1We use a large interspecific data set on diving variables for birds and mammals, and statistical techniques to control for the effects of phylogenetic non-independence, to assess evolutionary associations among different elements of diving behaviour across a broad and diverse range of diving species. Our aim is to assess whether the diving ability of homeothermic vertebrates is influenced by factors other than the physiology of the species.
2Body mass is related to dive duration even when dive depth is controlled for and thus for a given dive depth, larger species dive for longer. This implies that larger species have a greater capacity for diving than is expressed in their dive depth. Larger animals that dive shallowly, probably for ecological reasons such as water depth, make use of the physiological advantage that their size confers by diving for longer.
3Dive duration correlates with dive depth more strongly than with body mass. This confirms that some animals are poor divers for their body mass, either because of a lower physiological capacity or because their behaviour limits their diving.
4Surface duration relates not only to dive duration but also to dive depth, as well as to both independently. This indicates a relationship between dive depth and surface duration controlling for dive duration, which suggests that deeper dives are energetically more expensive than shallow dives of the same duration.
5Taxonomic class does not improve any of the dive variable models in the present study. There is thus an unsuspected consistency in the broad responses of different groups to the effects on diving of the environment, which are therefore general features of diving evolution.
Halsey et al. (2006b) found that the exponents of the relationships between mass and dive depth, dive duration and surface duration were close to one-third. This is as predicted by the oxygen store–usage hypothesis (see Butler & Jones 1982) assuming that oxygen stores scale isometrically with mass and oxygen usage scales closely to the two-thirds power of mass (as many studies show metabolic rate to do, e.g. Heusner 1991; White & Seymour 2003, 2005; McKechnie & Wolf 2004). Thus, the allometry of diving variables suggests that the diving of animals is in part influenced by physiology through the ability to store oxygen. However, this phylogenetic analysis also revealed considerable residual variation around the allometric relationships, as has been noted by certain previous allometric studies (e.g. Costa 1991; Schreer & Kovacs 1997). For example, the Grey Whale, Eschrichtus robustus, and Sperm Whale, Physeter catodon, have similar body masses, yet the Sperm Whale dives to much greater depths (e.g. Watkins et al. 1993 report a mean dive depth of 629 m) than the Grey Whale (e.g. Malcolm & Duffus 2000 report a mean dive depth of 13·8 m). If body mass is indeed a good surrogate for the physiological capacity to dive, as allometric studies imply, this suggests that factors other than physiology may also have an important influence on diving behaviour. If so, that influence may similarly be detectable through comparison of variables relating to diving across a broad range of aquatic birds and mammals.
Another obvious factor besides physiology that may influence diving behaviour is the ecology of the species. For example, the behavioural differences between Grey and Sperm Whales may say less about the physiological capacities of the species and more about differences in their feeding ecologies. Grey Whales feed in shallow coastal waters and so rarely have either the need or opportunity to dive to greater depths, whether or not they are capable of doing so (see Harvey & Mate 1984). An example of the difference between actual behaviour and potential in diving species was provided by Ridgway et al. (1984), who showed that Beluga Whales, Delphinapterus leucas, were capable of making much deeper and longer dives than they had been observed making naturally. On the assumption that diving animals remain submerged for as long as they can in order to maximize their foraging efficiency (Kramer 1988; Houston & Carbone 1992; Halsey & Butler 2006c), we might expect to see general relationships among diving variables that reflect the responses of behaviour to ecology as well as to physiology (see also Costa et al. 2004; Arnould & Costa 2006). The aim of the present study is thus to identify general evolutionary associations in the diving behaviour of birds and mammals, analogous to those already established between most metrics of diving behaviour and body mass (Trillmich et al. 1991; Costa 1993; Boyd & Croxall 1996; Schreer & Kovacs 1997; Watanuki & Burger 1999; Halsey et al. 2006b), and to interpret those associations in terms of potential determinants of diving behaviour.
