Energetic cost of hovering flight in nectar-feeding bats (Phyllostomidae: Glossophaginae) and its scaling in moths, birds and bats

ORIGINAL PAPER C. C. Voigt áY. Winter Energetic cost of hovering ¯ight in nectar-feeding bats (Phyllostomidae: Glossophaginae) and its scaling in moths, birds and bats Accepted: 10 November 1998 Abstract Three groups of specialist nectar-feeders covering a continuous size range from insects, birds and bats have evolved the ability for hovering ¯ight. Among birds and bats these groups generally comprise small species, suggesting a relationship between hovering ability and size. In this study we established the scaling relationship of hovering power with body mass for nectar-feeding glossophagine bats (Phyllostomidae). Employing both standard and fast-response respirometry, we determined rates of gas exchange in Hylonycteris underwoodi (7 g) and Choeronycteris mexicana (13±18 g) during hover-feeding ¯ights at an arti®cial ¯ower that served as a respirometric mask to estimate metabolic power input. The O 2 uptake rate _ Vo2in ml g )1 h )1 (and derived power input) was 27.3 (1.12 W or 160 W kg )1 ) in 7-g Hylonycteris and 27.3 (2.63 W or 160 W kg )1 )in 16.5-g Choeronycteris and thus consistent with measurements in 11.9-g Glossophaga soricina (158 W kg )1 , Winter 1998). _ Vo2at the onset of hovering was also used to estimate power during forward ¯ight, because after a transition from level forward to hovering ¯ight gas exchange rates initially still re¯ect forward ¯ight rates. _ Vo2 during short hovering events (<1.5 s) was 19.0 ml g )1 h )1 (1.8 W) in 16-g Choeronycteris, which was not signi®cantly dierent from a previous, indirect estimate of the cost of level forward ¯ight (2.1 W, Winter and von Helversen 1998). Our estimates suggest that power input during hovering ¯ight P h (W) increased with body mass M(kg) within 13±18-g Choeronycteris (n4) as P h 3544 (2057 SE) M 1.76 (0.21 SE) and between dierent glossophagine bat species (n3) as P h 128 (2.4 SE) M 0.95 (0.034 SE) . The slopes of three scaling functions for ¯ight power (hovering, level forward ¯ight at intermediate speed and submaximal ¯ight power) indicate that: 1. The relationship between ¯ight power to ¯ight speed may change with body mass in the 6±30-g bats from a Jtowards a U-shaped curve. 2. A metabolic constraint (hovering ¯ight power equal maximal ¯ight power) may in¯uence the upper size limit of 30±35 g for this group of ¯ower specialists. Mass-speci®c power input (W kg )1 ) during hovering ¯ight appeared constant with regard to body size (for the mass ranges considered), but diered signi®cantly (P< 0.001) between groups. Group means were 393 W kg )1 (sphingid moths), 261 W kg )1 (hummingbirds) and 159 W kg )1 (glossophagine bats). Thus, glossophagine bats expend the least metabolic power per unit of body mass supported during hovering ¯ight. At a metabolic power input of 1.1 W a glossophagine bat can generate the lift forces necessary for balancing 7 g against gravitation, whereas a hummingbird can support 4 g and a sphingid moth only 3 g of body mass with the same amount of metabolic energy. These dierences in power input were not fully explained by dierences in induced power output estimated from Rankine-Froude momentum-jet theory. Key words Hovering ¯ight áGlossophagine bats áHummingbirds áSphingid moths áAllometry Abbreviations Mbody mass áP f horizontal forward ¯ight power áP fmin minimum ¯ight power áP h hovering ¯ight power áP indRF minimum induced powers áP submax submaximal aerobic ¯ight power áRE respiratory exchange ratio áRQ respiratory quotient á_ Vo2oxygen uptake rate á_ Vco2carbon dioxide release rate J Comp Physiol B (1999) 169: 38±48 ÓSpringer-Verlag 1999 C.C. Voigt (&) 1 áY. Winter Institut fu Èr Zoologie II, Universita Èt Erlangen, Staudtstrasse 5, D-91058 Erlangen, Germany e-mail: [email protected], [email protected] Present address: 1 Boston University, Department of Biology, 5 Cummington Street, Boston, MA 02215, USA e-mail: [email protected] Introduction Hovering ¯ight is generally assumed to require more power than horizontal forward ¯ight because the aerodynamic induced power component, necessary for producing the lift force for resisting gravity, comes entirely from wing oscillations (cf.Norberg 1990). For three groups of nectar-feeders (sphingid moths, hummingbirds and glossophagine bats) which evolved their distinguished hovering ability as part of their ¯owerfeeding habits, the energetic cost of hovering has been determined from respirometry (see references given in Fig. 5). What makes a comparison especially interesting is that while individual power requirements for hovering overlap between the three groups, they all dier in hovering styles (von Helversen and von Helversen 1975; von Helversen 1986; cf. Norberg 1990; Norberg et al. 1993; Willmott and Ellington 1997; Liu et al. 1998). Within the two vertebrate classes, both hummingbirds (2±20 g) and glossophagine nectar-feeding bats (6±30 g) comprise species of relatively small body mass. This trend follows the scale eect that makes it easier for smaller species to mobilize power reserves for ¯ight. This is because power requirements for ¯ight increase with mass at a steeper slope than maximal metabolic rates (Pennycuick 1968, 1986; Norberg 1990; see also Ellington 1991). Despite this general scaling eect, details of the power requirements of hovering ¯ight, relative to forward ¯ight and their relation to metabolic and aerodynamic power capacities, were previously unknown. Hovering ¯ight in glossophagine bats diers from that of hummingbirds in a number of ways. 1. Hummingbirds oscillate their fully extended wings in a ®gure-eight motion with symmetrical half-strokes (Norberg 1990). The hovering kinematics of Glossophaga generally resemble slow ¯apping ¯ight with wings ¯exed during the upstroke. Lift is still generated during upstroke by supination of the handwing (the `distal wing triangle'), an adaptation that is possibly unique among bats (von Helversen and von Helversen 1975; von Helversen 1986). 