Periodic habitat loss alters the competitive coexistence between brown trout and bullheads in a small stream over 34 years

Authors


J.M. Elliott, Freshwater Biological Association, The Ferry House, Far Sawrey, Ambleside, Cumbria LA22 0LP, U.K. E-mail: jmel@fba.org.uk

Summary

  • 1Changes in the population density of juvenile sea trout Salmo trutta L. and bullheads Cottus gobio L. were compared in a small stream over 34 years. Both species have a similar diet and obviously live in the same general habitat. Habitat loss was most marked in seven summer droughts: severest in 1976, 1983, 1984, 1995, and less severe but followed by autumn droughts in 1969, 1989 and 1993. The contrasting effects of habitat loss on the two species were examined.
  • 2For both species, the Ricker curvilinear model significantly fit (P < 0·001) the relationship between initial egg density and survivor density for successive life stages, even though egg densities were much lower for bullheads than trout. These analyses provided evidence for density-dependent population regulation and also identified extreme outliers, most being for year-classes affected by summer droughts.
  • 3The variable effects of changes in habitable area (= % wettable area in sampling section) were quantified by using the residuals, each residual being the absolute value expressed as a percentage of the expected value from the Ricker curve. Significant relationships between the residuals and habitable area showed that habitat loss had a marked effect on survivor density, this being negative for 0+ and 1+ trout, and positive for 0+, 1+ and 2+/3+ bullheads.
  • 4Therefore, during periods of habitat loss in the summer months, bullhead density increased at the expense of trout density. Low flows and a decrease in wettable area were associated with a marked reduction in habitat quality for drift-feeding trout and an increase in habitat quality, and perhaps also quantity, for benthic-feeding bullheads. This case study shows that, during a major perturbation, the relationship between the densities of two species can change markedly in favour of the less numerous species. The competitive coexistence between the two species is therefore a dynamic process that changes through time with periodic changes in the environment.

Introduction

How animals cope with living in a patchy environment that varies in space and time is an essential part of their ecology and affects the life histories and behaviour of individuals, population dynamics of species, and community structure and function (Shorrocks & Swingland 1990). A large theoretical literature, supported by fewer practical studies, shows how habitat subdivision can allow two or more species, fugitive species and superior competitors, to coexist, and how spatial structure can stabilize host–parasite and predator–prey interactions (references in Tilman 1994; Hanski 1999; Hassell 2000). Nee & May (1992) developed a theoretical model for metapopulation dynamics and predicted that habitat loss can lead to an increase in the absolute abundance of an inferior competitor while decreasing the abundance of a superior competitor. Habitat loss can thus facilitate competitive coexistence between two species. Hanski & Gilpin (1991) note that many processes in metapopulation dynamics also operate in a parallel fashion in patchy distributions of individuals within populations. An intriguing question is how two or more species in the same locality react to marked changes in habitat, especially habitat loss. This question appears to have been rarely addressed by long-term field studies (> 20 years) in animal ecology. Therefore, the present study compares changes in the population density of juvenile sea trout Salmo trutta L. and bullheads Cottus gobio L. in Black Brows Beck over 34 years.

Bullheads and trout or salmon Salmo salar L. are often found together in small, stony, streams and share the same general habitat. Some workers have concluded that there is direct competition for food between the two species so that when bullheads are present, trout densities are lower than when bullheads are absent (Gaudin & Heland 1984; Bardonnet & Heland 1994; Gabler & Amundsen 1999; Gaudin & Caillere 2000; Olsen & Vøllestad 2001, 2003). Other workers have concluded that bullheads have no significant effect on densities of trout or juvenile salmon Salmo salar (Pihlaja et al. 1998; Carter, Copp & Szomlai 2004). Direct competition between bullheads and trout may occur when their habitat is restricted or when food is scarce because densities of benthic invertebrates are low. Such a situation could occur during a marked decline in habitat quality and quantity. Therefore, the contrasting effects of habitat loss on the two species were examined in detail.

