This study was conducted at Barberspan Bird Sanctuary (26° 35̕ S 25° 35̕ E), 307 km west of Johannesburg, between Sannieshof and Delareyville. This is a summer rainfall area with annual rainfall averaging around 500 mm. The daily mean minimum and maximum temperatures during summer are 15 °C and 30 °C, respectively, and during winter 0 °C and 20 °C (South African Weather Service data). Barberspan was one of the first (1975) wetlands in South Africa to be designated a Wetland of International Importance (Ramsar site) and represents a rare and unusual type of perennial wetland, that is, a large grassland pan. The water surface covers approximately 2000 ha and only has an overflow during high water levels, making the water of the pan alkaline because of evaporation. It functions as an important seasonal stopover site for migrating waterbirds and provides sanctuary for moulting waterbirds (Milstein 1975; Ndlovu et al. 2013). Of great importance is its provision of perennial water in an area characterised by seasonal wetlands (Milstein 1975). There is no hunting allowed in the reserve. A mean of 3052 tourists visit Barberspan annually but less than 10% are strictly birdwatchers, while the remaining 90% are anglers or tourists passing through the area to another destination (Van der Merwe, S., 2016, email, 08 September, barbersp@lantic.net). Between April 2013 and March 2014, 3243 people in 1393 vehicles visited the reserve.

More than 350 aquatic and terrestrial bird species have been recorded. The pan sometimes holds over 20 000 birds (Bouwman & Hoffman 2007). During the July 2014 Coordinated Waterbird Counts (CWAC), 6858 individuals representing 44 waterbird species were observed (CWAC report, July 2014, Van der Merwe, S., 2016, email, 05 September, barbersp@lantic.net). FIDs were measured in all accessible areas around the pan, including the frequently used general angling area on the eastern shore, the infrequently used competition angling area and the nature reserve covering the whole of the western shore. The latter area is a no-angling area and is strictly for the use of hikers and birders. These localities with varying numbers of visitors enabled us to get a general view of the birds’ habituation at Barberspan.

Flight initiation distances fieldwork was conducted in 5-day sessions, once in a month, for 3 months (March, April, and July 2014) as these autumn months include a high number of both resident and some migratory waterbird species. Observations were made, and FIDs were recorded from an hour after sunrise for 3 h (06:30–09:30), and again 2 h before sunset until sunset (16:00–18:00). FIDs were measured by identifying the bird from as far away as possible, and taking a starting distance with a Leica Rangemaster 1200 laser range finder. Two or three observers approached the birds, wearing dark-coloured clothing for uniformity and consistency, and walking abreast so as to not appear as only one person (Geist et al. 2005; McLeod et al. 2013). The mean height of the observers was 1.8 m. If birds were in a group, a focal, visible individual at the edge of the group was selected to measure starting distance and FID. The bird was approached head-on and in a straight line at a leisurely walking pace (± 2 km/h). The birds were not stalked but approached in the open, which the flat and open terrain of Barberspan allowed for. FID was taken as soon as the bird took flight because of the observers. In our case, we assumed that the obvious interest shown by the approaching humans by looking and pointing things at them (heads, binoculars, fingers, and the laser range finder; see also Bateman & Fleming 2011) was perceived as threats by the birds being approached. Any flight because of obvious relocation for foraging or other disturbances was not recorded. Mean body mass of each species was used as a reference to FID as it was previously found to correlate with FID (Møller & Erritzøe 2010; Møller, Vagasi, & Pap 2013b). The mean mass of the species was obtained from Del Hoyo, Elliot and Sargatal (1992, 1996, 2004).

Data for species with more than five FID observations were used (33 species, for a total of 525 records) for univariate analyses. The untransformed data were analysed in GraphPad PRISM v5.04 (http://www.graphpad.com) using linear regression and column statistics. Linear regression was used to investigate the correlation between mass and FID. One-phase decay was used to plot FID versus starting distance, with outliers eliminated (1% and 99% percentiles). A non-linear, one-phase decay model was chosen assuming that the rate at which FID increases is proportional to an increase in starting distance (http://www.graphpad.com).