2. Wing loading in hummingbirds is about twice as high as in Glossophaga (Greenewalt 1962; Winter 1998), as hummingbird wings are shorter and narrower. 3. The mass-speci®c power input during hovering in hummingbirds is 60% higher than in Glossophaga soricina (45 ml O 2 g )1 h )1 vs 27 ml O 2 g )1 h )1 or 261 W kg )1 vs 158 W kg )1 , Winter 1998) 4. While (at least the smaller) hummingbirds are able to sustain hovering ¯ight for up to several minutes (Pearson 1950), G. soricina does not sustain hovering ¯ight for extended periods. In the ®eld, duration of hovering of Glossophaga when feeding on nectar at ¯owers is almost always less than 1 s (C. Voigt, Y. Winter personal observation) and the longest hovering events observed in the laboratory at automatic feeders which delayed the nectar-reward (Winter 1998) were well below 10 s for this species (only the present study provides new evidence that, under special circumstances, a glossophagine bat may also sustain hovering ¯ight for longer periods). This suggests a limiting factor on hovering ability re¯ecting either aerodynamic or metabolic constraints. There are also similarities between the two groups. Measured rates of oxygen uptake _ Vo2by two small hummingbird species during forward ¯ight in a wind tunnel (Berger 1985) indicate that power input during hovering in hummingbirds may be equal or only slightly higher than for forward ¯ight at intermediate speeds (7ms )1 ). Likewise, metabolic data for G. soricina also indicate a factor of around only 1.1 between power inputs during hovering and level forward ¯ight at intermediate speed (Winter 1998; Winter and von Helversen 1998). In contrast, a previous quasi-steady aerodynamic analysis based on wing kinematic data and wake velocity measurements suggests that aerodynamic power output for hovering ¯ight in Glossophaga exceeds the corresponding value for horizontal forward ¯ight by a factor of 1.7 (perfect elastic storage) to 2.6 (zero storage) (Norberg et al. 1993). In view of the modest 10% increase in metabolic power from horizontal forward to hovering ¯ight, it is not obvious that metabolic power is limiting during hovering. In fact, the extrapolation of submaximal aerobic metabolic capacities (P submax ) of medium to largesized bats (95±800 g) during ¯ights in wind tunnels (Thomas 1987) to small body masses suggests that small bats may still have an 80% power margin above the power requirements for horizontal forward ¯ight at intermediate speeds (Winter and von Helversen 1998). For an 11.9-g Glossophaga, predicted P submax equals 3.0 W, compared to the power input of 1.88 W during hovering ¯ight and 1.65 W for horizontal forward ¯ight at intermediate speed. While the accuracy of this prediction of P submax is uncertain, it still indicates that metabolic requirements during hovering ¯ight may not constitute the limiting factor for a 12-g glossophagine bat. Similar to the prediction for Glossophaga, small hummingbirds have considerable metabolic (and aerodynamic) power reserves at their command. O 2 uptake at maximum ¯ight speeds in the wind tunnel has been measured at 1.6 times the rate during hovering (Berger 1985). Similarly, during experimental manipulations of metabolic and aerodynamic power requirements for hovering, hummingbirds have been proven to possess considerable power reserves (Wells 1993; Chai and Dudley 1995, 1996; Chai et al. 1996; Chai and Millard 1997). Data for the largest hummingbirds (20-g Patagona gigas, Greenewalt 1960), however, are not available. The present study was conducted to: (1) investigate the intraand inter-speci®c scaling of hovering power with body mass in glossophagine bats, and (2) compare the relationships between power input during hovering ¯ight and body mass between the glossophagine nectarfeeding bats, hummingbirds and sphingid moths. For 39 this, hovering power input was estimated from rates of gas exchange during hover-feeding at an arti®cial ¯ower that served as a respirometry mask for two species of glossophagine bat (7 g and 17 g), and (together with a previous value for 12-g Glossophaga) these results were compared with data available in the literature for sphingid moths and hummingbirds. Material and methods Animals and ¯ight cage Measurements for this study were performed with the two bat species Hylonycteris underwoodi Thomas 1903 and Choeronycteris mexicana Tschudi 1844 (Phyllostomidae: Glossophaginae). Prior to the experiments, animals were kept in greenhouses on an ad libitum food supply. A week before measurements were taken, we transferred the animals to an experimental room with a light: dark cycle of 12 h:12 h L:D, an average temperature of 24 °C, and a relative humidity of 75%, to acclimatize to the conditions. For a detailed description of the experimental protocol, see Winter et al. (1998). During each experiment, we introduced one individual into a plastic ¯ight tunnel (length 7 m, width 1 m and height 2 m). Within this ¯ight tunnel the only place the bat could rest was on a piece of cork that was suspended from an electronic balance (Mettler PM 100, accuracy 1 mg). Body mass and times of arrival at and take-o from the roost were recorded with a computer connected to the electronic balance. The animal was fed with a 17% (wt/wt) sugar solution (mixture of glucose, fructose and sucrose) provided from a computerized nectar-feeder. During