Thermoregulation in desert birds: scaling and phylogenetic variation in heat tolerance and evaporative cooling

The physiology of temperature regulation in wild birds has been of significant interest to comparative biologists for more than 175 years (reviewed by Kendeigh, 1944). The focus on thermoregulation in desert birds and the challenges that extreme heat and aridity impose on animal function largely began with a series of studies in the 1950s (Bartholomew and Dawson, 1953; Dawson, 1954, 1958). This classic early work recognized that animals were inseparable from their environments and that fitness and survival were contingent on the integration of behavior, morphology and physiology (Kendeigh, 1934; Dawson, 1954). These studies also recognized that understanding thermoregulatory performance required the integration of energy and water balance. Dawson (1954), for example, examined the comparative biology of two species of towhees, one from wetter coastal California and the second from hotter, drier inland deserts, and found distinct differences between the two species in their metabolic and thermal responses to heat exposure and relative ability to offset metabolic and environmental heat loads via evaporative cooling.

Evaporative heat dissipation is vital for avian thermoregulation in hot environments where air temperature (T_a) routinely exceeds normothermic body temperature (T_b) (Fig. 1), as well as in cooler environments where heat generated as a by-product of activity may cause T_b to approach lethal limits (Calder and King, 1974; Nilsson and Nord, 2018). Because evaporation is the only process whereby heat can be lost when environmental temperature is higher than T_b, birds in hot environments face fundamental trade-offs between water conservation and avoiding lethal hyperthermia via evaporative cooling. In addition, trade-offs between thermoregulation and foraging may severely constrain birds’ capacities to maintain body mass and successfully reproduce during hot weather (Conradie et al., 2019; Cunningham et al., 2013; du Plessis et al., 2012; van de Ven et al., 2020). Although sometimes overlooked in models of how heat dissipation may limit fitness (e.g. Nilsson and Nord, 2018; Speakman and Król, 2010; Speakman and Król, 2011), integrating the effects of evaporative heat loss and the effects on total water flux are essential for understanding the evolution of thermal physiology among animals living in hot, arid environments.

Our current understanding of physiological diversity and thermoregulatory patterns among birds exposed to heat lags behind other areas of avian thermal biology. Whereas large data sets have been compiled to examine the scaling of metabolic rates and evaporative water loss (EWL) at thermoneutral T_a (e.g. Lasiewski and Dawson, 1967; Londoño et al., 2015; Song and Beissinger, 2020; Williams, 1996), little is known about how avian evaporative cooling at high T_a and limits to heat tolerance scale with body mass (M_b). In two of the few studies that have taken place, Calder and King (1974) reported that maximum EWL=258.6M_b^0.80 (where EWL is in mg min⁻¹ and M_b is in kg) among 12 species, and Weathers (1981) combined his data for six species spanning M_b of 12–126 g with those of earlier studies to show that the slope of mass-specific resting metabolic rate (RMR) above the upper critical limit of thermoneutrality (T_uc) scales negatively with an exponent of approximately −0.65. More recently, literature data for 27 species were synthesized to model avian EWL at high T_a in the context of increasing water requirements associated with global heating (McKechnie and Wolf, 2010), and Song and Beissinger (2020) analysed the scaling of EWL at T_a=40°C, which these authors assumed to represent EWL at T_a above the thermoneutral zone. A related question that has also received relatively little attention concerns how scaling of EWL and RMR vary among groups differing in the relative importance of panting, gular flutter and cutaneous EWL for heat dissipation (Lasiewski and Seymour, 1972; Smit et al., 2018).

Comparative analyses of avian physiological responses to very high T_a are potentially confounded by methodological differences among studies. Early work on avian thermoregulation in the heat gave little hint of the effectiveness of evaporative cooling. Following work by Terroine and Trautmann (1927), Kendeigh (1939, 1944) and others, Dawson (1954) documented unexpectedly modest heat tolerance, accompanied by pronounced increases in T_b and RMR at air temperatures above 35°C, in two species of towhees (Dawson, 1954). Based on observations that (a) in both towhee species RMR increased substantially 2.5- to 3-fold between T_a=35°C and T_a=42°C, and (b) none of the species investigated by that time was able to evaporatively dissipate more than 50% of metabolic heat production, Dawson (1954) concluded that ‘Panting thus appears to produce too much heat to be an efficient process for heat dissipation at high air temperatures’. One of the major factors responsible for Dawson reaching this conclusion appears to have been chamber humidity; relative humidities were 45–50% at T_a=42°C, conditions that probably severely impeded evaporative heat dissipation. Subsequent studies of evaporative cooling varied substantially in terms of chamber humidity (Coulombe, 1970; Dawson and Fisher, 1969; Schleucher, 1999; Tieleman et al., 2002; Weathers and Greene, 1998), a variable that exerts a strong influence on interactions between T_b, RMR and EWL at high T_a (Gerson et al., 2014; Lasiewski et al., 1966; van Dyk et al., 2019).