The kinds of general relationship between diving variables that might be expected if ecology is influencing behaviour can be illustrated by means of examples. Consider first the determinants of dive duration. Dive duration is likely to increase with dive depth because it takes longer to reach deeper waters, so we would expect a positive relationship between these two variables (e.g. Costa & Gales 2003). However, this relationship is likely to be influenced by the physiology of the species concerned (e.g. Costa 1993). In particular, given that larger species have larger oxygen stores relative to the rate at which they use oxygen (Butler & Jones 1982), they should be able to dive deeper and for longer (and this has been confirmed by Halsey et al. 2006b). If the relationship between dive duration and dive depth is causal (deep dives take a long time regardless of body mass), then dive depth should be a better predictor of dive duration than body mass. This would suggest a general influence of ecology on dive duration, over and above the effect of physiology. On the other hand, if the relationship is largely driven by the covariation of both diving variables with mass (deep dives are of long duration because larger animals tend to dive deeper and tend to dive for longer during a dive to a given depth) then mass should be a better predictor of dive duration than dive depth. This would indicate a predominantly physiological influence on dive duration.
Some species that are capable of long dives may be restricted by their ecology to shallower waters (e.g. by the depth at which food is located and/or of the seabed; Wilson et al. 1993; Costa et al. 2004). In these cases, we might expect the excess diving capacity of those species to be expressed. For example, larger species restricted to shallow dives would have the capacity to dive for longer durations than smaller species at the same depth. More generally, we would expect that the time that a species can spend under water at any given depth will be positively related to its size and hence that by controlling for dive depth, there would be a positive correlation between dive duration and body mass. Given that depth is likely to be more influenced than duration by ecology, we would also expect to see a closer association between dive duration and mass than between dive depth and mass, across diving species.
Now consider the time that diving birds and mammals spend on the surface between dives. During this time, diving animals have to replenish oxygen stores and remove accumulated carbon dioxide (Kooyman et al. 1980; Halsey et al. 2003). Thus, surface duration is likely to increase with the duration of the proceeding dive (Kramer 1988; Costa & Gales 2000). We might also expect surface duration to be correlated with dive depth because depth and dive duration ought to be related. Furthermore, Halsey et al. (2006b) showed that surface duration is related to body mass. Again, whether body mass or dive duration is a better predictor of surface duration (i.e. whether physiological or ecological factors are more influential) depends on whether the relationship between surface duration and dive duration is causal (long dives take a long time to recover from and prepare for regardless of body mass) or largely driven by the covariation of both variables with mass (long dives take a long time to recover from and prepare for because larger species tend to dive for longer and tend to take longer to recover after a dive of a given duration).
Relationships between diving variables within and between species have been studied previously (e.g. Stonehouse 1967; Trillmich et al. 1986; Prince & Harris 1988; Schreer & Kovacs 1997; Watanuki & Burger 1999; Croll et al. 2001; Costa et al. 2001). However, while in some cases, the roles of physiology or ecology have been considered, the generality of such influences across a broad taxonomic range of diving species has not been investigated. Furthermore, the lack of explicit phylogenetic hypotheses in these previous studies means that the evolutionary basis of the associations is unknown. Here, we use the most extensive database compiled of the dive depth, dive duration, surface duration, dive : pause ratio and body mass of mammal and bird species to test the interactions between these variables, ensuring a robust analysis through the inclusion of phylogenetic information. The results from these analyses identify general evolutionary trends in the association between diving variables that are of interest in their own right, but also provide new insights into the broad-scale determinants of the diving behaviour of homeotherms.