hover-feeding ¯ight, the bat interrupted an infrared light beam at the front edge of the feeder mask. This was recorded by a computer that both activated the nectar pump and recorded the duration of the hovering event (Winter and von Helversen 1998). The feeder opening was constructed as a ¯owthrough respirometric mask (see drawing in Winter et al. 1998). The inner diameter of the mask was 2.3 cm during the measurements with H. underwoodi and 3 cm with C. mexicana. The feeder opening was oriented with an inclination of 30°to the ground and at a height of about 80 cm above the ground. Natural hovering durations of glossophagine bats are usually less than 1 s. In order to increase the duration of hovering ¯ights we delayed the injection of nectar into the feeder opening when a bat attempted to feed and provided the nectar reward in the form of successive small droplets (approximately 5 ll each) given every 0.8±1 s during a single hovering event. Typical nectar gains during a single feeding visit at a ¯ower in the wild is 10±50 ll (Howell 1979; von Helversen and Reyer 1984; Tschapka and von Helversen, personal observation). We carried out two types of respirometric measurements which focused on two dierent aspects of hovering gas exchange. Cumulative measurements We determined cumulative O 2 and CO 2 volumes for single hovering events. This allowed us to test for dierences in gas exchange rates between hovering events that diered in duration and to determine the respiratory exchange ratio (RE). These measurements were performed with two H. underwoodi (female 1: 7.1 g; female 2: 6.9 g) and four C. mexicana (see Table 1). We could not monitor changes in instantaneous rates of O 2 consumption and CO 2 release during a hovering event, due to the wash-out characteristics of the gas analysing system. Gas analysing system Room air was drawn through the respirometric mask with a membrane pump at a rate of 1000 ml min )1 . As shown previously for Glossophaga (Winter et al. 1998), this ¯ow rate made it possible to sample all respiratory gases while the bat was at the feeder mask. Downstream from the pump the air was dried using molecular sieve (3 A Ê) and the ¯ow regulated with a mass ¯ow controller (F 201 CFB, Bronkhorst, Ruurlo, Netherlands). Next, the airstream was split and led to both the O 2 analyser (S-3 A/II with a dual N-37 M sensor, Ametek now AEI, Pittsburgh, USA) and the CO 2 analyser (URAS 10 E, Hartmann & Braun, Frankfurt, Germany). Due to the high sensitivity of the CO 2 analyser (0±500 ppm), the sample air stream was diluted with dry, CO 2 -free air (NaOH as desiccant and CO 2 absorbent) by a factor of approximately ten. The dilution factor was monitored with a mass ¯ow controller (Bronkhorst F 201 C-FB) and a mass ¯ow meter (Bronkhorst F 111 C-HB) and held constant during the measurements. The analog signals of the gas analysers were recorded with a 386-MS-DOS computer at a rate of 3 Hz, using Sable Systems (Henderson, USA) hardware and software (16-bit A/D). The analog output signals of the three ¯ow meters were checked before and after an experiment with a digital voltmeter. Calibration and validation The CO 2 analyser was calibrated by injecting known concentrations of CO 2 with a gas mixing pump (Wo Èstho, SA27/2). For the O 2 analyser, no zero calibration was necessary because of the technical characteristics of its zirconium-oxide sensor. We validated system calibration regularly by simulating the O 2 consumption and CO 2 production of a hovering bat. We injected small volumes (100, 150 and 200 ll) of CO 2 (99.998%) into the respirometric mask and compared the measured with the injected CO 2 volumes. For the validation of the O 2 analyser we used the N 2 -dilution technique (Fedak et al. 1981). The average deviation between the measured and expected gas volumes was always less than 3.5%. Data analysis Due to the delay in wash-out of the gas-analysing system we were confronted with the problem of overlapping signals when the animal visited the feeder mask at short intervals. This problem was especially severe for the CO 2 signal, because of the large volume of the analyser's measuring chamber. During our initial measurements with Choeronycteris (female 1) and Hylonycteris (female 1), overlapping signals were separated following the procedure described in Winter et al. (1998). For all later measurements we applied a Z-transformation to separate overlapping signals (Bartholomew et al. 1981) using the Datacan program (Sable Systems). The Z-transformation is used to re-calculate the original pattern of O 2 depletion and CO 2 increase. Because of wash-out eects a stepwise change in O 2 or CO 2 concentration is usually measured as an exponential change of gas concentration. By applying Eq. 3 from Bartholomew et al. (1981) to the data, such an exponential curve can be transformed to its original, instantaneous shape: Meq MfÿMDf=ZDtMDt where M eq Z-transformed signal M, M t signal M at time t, M Dt signal M after Dthas passed, and Z Dt factor Z, describing the wash-out characteristic of the system. On the basis of the Z-transformed data, we calculated the integrals of the O 2 and CO 2 signals for each hovering event and multiplied them with sampling rate and ¯ow rate at the feeder mask. For the calculation of the CO 2 volume the dilution factor was taken into account. O 2 volumes were corrected with Eq. 3a in Withers (1977) as we used H 2 O absorbant, but no CO 2 absorbant for the sample gases. Gas exchange rates (ml s )1 ) were calculated by dividing the CO 2 and O 2 volumes by the corresponding hovering durations of the animal. The mass-speci®c gas exchange rates (ml g )1 h )1 ) were calculated by multiplying the gas exchange rate by 