Recently, data on heat tolerance limits and maximum evaporative cooling capacity have become available for 56 bird species spanning a M_b range of 8–670 g and representing 14 orders (Table S1), all collected under similar conditions of high flow rates and consequently low chamber humidities (<1 kPa, equivalent to a relative humidity of 13.6% at T_a=40°C, and often considerably lower). These studies have focused on eliciting maximum rates of EWL and thereby determining upper thermoregulatory limits in terms of T_b and T_a tolerance. The species are all arid-zone birds from southwestern USA (14 species), arid Australia (nine species) or the Kalahari Desert and Nama Karoo regions of southern Africa (33 species). We use these data to (a) examine the scaling of variables related to avian heat tolerance and quantify variation associated with phylogenetically correlated traits such as the primary avenue of evaporative heat dissipation (panting, gular flutter or cutaneous evaporation), (b) identify correlations between functional traits and (c) quantify the potential error associated with using allometrically predicted values to model water requirements for evaporative cooling.

An improved understanding of the scaling of physiological variables related to evaporative cooling and heat tolerance, and the degree to which these traits are evolutionarily conserved across taxa, is directly relevant to modelling both acute, lethal and chronic, sublethal consequences of exposure to increasingly frequent and severe heat waves. Recent analyses have revealed that the arid-zone avifaunas of several of Earth's major desert systems are likely to experience greatly increased risks of mortality during extreme heat events (Albright et al., 2017; Conradie et al., 2020). In addition, predictive equations for evaporative water requirements and energy demands during hot weather will be useful for modelling trade-offs between thermoregulation, foraging and nest provisioning (e.g. du Plessis et al., 2012; Sharpe et al., 2019; van de Ven et al., 2020) in species for which detailed physiological data are currently lacking.

We compiled data on avian resting metabolism, evaporative cooling, T_b and heat tolerance for one species of quail (Smith et al., 2015), five caprimulgids (O'Connor et al., 2017; Talbot et al., 2017), six columbids (McKechnie et al., 2016b; Smith et al., 2015), two owls (Talbot et al., 2018), two parrots (McWhorter et al., 2018), one sandgrouse (McKechnie et al., 2016a), one cuckoo and one roller (Smit et al., 2018) and 29 passerines (Czenze et al., 2020; Kemp and McKechnie, 2019; McKechnie et al., 2017; Smit et al., 2018; Smith et al., 2017; Whitfield et al., 2015) from recent literature (Table S1). In addition, we included unpublished data for one species of each of the following: owl, sandgrouse, barbet, swift, mousebird, kestrel, thick-knee and passerine (Table S1).

In brief, all these data were collected using flow-through respirometry under conditions of low chamber humidity (dewpoints typically ∼0°C) maintained using higher-than-typical flow rates (for details, see Whitfield et al., 2015). Birds were held in the dark and exposed to stepped profiles of increasing T_a in increments of 2°C when T_a>40°C, spending 10–30 min at each T_a value until stable T_b, EWL and RMR were reached. These conditions cannot be considered steady-state and the experimental protocol is analogous to the sliding cold exposure protocol used to elicit summit metabolism (Swanson et al., 1996). Steady-state measurements similar to those when quantifying parameters such as basal metabolic rate, where birds typically experience a single T_a value for 6–8 h, are not feasible when T_a is well above T_b and EWL may exceed 5% M_b h⁻¹. Measurements ended when birds showed a loss of balance, lack of coordination or thermoregulatory failure, after which they were allowed to recover with ad libitum access to water and food before being released at their sites of capture. One study in which individuals were monitored for several weeks after release revealed no obvious adverse effects of the experimental protocol (Kemp and McKechnie, 2019).