Materials and methods
A database was compiled on behavioural diving variables for as many diving avian and mammalian species as could be found in the literature, using as many published sources as possible, some unpublished sources and correspondence with authors. The database includes data for 195 species taken from 286 studies, and is included as an appendix to Halsey et al. (2006b). Variables from that database used for the present study were body mass, mean dive duration, mean surface duration and mean maximum dive depth. All were logarithmically (base 10) transformed for analysis. The analyses in Halsey et al. (2006b) and those in the present study necessitated only one value for each diving variable per species and so mean values were calculated in every case. While the data for diving behaviour of some species are not normally distributed or even unimodal (e.g. dive depth of Macaroni Penguins; Green et al. 2003), the error variance that this may introduce to the mean estimate for any given species will be small compared with the variation in body mass and dive variables across the species analysed. Halsey et al. (2006b) provide more details on how the database was compiled. We also categorized species as to the taxonomic class to which they belonged (bird or mammal).
The method of phylogenetic generalized least squares (PGLS) was used to control for phylogenetic non-independence (Grafen 1989; Martins & Hansen 1997; Garland & Ives 2000). A previous study demonstrated a phylogenetic component to dive variables (Halsey et al. 2006b), and hence that estimates of these variables for related species are not independent. This non-independence violates the assumption of standard statistics that errors are uncorrelated, and may result in biased parameter estimates and increased type I error rates if it is not accounted for in the analysis (Harvey & Pagel 1991). Non-independence was also confirmed in the analyses performed in the present study (see Results).
PGLS explicitly incorporates the expected covariance among species into a statistical model fit by generalized least squares. The correlation between error terms is thus altered to reflect the degree of phylogenetic relatedness among the species to which they relate. Under the assumption of a ‘Brownian motion’ model of evolution, the expected trait covariance between any two species is directly proportional to the amount of shared evolutionary history, which equals the length of the branches connecting the root of the phylogenetic tree to their most recent common ancestor. If this assumption is incorrect (e.g. if closely related species are not more similar in traits than two randomly chosen species), then a statistical model incorporating phylogenetic information may not fit the data as well as one assuming that traits evolved independently (phylogenetic independence). However, the covariance matrix can be modified in PGLS to accommodate the degree to which trait evolution deviates from Brownian motion, using a measure of phylogenetic correlation, λ, derived by Pagel (1999; see also Freckleton, Harvey & Pagel 2002). λ is a multiplier of the off-diagonal elements of the covariance matrix (i.e. those quantifying the degree of relatedness between species), and normally varies between 0 and 1. If the covariance matrix is constructed assuming a Brownian motion model of evolution then λ = 1 retains that model, while λ = 0 specifies phylogenetic independence. Intermediate values of λ specify models in which trait evolution is phylogenetically correlated but to a lesser extent than expected under the Brownian motion model. For each analysis, we estimated the maximum likelihood value of λ by fitting PGLS models with different values of λ and finding the value that maximized the log-likelihood. This best-fitting model can be used as a basis for inference, while the value of λ associated with it can be used as a metric of the degree of phylogenetic correlation in the data (Freckleton et al. 2002). The PGLS approach was implemented in the statistical computing language R (Ihaka & Gentleman 1996) using the Analysis of Phylogenetics and Evolution (APE) package (Paradis et al. 2004) and code written by R. P. Duncan.
PGLS requires a hypothesis about the phylogenetic relatedness of the species analysed. We used the same phylogenetic hypotheses for the birds and mammals in our database as Halsey et al. (2006b), where the phylogenies are illustrated and the references from which they derive are listed. Since we do not have consistent estimates of the branch lengths associated with our composite phylogeny, we assumed that all branches in the phylogeny were of equal length. This is equivalent to a punctuational model of evolution in which all change occurs at speciation events. Previous analyses suggest that a covariance matrix calculated from these phylogenies with equal branch lengths produces the best model fit for these dive data.