3600 and dividing it by body mass, obtained shortly before the hovering event. A correction for STP was not necessary as mass ¯ow meters give rates in STP. 40 Fast-response measurements Animals and experimental design We studied two individuals of C. mexicana (see Table 1). Air humidity in the ¯ight cage was increased to more than 95% during the measurements for the reason described below. Gas analysing system and validation For monitoring near-instantaneous rates of O 2 uptake during single hovering events, we modi®ed the analysis setup. To obtain a fast response from the O 2 analyser we minimized wash-out delay by installing the O 2 sensor without desiccant directly behind the feeder mask (Winter 1998). Without a desiccant, the evaporative water expired by a bat during respiration could deplete the O 2 concentration in the sampled air, which is not accounted for in the correction formula that we used (Eq. 3a in Withers 1977). This problem was minimized, however, by the high ambient relative humidity of 95% during measurements, which greatly reduced a bat's potential for further increasing the water content of the air. Downstream from the sensor the air was dried with a molecular sieve (3 A Ê) and the ¯ow rate was regulated with a mass ¯ow controller (Bronkhorst, F 201 C-FB) at 1000 ml min )1 .We validated the system by injecting dierent N 2 volumes into the feeder mask (N 2 -dilution technique; Fedak et al. 1981). The deviation between measured and expected O 2 volumes averaged less than 5%. Data analysis For calculating a mean curve of O 2 depletion, depletion curves for each bat were superimposed from their origin up to the point where the bat had left the feeder (i.e. without declining slopes). Mean values were then calculated for each sampling interval over all data points resulting in one mean O 2 depletion curve for each bat (Fig. 1). These mean O 2 depletion curves showed that rates of O 2 uptake were constant after 5 s of hovering (Fig. 1). A single mean value for the rate of O 2 uptake was therefore determined for an individual bat by averaging all values from the 5th s of hovering to the end. As before, this O 2 rate was corrected according to Withers (1977) using a respiratory quotient, RQ (RE) of 0.65. This value of RE was obtained in the following way. CO 2 was not measured during the fast-response measurements (due to the slow response time of the CO 2 analyser) but was only measured during the cumulative measurements. Cumulative (mean) RE was 0.76 for hovering events of 5±10 s duration (Fig. 2C). As the RE was around 1 at the beginning of a hovering ¯ight (Fig. 2C) it must Fig. 1 Mass-speci®c rates of O 2 uptake during hovering ¯ight of C. mexicana (individuals f 2 ,n38, m 1 n44) measured with a fast-response O 2 analysis system. The data show the change of rates during a hovering event. This is dierent to the data in Fig. 2 which are mean values calculated from total consumption during hovering events of diering durations. Data are mean values 1 SD. See Table 1 for overall mean Fig. 2A±C Mean rates of mass-speci®c gas exchange during hoverfeeding events of dierent durations in four Choeronycteris mexicana (means 1 SD). The data are AO 2 uptake, BCO 2 release and C respiratory exchange ratio (RE). Note that the higher values for individual f 1 in Aare associated with a higher body mass (see also Table 1). The trough in RE during hovering events of medium duration most likely re¯ects the time lag in the increase of CO 2 output after a stepwise increase in metabolic activity during the change from level forward to hovering ¯ight. All hovering events included in this analysis were preceded by a ¯ight interval of at least 4 s prior to hovering 41 have dropped to a value of 0.65 within 5±10 s of hovering to give an overall mean RE for the complete hovering event of 0.75. For conversion of mass-speci®c rates of O 2 uptake (ml O 2 g )1 h )1 )to metabolic power input (W), data were divided by 3600/ 21.1 170.64, assuming carbohydrate catabolism and a metabolic RQ of 1. Results Choeronycteris mexicana All bats were active throughout the whole night. Activity bouts normally lasted up to 15 min and were interrupted by resting periods of approximately equal duration. Total ¯ight activity during a night averaged 3.0 h (2.7 SD, minimum 0.58, maximum 9.3) and total hovering ¯ight time averaged 0.38 h (0.13 SD, minimum 0.10, maximum 0.58). On average a bat consumed 17.5 ml (11.8 SD, minimum 1.5, maximum 36.8) of sugar water per day. Bats prolonged their hovering durations at the feeder to varying degrees, due to the delayed food delivery (Table 1). One single individual with a very low body mass (m 1 ) hovered for up to 17 s. After the transition from resting to ¯ying, a bat needs several seconds to reach its new respiratory steady-state (Winter 1998). In this context we found that _ Vo2in Choeronycteris was signi®cantly depressed when the duration of the ¯ight bout preceding a hovering event was less than 4 s. For this reason data analysis was restricted to hovering events preceded by a ¯ight activity of at least 4 s duration. All Choeronycteris increased their _ Vo2during the ®rst 5 s of a hovering event. This is illustrated both by the results of the fast-response measurements of _ Vo2(Fig. 1) and by the cumulative measurements made during hovering events of diering durations (Fig. 2A). Of the four individuals, only two bats hovered for long enough to reach a plateau in _ Vo2(Fig. 2A). This increase in _ Vo2 during a hovering ¯ight was also shown in the mean rates of _ Vo2during both short (<1.5 s) and long hovering ¯ights (>5 s) (Table 1). In contrast to _ Vo2, the carbon dioxide release rate _ Vco2was almost constant during the ®rst 