From each study we obtained minimum values for T_b, EWL and RMR at thermoneutrality (typically 30–35°C), inflection T_a values above which each variable increased, the slopes of relationships at T_a above inflections and maximum values associated with thermal endpoints for calm birds, which were taken in all the studies as the heat tolerance limit (HTL), the T_a associated with loss of balance, coordination or rapid, uncontrolled increases in T_b, or sudden decreases in EWL or RMR (Whitfield et al., 2015). In these studies, the relationship between EWL and T_a was modelled using linear models at T_a above an inflection point. Although a number of authors have argued that the increase in EWL at high T_a is exponential rather than linear (e.g. Weathers, 1981, 1997), comparisons of the explanatory power of linear and polynomial models suggest little difference (Whitfield et al., 2015), and linear models greatly simplify comparisons among and within species.

A recent comparison of interspecific differences in thermal physiology between regularly drinking and non-drinking passerines from southern Africa revealed that HTL is correlated with evaporative scope, the ratio of maximum EWL to minimum EWL at thermoneutrality (Czenze et al., 2020). To test whether this correlation holds among passerines more broadly, we tested for significant relationships between evaporative scope and residual HTL (HTL scales significantly with M_b; Fig. 2) and between the maximum ratio of evaporative heat loss (EHL) to metabolic heat production (MHP). To evaluate relationships between T_uc and increases in EWL above minimal levels and the onset of panting, we also fitted linear models to residual T_uc and inflection T_a for EWL, and residual T_uc and the T_a associated with the onset of panting. We restricted these analyses to the 30 passerines in our data set to avoid the potentially confounding effects of EHL via gular flutter or rapid cutaneous evaporation that predominates in several non-passerine orders.

To quantify the potential error associated with predicting ecologically relevant parameters related to risk of lethal dehydration during extreme weather events on the basis of allometrically predicted rates of EWL rather than species-specific empirical data, we compared cumulative 6 h water loss and time to lethal dehydration [cumulative EWL >15% of M_b, following Albright et al. (2017) and Conradie et al. (2020)] using both approaches. We modeled evaporative water requirements using a T_a profile corresponding to an extremely hot day with a T_a maximum of ∼48°C (data from https://cals.arizona.edu/azmet/), for each of 27/30 passerines in our data set. We excluded curve-billed thrasher (Toxostoma curvirostre) on account of an atypically high EWL inflection T_a (Smith et al., 2017) and white-browed sparrow-weaver (Plocepasser mahali) and sociable weaver (Philetairus socius) on account of EWL inflection T_a values not being reported in the original study (Whitfield et al., 2015). For each species, we modelled cumulative EWL during the 6 h period between 09:00 and 15:00 h using the actual slope of EWL against T_a, and then using the slope predicted from a regression model fitted to all passerine data in our data set. We took the actual and predicted survival times for each species as the period over which cumulative EWL remained ≤15% M_b. We used linear regressions forced through the origin to model relationships between predicted and actual values, and calculated 95% prediction intervals for these models following Cooper and Withers (2006).

To quantify relationships between physiological variables and M_b and to test for correlations between variables, we initially fitted linear regression models in R (http://www.R-project.org/). For variables that scaled significantly with M_b, we used the R package ape (Paradis and Schliep, 2018) to test for phylogenetic signal using the parameter λ (Freckleton et al., 2002). We used a consensus phylogenetic tree generated from 1000 trees downloaded from http://www.birdtree.org (Jetz et al., 2012) using the Hackett et al. (2008) phylogeny as a backbone. From these trees, a 50% majority-rule consensus tree was calculated using the consensus function in ape, as recommended by Rubolini et al. (2015). As significant phylogenetic signal was present for all variables examined, we present phylogenetically informed models based on phylogenetic least squares regressions using the ‘gls’ function in ape. All phylogenetic analyses were performed following Garamszegi (2014). Coefficients are presented with standard error of the mean (s.e.m.) and sample sizes (N) refer to number of species.