The trout population in Black Brows Beck has been studied for 35 years (1966–2000). The results of the long-term study, and short-term studies to answer specific questions, have been described in detail (references in Elliott 1994; Elliott & Elliott 2005). Apart from a few young eels Anguilla anguilla (L.) (only 28 taken in whole study), brown trout and bullheads were the only fish species present with the trout dominant in terms of numbers and biomass. The long-term changes in the population density of the bullheads are described for the first time. Habitat loss occurred in seven summer droughts, which were severest in 1976, 1983, 1984 (also severe spring drought), and 1995, and less severe in summer but followed by autumn droughts in 1969, 1989 and 1993 (Elliott, Hurley & Elliott 1997). Therefore, the long-term study in Black Brows Beck provides a case study for the effects of habitat loss on two sympatric species.

Study area, life cycles and sampling methods

As these have been described in detail for the trout by Elliott (1994), only a summary of the essential points is provided here. The life cycle and sampling methods for bullheads are described for the first time.

study area

Black Brows Beck is a small stream (length 512 m, mean width 0·8 m) in north-west England, and serves as a nursery for the progeny of sea trout. Water temperature usually ranged annually from 0·1 to 18 °C with a mean of about 9 °C. Water velocity increased with discharge but never exceeded 60 cm s−1 at high discharges because the stream overflowed its banks and flooded the adjacent field. The stream bed was therefore never subjected to high water velocities and remained stable, this being important for the maximum survival of eggs and juveniles of both fish species. The conductivity of the stream was low at about 100 µS cm−1 (k25) with a low calcium concentration of about 0·4 m eq l−1 and a pH range of 6·7–6·9.

life cycles and sampling methods for trout

The trout spawned in November and December when the adults returned from the sea or estuary. The eggs were laid in a gravel nest, the hatched fish were called alevins (feed entirely on yolk and remain in nest), followed by the fry stage (short transition stage when the trout emerge from the nest and start to feed and disperse), and finally the parr stage (after the yolk sac has been absorbed and before smoltification occurs for seaward migration). Eggs usually hatched in February/early March and the alevin stage ended in late April/early May when the fry dispersed from the nest. Each year-class was named after the year in which the eggs hatched. The parr stage lasted about 2 years and most trout started to migrate downstream to the estuary in May at the start of their third year (age 2+ years). The standard convention for ageing trout was followed, e.g. 0+ trout were less than 1 year old, 1+ trout were between 1 and 2 years old, 2+ trout between 2 and 3 years old, etc. It was assumed that the age designation changed in February because this was usually the month in which most eggs hatched.

Since November 1966, an annual census has been made of the number of nests and hence the number of females returning to spawn within a study section of 120 m2, which included the section used to sample the juveniles. Excavations of nests elsewhere in the stream provided information on the number of eggs per nest. A total of 70 nests was excavated over 30 years and the mean number of eggs per redd (± 2 SE) was 1553 (± 36), 1034 (± 27) and 722 (± 24) for early (2–13 November), middle (14–25 November) and late (26 November−7 December) spawners, respectively (Elliott & Hurley 1998). It was thus possible to estimate egg density in each year from the number of nests and the time of spawning. Electrofishing was used to catch trout from known areas of stream at the end of May or early June and at the end of August or early September (1967–2000). The study section of 60 m2 was divided by block nets into six subsections, each with a surface area of 10 m2. All fish were removed, counted and returned live to the stream.

The total area of the study section was used in all estimates of population density so that values were comparable between sampling occasions, i.e. estimates were always number of fish per 60 m2. However, the area of the study section covered by water, and hence the habitable area available to the fish, varied between sampling occasions. When the stream was full to the banks and all large stones were submerged, the wettable area was 100%. On each sampling occasion, the wettable area was estimated to the nearest 5% and reached a minimum value of 30% in the severest summer drought. Percentage wettable area was thus used to estimate habitable area (h).

life cycles and sampling methods for bullheads

Eggs were laid in April and early May, and hatched in June. Bullheads were taken in the electrofishing samples together with the trout. Newly hatched bullheads (0+ fish) were not taken in the May/early June samples, but were first sampled in the following August/early September. A few first-year fish were found to be sexually mature and ready to breed at about 11 months old, but these represented < 3% of the 0+ fish. All surviving fish bred towards the end of their second (1+ fish), third (2+) and fourth (3+) years. The latter fish were rare in the samples, and as no older bullheads were taken, it was assumed that no 3+ fish survived for long after breeding.

