The southern yellow-billed hornbill is listed as the least concern by the International Union for Conservation of Nature (IUCN) but are thought to be declining (BirdLife International, 2021). The hornbill’s distribution includes most of southern Africa, with a large portion falling within the Kalahari Desert. They are socially monogamous and live up to 23 years in captivity (Strehlow, 2001), although the oldest known marked bird at our study site was 8 years old. Hornbills are cavity nesters; once sealed into the nest, the females moult all their flight feathers simultaneously. At our study site, pairs usually attempt a single brood per season (though they may be multibrooded in more mesic areas, Stanback et al., 2021), and the mean duration from the first egg lay to the female completing sufficient feather regrowth and breaking out of the nest is 53 ± 6.3 SD days (n = 56, range: 36–73) in nests that successfully produce at least one fledgling (van de Ven et al., 2020a). After the female breaks out of the sealed cavity, she aids in the subsequent provisioning of nestlings.

Research on hornbills has been ongoing at our study site (Kuruman River Reserve in the southern Kalahari Desert, Northern Cape province, South Africa; 26°58′ S, 21°49′ E; hereafter, KRR) since 2008, and breeding data for ten breeding seasons (2008–2019, excluding the summer of 2011–2012 when no data were collected) were used for this study. Hornbills at KRR typically breed between October and the end of March (van de Ven et al., 2020a). Monitoring allowed us to estimate the number of pairs at the site and track if and where they bred. During the decade monitoring period, some pairs were recorded breeding in natural cavities. No breeding data were recorded from these attempts. Breeding primarily occurred in nest boxes, and the breeding data analyzed in this study were exclusively from nest boxes. Wooden nest boxes were available at KRR throughout the study period (43 boxes during 2008–2011, 44 boxes during 2012–2015, 38 boxes during 2015–2017, 33 boxes during 2017–2018, and 36 boxes during 2018–2019). In seasons with low nest box occupancy, pairs were observed to still be present at the site but to skip breeding. We defined the hornbill breeding season as 19 October (the earliest recorded date a female entered a box during the decade of the study) to 24 March (the latest recorded fledge/failure date).

The KRR is an arid savanna dominated by grasses interspersed with large trees (mainly camelthorn, Vachellia erioloba; gray camelthorn, Vachellia haematoxylon; shepherds’ tree, Boscia albitrunca; and buffalo thorn, Ziziphus mucronata). Rainfall occurs primarily in the Austral spring and summer between October and April [mean annual rainfall: 213.9 ± SD 102.0 mm; 1993–2020, Van Zylsrus (VZ) weather station, South African Weather Service (SAWS)]. The spring and summer are hot; T_max [mean maximum daily air temperature (T_air)] between October and March is 34.9 ± SD 1.19°C (1991–2020, VZ), which is above the threshold T_air at which male hornbills showed a 50% likelihood of engaging in heat dissipation behavior, i.e., panting [T_thresh; T_air = 34.5°C](van de Ven et al., 2019).

In 2011, the study site was equipped with a weather station [Hot Birds Research Project (HBRP) Weather Station; Vantage Pro2, Davis Instruments, Hayward, CA, United States] that was set to record T_air (°C), wind speed (m.s^–1), relative humidity (%), and solar radiation (W.m^–2) at 5-min intervals. To create a continuous weather dataset appropriate to the study site spanning the entire hornbill monitoring period (2008–2019), we compared weather data from the HBRP station collected between 2011 and 2020 to those from SAWS stations at VZ, Northern Cape (∼30 km from the study site, data available 1992–2020), and Twee Rivieren (TR), Northern Cape (∼120 km from the study site, data available 1960–2020) to validate the use of VZ and TR weather data as proxies for the missing HBRP weather data prior to 2011.

