Study area

The study was conducted in two protected areas (Main Camp area of Hwange National Park (HNP) and Sikumi Forestry Area (FSA)) and the adjacent communal areas (CAs) as illustrated in Fig. 1. These areas are part of the southern Kavango Zambezi Trans-frontier Conservation Area (KAZA TFCA). The Main Camp area covers ca 1500 km², SFA ca 544 km2 and the CAs are under Tsholotsho and Hwange districts. The area receives about 600 mm of rainfall, falling between November and May (Chamaillé-Jammes et al. 2007) and is characterised by poorly drained Kalahari sands and patches of clay that form a system of more than 40 000 mostly shallow seasonal waterpans in HNP and surrounding areas (Childes and Mundy 2001). In addition to the seasonal waterpans, HNP and SFA have boreholes to supplement water to wildlife during the dry season, while in CAs some dams were constructed. For sites that retained water between the peak dry seasons of 2014–2016, the surface area averages for HNP, SFA and CAs waterpans were 1337, 1200 and 4085 m² respectively.

The waterpans are utilised by various waterbird species including waterfowl (e.g. ducks, geese and grebes), carnivores and piscivores (e.g. storks, egrets, herons, terns and kingfishers) and waders (e.g. sandpipers and lapwings). Since waterpans are small in size (Goudie and Wells 1995), numbers of waterbirds hosted on them are usually small as well, making observational studies easier when compared to large water bodies such as dams, floodplains and coastal areas. Waterpans are accessed by several mammalian species, chiefly the African elephant and African buffalo Syncerus caffer mostly in the PAs during the dry season, while those in CAs are also used by local cattle breeds Bos species (usually accompanied by cattle herders). Waterpans are also visited by a wide array of wild and domestic carnivorous species that potentially prey on waterbirds or their eggs.

Waterpan selection and observations

We purposively selected 13 waterpans that retained water throughout the year in the study area. We selected pumped and unpumped waterpans in HNP (6 and 2 respectively), SFA (1 and 2 respectively) and CAs (1 unpumped waterpan and 1 dam). Our waterpans were at least 5 km from each other to reduce risk of pseudo-replication when recording waterbird behaviour. Between July 2015 and March 2017 we monitored the activities of waterbirds at each waterpan during the day from 06:00h to 18:00h. Each site was visited at least twice in the wet season (between December and April) and also twice in the dry season (May–November), a total of 724 h of monitoring. Each time the monitoring team consisted of four people, one recorder and three observers (with 10 × 50 magnification binoculars) to cover the whole waterpan. This team size was sufficient in monitoring these small wetland systems. As it was unavoidable to cause some disturbances to waterbirds upon our arrival at a waterpan, we waited for at least 10 min to allow the waterbirds to resume their normal behaviours. We then noted all waterbirds present and their relative positions to the waterpan edge. We constantly scanned (after every minute) the waterpan area to note any changes as waterbirds arrived/left the waterpan area or when they changed their positions. We also constantly scanned (at one minute intervals) the sky for any predatory birds (or any other large bird that could possibly disturb waterbirds) that flew by the waterpan area, hereinafter a flyover. Upon seeing the flyover, we immediately recorded the original activity engaged by all the different waterbird species; any behavioural changes in response to the flyover danger (see the guide in Table 1); and any associated changes in relative position to water. However, in the cases that waterbirds did not show any immediate response after the flyover, we continued to monitor their activity for another one minute before concluding that there was no response. Likewise, we also recorded the position and activity changes when other forms of disturbances occurred (e.g. from carnivores, large herbivores, people and other noises). We also recorded the numbers of individuals that responded to each disturbance as it was possible to have multiple responses from one disturbance. The monitoring team took 30 min break between 10:00–10:30 h and 13:00–13:30 h to feed and rest. We believe that these intensive scans allowed us to record the waterbird responses timeously.

For the waterbirds that responded by flying away, we used a stopwatch to record the time taken (in seconds) before they returned to the waterpan (we considered individuals to have completely left the waterpan if they did not return within 2 h, a time in which local feeding conditions e.g. day time temperature and composition of herbivores visiting the waterpan is likely to have changed). We used group composition (e.g. numbers, sex and ages of individuals where possible) and any notable individuals (with distinct breeding plumage, broken limbs, rings etc.) to identify returning groups or individuals. With our experience of the study area and the predictable movements of species after disturbances, we could identify returning birds with a high degree of certainty. Disturbances triggered by our monitoring team were minimised as we observed the waterbirds from established observation platforms, or hiding places located at least 100 m from the waterpan. However, when we noted that our team had triggered the response, we recorded them accordingly.

