The study area comprises the KNP (S: −24.98252, E: 31.58189; ~2 000 000 ha) and Associated Private Nature Reserves (APNR; ~36 300 ha). Situated in the lowveld region of north-eastern South Africa, it forms a continuous, cooperatively managed conservation area supporting a historically accurate assembly of wildlife (South African National Parks 2018). The study area is bordered by areas of varying land-use intensity and includes the provincial nature reserves of Manyeleti, Letaba Ranch/Mthimkhulu Complex and Makuya Nature Reserve, as well as the Limpopo National Park in Mozambique and Gonarezhou National Park in Zimbabwe.

The KNP forms part of the extensive savanna biome (Rutherford et al. 2006) and varies from open tree savanna to dense woodland, extensive thickets, bushveld and some limited forest areas, usually along the major rivers (Gertenbach 1983). It is underlain by two broad geological zones (basalt and granite) with topography primarily consisting of undulating plains of low to moderate relief. There are mountainous areas in the extreme northern and south-western portions, along the entire eastern boundary, and following the course of the Olifants River (Venter, Scholes & Eckhardt 2003). The average elevation is approximately 300 m above sea level (min: 200; max: 840 m) (Venter et al. 2003). Variation in rock types and soils, and the resulting topography and vegetation, have led to the formation of 11 land systems in the KNP and APNR (Venter 1990) and 35 distinct landscapes for the KNP (Gertenbach 1983), four of which extend into the APNR (Peel, Kruger & MacFadyen 2007). For the purposes of our study, the 10 broad habitat units delineated by Mills, Juritz and Zucchini (2001) for the core KNP area were used, with one additional habitat type, bushwillow/mopane woodland, encompassing the land of the APNR (Figure 1).

Rainfall is highly concentrated in the period October to March (Rutherford et al. 2006) and varies across the study area, from an average total precipitation of < 400 mm to > 730 mm per annum, with a general trend of decreasing rainfall towards the east and north (South African National Parks 1999). For analysis, the KNP was divided into three regions, mirroring current management regions: North (north of the Olifants River), Central (between the Olifants and Sabie Rivers) and South (between the Sabie and Crocodile Rivers).

Yellow-billed oxpecker sighting data were obtained for two discrete periods consisting of: (1) historical data from the initial period of the species’ recolonisation of the KNP (1979–1994) and (2) contemporary data collected during our study (2005–2022). Historical data included unpublished internal reports and maps housed in the SANParks Archive in Skukuza. During the pre-1994 period, YBO sightings were recorded by KNP conservation staff and included in internal annual reports (Jaarverslag) under the section on rare and vagrant bird species. Sighting data (date, location, host species and number of birds seen) were extracted from these reports and mapped using inferred Global Positioning System (GPS) positions (accuracy: < 5 km) based on the detailed description of the location or the hand-drawn map provided with the record. Sightings were plotted onto a KNP map created in QGIS 3.4 LTR (QGIS Development Team 2018) using available KNP and South African shapefiles. This information was supplemented with and verified against three published articles (Hall-Martin 1986, 1987; Whyte et al. 1987) detailing the first 8 years of the recolonisation. Additionally, from 1986 to 1994, Dr Ian Whyte recorded buffalo herds with YBOs present, while undertaking the annual aerial buffalo census throughout the KNP (Whyte pers. comm.). The hard-copy census maps and datasheets from 1986 to 1994 were retrieved from the SANParks Archive, and the YBO locations were digitised in QGIS, along with the additional information provided by the original observer (year, number of birds seen, buffalo herd size and any additional comments).

Contemporary sighting data were obtained via direct field observations and through email and smartphone application submissions from citizen scientists. Information posters requesting the assistance of park visitors were placed at tourism locations throughout the KNP and on a dedicated Facebook (Meta Platforms, Inc. Menlo Park, California, United States) site. An official partnership with the BirdLasser^® Mobile Application owner Lejint (Pty) Ltd, Johannesburg, Gauteng, South Africa provided access to records logged by application users. Early records in the database (2014–2017) were made available, and from 2018, users were prompted to record additional details for their observations of YBOs through additional data fields in the species listing on the application. Additional YBO records from South Africa, for the period 2005–2022, were obtained from the online database of the Global Biodiversity Information Facility (GBIF) (GBIF.org 2022).

