoscope Pro software (version 4.5.5) and scrubbed for noise files. A call was defined as at least two recorded pulses within < 1 second of silence between (Fenton, 1970). Call files were run through the AutoID Southern African database (version 4.3.0) in Kaleidoscope for putative identifications. A total of 52,814 audio files was recorded over the autumn and spring sampling periods, of which 4,556 audio files were identified to species level. Unidentifiable calls (1,185) and noise files (47,073) were omitted from the statistical analyses. Each call file was manually identified to species level to mitigate AutoID misidentification owing to species overlapping call parameters (Taylor et al., 2013; Moir et al., 2020a). Spectrogram visualisation, zero crossing call parameters, and regional call libraries from the eastern regions of South Africa (Taylor et al., 2013; Moir et al., 2020a), were used to identify species. Key call parameters used for identification in this study were characteristic frequency (Fc), maximum (Fmax) and minimum frequency (Fmin), duration (Dur), and knee frequency (Fk). A conservative approach was taken with respect to overlapping Fc parameters between Tadarida aegyptiaca and Mops pumilus. Characteristic frequency calls that ranged between 18 — 22 kHz were assigned to T. aegyptiaca and while 23 — 27.7 kHz were assigned to M. pumilus, following the mean Fc stipulated in Monadjem et al. (2020). Additionally, Chaerephon ansorgei and Cnephaus hottentotus were removed as potential identifications based on their geographic distribution profiles (Monadjem et al., 2020) and lack of recording and active capture in the region (Moir et al., 2020a). To differentiate between Scotophilus dinganii and Laephotis botswanae, S. dinganii was assigned to calls with longer duration (3.5–4.6 ms) and greater knee frequency (36–37 kHz) while L. botswanae calls were of shorter duration (2–3 ms) and lower knee frequency (maximum 35 kHz) (Taylor et al., 2013; Moir et al., 2020a). Based on these assignments we recorded only Scotophilus dinganii at our research sites.
Activity was measured both as the total number of bat passes (henceforth referred to as ‘passes’) or call sequences recorded per night and the average number of passes recorded per hour per night. Species specific activity was standardised to Activity Index (AI) following Miller (2001) which utilises the presence of species per one time minute intervals. Nocturnal activity time graphs were constructed based on the number of passes per hour per night for autumn and spring separately, and for the pooled data.
Statistical Analysis
All statistical analyses were performed on corrected data (i.e., only confirmed call sequence identifications). Packages “nlme”, “vegan”, “MuMin”, “ggplot2” and “lme4” were used to perform statistical analyses in R software (version 4.3.1). EstimateS software (version 9.1.0) (Colwell, 2013) was used to generate Chao (1987) species accumulation curve estimates for standardised acoustic nights (pooled data = 22 sampling nights, 176 detector nights). Species accumulation curves were used to assess sampling saturation of species inventories within field categories during autumn and spring sampling (Moreno and Halfter, 2000). Diversity was quantified by species richness, Shannon-Weiner diversity and four functional diversity indices: functional richness (FRic), functional evenness (FEve), functional divergence (FDiv) and functional dispersion (Fdis). The functional diversity indices use quantitative values for traits, with species distributed in a multidimensional functional trait space (Villéger et al., 2008). FRic is the volume of functional space occupied by the community and calculated with the convex hull volume index (Villéger et al., 2008). FEve quantifies the regularity with which species abundances are distributed in a functional trait space (Mason et al., 2005). FDiv quantifies the divergence of species in their distances (weighted by abundance) from the centre of gravity in the functional space (Villéger et al., 2008). Lastly, FDis measures the mean distance of individual species to the centroid of all species in the multidimensional trait space, and is closely related to Rao’s quadratic entropy (Laliberte and Legendre, 2010).
In calculating functional diversity metrics, we followed procedures outlined by Moir et al. (2021). The functional traits used to plot species in functional space included: echolocation call type, characteristic frequency, forearm length, aspect ratio, and wing loading (Online Resource 1: Table S3); trait values for each species were taken from Moir et al (2020a) and mostly reflected measurements taken in the same province (Eastern Cape) of South Africa. Continuous traits were standardized to a mean of zero and standard deviation of one (Farneda et al., 2015). Relative abundance was characterised as mean passes per recording hour (Online Resource 1: Table S4). Mixed linear effect models using the package “glmmTMB” (Brooks et al., 2017) were used to investigate relationships between functional metrics (dependent) and site categories (predictor). Shannon-Weiner diversity and Peilou Evenness estimates were calculated using the package “vegan” (Oksanen et al., 2022) and the functional indices were calculated using the “FD” package (Laliberté and Legendre 2010) in R software.
Generalised Linear Models (GLM) and Generalised Least Squared (GLS) models were used to assess predictors of Species Richness (SR), Peilou Evenness, Shannon-Weiner diversity, and activity (total passes/night; average passes/hour/night). Depending on data normality either an ANOVA or Kruskal-Wallis test was performed to assess the significance of predictor variables. Predictor variables included distance to the nearest water body (m), distance to the nearest building (m), distance to the nearest road (m), distance to the nearest pristine forest fragment (m), and site aspect (Online Resource 1: Table S5 for values). Fire disturbance presence and acoustic detector model type (SM4 vs SM2+) were modelled for autumn and spring separately. The detector model was included to account for any microphone sensitivity differences during sampling. Lastly, logistic regression models were run to assess the significance of predictors on foraging guild presence and species presence. Three species were modelled, namely Pipistrellus hesperidus, Laephotis capensis, and Rhinolophus clivosus, based on their activity dominance, in the forest fragments, 1960s and 1980s sites, and 1970s sites, respectively.