Weather data from all three stations were highly correlated (see Supplementary Figures 1, 2), but the T_max data from VZ and TR were consistently higher than those from the HBRP (VZ mean: 1.38°C ± 1.54 SD higher; TR, mean: 2.13°C ± 1.79 SD higher; Supplementary Figure 1). Therefore, these values were adjusted (refer to Supplementary Figure 2) before being used to supplement the HBRP weather data collected at the study site. Daily rainfall was not correlated between the three weather stations, but seasonal rainfall was observed as follows: HBRP∼VZ r² = 0.97; HBRP∼TR r² = 0.86 (Supplementary Figure 3), allowing us to establish the occurrence/absence of drought at the study site using VZ rainfall data. The mean breeding season rainfall recorded by the VZ weather station was 151.71 mm (average rainfall recorded during the hornbill breeding season was 19 October–24 March each year from 1993 to 2019). We defined “drought” seasons as those in which rainfall recorded at VZ was < 80% of the long-term seasonal average following Bourne et al. (2020a). Rainfall has a lagged effect on prey abundance at the study site (Doolan and Macdonald, 1997), therefore the cumulative rainfall within the 2 months prior to the start of the breeding season was included in analyses as a proxy for food availability during breeding (Ridley and Raihani, 2007). To analyze long-term weather trends in southern Kalahari, we used unadjusted weather data from all three weather stations.

In the TR weather data, there was a significant inflection in the summer of 1996–1997, after which T_max (t₅₄ = 4.58, p < 0.001, ± 3.5 SE years) and the cumulative DaysT_thresh (t₅₄ = 3.9, p < 0.001, ± 4.1 SE years) began to increase (Figures 1A,B). After summer 1996–1997, TR mean seasonal T_max increased by 1.3°C per decade and cumulative DaysT_thresh by 19.7 days per decade. We did not test for an inflection point in the VZ T_max data; rather, considering the strong correlation between TR and VZ T_max data (rho.c = 0.94, r² = 0.89; Supplementary Figure 1), we applied the same inflection point identified for the TR data to the VZ data. Above the same inflection point, the VZ mean seasonal T_max increased 1.0°C per decade and the cumulative days on which T_max > T_thresh by 19.8 days per decade. Too few years of data were available for the HBRP station to model a trend in T_max; however, given the strong correlation between the VZ and HBRP data (rho.c = 0.94, r² = 0.94, Supplementary Figure 1), a similar warming trend of 1.0°C per decade can be expected for HBRP. No clear long-term pattern in rainfall was identified for either station; however, rainfall records only date back to 1993 (Figures 1C,D). No trends or inflection points were apparent in T_min or cumulative rainfall during the non-breeding season since 1960 and 1995 for TR, respectively (Supplementary Figure 4).

The percentage of nest boxes used (r = −0.80, p = 0.005) and the percentage of nesting attempts that succeeded (r = −0.73, p = 0.017) per season declined over the course of the monitoring period (Table 1 and Figure 2). There was a marginally non-significant decline in the mean number of fledglings produced per breeding pair per season (r = −0.63, p = 0.051, Table 1). We found no evidence of temporal autocorrelation in the number or percentage of nest boxes used, the percentage of nesting attempts that succeeded, or the mean fledglings per attempt, modeled by season.

The percentage of boxes used (r = −0.88, p < 0.001), percentage of successful attempts (r = −0.90, p < 0.001), and mean fledglings per attempt (r = −0.78, p = 0.008) all declined significantly with increasing cumulative DaysT_thresh and were all significantly higher in non-drought seasons compared to drought seasons (p = 0.014, p ≤ 0.001, and p = 0.005, respectively; Table 1 and Figure 3). However, neither percentage of boxes used (r = −0.43, p = 0.220; r = 0.27, p = 0.451), the percentage of successful attempts (r = −0.59, p = 0.075; r = 0.15, p = 0.670), nor the mean fledglings produced per attempt (r = −0.56, p = 0.092; r = 0.22, p = 0.537) were significantly related to the cumulative 2 months’ rainfall preceding the start of the breeding season or the T_min of the preceding non-breeding season, respectively (Table 1).