Yellow-billed oxpecker sighting data included date, GPS location, number of birds present, nesting status and, where possible, host species. Some direct submissions by park visitors only contained a written description of the location, which was then used to infer the GPS position (accuracy <2 km). Data were captured in an electronic database, with cross-checking between sources and elimination of duplicate or unreliable records. Sightings submitted by citizen scientists may contain errors, especially regarding the correct identification of the study species and the accuracy of the sighting location. Therefore, each direct submission was verified, using photographic evidence to confirm species identification. Sighting data from the direct submissions and the BirdLasser dataset were plotted using QGIS, and all sightings located > 0.5 km away from the KNP tourism infrastructure (e.g. restcamp boundary, tourist road network) were individually verified with the observer. Any unlikely or unconfirmed records were discarded. In addition, sightings located >10 km from the study area boundary were discarded, while those < 10 km from the study area were verified by overlaying sightings onto a satellite image (Google Earth Imagery^© 2022 TerreMetrics, Map Data^© 2022 AfriGIS [Pty] Ltd) imported into QGIS. Sightings logged at a highly unlikely location (e.g. in orchards or heavily built-up areas) were excluded. Where sightings were duplicated in citizen science and the BirdLasser records, only the latter records were retained.

The Mills Habitat Types (Mills et al. 2001) utilised for our study are based on a combination of the dominant plant species, vegetation structure, geology and topography, but do not include minor localised habitat differences, such as riparian zones, which are likely to introduce unnecessary complexity to the analysis of YBO habitat utilisation. Riparian zones constitute ~3.5% of the study area, and YBO are only likely to be observed there when accompanying large mammal hosts, which will be transient in this localised habitat as they access shade and water. The Mills Habitat Types classification does not include the adjacent APNR or other land use types, and these were thus not used in the habitat analysis.

All statistical analyses were performed using R (Version 4.2.2) statistical software (R Core Team 2022). Oxpecker host and habitat use data are categorical in nature so the Pearson’s chi-square test of independence was used to investigate whether a relationship exists between YBO host species selection and the geographical region in the study area, as well as the breeding (November–February) or non-breeding (March–October) season. All sightings were deemed independent, and any likely duplicates in the dataset were removed before running the analysis.

Contingency tables were created in R and inspected to ensure compliance with the test prerequisites (see McHugh 2012). As the contingency tables used for the analysis were larger than 2 × 2, pairwise comparison tests were used to further investigate the relationship between groups and reduce the possibility of Type I errors in the analyses (MacDonald & Gardner 2000). Specifically, the calculation of adjusted (corrected) p-values and critical z-values based on Bonferroni corrections was made. Using a critical p-value of 0.05, the following formula was used to calculate an adjusted (corrected) p-value, based on the rows and columns of the contingency table (Sharpe 2015): pCorrected = pCritical / (N.Rows x N.Columns). The adjusted critical z-value was then obtained using the R Package ‘stats’; Function ‘qnorm’ (R Core Team 2022). For sample sizes greater than 30, the standard residuals approximately follow a standard normal distribution, and any standard residual falling outside the critical z-value range is considered significant (MacDonald & Gardner 2000; Sharpe 2015). Standardised (Pearson) residuals for each cell in the contingency table were calculated in R so that those contributing the most to the chi-square test statistic could be determined (Sharpe 2015).