Burnham & Anderson's (2001) framework for model comparison was used to identify the most plausible model(s) of trait evolution based on Akaike's Information Criterion (AIC) as a measure of model fit (see also Burnham & Anderson 2002). The best out of all of the models tested to explain each dive variable was that with the lowest AIC. The probability that any given model i is actually the best fit out of those tested was measured by its Akaike weight (wi: Burnham & Anderson 2001), the relative likelihood of the model compared with all others (the likelihood of the model divided by the sum of the likelihoods of all other models). These weights can also be used to calculate the likelihood that one model (i) is a better fit to the data than a second model (j). This is termed the evidence ratio, and is calculated as wi/wj. An evidence ratio > 10 would be indicative that model i is superior to model j (Burnham & Anderson 2001). All the models for each diving variable are calculated on the same subset of species (that for which data were available for the response and all predictor variables), so that the AIC values for the models for each response variable are comparable.
Dive duration was highly positively correlated with both dive depth (slope estimate ± SE = 0·590 ± 0·050; Fig. 1a) and body mass (0·343 ± 0·055) in univariate tests. Dive depth was positively correlated with body mass (0·332 ± 0·086; see also Halsey et al. 2006b). However, the relationship between dive depth and duration is not a consequence of the relationship of both to body mass (Fig. 1b). Dive depth and body mass are both positively correlated with dive duration in a multivariate model that includes both as predictor variables (dive depth: 0·499 ± 0·049, Fig. 1b; body mass: 0·175 ± 0·038, Fig. 1c). If taxonomic class is added to this multivariate model, the standard error for the class term embraces zero, i.e. the effect of class is not significantly different from zero (0·153 ± 0·220). The same is true for the interaction terms between class and dive depth and class and body mass (results not shown). Akaike weights indicate that the most likely of the above models is that with dive depth and body mass as predictors, followed by the same model with the class term added (Table 1). The evidence ratio comparing the fit of the univariate models for dive duration in terms of dive depth and body mass is 5014, indicating that dive depth is a greatly superior fit to dive duration than is body mass. Nevertheless, neither of the univariate models is likely to be the best of the set of fitted models (Table 1). λ is > 0·9 for all the models fitted to dive duration (Table 1), indicating a strong phylogenetic correlation in these analyses, and hence confirming the need to incorporate phylogenetic information via GLS.
Table 1. Summary of fitted models for dive duration in terms of dive depth, body mass, taxonomic class (class; bird or mammal), and the interaction terms between dive depth and class (Dd : Class) and body mass and class (Mass : Class). λ = the measure of phylogenetic correlation estimated for the model (see Methods). AIC = Akaike's information criterion. wi = Akaike weight for the model (the likelihood that the model is the best fit out of those tested). Dive duration, dive depth and body mass were all log10-transformed for analysis
Dive depth + Mass
Dive depth + Mass + Class
Dive depth + Mass + Class + Dd : Class + Mass : Class
Surface duration was positively correlated with dive duration (0·620 ± 0·115; Fig. 2a), dive depth (0·449 ± 0·083; Fig. 2b) and body mass (0·179 ± 0·068) in univariate tests. When all three predictors are included together in a multivariate model, only the confidence intervals for dive depth fail to encompass zero (confidence intervals: dive duration: −0·015–0·773; dive depth: 0·011–0·501; body mass: −0·154–0·112). However, if body mass is removed from this model, the resulting estimates for both dive duration (0·348 ± 0·172; Fig. 2c) and dive depth (0·257 ± 0·124; Fig. 2d) marginally differ from zero. If taxonomic class is added to this model, or class and its interaction terms with both dive depth and dive duration, neither class nor its interaction terms differ significantly from zero (results not shown). Akaike weights indicate that the most likely of the above models is that with dive depth and dive duration as predictors (Table 2), although no model is clearly better than any of the others in this set. Ignoring the models with class, the best model for surface duration is more likely to include dive duration (Σ wi = 0·66) and dive depth (Σ wi = 0·56) than body mass (Σ wi = 0·24). Adding taxonomic class to the model with dive depth and dive duration raises the model AIC, although this is the second most likely model as judged by its Akaike weight (Table 2). λ is > 0·8 for all these models, again indicating a strong phylogenetic correlation in these analyses.