Although deaths of trout were rare during electrofishing, more bullheads died; three to six per sample to provide a total of 343 fish over the 34 years of sampling. Otoliths (sagitta) were extracted from these fish for age determination and stored dry. After soaking for a few minutes in cedarwood oil, they were examined by reflected light against a dark background. Length for age relationships were determined from the number of growth checks. Check formation occurred in late May, June and early July, and no change with fish age was detected. Therefore, the mean birth date was taken as the 15 June. The 343 bullheads used for age determination were also sexed and measured (total length to nearest mm), so that lengths for age could be obtained for males and females (Table 1). Lengths could not be used to separate 2+ and 3+ fish, but facilitated the separation of 0+, 1+, and the combined 2+/3+ age groups in the field samples. In the late August/early September samples, 0+ fish were < 33 mm, 1+ fish were 45–55 mm, and 2+/3+ fish were 57–65 mm. In the late May/early June samples, 0+ fish were < 45 mm, 1+ fish were 48–58 mm, and 2+/3+ fish were 60–68 mm (Table 1).

Table 1.  Age of female and male bullheads killed during sampling (total n = 343), together with their mean length (mm ± SE) and length range (mm) (n = number of fish measured for each estimate)
Sampling timeAgeFemale length (mm) Mean ± SE (range)nMale length (mm) Mean ± SE (range)n
Late Aug/early Sept0+ (c. 2 months)30 ± 2 (26–33)2830 ± 2 (26–33)28
Late May/early June0+ (c. 11 months)38 ± 2 (34–41)2043 ± 2 (38–45)22
Late Aug/early Sept1+ (c. 14 months)47 ± 2 (45–50)2252 ± 2 (50–55)27
Late May/early June1+ (c. 23 months)49 ± 2 (48–51)2956 ± 2 (53–58)27
Late Aug/early Sept2+ (c. 26 months)59 ± 1 (57–60)3061 ± 1 (59–64)28
Late May/early June2+ (c. 35 months)61 ± 1 (60–63)2863 ± 1 (61–66)32
Late Aug/early Sept3+ (c. 38 months)59 ± 2 (57–62) 662 ± 2 (58–65) 4
Late May/early June3+ (c. 47 months)62 ± 2 (60–64) 564 ± 2 (61–68) 7

A total of 112 females were also caught by electrofishing in early April, just prior to the spawning period, over the first 10 years of the study. The fish were taken from elsewhere in the stream, outside the normal study section, and were used to determine fecundity. The mean number of eggs per female (± 2 SE) was 39 ± 4 for 0+ fish (n = 3), 64 ± 3 for 1+ fish (n = 53), 76 ± 3 for 2+ fish (n = 41), and 78 ± 6 for 3+ fish (n = 15). There was no significant difference between the mean number of eggs for 2+ and 3+ females. It was thus possible to determine egg density at the start of each year-class from the number of spent females taken in the May/early June samples and the mean number of eggs per female in each age class. As there were so few mature 0+ females, these were ignored in the estimates of total egg production.

statistical methods

Of six stock-recruitment models tested earlier on the data from Black Brows Beck, only the following two-parameter model was found to be a suitable general model for the different life stages of the trout (Elliott 1985):

image(eqn 1 )

where R = numbers of survivors at different stages in the life cycle, S = number of eggs at the start of each year-class, and a and b are parameters. This model is often is named after Ricker (1954), but is a common density-dependent model that has been used in many terrestrial studies (Elliott 1994). The parameters of the Ricker curve were estimated by nonlinear least squares (Elliott 1985). Outliers that adversely affected the fit of the model were identified by a step-wise procedure based on analyses of standardized residuals, and were excluded from the fitting procedure when Cook's D statistic was greater than 1 (Cook & Weisberg 1982). For all other analyses, standard statistical methods were used. Non-significant relationships were all at the 5% level of probability (P > 0·05) and probability values are given for all significant relationships, together with coefficients of determination adjusted for sample size (r2).