Buffer zones were calculated by using the maximum FID for species in arbitrarily chosen mass classes, namely small (less than 200 g), medium (201 g – 1000 g), and large birds (more than 1000 g). Although several authors (Borgmann 2011; Rodgers & Smith 1995) suggest using only or mostly the most sensitive or skittish species to establish buffer zones (or set-back distances), we combined data for all species as this represents the actual situation on the ground, as Barberspan may be merely a stopover site for some of the skittish and rare birds. Using this combination allowed for better accommodation of all species when proposing a buffer zone.

For the relatively small area sampled and the good visibility, we assumed that our sample (consisting of the 33 species recorded more than five times) was representative of the occurrence of the species at Barberspan. However, larger, more conspicuous birds, such as the Goliath Heron, would attract more attention per individual than smaller birds per sampling effort and there may be a bias towards the larger birds. The smaller waders and ducks are not restricted to open habitat, as some (such as the red-billed teal Anas erythrorhyncha and three-banded plover) can also be found in or near areas with more vegetation along the shoreline. A shorter FID, allowed for by the ability to quickly escape, makes it safer for them to venture here. The taller birds would be less restricted by obstructive vegetation, because a higher eye height would allow the farther detection of bigger prey items and threats (Blumstein 2006).

Some of the species we observed, especially those favouring the frequently used angling area, seemed to be partially habituated to humans. FIDs reported for Australian, European and American conspecifics and congenerics (Blumstein 2006) were consistently shorter than at Barberspan, and some by a quite a large margin (Table 2). The large FIDs measured at Barberspan Bird Sanctuary may be accounted for by two interacting reasons. Firstly, black-backed jackal Canis mesomelas, Cape clawless otter Aonyx capensis, caracal Felis caracal, small-spotted genet Genetta genetta, a variety of mongooses, and many raptors (including African Fish Eagle Haliaeetus vocifer, taking ducks and flamingos; H. Bouwman, pers. obs.) also occur at the Barberspan Bird Sanctuary. These are all known predators of birds. This occurrence of fast-moving predators might contribute to the longer FIDs when compared with congenerics and conspecifics elsewhere (Blumstein 2006; Table 2). Møller and Liang (2012) predicted and found that species in tropical areas take smaller risks (therefore longer FIDs) than the same or closely related species in temperate areas. According to them, this effect was not only related to predation pressure, possibly being greater in tropical areas, but also to life-history traits, especially clutch size. Tropical birds live longer, produce smaller clutches and start breeding at a later age. Therefore, they should be more risk-averse than their temperate counterparts (Møller & Liang 2012).

The species CVs were normally distributed (Figure 1c), and the regression between CV and mass was not significant (p = 0.0711), indicating a set of common factors that birds, irrespective of species, consider when perceiving and reacting to threats. Describing these factors will need more research.

Our results confirm those of Blumstein (2003) and Cooper (2005), namely that FID and starting distances are not linearly related (see Figure 1d, R² = 0.6581). Beyond a certain distance, an increase in mean starting distances does not seem to affect FID, and therefore, one-phase decay was an appropriate model. Distances longer than 246 m (the plateau) is very long and visual acuity and observation skills of the observers may become a factor.

Nature reserves dedicated to mostly birdwatching such as Barberspan should be managed with least possible disturbance to the birds, while allowing birdwatchers good and unobstructed views and minimising disturbance to birds.

Figure 1a shows that the upper standard deviation of the mean of all observations was 125 m. Not approaching any bird by less than 125 m would protect against most disturbances, but would still affect the larger birds such as the Goliath Heron. However, birds are not distributed evenly around Barberspan because of large differences in vegetation, shallows, wind and bottom conditions (mud, gravel, sand, etc.). Certain shallower areas with muddy bottoms have mainly smaller birds where the minimum approach distance may be shortened. Smith and Bouwman (unpublished) conducted a study in 2004 to establish distribution patterns along the edge of the pan for conservation purposes. Their findings are presented in Appendix 1 (Figure 1-A1) and used to derive appropriate buffer zones.