Table 2. Summary of fitted models for surface duration in terms of dive duration, dive depth, body mass, taxonomic class (class; bird or mammal) and the interaction terms between dive duration and Class (Td : Class) and dive depth and Class (Dd : Class). λ = the measure of phylogenetic correlation estimated for the model (see Methods). AIC = Akaike's information criterion. wi = Akaike weight for the model (the likelihood that the model is the best fit out of those tested). Surface duration, dive duration, dive depth and body mass were all log10-transformed for analysis
Dive duration + Dive depth
Dive duration + Mass
Dive depth + Mass
Dive duration + Dive depth + Mass
Dive duration + Dive depth + Class
Dive duration + Dive depth + Class + Td : Class + Dd : Class
Mean dive : pause ratio is not related to either dive depth (0·107 ± 0·083, AIC = 28·00) or body mass (0·077 ± 0·046, AIC = 25·77) in univariate tests. Combining both variables in a multivariate analysis increases the AIC (to 27·69) over a model with mass alone. Adding class to the multivariate or univariate models, or class and its interaction terms with other predictors also increases AIC (results not shown). Akaike weights indicate that the most likely model for dive : pause ratio is in terms of body mass alone (wi = 0·51), but the slope estimate for body mass does not differ from zero. The next most likely model is that with mass and dive depth (wi = 0·20). λ is either 0 or close to it for these models except that for mass (λ = 0·99), indicating that for the most part dive : pause ratio is unrelated to phylogeny, as well as to the predictor variables.
It is now well established that metrics of diving behaviour, such as depth and duration, are associated with body mass across homeothermic vertebrate species (Trillmich et al. 1991; Costa 1993; Boyd & Croxall 1996; Schreer & Kovacs 1997; Watanuki & Burger 1999; Halsey et al. 2006b), that these relationships show an evolutionary component (Halsey et al. 2006b), and that the general forms of such relationships are consistent with the oxygen store–usage hypothesis (Halsey et al. 2006b). Here, we have shown that consistent evolutionary relationships exist not only between diving variables and body mass, but also between the diving variables themselves. Most notably, we show associations between dive depth and dive duration independently of body mass, between dive depth and surface duration controlling for dive duration, and between body mass and dive duration controlling for dive depth. To our knowledge, such general associations between diving variables across birds and mammals have not previously been demonstrated, and are indicative of influences other than physiology on the evolution of diving behaviour.
Furthermore, the models for the inter-relationships among diving variables presented in the current study underline a major conclusion drawn by Halsey et al. (2006b) in the context of diving allometry. While diving birds and mammals clearly differ in aspects of behaviour, ecology, life history, anatomy and physiology (e.g. Costa 1991), these differences do not translate into significant differences in the inter-relationships among diving variables for these groups. Taxonomic class does not improve any of the models for dive variables presented here. Much is known about the differences in physiology between taxonomic groups of divers (for recent reviews, see Butler & Jones 1997; Costa & Sinervo 2004; Green et al. 2005b), for example that the size of the relative oxygen stores of pinnipeds, cetaceans and penguins varies greatly (Kooyman 1989). However, the broad scale influences of physiology, ecology and behaviour on diving appear to be similar among different taxa. This provides the opportunity for a general description to be made of the relative influences of physiology, ecology and behaviour on homeotherm diving behaviour.
The fact that diving variables scale allometrically as predicted by the oxygen store–usage model (Butler & Jones 1982; Halsey et al. 2006b) suggests that while neither oxygen storage capacity nor rate of oxygen usage is likely to depend on body mass alone, body mass may be used as a reasonable proxy for the physiological capacity of a diving species. Nevertheless, residual variation around these allometric relationships implies that body mass is not the only driver of diving behaviour.