8 s of hovering ¯ight (Fig. 2B). Only in m 1 did _ Vco2increase during hovering ¯ights of longer than 9 s. As a consequence of the uneven change in _ Vo2 and _ Vco2during a hovering event, the RE showed a marked deviation from 1 during hovering events of 4± 10 s duration (Fig. 2C). For calculating a species mean for metabolic hovering power input in Choeronycteris we only included data from longer hovering ¯ights (>5 s, Table 1). Mean O 2 uptake for all three individuals was 25.2 ml O 2 g )1 h )1 at a mean body mass of 15.4 g, which corresponds to a power input of 2.25 W or 146 W kg )1 .If the values for m 1 are excluded because of its atypically low body mass (see Discussion) then mean O 2 uptake was 27.3 ml O 2 g )1 h )1 at a mean body mass of 16.5 g, which converts to a power input of 2.63 W or 160 W kg )1 . Hylonycteris underwoodi The two female Hylonycteris underwoodi showed a mean total ¯ight activity of 1.26 h per night (0.17 SD, minimum 0.22, maximum 1.55) and a mean total hovering ¯ight time of 0.22 h per night (0.05 SD, minimum 0.07, maximum 0.87). We measured gas exchange rates during 206 hovering events of female 1 (2.11  1.02 s, minimum 0.15, maximum 8.77; mean  SD) and 38 hovering events of female 2 (3.08  1.31 s, minimum 0.88, maximum 4.97; mean  SD). A Kruskal-Wallis test revealed no signi- ®cant dierence in mass-speci®c gas exchange rates between the two individuals (P> 0.05), and we therefore pooled the data. The mass speci®c rates of O 2 uptake are shown in Fig. 3. Interestingly, in the small Hylonycteris we found no change in _ Vo2for hovering ¯ights of differing durations (slope of linear regression did not dier from zero, r 2 0.008, t)1.365, P0.17, n242). The mean O 2 uptake for female 1 (7.1  0.1 g; mean  SD) was 28.8 ml g )1 h )1 (7.1 SD) and for female 2 (6.9  0.08 g; mean  SD) it was 25.7 (9.6 SD). Converted to power this gives a species mean of 1.12 W for a 7.0-g bat or 160 W kg )1 . Discussion Power input during hovering The power input during hovering ¯ight (P h ), as estimated in this study from oxygen consumption, was 1.12 W or 160 W kg )1 in the small H. underwoodi (7.0 g) and between 1.69 W and 2.84 W (or 128±162 W kg )1 )in the larger C. mexicana (13.2±17.5 g). These values are in line with our earlier ®nding for Glossophaga soricina Fig. 3 Mean rates of mass-speci®c gas exchange during hovering events of diering durations of two Hylonycteris underwoodi bats. The data combine measurements (n242) from two individuals (6.9 g and 7.1 g) that did not dier signi®cantly from each other (Kruskal-Wallis test, P> 0.05). Rates of O 2 and CO 2 exchange diered signi®cantly for hovering events lasting less or longer than 2s (t-test, P< 0.05), but the two overall means did not dier signi®cantly from each other (t-test, P> 0.7). O 2 ®lled symbols, CO 2 open symbols 42 (1.88 W at 11.9 g or 158 W kg )1 , Winter 1998). This strengthens the evidence for the conclusion that metabolic power input increases only moderately during hovering ¯ight as compared to horizontal forward ¯ight in glossophagine bats. According to Winter and von Helversen (1998), metabolic ¯ight power during level forward ¯ight (P f ) at intermediate speed in small bats follows the relation P f (W) 50.2 M(kg) 0.771 where M is body mass which yields 1.09 W and 2.22 W for a 7.0 g and 17.5 g bat, respectively. Within the range of body masses investigated here, power input during hovering ¯ight amounted to only 1.03±1.28 times the cost of horizontal forward ¯ight at intermediate speeds. Considering that this regression equation may overestimate minimum ¯ight power (P fmin ) by 10% (Winter and von Helversen 1998), then hovering ¯ight power input would amount to 1.14±1.42 times P fmin for the 7.0 g and 17.5 g bats. This dierence in power input between hovering and horizontal forward ¯ight is well below the predicted aerodynamic power output based on an aerodynamic and kinematic analysis in Glossophaga soricina, assuming quasi-steady conditions (Norberg et al. 1993). Norberg and co-workers predicted that aerodynamic power output during hovering ¯ight would exceed the corresponding value for horizontal forward ¯ight by a factor of 1.7 (perfect elastic storage) to 2.6 (zero elastic storage). This is markedly higher than the range of 1.1±1.4 obtained in the present study and the factor of 1.2 obtained previously for Glossophaga soricina (Winter 1998). Norberg et al. (1993) used a quasi-steady approach based on wing kinematic data and wake velocity measurements for Glossophaga soricina to derive aerodynamic parameters. However, if unsteady eects are involved, a quasi-steady approach may underestimate the induced power, while the other power components remain unaected (Norberg et al. 1993). Since according to theoretical calculations the induced power component in hovering ¯ight is three times higher than in horizontal forward ¯ight (Norberg et al. 1993), the consideration of possible unsteady eects in the aerodynamic models might lead to an even larger dierence in the theoretically expected aerodynamic power output between these two modes of ¯ight. It seems unlikely that dierences in muscular eciencies between hovering and horizontal forward ¯ight can account for this deviation as this would mean that muscle eciency at the regular foraging ¯ight speed which accounts for more than 90% of daily ¯ight energy expenditures (Winter and von Helversen 1998) would be especially low. Thus a distinct and presently unresolved dierence between the results of metabolic measurements and those of the quasisteady aerodynamic analysis exists (cf.Ellington 1991). Power input during forward ¯ight The cost of horizontal forward ¯ight in small bats (<35 g) has thus far only been determined