For the habitat selection analysis, we used the Neu, Byers and Peek (1974) method to determine if the number of observed sighting records of YBOs within the various core KNP habitat types was different to the expected number of sightings (i.e. did YBOs utilise habitat types in proportion to each habitat’s availability within the study area relative to other available habitats?). The size of each habitat type was extracted from the Mills Habitat Classification shapefile (South African National Parks 2020), and the number of sightings per habitat type was counted. The chi-square goodness-of-fit test was used to determine if YBOs utilised available habitat types in proportion to the habitat’s availability. Bonferroni Confidence Intervals for each habitat type were then generated, and the expected proportion of utilisation was evaluated against the confidence interval to determine preference, avoidance or expected use for each habitat. As habitat choice may be influenced by the breeding status of a species, the chi-square test and subsequent pairwise comparison tests of independence were again used to investigate whether YBOs displayed differences in habitat selection between the breeding and non-breeding seasons.

Due to the likely interaction between host choice, habitat and other factors that influence YBO occurrence in the landscape, a multiple correspondence analysis (MCA) (Husson, Lê & Pagès 2017) approach was used to further investigate the dataset. The MCA was performed using the YBO observations that included reliable host species data (n = 1275) from within the KNP boundary.

Host species, breeding season and geographical region were included as active variables. Due to its known association with geographical region, habitat type was used as a supplemental variable. Supplemental variables do not contribute to the calculation of the principal dimensions, but their coordinates are plotted to illustrate how they may relate to the active variables. Variable imbalance in host species data required the collapse of all but the two most frequent host species, each of which comprised less than 10% of sightings, into a single response category, ‘other’. Eigenvalues, % inertia and eta² were calculated for each variable, and category contributions cos² and v.test were calculated for each variable in each dimension.

The MCA was conducted using the Burt matrix approach, which is better suited for analysing relationships among categorical variables rather than focussing on individuals. Typically, only the first two or three dimensions are interpreted in MCA (Ayele, Zewotir & Mwambi 2015; Husson et al. 2017) as the variable categories that contribute the most to dimension 1 and dimension 2 are always the most important in explaining the variability in the dataset (Kassambara 2018). In MCA, the cos² value corresponds to the quality of representation of the individual or category on the dimension, such that if the value is close to 1, the individual or category is well projected on the dimension (Kassambara 2018). In this analysis, categories with a cos² value ≥ 0.3 were deemed to be well represented on the corresponding dimension.

The v.test values determine whether a category has a coordinate which is significantly different from zero. As a v.test follows a normal distribution, values greater than |1.96| were considered significant. The measure of influence of each categorical variable on the dimension is indicated by eta². An eta² value close to 1 indicates a strong link between the dimension and the categorical variable.

Hierarchical clustering on principal components (HCPC), using Wards criteria, was then used to group YBO observations based on similar patterns in the categorical response variables. The number of clusters was chosen by maximising the relative loss of inertia when two clusters are merged. The MCA and HCPC were performed in R Studio using the FactoMineR (Le, Josse & Husson 2008) and factoextra (Kassambara & Mundt 2020) packages.

Historical sighting data (n = 118 records, period: 1979–1994) included those obtained during the KNP annual buffalo census (Table 1). Contemporary data included the 7-year period (01 April 2015–31 March 2022) for which regular and accurate sighting data were available. This resulted in 6436 suitable contemporary sightings from the study area and surroundings for analysis, of which 1394 contained reliable wild mammal host data (Table 1).

Sighting data (Figure 2) showed a park-wide expansion in the distribution of YBOs, progressing in a southerly and westerly direction with concurrent increases in reports, especially in the northern areas where recolonisation was first recorded. The increase in YBO densities across KNP and its surrounding areas (1979–2022) was evident in the sighting records. Sightings of YBOs in each of the three regions of the KNP, during the period 01 January 1979–31 December 1994, were 90.7% (n = 107) from the north region, 5.1% (n = 6) from the central region and 4.2% (n = 5) from the south region. In comparison, 6436 sightings from 01 April 2015 to 31 March 2022 were 72.6% (n = 4672) from the north, 19.7% (n = 1268) from the central and 7.7% (n = 496) from the south regions. As found with the sighting records from 2005 to 2022 (Figure 2), YBO was observed further south into the Mjejane Game Reserve and west into APNR. The mean number of birds recorded per sighting = 3.89 (n = 1275), and the maximum flock size recorded was at least 73 individuals on a herd of buffalo in the vicinity of the João Windmill, which lies approximately 8.0 km south-west of Shingwedzi Rest Camp (14 August 2018).