The present study shows that body mass is related to dive duration even when dive depth is controlled for (Fig. 1c) and thus larger species also dive for longer when diving to a given dive depth. This suggests that the ecology of a species, or the environment it inhabits, affects some aspects of diving behaviour, in this case dive depth. This supports the conclusion of Costa et al. (2004), who found that within the otariids, species that forage benthically in deep water (and thus dive deeply) use a greater proportion of their oxygen stores during diving than do epipelagic foragers. A particularly clear example of the environment affecting dive depth can be seen in Fig. 3. The figure shows a Northern Elephant Seal (Mirounga angustirostris) foraging over and around the Cortez Bank in the Southern California Bight, periodically making dives that reach the seabed. Where the depth of the seabed is greater, so are the depths of the dives.
That larger species dive for longer than smaller species at a given depth implies that larger species have a greater capacity for diving than is expressed in their dive depth. Further evidence of ecological effects on dive depth is also expressed by the tighter relationship between body mass and dive duration than between body mass and dive depth. Clearly, the depth to which an animal dives can be dictated by the depth of the river, lake or seabed, or strongly influenced by the location of prey in the water column, but the duration of the dive is instead affected by the physiology of the animal. Therefore, larger animals that dive shallowly for ecological reasons make use of the physiological advantage that their size confers by diving for longer and increasing their foraging efficiency (Kramer 1988; Houston & Carbone 1992). This in turn suggests that differences between actual diving behaviour and diving potential, such as that shown experimentally in Beluga Whales by Ridgway et al. (1984), and inferred through estimates of aerobic capacity compared to dive duration in species of otariids (Costa et al. 2001) and phocids (Thompson & Fedak 2001), may be a more general feature of diving birds and mammals.
However, such an extension of dive duration at shallow depths by larger animals does not outweigh the influence of dive depth on dive duration. As expected, dive depth and dive duration are strongly positively correlated, i.e. deeper dives take longer to perform (Fig. 1a). While this result is not surprising, it may have been the case that this trend was due to the relationship between body mass and both of these diving variables, as larger species dive deeper and dive for longer (the present paper and Halsey et al. 2006b). However, our analyses show that dive depth and duration are related independently of body mass (Fig. 1b). If animals of the same mass always had the same physiological capacity for diving, we might have expected the relationship between dive duration and mass to be stronger than that between duration and depth. This is because if diving capacity was related to physiology alone, duration ought not to depend on depth but only on mass. In fact, the evidence ratio for the two univariate models shows that the model with dive depth alone is a greatly superior fit to dive duration than is the model with body mass alone.
There are three, not mutually exclusive, reasons why dive duration may be more weakly related to body mass than to depth. First, some animals are simply poor divers for their mass. Some species that do not have to dive deeply may have a reduced diving capacity relative to deep diving species of the same size, albeit that they can dive for longer than smaller species utilizing equivalent depths. A possible example is the Bearded Seal, Erignathus barbatus, which weighs 350 kg but dives to an average of 17 m in shallow Arctic waters, and remains submerged for only 120 s (Krafft et al. 2000). Second, and related to the first point, body mass may be a limited surrogate for diving capacity. Relative body oxygen stores are known to vary widely across diving species within taxa, and across taxa (Kooyman 1989), and probably intraspecifically as well (Costa et al. 2004). To give an example, pinnipeds with larger spleens or larger blood volumes for their size tend to dive more deeply and for longer (Mottishaw, Thornton & Hotchachka 1999). This means that size and oxygen storage ability (diving capacity) are to some extent uncoupled across species, and probably within species too. Third, it is possible that the diving capacity of certain species has simply not been expressed in the field observations from which our data are derived. Some species may work well within their physiological limits because food is relatively easy to obtain (e.g. the Tufted Duck, Aythya fuligula; Halsey, Butler & Woakes 2005a; Halsey et al. 2005b; epipelagic otariids; Costa et al. 2004) or there is little competition for food (Halsey et al. 2006a). Certainly some studies reveal the truer capacity of certain diving species in undertaking foraging dives. Benthic feeding otariids often undertake dives around or even beyond their calculated aerobic dive limits (Costa et al. 2004). Stephenson, Butler & Woakes (1986) found that Tufted Ducks preferred to dive under natural conditions for 20 s but could be trained to dive repeatedly for over 40 s to obtain food.