Categorisations of species, their activities, source of disturbances, time and waterpans

All waterbirds recorded during the study were grouped according to their taxonomy, size and feeding guilds or their nearest group as generalist medium and large (herons, ibises, storks and egrets), large wildfowl (Egyptian geese Alopochen aegyptiaca, knob-billed duck Sarkidiornis melanotos and spur-winged geese Plectropterus gambensis), small wildfowl (teal, grebes, moorhens and white-faced duck Dendrocygna viduata) and waders (lapwings, sandpipers, stilts and jacanas) as shown in Table 2. We categorised original species activity, distinguishing if activities could allow for some form of vigilance, e.g. swimming with head up, or grazing in short grass (Guillemain et al. 2001); or preventing any vigilance, e.g. underwater feeding, sleeping with head under wings (Thomas 1982, Guillemain et al. 2007) and other responses (continued in non-vigilant activities or became more alert and escaping from the danger) as shown in Table 1. We categorised the type of disturbance as aerial (raptors and any flying birds that initiated a response), terrestrial (including wild carnivorous mammal species that potentially eat waterbirds, crocodiles Crocodylus niloticus, people, domestic dogs Canis familiaris and noise from vehicles) and herbivores (wild herbivores and livestock). We categorised the time of day as early morning (06:00–09:00 h), mid-morning (09:01–12:00h), early afternoon (12:01–15:00h) and late afternoon (15:01–18:00h). Observations done between December and April were categorised as wet season and the rest as dry season.

Figure 1.

Distribution of waterpans across the three land-use types. The abbreviated waterpan names are defined in Table 4.

Table 1.

Description of behaviours used to categorise the original activity and responses of waterbirds to disturbances. In the original activity category, F.V = Feeding allowing vigilance, F.nV = feeding not allowing vigilance, nF.V = non-feeding allowing vigilance, nF.nV = non-feeding not allowing vigilance; and in the response categories, Vig = vigilance, noRes = no response and Esc = escape.

Table 2.

Waterbird species (alongside allocated guilds) and frequency of disturbances recorded from 2015 to 2017 in and around Hwange National Park.

As habitat type may influence responses by waterbirds, we used vegetation characteristics and waterpan size data collected in July 2014 and January 2015 as part of a broader survey of waterpans in our study area. We conducted assessments of the surrounding vegetation at each waterpan by establishing 30 m long transects along the cardinal and inter-cardinal directions from the water edge. On each transect we placed a 1-m² quadrat about 10 cm from the water's edge and recorded the grass cover (visual estimate) and representative grass height (using a standard calibrated wooden ruler). We repeated these measurements at 10, 20 and 30 m away from the water's edge for all the eight transects, which is a modification from Trash (2000). We also used a range-finder to measure the distance from the water's edge to the nearest tree line (bush edge) for each transect direction. Surface area and perimeter of each waterpan were calculated by walking right round its shores while marking our tracks on a hand-held GPS. Visual estimations of the percentage of the waterpan covered by emergent and submerged aquatic vegetation approximately 3m from the shore were made (Raeside et al. 2007). The vegetation and waterpan size variables were used to distinguish four groups of waterpans using a Euclidean distance based cluster analysis to produce groups described in Table 3.