From the historical observations (pre-1995), six host species were recorded being used by YBOs, with buffalo accounting for 92.4% (n = 105 sightings) of the host observations. In the contemporary sightings (n = 1397 sightings on hosts), 14 wild host species were recorded, with an additional two domestic host species (cattle and donkeys). The three observations on domestic hosts occurred in rural areas immediately adjacent to the KNP (maximum perpendicular distance from fence = 4.6 km).

Host selection (Figure 3) differed significantly across the three geographical regions of the KNP (chi-square test: χ² = 45.89, df = 12, p < 0.01, Bonferroni corrected values: p ≤ 0.01, z-critical = ± 3.04), with the two most frequently utilised host species showing significant differences between the regions (Online Appendix 1, Table 1-A1). Giraffes were utilised significantly more than other hosts (standard residual = 4.48) in the central region, while buffalo were significantly underrepresented in the central region (standard residual = −4.91). Buffalo were significantly selected as hosts in the southern region (standard residual = 3.21). For all other common host species, there were no significant differences in selection between regions.

Host species selection differed between the breeding (November–February) and non-breeding (March–October) seasons of the YBOs (chi-square test: χ² = 34.10, df = 6, p < 0.01, Bonferroni corrected values: p < 0.01, z-critical= ± 2.91), with hippopotamus and kudu selected for significantly less in the breeding season (standard residual= −3.01 and −3.49, respectively) (refer to Online Appendix 1, Table 2-A1).

Habitat selection by YBOs within the KNP boundary (Table 2) was not proportional to habitat availability (χ² = 4362.40, df = 9, p < 0.001). Only the Senegalia (Acacia) nigrescens/Marula Savanna habitat type was used in proportion to its availability. In the Pafuri Rugged Veld, Punda Maria Sandveld and Shrub Mopane, YBOs were encountered more frequently than expected (preference), while in all other habitat types, YBOs were encountered less frequently than expected (avoidance).

As YBOs were more commonly encountered north of the Olifants River and were comparatively sparsely distributed south of this natural boundary, the above analysis was repeated using only the habitat types and extents north of the Olifants River. Results were again significant (chi-square test: χ² = 2223.80, df = 6, p < 0.01). In the Pafuri Rugged Veld, Punda Maria Sandveld and Shrub Mopane, YBOs were encountered more frequently than expected (preference), while in the Lebombo, Mopane/Senegalia (Acacia) nigrescens Bush and Mopane/Combretum Bush, YBOs were encountered less frequently than expected (avoidance). In the Senegalia (Acacia) nigrescens/Marula Savanna habitat type, YBO were encountered in proportion to what was expected.

A significant difference in habitat choice between the breeding and non-breeding season was found (chi-square test: χ² = 39.36, df = 9, p < 0.01, Bonferroni corrected values: p < 0.01, z-critical = ± 2.81). The Punda Maria Sandveld and Shrub Mopane habitat types were utilised significantly more than expected (standard residual = 3.71 and 3.54, respectively) during the breeding season. There were no significant differences in habitat choice between the breeding and non-breeding seasons for any other habitat type (refer to Online Appendix 1, Table 3-A1).

The calculated eigenvalues and contribution (% of variance) of each dimension in the MCA (Table 3) indicate that five dimensions explain the total inertia (variation) in the dataset. The variables biplots (MCA factor maps) for the first two dimensions (Figure 4), describe 48.93% of the total, unadjusted inertia. The greatest influence by a variable on dimensions 1 and 2 was host species (eta² = 0.62) and breeding period (eta² = 0.57), respectively. The variable region had a moderate influence on both dimensions 1 and 2 (eta² = 0.49 and 0.39, respectively). The v.test values were significant for all active variables.