Surface duration relates to both dive duration (Fig. 2a) and dive depth (Fig. 2b), and to both independently (Fig. 2c,d). Thus, the amount of time a species spends recovering from a dive, and/or preparing for the next one, depends not only on the time spent underwater (absolutely and for a given depth of dive), but also on the depth of a dive of a given duration. A relationship between dive depth and surface duration would have been expected from the strong relationship between dive duration and dive depth (Fig. 1a). A relationship controlling for dive duration is more surprising, as it implies that deeper dives require more time subsequently at the surface than shallower dives lasting the same length of time. This increased surface duration could conceivably be associated with the extension of behaviours such as the replacement of air in the fur or plumage, or to preen. However, it is most likely to be primarily an elongated period of recovery required in response to an energetically more costly dive. While some previous studies suggest that deeper dives are more costly (e.g. Croxall et al. 1988), deeper dives are generally also longer dives and thus the independent effect of dive depth on costs cannot be discerned.
The metabolic cost of a dive is the result of a complex interplay between the time spent during each phase of the dive (where a dive is simplistically divided into the descent and ascent portions and the foraging, or bottom phase) and the changing energetic cost of the dive during those phases. The volume of air within the body and the ratio of lipid to lean tissue (Biuw et al. 2003) are the two main factors determining buoyancy and thus the energy costs of the different phases. Different species are affected by these factors to differing degrees. For example, diving ducks are very positively buoyant during dives (Stephenson 1994, 1995), while some mammalian species exhale before they start a dive, which reduces buoyancy dramatically (see Kooyman 1989). For animals that are always positively buoyant, such as the majority of diving birds (Wilson et al. 1992), the descent phase of a dive is usually the most expensive (Lovvorn et al. 1991). However, the deeper these animals descend, the less buoyant they become owing to the compression of air. If the depth to which they descend is so great that negative buoyancy ensues, the animal must work against this force during part of the ascent phase (Beck, Bowen & Iverson 2000). This relationship is also complicated by the fact that some animals adjust the oxygen stores in their lungs depending upon the duration of the dive they anticipate undertaking (Sato et al. 2002; Wilson 2003), which in turn can affect their buoyancy and hence the cost incurred by it throughout the dive (Halsey et al. 2005b). Furthermore, animals may also adjust the angle or the speed at which they descend (Wilson, Ropert-Coudert & Kato 2002; Halsey et al. 2005b), which again affects metabolic costs. The ascent phase becomes less costly as the animal reaches shallower waters and it becomes more positively buoyant. However, the overall ascent will be more costly if positive buoyancy has decreased during the course of the dive because of a loss of trapped air from the surface of the skin (Stephenson 1993), and this decrease may be greater during longer dives (Knower Stockard et al. 2005). Maintaining a depth in the water column during the bottom phase of the dive is generally less energetically expensive than the descent phase for birds (Lovvorn et al. 1991; Wilson et al. 1992) but depends upon the buoyancy at the depth of this phase (Wilson et al. 2003) and could be costly for mammals if they have negative buoyancy at this point (Sato et al. 2003). Furthermore, this phase will tend to change in duration in response to the duration of the descent phase, perhaps becoming longer as predicted by the model of Houston & Carbone (1992), or sometimes shortening as observed in Tufted Ducks (Halsey et al. 2005b). With such a multitude of variables that change with depth, it is therefore difficult to model how the overall energetic costs of a dive differ as the depth of the dive changes.