Analysis

We excluded from our analysis observations where waterbirds responded to anything that we did not see and those when our monitoring team had initiated the response (these comprised 2.2% of the original dataset). We wanted to specifically test for a group vigilance effect (H2) but also tested for a possible confounding effect with the waterbird guilds, as group size may vary widely among waterbird species and guilds. Although we noted that group sizes were significantly different across waterbird guild (using one way ANOVA), the average group sizes fell within the same order of magnitude (between 12 and 21), and thus any effect of group size was unlikely to be explained uniquely by our waterbird guild. We therefore used a set of independent variables (land use, season, waterbird guild, group size, original activity, disturbance type, waterpan cluster group, waterpan identity and time of day) that we thought could influence the probability of reacting to disturbance and length of time spent flying after a disturbance. Accordingly, we used a mixed-effects logistic regression model in which the response variable was treated as ‘no response' (when waterbirds showed non-vigilance behaviour) or ‘response' (vigilant and/or an escape of some form). The site identity was treated as a random variable, and also in all subsequent mixed-effects models used in this study. Since various response types have different energetic requirements and associated costs in lost feeding or resting opportunities, we created two subsets of our main data to explicitly test for factors associated with each reaction category. We termed the first ‘mild', which excluded observations with escape responses (i.e. restricted to vigilance); and the second ‘extreme’, included individuals that swam/walked/ran or flew away. We used our set of independent variables to investigate the factors that resulted in waterbirds showing ‘mild' responses (i.e. becoming vigilant without moving away from their original position) and ‘extreme’ responses (i.e. when waterbirds engaged in escaping activities) using mixed-effects logistic regressions.

Table 3.

Characteristics of waterpan clusters as distinguished through cluster analysis on waterpans in the study area.

We wanted to further investigate which disturbance would trigger flying, as it is the most energetically costly response. We thus also performed a mixed-effects logistic regression (with our set of independent variables) to determine factors that caused waterbirds to respond by flying away as opposed to remaining at the waterpan. To assess the consequences of this extreme response, we also distinguished waterbirds that responded by flying away and then returned to the waterpan, as opposed to those that did not return in 2 h. For the former, we used our set of independent variables to investigate factors that affected the time spent flying (in seconds) using a generalised mixed-effects model. For the latter, we also used a mixed effect logistic regression to determine the probability of flying away and not returning (given its guild and disturbance type). As there is a conservation issue related to anthropogenic disturbance, not only outside PAs but also mildly inside, we specifically tested for the anthropogenic disturbance effect for terrestrial threats (with the re-categorised disturbance source – people/not people), and we used a generalised mixed-effects model on our set of independent variables to investigate the relative influence of anthropogenic disturbance (across the guilds) on the probability of flying, and time spent flying. Our analysis for time spent flying only included comparisons between HNP and the CA since we did not have enough records for SFA.

For all our analyses, we included single variables and two-way interactions (between waterbird guild and disturbance type, original activity, season, land use and waterpan cluster) according to our main predictions in the models. We used the ‘dredge’ function in the MuMln package (Barton 2011) to select candidate models where delta AIC was less than 2 and then considered the best model from the list as the one with the lowest Akaike information criteria (Burnham and Anderson 2002). All analysis was done in the R package for statistical computing (< www.r-project.org>).

The probability of responding to any disturbance

The best model describing the probabilities of overall responses (i.e. response or no response) only retained the single effects of waterbird original activity, guild and the type of disturbances. Waterbirds were more likely to respond if they were engaged in non-feeding allowing vigilance activities before the disturbance when compared to the rest of the states (Fig. 2a, Supplementary material Appendix 1). Small wildfowl were least likely to respond (Fig. 2b) when compared to the other guilds. Terrestrial disturbances had greater chances of stimulating responses compared to herbivores and aerial disturbances (Fig. 2c).

Probability of mild, extreme and escape responses after a disturbance

The single effects of waterbird original activity, guild and the type of disturbances constituted the best model explaining mild responses (probability of responding by vigilant activities). Waterbirds were likely to respond using mild behaviours if they were engaged in non-feeding allowing vigilance activities prior to the disturbance (Fig. 2d) compared to other activities. Small wildfowl were also less likely to respond with mild responses (Fig. 2e) when compared to the rest of the guilds. Terrestrial disturbances were significantly more likely to stimulate mild responses (Fig. 2f) compared to aerial and herbivores disturbances.

Table 4.

The mean waterbird guild size (± standard deviation) and ranges of across the three areas in and around Hwange National Park. The dash (–) denotes cases where no observation was made.

For extreme responses, waterbirds were most likely to escape if they were engaged in non-feeding allowing vigilance activities prior to the disturbance but least likely in feeding allowing vigilance states (Fig. 2g, Supplementary material Appendix 1). The likelihood of escaping was greater in waders and generalists when compared to small and large wildfowl (Fig. 2h); and when they were disturbed by terrestrial threats (in comparison to herbivores and aerial ones, Fig. 2i). We did not detect any significant effects of land use, seasonality and habitat on the probability of waterbirds to react to disturbances in our study area.