The MCA grouped giraffe host use and the central region as variables highly correlated with dimension 1. Buffalo hosts were also highly correlated with dimension 1 but opposed the giraffe hosts and the central region across the vertical axis. The breeding and non-breeding periods and the north and central regions were variables highly correlated with dimension 2. The breeding period and the north region oppose the non-breeding and the central region across the horizontal axis.

Hierarchical clustering identified four sighting clusters based on observations with similar profiles regarding habitat, region, breeding period and host selection (Figure 4).

Cluster 4, the only distinct cluster, corresponds to all sightings of YBO on giraffe hosts (v.test = 32.74, p < 0.001), sightings from the central region (v.test = 5.12, p < 0.001) and within the Senegalia (Acacia) nigrescens/Marula Savanna (v.test = 6.80, p < 0.001) and Shrub Mopane (v.test = 3.02, p < 0.001) habitat types. Cluster 2, which exhibits overlap with both cluster 3 and 4, groups all sightings on buffalo hosts (v.test = 33.39, p < 0.001) and is strongly associated with the north region (v.test = 10.19, p < 0.001) and the following habitat types: Mopane/Combretum Bush (v.test = 4.97, p < 0.001), Mopane/Senegalia (Acacia) nigrescens Bush (MAB) (v.test = 4.00, p < 0.001), Pafuri Rugged Veld (v.test = 3.35, p < 0.001), Shrub Mopane (v.test = 2.82, p < 0.001), Punda Maria Sandveld (v.test = 2.53, p < 0.001) and Lebombo (v.test = 2.47, p < 0.001) habitat types. Cluster 3 contains all sightings on ‘other’ hosts (v.test = 31.49, p < 0.001) and is strongly associated with the north region (v.test = 2.06, p < 0.001) and the Punda Maria Sandveld habitat type (v.test = 2.88, p < 0.001). Cluster 1 corresponds to all sightings in the south region (v.test = 29.75, p < 0.001) and is strongly associated with sightings on buffalo hosts (v.test = 3.25, p < 0.001).

The arrival of YBOs in the KNP, after an absence of at least seven decades, signalled the start of an expansive and ongoing recolonisation of the KNP and surrounding areas of South Africa.

Although the initial return of YBOs was well-documented, these records remained largely inaccessible and unpublished until now. Few reliable YBO sighting records, from both internal KNP sources and citizen scientists, exist after 1994, up until the initiation of our study in late 2014. Citizen scientist records contained in global online biodiversity datasets were scarce and of limited value, averaging less than one YBO record per month for the study area prior to 2014. Thus, the period 1994–2014 effectively constitutes a 20-year gap in the available data.

By 1987, YBOs were uncommon but considered to be a widespread breeding resident in the north-western savannas of the KNP, with a rapidly expanding range and population densities (Whyte et al. 1987). However, records obtained from historical buffalo census include occasional YBO sightings in the southern and central regions of the KNP as early as 1989, some as far south as the Crocodile Bridge Ranger Section. It is now reasonable to expect that YBOs, like the co-occurring RBO, may be encountered throughout the KNP and APNR. The YBO may also be encountered in adjoining conservation areas, including the Mjejane Game Reserve, and occasionally in adjacent communal grazing lands and other areas (e.g. wildlife eco-housing estates), provided suitable hosts are present.

Further range expansion by YBOs into farmland, as has occurred with RBOs (Craig 2005), may still occur, although the more specialised host requirements of YBOs may be restrictive (Stutterheim, Bezuidenhout & Elliott 1988). Cattle are deemed to be suitable hosts (Feare & Craig 1999) if present in extensive systems and at sufficient densities, so long as other ecological requirements (sufficient food and nest sites) of YBOs are met. Communal cattle herds may even have played a role in facilitating the return of the YBO to the KNP. YBOs likely dispersed from an existing population in Gonarezhou National Park, Zimbabwe, via the Sengwe Communal Lands, after cattle dipping in the communal area ceased (Hall-Martin 1987). This allowed YBOs to utilise undipped cattle and other wild hosts (buffalo, kudu, eland and roan) while crossing the communal lands until they reached the wild game populations of the northern KNP. Recent vagrant records of YBOs (Hausler 2023) would suggest that long-distance movements outside current range areas are occasionally undertaken, although typically by singletons.