Results

During the trout spawning period in November and December, the wettable area, and hence the habitable area (h%), was less than 80% in only 1 year (1993) and was bank full (100%) in 28 of the 35 years of observations (Fig. 1a). When the samples were taken in late May/early June, the habitable area was 100% in 17 of the 34 years and varied in the range 70–95% for the remaining years (Fig. 1b). The lowest h-values were recorded during the late August/early September sampling with a habitable area of 100% in only six of the 34 years (Fig. 1c). Values were in the low range 55–70% in 1969, 1976, 1990 and 1993, and the even lower 30–50% in 1983, 1984, 1989 and 1995. All these low values, except that for 1990, were associated with the seven summer droughts.

Figure 1.

Habitable area (h%) within the sampling area in each year: (a) during November and December (minimum values only, maximum values always 100%); (b) during sampling period in late May/early June; (c) during sampling period in late August/early September

relationships for trout densities

The Ricker curvilinear model (eqn 1) significantly fit (P < 0·001) the relationship between initial egg density (S eggs 60 m−2) and survivor density (R fish 60 m−2) for: parr aged 0+ years sampled in late May/early June (R1), and late August/early September (R2); and 1+ parr sampled in late May/early June (R3), and late August/early September (R4) (Fig. 2a–d). There were no analyses for 2+ parr because few were taken in the May/June samples, most having migrated downstream to the sea. These analyses provided strong evidence for density-dependent population regulation and also identified extreme outliers (labelled year-classes in Fig. 2a–d), most being for year-classes affected by summer droughts. There were no significant relationships between survivor densities and habitable area because of the strong density dependence. However, significant relationships were obtained when the variable effects of changes in habitable area were quantified by using the residuals, each residual being the absolute value (Obs) expressed as a percentage of the expected value (Exp) from the Ricker curve (100(Obs − Exp)/Exp in Fig. 2(e–h).

Figure 2.

(a–d) Relationship between egg density (S eggs 60 m−2) and survivor density (R fish 60 m−2) for: (a) trout parr aged 0+ years sampled in May/June (R1), and (b) August/September (R2); and (c) trout parr aged 1+ years sampled in May/June (R3), and (d) August/September (R4). Ricker curves estimated from eqn 1 with: a (SE) = 0·437 (0·021), b (SE) = 0·00039 (0·00001) for R1; a = 0·086 (0·006), b = 0·00028 (0·00001) for R2; a = 0·033 (0·029), b = 0·00026 (0·00002) for R3; a = 0·0344 (0·029), b = 0·00032 (0·00002) for R4. (e–h) Relationship between the residuals ((Obs − Exp)/Exp as Y%) and habitable area (h%) for: (e) R1; (f) R2; (g) R3; (h) R4. Regression lines given by: Y = bh − a where b = 0·86, a = 77·80 for R1; b = 1·23, a = 104·12 for R2; b = 0·90, a = 77·66 for R3; b = 1·23, a = 104·45 for R4. Year-classes identified as outliers and excluded from analyses are given for each figure.