by metabolic energy balance methods. Using this protocol, the total energy turnover of an animal was ascertained over a length of time consisting of periods of both ¯ight and rest. Flight cost was determined by subtracting the estimated cost entailed during the non-¯ight period from the total energy turnover (Speakman and Racey 1991; Winter et al. 1993; Winter and von Helversen 1998). The results of the measurements carried out to date are ca. 20±25% lower than predicted by the majority of the allometric equations for bird ¯ight (Masman and Klaassen 1987; Rayner 1990), which indicates a metabolically less costly forward ¯ight by small bats at their slower speeds. Approximations of forward ¯ight cost can also be derived from measurements conducted in the present study, allowing estimates of ¯ight cost obtained to date to be compared with results of respirometry. Immediately following the change in performance level from forward ¯ight to hovering, rates of gas exchange measured at ®rst correspond to those of forward ¯ight because of the respiratory delay. Such an eect was observed in Choeronycteris (Fig. 2, Table 1). The rate of oxygen uptake during short hovering bouts (<1.5 s) diered clearly from the value following longer bouts of hovering. This value, at 19.0 ml O 2 g )1 h )1 or 1.8 W (average mass of 16.0 g), was lower than, but not signi®cantly dierent from the expected value of 2.1 W for the cost of forward ¯ight. The respirometric data are thus consistent with the forward ¯ight costs obtained using the energy balance method, as already shown in previous studies of Glossophaga soricina (Winter et al. 1998; Winter 1998). Moreover, these results support the ®ndings that forward ¯ight costs of small bats are relatively low. Only in Hylonycteris did we ®nd no dierence in the _ Vo2between short and longer hovering ¯ight (Fig. 3). However, this is not expected since in this species (at 7 g, the smallest of those tested) the dierence in ¯ight power between hovering and forward ¯ight was relatively small (see Fig. 4B). The duration of the respiratory transition period is assumed to be particularly short, so that adjustment to the hovering metabolic level occurs very quickly. Scaling of hovering power with body mass Intra-speci®c scaling Hovering power increases with body mass. Here, hovering power input in Choeronycteris increased from 1.69 W at 13.2 g to 2.84 W at 17.5 g (Table 1), proportional to M 1.76 (0.21 SE) (n4, Fig. 4A). The theoretically expected change of ¯ight power with body mass depends on the associated change of body geometry (i.e. wing area and body shape). When body mass increases but wing area remains constant (within a single individual) then ¯ight power increases more than when both mass and wing area increase together (as between 43 dierently sized individuals) (Norberg and Rayner 1987). In the present case, the three Choeronycteris included in this analysis (hovering >5 s, Table 1) were of nearly identical linear body dimensions despite the 40% variation in mass during the measurements. Both forearm length (a suitable intra-speci®c index of body size in bats) and wing area varied only by 1% between the three individuals measured (Table 1). Power input for hovering in this case should scale with mass according to the expectations for intra-individual body mass changes. For this reason, we used the two measurements for m 1 at dierent body masses as two data points. A theoretical expectation of the increase in aerodynamic power output is dicult to derive without knowing how kinematic and aerodynamic parameters changed. While the induced power component of hovering ¯ight should increase with an exponent of 1.5 to body mass, the change in inertial power depends (besides the unresolved question of elastic storage) on wing tip velocity and wing beat amplitude (Norberg et al.1993) which were not measured here. Beside these two power components, pro®le power is of minor importance during stationary hovering (Norberg et al. 1993). A lower mass exponent than measured in the present study was obtained from an analysis of aerodynamic power output of arti®cially loaded hummingbirds during hovering ¯ight (Chai and Millard 1997). When the masses of four species of 3±7-g hummingbirds were increased to 2±3 times body mass, estimated aerodynamic power output changed with an exponent to mass of around 1.4. To obtain an estimate of typical body masses of our experimental animals, we recaptured and weighed them 6 months after the measurements. All four Choeronycteris had increased their body masses and weighed between 16.8 g and 18.3 g; m 1 increased in mass by 30%. These values corresponded to values recorded in the animal's natural habitat. Six Choeronycteris with a mean forearm length of 45.7 mm had a median mass of 17.0 g, with ®ve individuals between 16.8 g and 17.5 g and one individual at 14.8 g (Mexico, Tapanti, O. von Helversen personal observation). Flight morphological and physiological parameters in Choeronycteris may have become adapted during evolution to support adult body masses between 16 g to 18 g. Only the two individuals, f 1 and f 2 , for which data were obtained from hovering ¯ights of longer than 5 s fell in the mass range that is typical for this species. Therefore, the species mean value for Choeronycteris used for the interspeci®c comparison below was based on these two individuals only to ensure that it more likely re¯ected the value expected under natural conditions. Inter-speci®c scaling Data available for glossophagine bats provide the opportunity to determine a ®rst provisional estimate of the inter-speci®c scaling of hovering ¯ight cost, and possible energetic limits to hovering in bats. Hovering power