Time spent flying

The best model explaining variation in time spent flying included the interacting effects of guild with land use and disturbance type. Overall, terrestrial and aerial threats stimulated significantly longer periods (35.8 (29.6–43.3 CI) and 27.0 (22.6–32.2 CI) seconds respectively) when compared to herbivore threats (13.7 (11.1–16.8 CI) seconds). Across these disturbance types, large and small wildfowl took significantly more time to return to waterpans after a disturbance when compared to waders and generalist species (Fig. 3a, Supplementary material Appendix 2). On average, waterbirds took significantly longer periods to return to waterpans in the CAs (30.7 s (22.2–42.4 CI)) when compared to those in HNP (18.2 s (15.6–21.0 CI)). However, in both land uses (CAs and HNP), large and small wildfowl took significantly longer periods to return to waterpans after disturbances when compared waders and generalist species (Fig. 3b).

Overall, disturbances by people elicited longer periods of flying compared to those stimulated by non-human sources. Time taken flying after humans and non-human disturbance sources was not different in generalist species; waders and small wildfowl had significantly longer flight periods after disturbances by humans compared to non-human sources (Fig. 4a, Supplementary material Appendix 2). Waterbirds took significantly less time to return to waterpans inside HNP (8.1 (5.7–11.4 CI) s) compared to CAs (20.4 (14.7–28.4 CI) s) as illustrated in Fig. 4b.

The probability of flying away from a disturbance and not return

For the majority of instances (94.5%, 398 cases out of 421 cases of fly-aways), waterbirds generally returned to the waterpans. For most cases when birds did not return to the waterpan (82.6%, 19 out of 23) the disturbance type was aerial. The probability of not returning was significantly higher in large wildfowl (0.357 (0.208–0.341 CI)) compared to the rest of the guilds all of which had probabilities lower than 0.09.

Table 5.

Summary of waterbird disturbances at waterpans in and around Hwange National Park during the study period. The asterix (*) on waterpans denotes those that are pumped. The waterpan names in brackets refer to those used in Fig. 1.

Types of disturbances and associated responses

Terrestrial disturbances triggered the most responses (including escape reactions), although these threats were not the most frequent at our waterpans (Table 5). For the instances that waterbirds flew away from terrestrial threats, individuals took significantly longer time to come back to the waterpan area when compared to aerial and herbivore disturbances. This suggests that terrestrial threats are the least tolerable in our study area and indicates the high levels of risk associated with them. Since our terrestrial disturbance class mostly included humans, noise from vehicles and domestic dogs (all anthropogenic sources contributing to 58.1% of cases), the perceived threats of people seem to be high as was recorded elsewhere (Klein et al. 1995, Banks and Bryant 2007, Krüger 2016, Moore et al. 2016). Probably the most threatening human activities for waterbirds in our study area is domestic dogs and their use in subsistence hunting and noises around waterpans, especially in the CAs (Tarakini et al. 2018). In line with our anthropogenic disturbance effect hypothesis (H3), we expected waterbirds from the PAs (where hunting is prohibited and therefore some possibility of habituation) to have less escape responses from people, but this was not so. Our results suggest that risk levels associated with terrestrial threats are similar in CAs and HNP, and could imply that waterbirds from the park sometimes visit and utilise waterpans in CAs where they are also exposed to human predation (Borgmann 2011). However, for the waterbirds that flew away, those in CAs took significantly longer periods to return to waterpans, thus greater fear could be associated with returning to them when compared to HNP. We also think that the effect of disturbance type overshadowed the habitat type effect (H4) which did not significantly influence the likelihood of responding and the time spent flying.

Figure 2.

Schematic presentation of the likelihood of waterbirds to respond to disturbances plotted from model estimates. The top panel shows the probability of waterbirds to respond, the middle one presents the probability of reacting with a mild response (i.e. becoming vigilant without change of position) and the bottom panel shows the likelihood to show extreme response (i.e. flying away). The plots a, d and g illustrates comparisons across original activities classes; b, e and h shows comparisons across guilds; and c, f and i illustrates differences across disturbance types. Different numeric superscripts (i,ii,iii) represent groups that are statistically different.