The best fits of the Ricker curve were for R1 and R2 (excluding the outliers for R2) with r2 values of 0·87 and 0·72, respectively. Only the 1996 value for R1 differed by more than 20% from the expected value (Fig. 2a), this being due to the samples being taken during a flood that impeded the catching of trout while electrofishing. When this value was excluded, there was a significant positive relationship (P < 0·001) between the residuals and habitable area, the latter explaining 76% of the variation in the data (Fig. 2e). However, the difference between observed and expected values was never very large (< 20%). The 1993 value for R2 was much higher than expected (Fig. 2b). There was no definite reason for this but it was possible that egg density at the start of the year-class was underestimated. The 1983, 1984, 1989 and 1995 values were well below expected, all being associated with summer droughts. Apart from the 1993 year-class (excluded from the analysis), all values fitted the significant positive relationship (P < 0·001) between the residuals and habitable area, the latter explaining 83% of the variation (Fig. 2f). This relationship covered a much wider range of h-values than that for R1 and showed clearly the marked effect of habitat loss. The habitable area was ≤50% for the four values most affected by summer droughts (h = 35% for 1983, 30% for 1984, and 50% for 1989 and 1995 year-classes).

It was difficult to interpret the relationships for R3 because this was the first sample taken after the first winter of the life cycle and several factors could have affected survivor density during the period from late August/early September to late May/early June The Ricker curve explained just over half the variation in survivor density (r2 = 0·56) when the three year-classes most affected by the previous summer droughts were excluded from the analysis (labelled year-classes in Fig. 2c). The low value for the 1995 year-class was also due to the difficulty in catching fish during the 1996 flood. The relationship between habitable area and the residuals (Fig. 2g) was not significant (r2 = 0·11).

In interpreting the relationship for R4, the immediate effects of a summer drought were separated from those of the previous summer. The Ricker curve explained just over half the variation in survivor density (r2 = 0·58), excluding the five year-classes most affected by droughts (labelled year-classes in Fig. 2d). For two of these year-classes, the percentage difference from the expected value was the same as that for R3 (−68% for 1984, −73% for 1995 year-class), indicating that the low survivor densities were due to the drought effects of the previous summer. The remaining three year-classes were all affected by the summer droughts of 1969 (1968 year-class), 1984 (1983 year-class, already reduced by 1983 drought), and 1995 (1994 year-class). As the drought effects were immediate for these year-classes, they were included in the significant positive relationship (P < 0·001) between the residuals and the habitable area, the latter explaining 53% of the variation in the data (Fig. 2h). Therefore, habitat loss had a marked negative effect on the survivor density of 1+ trout as well as 0+ trout.

relationships for bullhead densities

Egg densities for bullheads were much lower than those for trout; they were always < 700 eggs 60 m−2 whereas only one value for trout was below this density. Although the curve was not dome-shaped, the Ricker model (eqn 1) significantly fit (P < 0·001) the relationship between egg density (S eggs 60 m−2) and survivor density (R fish 60 m−2) for: bullheads aged 0+ years sampled in late August/early September (R1), and late May/early June (R2); 1+ fish sampled in late August/early September (R3), and late May/early June (R4); and 2+/3+ fish sampled in late August/early September (R5) (Fig. 3a–e). The number of 2+/3+ bullheads taken in the following May/June was too low for eqn 1 to be fitted. A simple linear model was also a significant fit (P < 0·05) to the data. However, the Ricker model produced a consistently lower residual variance and coefficient of determination: values of r2 for R1, R2, R3, R4 and R5 were 0·85, 0·83, 0·78, 0·65 and 0·54 for the Ricker model and 0·60, 0·65, 0·63, 0·53 and 0·50 for the linear model, respectively. Therefore, there was evidence for density-dependent population regulation, but it was not as strong as that for trout. Extreme outliers were also identified in the analyses and these were all associated with summer droughts (labelled year-classes in Fig. 3a–e). Slightly lower numbers than expected were taken during the flood in May 1996, but the effects were not as marked as those for trout. As there was no direct relationship between survivor density and habitable area, the same method as that for trout was used to examine the variable effects of changes in habitable area (Fig. 3f–j).

Figure 3.