One best-performing model was identified for variation in breeding success (model weight: 0.96, Supplementary Table 2). This model indicated that the probability of success was negatively correlated with later female entry dates (estimate: −1.10 ± 0.41; p = 0.007; Table 2 and Figure 4). Additionally, there was a significant interaction between the percentage DaysT_thresh during the attempt and the drought occurrence on breeding success (estimate: −2.65 ± 1.04; p = 0.011, Table 2). Post hoc models of breeding success in drought and non-drought years revealed a significant negative correlation between the percentage DaysT_thresh during the attempt and breeding success in non-drought seasons (estimate: −1.87 ± 0.65; p = 0.004), but not in drought seasons (estimate: −0.20 ± 0.67; p = 0.771, Table 2 and Figure 4). In drought seasons, breeding success was uniformly low; only four of 23 (17.4%) attempts succeeded in fledging at least one chick, compared to 40 of 65 (61.5%) in non-drought seasons. Of the 115 breeding attempts, we recorded all 18 attempts that experienced ≥ 72% of the days on which T_max > T_thresh (corresponding to T_max during the attempt ≥ 35.7°C) failed (Figure 4). Based on current warming trends of 1°C per decade (Figure 1), this T_max threshold of 35.7°C will be exceeded throughout the entire breeding season by approximately 2027 at our study site.

One top model was identified for the probability of a breeding attempt successfully progressing from female entry to hatch (model weight: 0.66, Supplementary Table 2), with nests significantly less likely to progress to hatching when females entered the nest later (estimate: −1.24 ± 0.44; p = 0.005, Table 2). The interaction between the percentage DaysT_thresh during the attempt and drought occurrence was significant (estimate −1.55 ± 0.77; p = 0.005, Table 2); the probability of a breeding attempt successfully progressing from female entry to hatch declined significantly with an increasing percentage DaysT_thresh during the pre-hatch period within non-drought seasons (estimate: −1.22 ± 0.00; p = 0.013), but not within drought seasons (estimate: −0.91 ± 0.97; p = 0.310; Table 2) where successful hatches were low; 15 out of 23 (65.2%) attempts hatched in drought seasons, compared to 56 out of 65 (86.2%) in non-drought seasons.

During the nestling period, four top models for breeding attempt success were identified (model weights: 0.30, 0.20, 0.13, and 0.11 respectively; Supplementary Table 2). The model average of these four top models revealed that survival during the nestling period was positively correlated with an increased number of days post-hatch spent in the nest by the female (estimate: 1.08 ± 0.38; p = 0.005) and negatively correlated with an increased percentage DaysT_thresh during the nestling period (estimate: −0.86 ± 0.37; p = 0.021; Table 2 and Supplementary Figure 5). The effect of drought occurrence on survival during the nestling period was not significant in the averaged model (estimate: 0.71 ± 0.92; p = 0.444, Table 2), although 4 of 15 chicks (26.7%) that hatched in drought seasons fledged, compared to 40 of 56 chicks (71.4%) in non-drought seasons (Supplementary Figure 6).

There were two top models identified for the effect of drought occurrence on female entry date, including the null model (model weights: 0.56 and 0.44 respectively; Supplementary Table 2), and on the number of days post-hatch spent in the nest by the female, also including the null model (model weights: 0.55 and 0.45 respectively; Supplementary Table 2). Model averages indicated that drought occurrence did not significantly affect the female entry date (estimate: −0.07 ± 0.11; p = 0.552) or the number of days post-hatch spent in the nest by the female (estimate: −0.03 ± 0.05; p = 0.550) for breeding attempts that hatched at least one chick (Table 2).

A single top model was identified for T_nest, indicating that T_nest inside both nest boxes and natural cavities showed a significant positive relationship with outside T_air (estimate: 1.03 ± 0.00; p = 0.001), but that there was a significant interaction between T_air and the type of nest (estimate: −0.49 ± 0.00; p = 0.001; Supplementary Tables 3, 4). The slope of the relationship between T_nest and T_air was 0.55 T_nest °C.T_air^–1 for natural cavities (estimate: 0.55 ± 0.00; p = 0.001) compared to 1.03 T_nest °C.T_air^–1 for nest boxes (estimate: 1.03 ± 0.00; p = 0.001), suggesting that T_nest within natural cavities was buffered against T_air, while T_nest within nest boxes tracked T_air (Supplementary Table 4 and Supplementary Figure 7). The difference between T_nest within nest boxes and natural cavities was most pronounced at higher T_air; at T_air > T_thresh, the mean T_nest within nest boxes was 39.65°C ± 3.23 SD, while the mean T_nest within natural cavities was 34.91°C ± 3.68 SD (Supplementary Figure 7).