The upper limits of T_b associated with the loss of balance or capacity for coordinated locomotion during acute heat exposure among arid-zone birds ranged from 40.7°C in freckled nightjars to 45.2°C in fork-tailed drongos, varying substantially among taxa (Table 1; Fig. 2). There was no overall effect of M_b on maximum T_b, but within passerines maximum T_b scaled significantly and negatively with log₁₀ M_b (coefficient=−0.814±0.276, t=−3.045, P=0.005, λ=0.148) (Fig. 2). The difference between normothermic and maximum T_b (i.e. ΔT_b) ranged from 0.9°C (mourning dove) to 6.2°C (Australian owlet-nightjar and little swift) and did not scale significantly with M_b in the overall data set (t=−1.952, P=0.056), but showed significant negative scaling within passerines (coefficient=−1.374±0.401, t=−3.429, P=0.002, λ=0.46). The slope of T_b as a function of high T_a varied ∼14-fold from 0.041 to 0.576°C T_b per °C T_a and was not significantly related to M_b.

Higher maximum T_b among passerines is consistent with the more pronounced use of facultative hyperthermia by songbirds compared with other orders (Gerson et al., 2019). A somewhat unexpected pattern to emerge from our analysis is that increases in T_b above normothermic levels are similar in magnitude in caprimulgids and passerines (Table 1); the fact that maximum T_b in the songbirds remains well above that of the nightjars reflects the nightjar’s considerably lower mean normothermic T_b (38.5 versus 40.7°C). We also found passerine maximum T_b scaled negatively with M_b (Fig. 2) as a result of larger T_b increases in smaller species, with the expected maximum value for a 100 g songbird ∼1°C lower compared with a 10 g species.

The upper limits to T_b among arid-zone species we report here were associated with loss of balance and coordinated locomotor function. They are thus broadly consistent with criteria for critical thermal maxima (Lutterschmidt and Hutchison, 1997), although they should be viewed as conservative estimates of maximum values that avoid lethal consequences (McKechnie and Wolf, 2019). Maximum T_b values among these arid-zone species are lower than the known range of avian lethal T_b, 46–48°C (Arad and Marder, 1982; Dawson, 1954; Randall, 1943). A few authors have reported tolerance of T_b approaching 46°C (Dmi'el and Tel-Tzur, 1985), 47°C (Weathers, 1997) or even 48–49°C (Freeman et al., 2020) in passerines without any apparent adverse effects, but arid-zone species generally appear not to attain such high values.

Heat tolerance limits (i.e. the T_a associated with compromised locomotor ability or thermoregulatory failure) varied by 16°C from 46 to 62°C (Table 1), with significant positive scaling in the overall data set (coefficient=4.867±1.038, t=4.687, P<0.001, λ=1.04; Fig. 2). HTL also scaled significantly within passerines (coefficient=3.873±1.118, t=3.463, P=0.002, λ=0.88; Fig. 2). The HTL values we analyse here represent the maximum T_a values that birds endured for short periods. In addition to confirming that HTLs are consistently much higher in caprimulgids and columbids compared with songbirds, our analysis reveals that HTL scales positively with M_b in both the overall data set and within passerines. This finding is surprising as, a priori, the higher surface area to volume ratios of smaller species might be expected to permit more rapid heat dissipation and hence tolerance of higher environmental temperatures. One possibility is that this pattern arises from the non-steady-state nature of the stepped profile of progressively higher T_a used in all the studies included here and the greater thermal inertia of larger birds. However, the fact that the largest species for which data are available, a 434 g thick-knee and 668 g owl, had ΔT_b and maximum T_b values similar to those of much smaller species (Fig. 2) argues against the possibility of these scaling patterns representing an experimental artefact.

The evolutionarily conserved differences in HTL among taxa are, we suspect, functionally related to broad ecological differences. For instance, the high HTLs of several nightjars are consistent with these nocturnal birds sometimes spending the entire diurnal period on the ground exposed to intense solar radiation and operative temperatures 10–15°C above T_b, particularly when breeding (Grant, 1982; O'Connor et al., 2018). The even higher average HTL of doves and pigeons, reflecting their capacity to rapidly dissipate heat cutaneously at little metabolic cost (Calder and Schmidt-Nielsen, 1966; Webster and Bernstein, 1987), is consistent with the ground-foraging habits of granivorous species, their ability to depress T_b to cool their eggs during incubation (Russell, 1969; Walsberg and Voss-Roberts, 1983) and their strong dependence on surface water (Fisher et al., 1972; Willoughby and Cade, 1967).