Contemporary sightings utilised in this study were clustered along the tourist infrastructure, a consequence of the citizen scientist data collection method, although direct YBO observations were recorded in more remote areas of the KNP where possible. No contemporary sightings were obtained from Mozambique, but the presence of suitable resources for YBOs in some conservation areas of Mozambique that abut the KNP (e.g. Limpopo National Park; Sabie Game Park) suggests that YBOs may be present but overlooked or may potentially colonise these areas in the future. However, the occurrence of YBOs in Mozambique south of the Zambezi River was previously deemed unlikely (Feare & Craig 1999; Hall-Martin 1987), and no records exist in the current South African Bird Atlas Project (SABAP2) database (Brooks & Ryan 2022b).

The six most frequent wild mammalian host species recorded in our study have, with the exception of impala, all been repeatedly recorded as frequent host species for YBOs (Feare & Craig 1999; Mundy 1992). In KNP, we found that buffalo were the most frequently utilised host irrespective of YBO breeding season or region, a similar finding to previous studies where buffalo were present (Grobler & Charsley 1978; Hustler 1987; Jubber 2014; Mundy 1983).

As found in our study, giraffes were the second most frequent host in the Masai Mara Reserve, Kenya (Koenig 1997) and Moremi Game Reserve, Botswana (Stutterheim & Panagis 1985) and appear to be the primary alternative host if buffalo were absent (Dale 1992; Hustler 1987).

Previous studies indicate that for all hosts other than buffalo, host selection of YBOs differs by location (Feare & Craig 1999) and is likely influenced by the availability of hosts in the landscape (Attwell 1966; Buskirk 2012; Grobler & Charsley 1978; Koenig 1997). Among the infrequent host species that we recorded, all but nyala (n = 3 sightings) have been recorded as hosts elsewhere. Observations of YBOs on domestic hosts were likely individuals that typically utilise wild mammal hosts from the adjacent conservation areas. It is, however, important that studies from elsewhere in Africa have indicated that cattle, donkeys and horses are suitable and frequent hosts in some areas (see Feare & Craig 1999).

The use of sighting data of the type collected during this study gives some indication of host choice, especially as YBOs actively seek out and remain with specific host species (Palmer & Packer 2018; Stutterheim & Panagis 1985). However, accurate measures of host availability were difficult to obtain for our study area due to the extensive landscape of the KNP and APNR and the lack of host population and seasonal distribution data. Only a single large-mammal census was undertaken by SANParks during the course of this study (South African National Parks 2017), and unfortunately, this provided a single dry-season snapshot of host presence at the end of a 2-year drought period. This census did not reflect the host availability across the study period, which was mostly characterised by above-average rainfall. The six common host species are known to occur at varying densities across the study area, and undertake seasonal movements based on resource availability (ed. Apps 1996; Du Toit 2003; Smuts 1974), with the exception of impala, which are more sedentary (Smuts 1974). Previously, host preference studies for oxpeckers have been restricted to small land parcels where host availability is easily determined (see, for example, Dale 1992; Mooring & Mundy 1996). In the KNP, if host preference is to be further investigated, localised studies within the preferred habitat types would be more suitable, especially if individual YBO could be monitored and local host presence quantified.

Differences in host use patterns between breeding and non-breeding seasons may be a result of YBOs altering their foraging approach due to the demands of breeding attempts (e.g. nest guarding, incubation and provisioning duties). When breeding, YBOs may be restricted to foraging in a certain limited range around the nest site and are unable to follow and remain with preferred hosts, such as large buffalo herds, in a nomadic fashion. Although buffalo and giraffe remained the most frequently utilised hosts during YBO breeding seasons, buffalo were utilised less in this period compared to the non-breeding season. Various studies (Dowsett 1968; Palmer & Packer 2018; Stutterheim & Panagis 1985) observed that buffalo are preferred and utilised whenever possible, but especially in the non-breeding season when YBOs are able to remain with the preferred hosts as they travel through the landscape, thus eliminating the risk of losing their hosts overnight.