The decline in the percentage of nest boxes used over the decade of monitoring could be a result of fewer potential breeding pairs in the study population as the monitoring period progressed (e.g., Rioux Paquette et al., 2014; Cruz-McDonnell and Wolf, 2016). However, the number of pairs present was continuously estimated to be approximately 20–25 each year, and in the majority of years since 2012, pairs were seen inspecting nest boxes between October and March but failed to ultimately breed (Pattinson and van de Ven, unpublished data). Moreover, there was a significant link between higher seasonal T_air and reduced nest box occupancy. Therefore, the declining breeding effort likely reflects an increasing number of resident pairs skipping breeding in response to increasingly challenging environmental conditions (i.e., higher T_air during the summer breeding season); skipping breeding is common in another southern African bucerotiforme, the southern ground-hornbill (Bucorvus leadbeateri; Carstens et al., 2019), and is suspected to occur in response to poor breeding conditions in various avian, mammalian, and amphibian taxa (e.g., Pietiainen, 1989; Pilastro et al., 2003; Kinkead and Otis, 2007; Keynan and Yosef, 2010; Cayuela et al., 2014; Griffen, 2018). Nesting hornbills experience low rates of predation because the birds seal up the entrance of the nest, leaving only a tiny slit through which food can be passed from the male to the female and chicks inside (Moreau and Moreau, 1941), so changes in predation rates with T_air or drought are unlikely responsible for variation in breeding performance.

Nest boxes were less buffered against changes in T_air compared to natural cavities. Therefore, our findings may represent a more severe response to high T_air than would be evident in birds breeding in natural cavities, especially given the strong effect of T_nest on fledging conditions (van de Ven et al., 2020a). However, the T_air recorded at our field site was consistently lower than that recorded by weather stations in surrounding areas, suggesting that the trends we measured are likely at least indicative of what is happening to populations of hornbills breeding in natural cavities in the hottest parts of their range. Moreover, variation in nest success is also driven by the effects of T_air on the provisioning behavior of the parents (van de Ven et al., 2019; van de Ven et al., 2020a) and drought, which are effects independent of nest type.

Inter-annual and within-season declines in breeding success were strongly associated with both high temperatures and drought. Of the 115 breeding attempts we recorded, none were successful when T_max exceeded T_thresh ≥ 72% or more days during an attempt. The hottest period during which an attempt was successful involved 55 of 76 days (72%) with T_max > T_thresh, but all 18 attempts during hotter conditions failed, indicating the potential for the precipitous decline in breeding success at high T_air. Additionally, significantly more breeding attempts succeeded (fledged at least one chick) in non-drought seasons (61.5%) compared to drought seasons (17.4%). The near-ubiquitous nest failure across a range of T_air in drought years presumably occurred because low rainfall leads to low food abundance in arid and semi-arid systems since rainfall drives primary productivity and the energy available through the entire trophic cascade (Mares et al., 2017; Carstens et al., 2019). These results suggest that breeding performance in this population of hornbill is susceptible to both low rainfall and high T_air independently, compared to studies suggesting vulnerability only when low rainfall and high T_air co-occur (Flesch, 2014; Cruz-McDonnell and Wolf, 2016; Iknayan and Beissinger, 2018; Bourne et al., 2020b). Although our models did not indicate a statistically significant effect of drought during the nestling period, the much lower percent of nestlings that fledged in drought (26.7%, n = 4 out of 15) compared to non-drought (71.4%, n = 40 out of 56) seasons suggested that drought was associated with reduced nest success during both pre-hatch and nestling periods. In fact, the low number of chicks that hatch or fledge during drought seasons creates a limited sample size for analyses, likely reducing our power to statistically demonstrate the effect of drought.

Variation in breeding success within breeding seasons was also correlated to intrinsic factors including female entry date (when the female sealed herself in the nest box) and the number of days post-hatch the female was incarcerated in the nest. Later female entry date and fewer days post-hatch spent incarcerated by the female both correlated with lower breeding success. Delayed nest initiation is associated with drought and low resource availability in some species (McCreedy and van Riper, 2015; Carstens et al., 2019). However, rainfall in the 2 months preceding the breeding season had no significant effect on seasonal breeding output, and neither female entry dates nor the number of days post-hatch spent in the nest by the female parents were significantly related to the occurrence or absence of drought, suggesting that their effects were likely not artifacts of rainfall and food availability (Harriman et al., 2017). Rather, we suspect that lower-quality parents delay the onset of breeding and/or that lower-quality females cannot stay incarcerated for extended periods. Fewer stored body reserves in either parent may preclude beginning breeding early or the ability to sustain continuous mass loss during the breeding attempt (van de Ven et al., 2020a). This pattern reflects widespread interactions between the timing of breeding and the quality of the parents (for review, see Verhulst and Nilsson, 2008) and supports the well-established concept of earlier breeding and higher parental quality positively affecting reproductive performance (Moreno et al., 2005; Verhulst and Nilsson, 2008; de Zwaan et al., 2019).