(a–e) Relationship between egg density (S eggs 60 m−2) and survivor density (R fish 60 m−2) for: (a) bullheads aged 0+ years sampled in August/September (R1), and (b) May/June (R2); and (c) bullheads aged 1+ years sampled in August/September (R3), and (d) May/June (R4); and (e) bullheads aged 2+/3+ years sampled in August/September (R5). Ricker curves estimated from eqn 1 with: a (SE) = 0·161 (0·004), b (SE) = 0·00065 (0·00002) for R1; a = 0·060 (0·002), b = 0·00042 (0·00001) for R2; a = 0·054 (0·019), b = 0·00042 (0·00002) for R3; a = 0·033 (0·029), b = 0·00061 (0·00002) for R4; a = 0·029 (0·022), b = 0·00066 (0·00002) for R5. (e–h) Relationship between the residuals ((Obs − Exp)/Exp as Y%) and habitable area (h%) for: (f) R1; (g) R2; (h) R3; (i) R4; (j) R5. Regression lines given by: Y = a − bh where b = 1·14, a = 99·59 for R1; b = 0·49, a = 52·36 for R2; b = 0·93, a = 82·59 for R3; b = 0·44, a = 44·65 for R4; b = 1·04, a = 84·09 for R5. Year-classes identified as outliers and excluded from analyses are given for each figure.

The best fits of the Ricker curve were for R1 and R2 with r2 values of 0·85 and 0·83, respectively (Fig. 3a,b). Progressively poorer fits were obtained for the other life stages with r2 values of 0·78 for R3, 0·65 for R4, and 0·54 for R5. The sampling dates for R1 were the same as those for R2 for trout and the same four year-classes were affected by summer droughts. In contrast to those for trout, the values for bullheads were much higher than expected from the Ricker curve (labelled year-classes in Fig. 3a). All values fitted the significant relationship (P < 0·001) between the residuals and habitable area, the latter explaining 90% of the variation in the data (Fig. 3f). This relationship was negative, in contrast to that for trout (cf. Figs 2f and 3f).

It was difficult to interpret the relationships for R2 because this was the first sample taken after the first winter of the life cycle and several factors could have affected survivor density, as also for trout. The relationship between habitable area and the residuals (Fig. 3g) was not significant (r2 = 0·04), a result similar to that for R3 for trout. This pattern was repeated in the following life stages with significant negative relationships (P < 0·001) between habitable area and the residuals for R3 (r2 = 0·67, Fig. 3h), and R5 (r2 = 0·77, Fig. 3j), and a nonsignificant relationship for R4 (r2 = 0·02, Fig. 3i). Therefore, habitat loss had a marked positive effect on the survivor density of 0+, 1+ and 2+/3+ bullheads during the summer.

comparison of trout and bullhead densities

There was no significant relationship between the densities of bullheads (R1) and trout (R2) at the end of the first summer of their life cycles (0+ fish, Fig. 4a), or between the densities of bullheads (R3) and trout (R4) at the end of the second summer (1+ fish, Fig. 4b). A similar absence of a relationship was obtained when mortality rates for bullheads and trout were plotted against each other for the first (0+ fish) and second (1+ fish) summers of their life cycles. When density-dependent effects were removed by comparing the residuals, significant negative relationships (P < 0·001) were evident (Fig. 4c–f). The strongest relationship (r2 = 0·84) was for 0+ fish, comparing bullheads (R1) and trout (R2) sampled at the end of the first summer (Fig. 4c). The significant relationships were weaker for 1+ fish (r2 = 0·39), comparing bullheads (R3) and trout (R4) sampled at the end of the second summer (Fig. 4d), and for comparisons between 0+ bullheads (R1) and 1+ trout (R4) (r2 = 0·33, Fig. 4e), and 1+ bullheads (R3) and 0+ trout (R2) (r2 = 0·39, Fig. 4f). Therefore, there was clear evidence that bullhead densities increased at the expense of trout densities, but this was only apparent after correction for density dependence.

Figure 4.