The shores of Barberspan are not homogenous. Combined with seasonal changes in water levels, a single approach to minimising disturbances will not be feasible. We propose the use of different approach distances for differing bird sizes, and buffer zones based on the average water level following the wet-season (September–March), assuming that most birds congregate at this interface. Buffer zones are set at the maximum FID per mass group, which is 104 m for small, 213 m for medium and 300 m for large birds. Rodgers and Smith (1995) suggested the buffer zones (or set-back distances) in Florida be set at 100 m for wading-bird colonies, and 180 m for tern and skimmer colonies. Their distances were calculated by adding the standard deviation to the mean from their sample, and then adding another 40 m to allow for the distances where the birds usually become agitated by a disturbance. Using a similar formula, Erwin (1989) suggested roughly the same distances. These distances were, however, for breeding colonies. Around foraging and loafing sites for most waterbirds in Florida, Rodgers and Smith (1997) suggested a buffer zone of about 100 m. Borgmann (2011) suggested a buffer zone of 250 m for waterfowl, diving ducks, wading birds and shorebirds, to lessen the impact of human disturbance on sensitive or skittish species.

Considering what Smith and Bouwman (unpublished) found at Barberspan in 2004 (Appendix 1), we suggest that the medium- and large-bird buffer zones (213 m and 300 m, respectively) be used in areas with greater species richness, such as hotspots A, B and E (Figure 1-A1). Elsewhere, the small-bird buffer zone can be applied. It might, however, be somewhat impractical as birders might not be able to see the plovers and sandpipers at 104 m. Therefore, we suggest that the small-bird buffer be set at the highest species mean FID for the group smaller than 200 g, which is 62 m for the Blacksmith Lapwing Vanellus armatus. During the breeding season, the original buffer zone of 104 m may still be implemented in areas surrounding the breeding colonies. For practical reasons, the 300 m buffer for large birds is only applicable in areas where threatened and sensitive species occur regularly, as the highest mean FID for birds larger than 1000 g is still accounted for under the medium buffer. For the Goliath Heron and flamingos, being the largest species at Barberspan, we propose an individual minimum approach distance of the median FID (127 m) plus one standard deviation (56 m) be used for a more practical buffer of 183 m.

There are several bird hides along the western shore of Barberspan and it is still preferable for tourists to use these facilities. Walking along the shore of the pan is allowed if size-class buffer zones are adhered to, while not disturbing the birds. The practical implementation of our proposals will need integration with the Barberspan management plan following consultations with the management and stakeholders. Information on appropriate buffer zones and sensitive areas can be provided at reception, and the distances can be indicated by representation of bird cut-outs at appropriate distances for birders to practice their distance estimations. The effectiveness of the memory retention of distances should, however, be tested.

Several factors in our study may be perceived as constraints. Firstly, we approached the birds directly on foot (simulating a threat), while a more oblique approach might result in shorter FIDs (Bateman & Fleming 2011). Approach by vehicle, which may often be the case for birdwatchers in open habitat, would presumably have an influence on FID (McLeod et al. 2013). Another constraint may be that our observations were made outside the breeding season. Species that forage close to their nests (such as the Kittlitz’s Plovers Charadrius pecuarius and Blacksmith Lapwings) may adjust their anti-predator behaviours and FIDs, while birds that forage away from their nests, such as the herons, presumably would not or very little. For example, Møller et al. (2013a) found that FID was shortened for European bird species following harsh winter. We are also not aware of the effects of drought on FIDs, which could be a valuable future research area. The species distribution and biomass study (Smith & Bouwman unpublished) was conducted 10 years prior to the flight initiation study. Distribution patterns and biomass could have changed over time. However, we believe the changes are not so profound as to disregard the information, as the conditions around the pan have not drastically changed.