Our analysis of thermoregulatory traits among passerines confirms the correlation between evaporative scope and HTL recently reported for 17 southern African species by Czenze et al. (2020) is also apparent in a data set spanning two other arid regions, Australia and North America. The weaker but nevertheless significant relationship between maximum EHL/MHP and HTL further suggests that the comparative efficiency of evaporative cooling processes is also a determinant of the highest environmental temperatures songbirds can tolerate (Fig. 3).

Minimum EWL at thermoneutral T_a scaled positively with log₁₀ M_b (Fig. 4; coefficient=0.825±0.061, t=13.423, P<0.001, λ=0.46). Significant scaling was also evident for passerines (coefficient=0.808±0.102, t=7.872, P<0.001, λ=0.58) and columbids (coefficient=1.108±0.255, t=4.349, P=0.012, λ=0.77). Overall scaling of maximum EWL involved a very similar exponent (Fig. 4; coefficient=0.801±0.033, t=24.004, P<0.001, λ=−0.12). Maximum EWL of passerines had a shallower slope (coefficient=0.708±0.058, t=11.916, P<0.001, λ=−0.20), as did that of doves (coefficient=0.740±0.134, t=5.537, P=0.005, λ=−0.12) (Fig. 4). Both maximum EWL and minimum EWL expressed as percentages of M_b decreased with increasing M_b (Fig. 5).

Inflection T_a values above which EWL increased varied by >10°C, with values for most species between 36 and 43°C. These inflection T_a values did not scale significantly with M_b. The slope of increasing EWL above these inflection points was strongly dependent on M_b in the overall data set (coefficient=0.631±0.051, t=12.426, P<0.001, λ=0.76), with significant effects of M_b for passerines (coefficient=0.580±0.075, t=7.770, P<0.001, λ=0.54) and columbids (coefficient=0.730±0.163, t=4.466, P=0.011, λ=0.44). Evaporative scope varied widely, from 3.1 in pyrrhuloxias to 18.4 in sociable weavers (P.socius), with considerable overlap among taxa (Fig. 6).

The remarkably similar scaling of thermoneutral EWL and maximum EWL during acute heat exposure (exponents of 0.825 and 0.801, respectively; Fig. 4) occurs despite large differences in evaporative scope among and within taxa (Fig. 6). Minimum EWL predicted from a regression fitted to our data is considerably higher (75% for a 10 g species, 108% for a 100 g species) than values predicted for desert birds by Williams (1996). This difference, we suspect, reflects several factors. First, data included in the present analysis were typically active-phase measurements at T_a≈35°C, compared with the combination of active- and rest-phase measurements at T_a≈25°C analysed by Williams (1996). Second, although Williams (1996) excluded measurements from studies involving low flow rates and hence elevated chamber humidities, a cursory examination of his sources suggests that, in general, humidity was often well above the very low levels characteristic of the present data set.

Calder and King (1974) presented an analysis of maximum EWL in 12 species ranging in M_b from 11 g zebra finches to 88 kg ostriches. The scaling exponent these authors reported (0.80) is identical to ours, but the intercept of our equation is 70% higher (Fig. 4). The negative scaling of maximum EWL expressed as a percentage of M_b (Fig. 5), such that a 5 g bird loses the equivalent of 7.6% M_b h⁻¹ compared with 3.0% M_b h⁻¹ in a 500 g bird, is the physiological basis for smaller species being more vulnerable to lethal dehydration during extreme heat events compared with larger species (Albright et al., 2017; McKechnie and Wolf, 2010). Our analysis of EWL slope at high T_a supports the findings of Song and Beissinger (2020), but also reveals taxonomic variation whereby EWL increases more rapidly in passerines compared with similarly sized columbids (Fig. 4), reflecting the lower efficiency of panting compared with predominantly cutaneous evaporation. For a given incremental increase in T_a, a greater increase in EWL is necessary in songbirds to dissipate environmental heat loads and the additional metabolic heat generated by panting.