Evidence of YBO roosting on host mammals in the KNP would confirm whether YBO remains with preferred hosts in the non-breeding season or departs to roost elsewhere. Although this could not be definitively confirmed in our study, groups of YBOs were observed on hosts well before sunrise and after sunset. All of these observations were during the non-breeding period (June) and included sightings on kudu (four and six birds, 32 min after sunset), giraffe (four birds, 50 min before sunrise) and buffalo (minimum four birds, approximately 20 min before sunrise). We found a single historical record of this behaviour in the KNP of six birds roosting on a giraffe at Maketekete, Shangoni, on 01 May 1993 at 20:00 (Nasionale Parkeraad 1994).

Although additional attempts were made to locate roosting YBOs on suitable hosts during night-time travelling, in all seasons, no additional observations of this behaviour were obtained.

However, roosting YBOs may be easily overlooked, especially as they are likely tucked into inconspicuous locations on the body of the host when roosting, including the axillae and underbelly, which are difficult to view at night (Palmer & Packer 2018). Evidence of this behaviour would be an important component of future studies in the KNP. Such behaviour has been observed elsewhere in Africa (Dowsett 1968; Palmer & Packer 2018; Stutterheim & Panagis 1985).

Yellow-billed oxpeckers were recorded in all habitat types in the KNP, with a preference for the Pafuri Rugged Veld, Punda Maria Sandveld and Shrub Mopane. The apparent avoidance of various habitats in the southern region of KNP may be a result of the lower YBO densities here, when compared to northern regions, and these patterns would likely change should YBO densities increase to similar levels as those currently present in the northern regions.

The Pafuri Rugged Veld and Punda Maria Sandveld are largely encompassed by the early YBO range described in 1987, while the remainder of the range at that time was largely comprised of Mopane/Senegalia (Acacia) nigrescens Bush and Mopane/Combretum Bush (Whyte et al. 1987). Sightings in the period up until 1994 continued in this pattern and were most frequent in these four habitat types, with scattered records from elsewhere. The Shrub Mopane habitat type made up a very small portion of the YBO range in 1987 (Whyte et al. 1987), but by 1994, sightings were recorded with increasing frequency throughout this habitat type.

Presently, sightings in Shrub Mopane are commonplace, but YBOs appear to be selecting this open habitat based on the availability and visibility of suitable host species, and not the availability of trees for nesting or roosting. Grazers, like buffalo and zebra, utilise the Shrub Mopane areas extensively due to the dense grass sward that is typical of this habitat type, while kudu, giraffe and impala are typically scarce (Gertenbach 1983). Early YBO studies in the KNP indicated that YBO distribution in the KNP was determined by habitat rather than the distribution of symbiont hosts, and YBO range was limited to the taller tree savannas of the western portion of the northern areas of the KNP, where breeding and roosting sites were readily available (Whyte et al. 1987). However, this seems to have changed over time, possibly a result of increasing YBO presence. Shrub Mopane habitat is used significantly more frequently in the breeding period compared to the non-breeding period, even though this habitat has few mature trees that are an essential requirement for breeding. Given that YBO must frequently leave preferred hosts during the breeding period (due to nest construction and guarding and chick provisioning activities) (Hausler 2023), they may find it easier to relocate hosts in open habitats compared to the denser woodland habitats in which they breed. There is most likely some commuting between nest sites and foraging areas, although the typical distances covered in a single foraging trip are currently unknown. It is also important to note that sightings of YBO recorded in open habitats like the Shrub Mopane may be inflated compared to other habitats due to the hosts and accompanying YBOs being more conspicuous to the citizen scientist observers.