Our analyses of T_air and rainfall trends in the southern Kalahari revealed trends consistent with those reported in recent studies (Kruger and Sekele, 2013; van Wilgen et al., 2016; Mbokodo et al., 2020); T_max and DaysT_thresh during the breeding season of the hornbills have increased by more than 1.0°C and almost 20 days per decade, respectively, over the last ∼25 years, whereas no long-term trend in rainfall/drought recurrence was apparent. This trend indicates that hornbills face severe challenges to their persistence across the seasonally hot, arid parts of their range. The negative effects of high temperatures will increasingly cause reduced breeding success even in non-drought years as global warming advances. Therefore, while the magnitude of population decline is limited by the severity of droughts and heatwaves, population recovery and persistence will be limited by a decreasing capacity of the hornbills to breed successfully in non-drought seasons due to increased mean T_air (Williams et al., 2016; Albright et al., 2017; Palmer et al., 2017; Conradie et al., 2019). Based on (1) the rapid rate of warming, (2) the fact that no breeding attempts succeeded if ≥ 72% days during the attempt had T_max > T_thresh (corresponding to a T_max during the attempt ≥ 35.7°C, a threshold which will be exceeded across the entire breeding season by ∼2027 at our study site under current warming trends), and (3) 8 years is the longest a wild, color-ringed hornbill has survived in our study population, we arrive at the grim prediction that this population of hornbills could be extirpated by 2040. Moreover, model predictions suggest that the majority of the hornbill’s range will approach the T_air threshold, above which breeding success is < 50% by the turn of the century (Conradie et al., 2019).

Many birds in seasonally hot, summer rainfall, arid zones are constrained to breed in response to, or to coincide with, rainfall, making it difficult for them to shift breeding dates outside of the hottest periods of the year which correspond with the rainfall season (Wolf, 2000; McCreedy and van Riper, 2015; Mares et al., 2017; Iknayan and Beissinger, 2018; van de Ven et al., 2020a). Therefore, even small increases in summer maximum T_air could drive large consequences during breeding (Sinervo et al., 2010; Alagaili et al., 2017; Albright et al., 2017; Riddell et al., 2019; McKechnie et al., 2021). Moreover, an increase in the frequency and severity of sub-lethal, suboptimal conditions that reduce parental quality may also reduce fledgling conditions, generally resulting in lower survival probability, recruitment, and lifetime fecundity (Dawson et al., 2005; Gardner et al., 2016; Conradie et al., 2019; de Zwaan et al., 2019; Bourne et al., 2020a; Cunningham et al., 2021) (but see McLean et al., 2020). In the case of hornbills, there is evidence that the sub-lethal consequences of high T_air (regardless of high rainfall) and drought on the parents affect offspring quality (van de Ven et al., 2020a), and in this study, the probability of successfully fledging offspring or even attempting to breed at all. While the unusual breeding strategy of the hornbills could make them especially vulnerable, high temperatures and drought have negative impacts on breeding output in a variety of taxa across the globe (Welbergen et al., 2008; Sinervo et al., 2010; Gardner et al., 2014; Andreasson et al., 2020; van de Ven et al., 2020b; Cunningham et al., 2021). Therefore, we suggest that our findings are likely applicable to a range of species and support the proposition that even for species where catastrophic heat-related mortality events remain unlikely, climate change can drive rapid declines and potentially local extinctions (Conradie et al., 2019; McKechnie and Wolf, 2019; McKechnie et al., 2021). In fact, both Iknayan and Beissinger (2018) and Riddell et al. (2019) recently demonstrated that this extirpation process in arid zones is already taking place and is set to continue under future climate change.