Evaporative scope broadly overlaps across taxa, with both the highest and lowest values in our data set occurring in passerines (Fig. 6), a taxon that lacks gular flutter or major cutaneous contributions to evaporative cooling (Dawson, 1982; Wolf and Walsberg, 1996). This high degree of overlap suggests that the primary avenue of evaporation per se (i.e. panting, gular flutter or cutaneous evaporation) is not the major determinant of the maximum rate of EWL a bird can achieve. The only group whose average evaporative scope is noticeably higher than those of other taxa is the caprimulgids (Fig. 6), in which gular flutter predominates (Cowles and Dawson, 1951; Dawson and Fisher, 1969; Talbot et al., 2017), but this probably reflects the disproportionately large gapes of nightjars. The broad similarities in evaporative scope across taxa varying in the primary avenue of evaporative heat dissipation contrast markedly with taxonomic variation in maximum EHL/MHP (Fig. 6) and HTL (Fig. 2), revealing that the relative metabolic costs of heat dissipation pathways are the primary determinant of upper limits to evaporative cooling capacity and heat tolerance.

Recent studies predict large increases in the risks of lethal dehydration during hot weather for birds inhabiting the deserts of southwest North America and arid Australia when the cumulative evaporative cooling requirements of resting birds in shaded microsites exceed lethal dehydration limits (Albright et al., 2017; Conradie et al., 2020). Our comparison of cumulative EWL and associated survival times based on species-specific empirical data with those estimated using allometrically predicted values suggests that predicted values are often suitable for estimating water requirements of species for which neither empirical data nor detailed biophysical models (e.g. Kearney et al., 2016) are available. In a few cases, however, species deviate substantially from these predicted values. For example, using allometrically predicted values (Fig. 5) resulted in a substantial under-estimate of cumulative water losses and an over-estimate of survival time by common fiscals and an under-estimate of survival time in the red lark (Calendulauda burra), a threatened species restricted to a small area in southern Africa (Dean and Ryan, 2005).

Minimum RMR at thermoneutral T_a scaled positively and significantly with M_b (Fig. 7; coefficient=0.688±0.032, t=21.775, P<0.001, λ=0.98) in the overall data set, as well as within passerines (coefficient=0.715±0.043, t=16.496, P<0.001, λ=0.77) and columbids (coefficient=0.827±0.046, t=18.062, P<0.001, λ=1.12). Maximum RMR at high T_a scaled positively with M_b (Fig. 7, coefficient=0.748±0.052, t=14.275, P<0.001, λ=0.92). Maximum RMR also scaled significantly with M_b in passerines (coefficient=0.738±0.055, t=13.34, P<001, λ=0.63) and columbids (coefficient=0.788±0.000, t=309,704.05, P<0.001, λ=1.98).

Minimum daytime RMRs for passerines were 30–50% higher than values predicted from the overall data set, qualitatively consistent with the ∼12% significantly higher basal metabolic rates of passerines compared with non-passerines (Londoño et al., 2015). Maximum RMR shows much greater variability around the phylogenetically independent regression line (Fig. 7). It is striking, for instance, that the maximum RMR values of several caprimulgids were below the minimum RMR of most species. The maximum RMR of 0.317 W in the 67 g freckled nightjar is equivalent to just 19.6% of the maximum RMR of the 71 g curve-billed thrasher, reflecting the extremely high efficiency of gular flutter in caprimulgids compared with panting in passerines.

Inflection T_a values for RMR, which we interpret as T_uc, varied widely from ∼29°C in lark-like buntings (Emberiza impetuani) to ∼48°C in spotted eagle-owls (Bubo africanus; Fig. 7). In addition, significant inflections in RMR were evident only at T_a>50°C in three species of caprimulgids. T_uc scaled positively and significantly with M_b (Fig. 7; coefficient=7.045±1.638, t=4.300, P<0.001) and although there were differences in slope among orders, there was low phylogenetic signal (λ=0.99). In passerines, T_uc scaled significantly and positively with M_b (coefficient=6.130±2.078, t=2.952, P=0.006, λ=0.34), and the same was true for columbids (coefficient=16.405±3.200, t=5.127, P=0.007, λ=1.33).