The MCA and HCPC divided YBO observations into four groups. A distinct cluster (cluster 4) associated with dimension 1 captured sightings with giraffe hosts, primarily in the central region and the Senegalia (Acacia) nigrescens/Marula Savanna and Shrub Mopane habitat types. This cluster is not strongly associated with the breeding period and occurs year-round.

Giraffe may favour this habitat type, while buffalo may avoid it, especially the rugged areas close to the Olifants River, which have sparse grazing. Most sightings with buffalo hosts in the central region are from the open grassland areas around the Satara rest camp and further south on the eastern grasslands. Clusters 1–3 are not strongly associated with a single dimension.

Cluster 2, which contains all buffalo host observations, is positioned at the origin, which can be explained by the fact that buffalo are the most frequent host species irrespective of region and ubiquitous throughout the study area. The partial overlap of clusters 1 and 3 with cluster 2 is similarly due to the presence of buffalo hosts in the profiles. Buffalo hosts comprise 100% of sightings in cluster 2 and make up the majority of sightings in all regions and seasons. There is no clustering of the centroids of the three overlapping clusters, which indicates there is no shared underlying profile on the main dimensions. Cluster 1, characterised by observations from the south region and habitat types and containing buffalo hosts in the sightings profile, represents the strong association between YBO and buffalo in the south region. Cluster 3 represents the sightings of YBO utilising the wider array of hosts, as is more commonly observed north of the Olifants River.

The increased proportion of sightings on buffalo in the southern areas is likely influenced by the lower densities of YBOs in this region. All large buffalo herds (> 100 individuals) encountered south of the Sabie River during the collection of the contemporary data (n = 86 observations) had accompanying YBOs. It is probable that in areas recently recolonised by YBOs, where their population densities are lower, the preferred host is buffalo, and as population densities increase (as is the case in the northern regions of the study area), there is a switch to a wider range of hosts. This suggestion is supported by Whyte et al. (1987) who indicated that while the YBO population density is low, the birds tend to concentrate on buffalo, their primary symbiont host. Further evidence of this comes from the historical sighting data which reveal that the majority of sightings (92.4%, n = 105) during the period 1979–1994, after the recolonisation of the KNP began, were of YBOs on buffalo. Furthermore, during this period, only a single record for the central and southern regions involves YBOs utilising a host other than buffalo. The switch to other host species may be partially due to increased intraspecific competition at higher population densities and a more homogeneous occurrence of YBOs in the available landscape.

Field observations show that the number of YBOs was generally higher in the northern part of the study area and was the highest when buffalo were present. However, reliable YBO numbers per host group could not be derived from the citizen science records. These sighting records commonly underestimate the true number of YBOs, particularly for hosts in large herds such as buffalo, where YBOs are dispersed across multiple hosts, and not all individuals are immediately visible.

The extensive range expansion by the YBO will likely have restored an important element of the KNP and APNR savanna ecosystem but may also have had an impact on the co-occurring RBO, previously the only oxpecker in the study area. Interspecific competition between the two species is well-documented, with YBOs typically dominating RBOs when accessing food and favourable foraging locations (Stutterheim et al. 1988) and nest sites (Hausler 2023).

Comparisons between the species reporting rate for RBOs between the two Southern African Bird Atlas Projects (SABAP1 [Period: 1987–1991] and SABAP2 [Period: 2007–Present]) indicate reductions in RBO reporting rate over much of the study area (Brooks & Ryan 2022a), while YBO reporting rate has increased. This may signal a reduction in RBO numbers in the wake of YBO population increases, although extirpation of RBO is unlikely given the broader host species utilisation by RBOs (Chittenden & Whyte 2009; Stutterheim et al. 1988) and that the two species can occur alongside one another on the same host individuals (Koenig 1997).

Currently, no suitable population density data exist for comparing the two oxpecker species in the study area or for determining how density varies by location. Informal observations in the form of daily sighting records suggest that YBO outnumbers RBO in the northern regions of the study area, but this needs to be verified empirically and would constitute valuable future study and monitoring.