Nevertheless, the results from the present study imply that a deep dive will also be more expensive than a shallow dive for a given dive duration. Furthermore, this relationship is general across diving birds and mammals. This has significant implications for the depth to which a pelagic feeder should dive when foraging. While a number of factors are likely to affect the energetic costs of dives (e.g. swim speed; Thompson et al. 1993), the present findings suggest that rather than diving to the depth of highest prey density (presumably where rate of prey capture is highest), divers may tend to dive to a shallower depth because of the trade-off between rate of prey capture and the increased metabolic expense of deeper dives. This agrees with the model of optimal foraging depth by Mori (1998), which predicts that diving animals should choose to forage at a depth that is always shallower than the depth at which prey density is highest. Thus, the depth at which prey are dense is an ecological factor affecting the diving patterns of a foraging species. Even if that species is physiologically capable of reaching the depth of densest prey, its diving behaviour might limit dive depth further to optimize foraging efficiency.
Mori (1998) offers several reasons for why larger animals should be expected to spend longer at the surface recovering after dives of a certain duration and depth, based on the idea that the rate that oxygen can be loaded to the body stores correlates negatively to mass. These include a smaller arteriovenous oxygen difference and slower blood circulation in larger animals. However, evidence presented here suggests that the best model for surface duration is unlikely to include body mass (Table 2 and Results), suggesting that in fact body mass is less likely than dive duration and dive depth to influence the recovery time required by an animal at the surface. In other words, the allometric relationship for surface duration shown by Halsey et al. (2006b) can be best explained by differences in recovery time dictated by the depth to, and duration for, which different species dive. In turn, this indicates that the surface duration after a dive is unlikely to be dictated directly by physiology as argued by Mori (1998), but rather by the behaviour during the dives.
While Halsey et al. (2006b) and the present study demonstrate that many diving variables are correlated with each other and with body mass, in contrast, the dive : pause ratio does not significantly correlate to either body mass or dive depth. Thus, it seems that increases in dive duration and surface duration scale equally with body mass and dive depth, although there is no obvious reasons why this should necessarily be so.
The models for inter-relationships among diving variables presented here reinforce an important conclusion drawn by Halsey et al. (2006b) in their study of the allometry of diving. That is, failure to account for phylogenetic autocorrelation can lead to misleading parameter estimates and inflated type I error rates. For example, ignoring phylogeny in the analyses of the present study would have led to the conclusion that dive duration was positively related to surface duration (estimate ± SE = 0·474 ± 0·162), but that dive depth was not (0·252 ± 0·134).
In summary, the behaviour expressed by species of aquatic birds and mammals reveals general evolutionary associations between diving metrics that provide useful evidence about the general roles of physiology and ecology in determining that behaviour, and allows some unexpected conclusions to be tentatively drawn. Larger species dive for longer than smaller species during dives of a given depth, indicating that even if dive depth is influenced ecologically, larger species still express their physiological diving advantage in such environments. This influence is also suggested by the stronger relationship of dive duration to body mass than dive depth to body mass. Nevertheless, dive duration correlates with dive depth more strongly than with body mass, suggesting that some animals are poor divers for their body mass, either because of a lower physiological capacity or because it is their behaviour that limits their diving rather than their physiology. Larger animals do not appear to spend longer at the surface between dives of a given depth and duration than smaller species, suggesting that, in general, diving behaviour rather than diving physiology dictates recovery time. For example, surface duration is longer between longer dives and also between deeper dives. More interestingly, surface duration also increases between deeper dives of a set duration. This is likely to suggest that deeper dives are energetically more costly. The influences of physiology, ecology and behaviour on diving appear to be similar within different taxa of diving birds and mammals despite the fact that diving behaviour has undoubtedly evolved repeatedly in birds and mammals and thus the taxa we compare will have followed different evolutionary paths to become divers.
We thank R. P. Duncan for providing us with his PGLS code, and for useful discussions on phylogenetically informed comparative methods. We are grateful to the anonymous referees who commented on the manuscript and in particular to D. Costa who was especially helpful in providing feedback and additional citations for the paper.