The positive scaling of T_uc among passerines and columbids (Fig. 6) supports Weathers’ (1981) observation that the interval between T_uc and the T_a at which T_a=T_b decreases with increasing M_b. Our analysis also offers an opportunity to evaluate recent suggestions that the T_uc of endotherms provides an index of upper thermal tolerance appropriate for global-scale analyses of climate change vulnerability (Araújo et al., 2013; Khaliq et al., 2015, 2014). Even after excluding caprimulgids, which have exceptionally shallow slopes of RMR against T_a and discernible inflections only at T_a≥50°C, there remains interspecific variation in T_uc of ∼20°C (Fig. 7). The magnitude of this variation among species occupying similar climates and similar latitudes, together with the marked differences among taxa in evaporative cooling efficiency at T_a>T_uc, argues against the notion that T_uc is a meaningful proxy for avian upper thermal tolerance limits (Khaliq et al., 2014, 2015, 2017). Moreover, ∼40% of species in our data set have T_uc>40°C. This fraction is far greater than that shown in fig. 1 of Khaliq et al. (2014), reflecting the inclusion by these authors of a large number of data from studies where RMR was not measured over a wide enough range of T_a to reliably estimate T_uc (Wolf et al., 2017), as well as the combination of active- and rest-phase values included in the former study. Our analysis also reveals that T_a=40°C often does not fall above avian T_uc, as recently assumed by Song and Beissinger (2020).

One of the clearest patterns of evolutionarily conserved variation to emerge from our analysis concerns the slope of RMR at T_a>T_uc. Models fitted to slopes of RMR as a function of T_a>T_uc revealed significant variation in intercepts, indicating that separate scaling relationships are needed for different taxa (Fig. 7). RMR slope scaled significantly and positively with M_b in passerines (coefficient=0.856±1.121, t=7.070, P<0.001, λ=0.78) and columbids (coefficient=1.379±0.324, t=4.255, P=0.013, λ=1.62). Slopes are five to 10 times higher in passerines compared with columbids, with virtually no increase in RMR above a T_uc in several caprimulgids. In a comparative analysis of thermoregulation in 19 passerines and seven non-passerines, Weathers (1981) found that the slope of mass-specific RMR at high T_a (in mW g⁻¹ h⁻¹ °C⁻¹), which he designated the coefficient of heat strain, scaled to 12.5M_b^−0.65. Fitting a conventional regression to mass-specific slopes for the passerines in our data set revealed significant negative scaling to M_b^−0.36, with predicted slopes 2.4- to 4.7-fold higher over the M_b range of 10–100 g compared with those predicted by Weathers’ equation. We suspect one reason for these marked differences concerns the ranges of T_a over which RMR was measured: Weathers’ (1981) data for six species involved maximum T_a values between 42 and 44.5°C, with similar T_a ranges in many of the published studies from which he obtained data. Consequently, few of the slopes included by Weathers (1981) are based on conditions where birds were maintaining T_b<T_a.

Maximum EHL/MHP varied from 1.20 in yellow-plumed honeyeaters (Lichenostomus ornatus) to 5.49 in Namaqua sandgrouse (Pterocles namaqua). Considerable among-taxa variation was evident, with values for caprimulgids and columbids consistently and substantially higher than those for passerines (Fig. 6). The range of maximum EHL/MHP values for other orders (1.39–5.49) exceeded the combined range for passerines, owls, nightjars and doves (1.20–5.15; Fig. 5).

Among the passerines in our data set, residual HTL was significantly correlated with both evaporative scope (coefficient=0.352±0.077, t=4.592, P≤0.001, λ=0.487) and maximum EHL/MHP (coefficient=2.976±1.026, t=2.901, P=0.007, λ=0.361; Fig. 3). The T_uc occurred at T_a values 3.2±3.6°C lower (N=20, range 10.8°C lower to 3.2°C higher) than the onset of panting, and 1.9±3.2°C (N=27, range 9.4°C lower to 3.5°C higher) lower than the inflection T_a above which EWL increased above baseline levels. Both T_a associated with the onset of panting (coefficient=0.903±0.377, t=2.396, P=0.028, λ=1.066) and EWL inflection T_a (coefficient=0.795±0.203, t=3.923, P<0.001, λ=0.648) were significantly correlated with residual T_uc (Fig. 3).

Comparisons of cumulative EWL over a 6 h period on an extremely hot day estimated using actual EWL slopes and predicted slopes for passerines (Fig. 4) revealed that actual and predicted values were typically similar, but in a few instances differed substantially (Fig. 5). Predicted cumulative EWL values for the 27 species we evaluated were all within the 95% prediction intervals, with the exception of common fiscal (Lanius collaris) for which the actual value was equivalent to 175% of that estimated using an allometrically predicted slope (Fig. 5). For estimated survival times before cumulative EWL exceeds 15% M_b, predicted versus actual values fell within the 95% prediction intervals except for common fiscal (actual=72% of predicted) and red lark (C.burra; actual=140% of predicted) (Fig. 5).