of three species with mean body masses ranging from 7 g to 16.5 g increased in proportion to M 0.95 (Fig. 4B). Based on n3 species, this mass exponent, however, is neither signi®cantly dierent from the previously derived slope for the cost of horizontal forward ¯ight (0.771, Winter and von Helversen 1998) nor from the value of 1, thus any inferences based on this dierence must be considered provisional. An interesting comparison can be made with the relationship established by Thomas (1987) between body size and submaximal aerobic performance of bats (95± 800 g). He used maximum metabolic rates measured during wind tunnel ¯ights to derive the equation (recalculated to SI-units) P submax (W) 65.7 M(kg) 0.70 (Fig. 4B). For an 11.9-g Glossophaga predicted P submax equals 3.0 W, compared to 1.88 W hovering power input. While it must be kept in mind that the accuracy of Table 1 Oxygen uptake during hover-feeding ¯ights in four individuals of Choeronycteris mexicana bats determined by two different methods (cumulative and fast-response, see Materials and methods). Values are means  1 SD. The species mean metabolic hovering power input for C. mexicana was determined as 27.3 ml O 2 g )1 h )1 (2.63 W or 160 W kg )1 ) at 16.5 g. This is the mean of the values for f 1 and f 2 during hovering ¯ights of longer than 5 s; m 1 was not included as its mass during the measurements was untypically low for this species (see Discussion). In both f 1 and m 1 , O 2 uptake rate was signi®cantly dierent between short (<1.5 s) and long (>5 s) hovering ¯ights (Mann-Whitney-Rank-Sum-Test, P< 0.05). Wing areas (A) were determined according to Norberg and Rayner (1987) and include body and tail membrane areas which make up about 7% of the total value. Rates of O 2 uptake (ml O 2 g )1 h )1 ) were converted to power (W) by multiplying by mass and dividing by 3600/21.1 = 170.64, assuming carbohydrate catabolism and the respiratory quotient = 1. All hovering events included in this analysis were preceded by a ¯ight interval of at least 4 s prior to hovering. (bwing span, FA forearm length, Ind. individual bat, _ Vo 2 oxygen uptake rate) Ind. Body mass FA mm b m A m 2 Hover events Duration _ Vo 2 < 1.5 s _ Vo 2 > 5 s Power > 5 s W Type of measurement gnmean max. ml g )1 h )1 nml g )1 h )1 n ss f 1 17.50.3 45.4 0.304 0.0145 252 2.61.2 8.8 18.16.8 40 27.75.1 23 2.84 0.52 Cumulative f 2 16.31.3 46.0 0.312 0.0147 292 2.90.8 4.9 18.17.5 71 Cumulative f 2 15.40.1 46.0 0.312 0.0147 38 3.62.3 8.6 26.82.3 38 2.420.21 Fast-response m 1 14.40.6 45.6 0.315 0.0145 414 6.74.6 16.7 21.75.9 71 24.32.7 236 2.05 0.23 Cumulative m 1 13.20.2 45.6 0.315 0.0145 44 6.41.7 10.1 21.82.1 44 1.69 0.16 Fast-response m 2 15.70.5 45.0 0.314 0.0153 224 2.11.2 5.3 18.28.5 6 Cumulative 44 this extrapolation of P submax to the small body sizes of glossophagine bats is uncertain, the wide power margin predicted may indicate that hovering ¯ight capacity of the majority of small to medium sized glossophagine species is not limited by metabolic requirements. In this context it is interesting to note that even pregnant females of Glossophaga at 130±140% adult body mass feed from ¯owers while hovering (Voigt 1998). The dierent mass exponents between P submax (0.70) and P h (0.95) lead to the expected result that larger animals must invest proportionally more in hovering ¯ight than small animals. However, as can be seen from Fig. 4B, all hovering glossophagine bats should be able to sustain aerobic hovering ¯ight up to their maximum size of 30± 35 g. Eventually, metabolic constraints may limit the body size for a hovering glossophagine bat and, incidentally, the largest glossophagine species, Leptonycteris nivalis, has a body mass of up to 30±35 g. Sustained hovering After we observed that glossophagine bats mostly hovered for very short periods, or for a few s at most, we assumed that they were fundamentally unable to maintain hovering ¯ight for longer periods. However, during the present investigation we observed for the ®rst time that a glossophagine could hover for a time intervall of 17 s (Table 1); in addition, 7.5-g Saccopteryx bilineata males (Chiroptera, Emballonuridae) were observed to hover for 14.4 s during display ¯ights in front of females (C. C Voigt and O von Helversen personal observation). This demonstrates that the basic chiropteran structure permits sustained hovering. The sustained hovering of a Choeronycteris was possible under the following conditions: while Choeronycteris normally weighs ca 17 g, individual m 1 had a temporarily reduced body mass of 13.2 g or 78% of this value. Wing loading was therefore reduced by the same amount, from 11.5 N m )2 to 8.9 N m )2 (these values increase by 7% when body area and tail membrane are excluded from the `wing area'; see Table 1). Similarly, wing disc (pb 2 /4) loading decreased from 2.15 N m )2 to 1.67 N m )2 . Hovering power decreased from 2.7 W to 1.7 W, a value well below the cost of forward ¯ight at intermediate speed (2.2 W), expected for a typical 17-g Choeronycteris. Thus, energetic and aerodynamically relevant parameters changed in concordance and it presently remains unresolved which was most critical for sustained hovering. That sustained hovering is not observed in bats more often is probably explained by the dominance of other selection pressures on ¯ight abilities. Birds, bats and moths Three groups of nectar-feeders with nearly overlapping size ranges within the Insecta, Aves and Chiroptera have evolved specialized adaptations for hovering ¯ight: glossophagine bats, hummingbirds and hawkmoths. Figure 5B compares the mass-speci®c metabolic energy expenditure required for hovering in the three groups. Although all regression slopes