Figure 3.

Plots from raw data of time taken flying by the different waterbird guilds (a) after being disturbed by various threat types and (b) across land uses.

Figure 4.

A schematic presentation of raw data illustrating time spent flying by waterbirds after disturbances by humans or not across (a) guilds and (b) land uses..

The waders (which normally forage on the shores and do not swim) responded more to threats compared to those that do swim (small and large wildfowl) and those that normally forage away from shores (generalist). Also, when disturbed by people, waders, which mostly comprised the blacksmith lapwing Vanellus armatus (83.2%, Table 2), took the longest time to return to the waterpans (compared to other guilds). This could be associated with the lower energetic requirements (because of light weight) by these species (Grand 1999) when compared to small wildfowl and generalists. In addition, several waders are known for prolonged territorial flight behaviour especially close to their feeding or nesting areas (Goss-Custard 1970, Coetzer and Bouwman 2017). The tendencies of flying for long periods before returning were also observed in large and small wildfowl, particularly in the CAs where they are normally targeted for meat by humans (Tarakini et al. 2018), but those facing less human predation pressure such as generalists and waders (Hancock et al. 2010) took less time in flying. Our results therefore failed to show a clear waterbird size effect (H1) on waterbird responses to threats.

The disturbances (especially the terrestrial ones) triggered more flight responses from waterbird guilds that do not usually swim (waders and generalists in our case) when compared to others. Previous studies have observed that swimming to safer or unreachable areas is commonly used by waterbirds, especially for small wildfowl, which is an energetically less demanding escape option (England et al. 2015, McFadden et al. 2017). However, the long periods taken by large wildfowl to return after flying away suggest that risk-taking is low (Grand 1999). If these species are constantly under threats that trigger flight, a significant proportion of foraging time could be lost making such sites unfavourable for them (Klein et al. 1995, Blumstein 2006). Therefore, if all the other factors are constant, sites that experience constant disturbances will progressively be selected by communities skewed towards species that can swim (Rajashekara and Venkatesha 2017). Such biases will not meet tourist expectations for greater species diversity (Ian 2000), and may ultimately result in skewed ecosystem processes, for example, some food resources may remain underutilised (Gill et al. 2001, McFadden et al. 2017) or waterbirds may shift their feeding times and locations accordingly (Guillemain and Fritz 2002). The implications of such a scenario could be worse inside the park, especially if tourist numbers visiting waterpans negatively affect waterbird communities.

Returning to a site after a fly away

Only about 5.5% of flying responses did not return to the same waterpan in 2 h after being disturbed. Whether or not waterbirds return to the original site can also depend on the quality of habitat (such as food abundance, cover, the frequency of disturbances when compared to other sites) (Ian 2000, Coetzer and Bouwman 2017). Some breeding waterbirds (constructing nest, tending to eggs or chicks) may be compelled to return to sites after disturbances more than the non-breeding ones (Hernández-Matías et al. 2003, Gayet et al. 2011). However, our study did not reveal differences across seasons (i.e. in contrast to our seasonal effect hypotheses (H5)) but we acknowledge that our sample size was too small to test for the breeding status of each species. Some studies used rates of disturbances and food acquisition to relate to the fitness consequences for waterbirds (Goss-Custard et al. 2006), and have suggested threshold disturbance levels for different scenarios of food availability. Thus further research is needed in our study area to investigate how waterbirds tradeoff reacting to various stimuli under varying food availability and such work will be critical as the area is important for breeding waterbirds (Godfrey 1992).

Our study has shown that in the few instances that waterbirds did not return, it was due to aerial disturbances. The aerial threats frequency was high at Nengasha, Guvalala, Nyamandlovu and Ngweshla (Table 5). The large wildfowl had the highest probability of flying away and not return, which could be related to the greatest flight energy costs related to their body weight. It could be argued that they would be the least likely to engage in a return flight when they have escaped to an alternative waterpan after disturbance (in comparison to the rest of the guilds). Although we found no significant interactive effects of disturbance type and waterbird guild on responses to threats, we think that if raptor pressure increases on waterbirds such species will abandon the high risk sites (Guillemain et al. 2007). This may not only affect the large species but also the small that can be carried away on the wing (Goss-Custard et al. 2006).