Our analysis of thermoregulation during acute heat exposure among 56 arid-zone birds reveals several novel scaling patterns, including the positive scaling of HTL, and confirms scaling patterns identified previously for variables such as maximum EWL (Calder and King, 1974), slope of EWL (Song and Beissinger, 2020) and slope of RMR (Weathers, 1981). However, the present study also reveals large differences among orders in these variables, particularly the slope of RMR. Moreover, because the studies from which we obtained data involved methods designed to elicit the upper limits of heat tolerance and evaporative cooling, this review confirms the presence of consistent taxonomic differences in HTL and maximum EHL/MHP. Caprimulgids and columbids emerge as the most heat-resistant taxa, although there is evidence for similarly pronounced heat tolerance in representatives of some other non-passerine taxa. The broad overlap among taxa in terms of evaporative scope, but a lack of such overlap in the scaling of the slope of RMR and the large differences in maximum EHL/MHP reveals that, on a broad scale, upper limits to avian heat tolerance are constrained by the metabolic costs of heat dissipation mechanisms to a much greater extent than by limitations on the capacity to increase EWL above baseline levels.

These patterns of scaling and phylogenetic variation among physiological variables related to evaporative cooling and heat tolerance have several implications for modelling physiological aspects of avian responses to climate change. First, they reiterate the size dependence of cumulative EWL and hence survival time on extremely hot days, with smaller species reaching dehydration tolerance limits sooner than larger species (McKechnie and Wolf, 2010). The relationship between heat tolerance limits (i.e. maximum T_a tolerated during acute heat exposure) and M_b, among passerines in particular, also suggests that smaller species are more vulnerable to lethal hyperthermia during extreme heat events. In addition to these general scaling effects, our analysis of evolutionarily conserved variation suggests that lethal effects of extreme heat will vary across taxa. Passerines, on account of the relatively low efficiency of panting, may be expected to be disproportionately affected during heat-related mortality events, a prediction supported by the prominence of passerines and parrots (which also lack gular flutter or pronounced cutaneous evaporation) in historical accounts of mortality among Australian species (e.g. Finlayson, 1932; McGilp, 1932; see also Conradie et al., 2020). Taxa such as columbids and caprimulgids are better able to tolerate extreme T_a during such events, particularly if water is available. Even in the absence of drinking, however, estimated survival times are substantially longer for columbids and caprimulgids compared with comparatively sized passerines.

One important limitation of our analysis is that it focuses solely on arid-zone birds, and involves data collected at low humidities. Avian evaporative cooling in more mesic habitats may be severely constrained by higher humidity levels, particularly in humid tropical lowlands (Weathers, 1986, 1997). Relatively few studies have investigated interactions between evaporative cooling and experimentally manipulated humidity (Gerson et al., 2014; Powers, 1992; van Dyk et al., 2019) and much remains to be explored in this area. Similarly, few data are available on limits to evaporative cooling and heat tolerance along non-desert species or how these traits vary across broad biomes (Milne et al., 2015). It is also noteworthy that there are a small number of reports of maximum avian T_b being substantially higher than the range for arid-zone species. Weathers (1997) documented tolerance of T_b=47°C in a passerine from Panama's lowlands and argued that tolerance of such a high T_b may evolve in response to the constraints imposed by high humidity on evaporative cooling. Tolerance of even more extreme hyperthermia has recently been documented in an African ploceid passerine, the red-billed quelea (Quelea quelea), with maximum T_b averaging 48.0°C and individual values of up to 49.1°C (Freeman et al., 2020). Data from species across a range of habitats are needed to identify global patterns in limits to avian evaporative cooling and inform models of species’ sensitivity (Williams et al., 2008) to climate change.

We thank all the students and field assistants who, over the last decade, contributed to the collection of the data included in this article. We also thank Craig Franklin and Hans Hoppeler for the invitation to contribute to this special issue, and two anonymous reviewers who provided constructive comments on an earlier version of the manuscript.