appeared to be slightly negative, in no case did the mass exponent dier from zero to a statistically signi®cant degree (P> 0.05). Thus, for the range of body masses considered here, the data did not indicate any changes in the mass-speci®c costs of hovering within groups. Mean mass-speci®c Fig. 4A, B Scaling of metabolic power input for hovering with body mass in AC. mexicana and Bthree species of glossophagine bats. Data in Aare from Table 1 (hovering durations >5 s). Data in B,solid line with ®lled symbols, combine the species means from the present study of Hylonycteris (n2) and Choeronycteris (n2, only f 1 and f 2 , because of low body mass of m 1 , see Discussion) and a published value for Glossophaga soricina (11.9 g, 1.88 W, Winter 1998). Additional functions in Bwere included for comparison with hovering ¯ight costs. Dashed line energy cost of level forward ¯ight at intermediate speed (P f ) (W) 50.2 body mass (M) (kg) 0.771 (Winter and von Helversen 1998); dashed-dotted line estimated minimum power ¯ight cost ( 0.9 ´dashed line); dotted line: extrapolation of submaximal aerobic ¯ight capacity (P submax ) in 95±800-g bats during wind tunnel ¯ights, P submax (W) 65.7 M(kg) 0.70 (Thomas 1987, recalculated to SI-units). Open symbols are, for comparison, the individual data points for Choeronycteris from A. The regressions given in the ®gures were based on mass in kg; standard errors of regression coecients were A3544  2057 and 1.76  0.21, n4, B128  12.4 and 0.950  0.037, n3 45 hovering ¯ight costs were 393 (98.8 SD, n13 species) W kg )1 in sphingids, 261 (23.6 SD, n10 species) W kg )1 in hummingbirds, and 159 (0.76 SD, n3 species) W kg )1 in glossophagine bats. These values were signi®cantly dierent from each other (ANOVA on ranks, P< 0.001). Thus, glossophagine bats expend only 40% of the energy of a hawkmoth and 61% of a hummingbird to maintain hovering ¯ight. At a metabolic power input of 1.1 W a glossophagine bat can generate the lift forces necessary for balancing 7 g against gravitation, whereas a hummingbird can support 4 g and a sphingid moth only 3 g of body mass (Fig 5 A). With the same expenditure of metabolic energy, a glossophagine bat can therefore generate a considerably greater lift force than members of the other two groups. Minimum induced power The essence of hovering ¯ight is the production of a vertical force to balance the body mass of an animal against gravitation. The main power drain for hovering is the induced power, and the requirements for the induced power can be reduced by increasing the wing span. This is a classical solution for both animals and helicopters, and glossophagine bats with their long wings make use of it. For a relative comparison between the three groups, minimum induced powers (P indRF ) were calculated from Rankine-Froude momentum-jet theory (Norberg 1990). We determined the values for those representatives of the three groups with an identical metabolic power input for hovering of 1.1 W (Fig. 5A, Table 2). This ®rst approximation of a comparison between power input and output during hovering indicates that dierences in induced power can partly explain the dierence in metabolic power requirements between bats and hummingbirds but none between bats and moths. With the same energy input, bats appeared to be able to produce an aerodynamic power that is 30% higher than in hummingbirds and 130% higher than in sphingid moths. To explain these dierences, an aerodynamic model that is more detailed than the classical quasi-steady aerodynamic approach to hovering ¯ight may be required. Manoeuvrability One consequence of the dierences in ¯ight kinematics between the groups is an apparently reduced ¯ight acceleration from hovering in glossophagine bats as compared to hummingbirds or sphingid moths (although this has not been quanti®ed yet). Selection for high manoeuvrability in these two groups has been related to its importance for interference in competition and territoriality in hummingbirds (Feinsinger and Chaplin 1975; Feinsinger et al. 1979) and to predation pressure during nectar-feeding in sphingid moths (Wasserthal 1993, 1997, 1998). Territoriality of the type observed in hummingbirds is unknown and unlikely in glossophagine bats. In addition, their main anti-predator strategy during nectar-feeding appears to lie in the brevity of their ¯ower visits (typically less than 0.75 s), as their ability to react to an ambush predator is severely limited when eyes (and ears) are inserted in the ¯ower corolla during feeding (see photographs in Tuttle 1991 and Winter 1998). Fig. 5A, B Scaling of metabolic power input for hovering with body mass in sphingid moths, hummingbirds (Trochilidae) and nectar-feeding bats (Phyllostomidae: Glossophaginae). Data in A are individual and data in Bmass-speci®c values. The dashed line in Aconnects the body masses with equal metabolic power input for hovering of 1.1 W. Data for sphingid moths from Bartholomew and Casey (1978); data for hummingbirds from Berger and Hart (1972), Berger (1985), Epting (1980), Bartholomew and Lighton (1986), Suarez et al. (1990), Chai and Dudley (1996); data for glossophagines as in Fig. 4. When several values were reported for a species, data were combined into a single mean value for each species. Regression equations in Awere: P(W) aM(kg) b Sphingidae: a 185 (17.7 SE), b 0.882 (0.104 SE) (n13 species), Trochilidae: a 186 (17.8 SE), b 0.933 (0.049 SE) (n10 species), Glossophaginae: a 128 (12.4 SE), b 0.950 (0.037 SE) (n3 species), all exponents signi®cantly dierent from zero (P< 0.001) but not signi®cantly dierent from 1 (P> 0.05). In Bmass exponents b did not dier signi®cantly from zero (P> 0.05) for the mass ranges considered here 46