Relationship between: (a) densities of bullheads (R1) and trout (R2) at the end of the first summer of the life cycle (0+ fish); (b) densities of bullheads (R3) and trout (R4) at the end of the second summer (1+); (c) % differences for 0+ bullheads (R1) and 0+ trout (R2); (d) % differences for 1+ bullheads (R3) and 1+ trout (R4); (e) % differences for 0+ bullheads (R1) and 1+ trout (R4); (f) % differences for 1+ bullheads (R3) and 0+ trout (R2). Regression lines given by: Y = a − bX where for: (c) a = 3·39, b = 0·82; (d) a = 2·13, b = 0·47; (e) a = 2·74, b = 0·38; (f) a = 4·05, b = 0·63 (1993 year-class was excluded from the analyses in (c) and (f) for the reasons given in the text).

Discussion

This study has shown clearly that as habitable area decreased in Black Brows Beck, especially during the summer, bullhead density increased at the expense of trout density. However, it was necessary to correct for density-dependent effects before the relationships with habitable area became clear. A summary of the relationships showed that they were very similar for 0+ fish, 1+ fish and older bullheads (Fig. 5). Values were identical for bullheads and trout when habitable area was 86% for 0+ and 1+ fish, and 83% for older bullheads (point where lines cross in Fig. 5). Bullheads and trout coexisted over a wide range of habitable area values in Black Brows Beck, at least in the range 30–100%. Although bullheads were never as numerous as trout, their relative density increased, while that of trout decreased, when habitable area decreased progressively from 83–86% to 30%.

Figure 5.

Summary of the effects of changes in habitable area (h%) on: (a) 0+ bullheads and 0+ trout at the end of their first summer; (b) 1+ bullheads and 1+ trout at the end of their second summer; (c) 2+/3+ bullheads and 1+ trout.

These relationships are very similar to those predicted by the theoretical metapopulation model of Nee & May (1992) (see their Fig. 2). They first proposed that habitat loss can lead to an increase in the absolute abundance of an inferior competitor while decreasing the abundance of a superior competitor. At first, this appears to be counterintuitive because inferior competitors persist by colonizing empty patches, so it is surprising that patch removal increases their abundance. However, the direct negative effect of habitat loss on the inferior competitor is more than compensated by the reduced competition with the superior species. The present study suggests that these processes may also facilitate the competitive coexistence of two species living in a changing mosaic of microhabitats in the same locality.

Bullheads and trout are often found together in small, stony, streams and share the same general habitat, with both species feeding on stream invertebrates. However, there are differences in their habitat preferences and feeding behaviour, and these may help to explain their different responses to a reduction in habitable area.

Juvenile 0+ trout are usually found in small groups in open water, but each fish within the group has its own feeding territory (Elliott 1994). They are often located in the most profitable feeding positions where they are sheltered by a large stone from the direct effects of the current but can swim rapidly to intercept invertebrates drifting in largest numbers where the current is highest. They feed on both aquatic and terrestrial invertebrates in the drift. Shadow competition occurs within each group with the dominant trout at the upstream end of the group (Elliott 2002). The 0+ trout prefer a coarse, cobble, substratum, and are usually found in shallow stream areas (< 20–30 cm depth) and at moderate water velocities (20–50 cm s−1) (Heggenes 1996). Older trout are usually found in deeper water but require cover during the day, usually under the banks or large stones (Heggenes, Bagliniére & Cunjak 1995; Heggenes 1996). Although they continue to feed on drifting invertebrates, they also feed on larger invertebrates on the stream bed (Elliott 1994). Bullheads prefer a substratum of larger stones with crevices large enough to provide shelter; hence there is a strong relationship between bullhead size and preferred stone size (Smyly 1957; Welton, Mills & Rendle 1983; Welton, Mills & Pygott 1991). Bullheads also prefer low water velocities (< 20 cm s−1), sites where aquatic macrophytes are absent, and shaded areas, but are indifferent to water depth (Gaudin & Caillere 1990). They are most active at night, especially at dusk and dawn, when they forage for benthic invertebrates (Elliott & Elliott 1995). As noted in the introduction, direct competition between bullheads and trout may occur when their habitat is restricted or when food is scarce because densities of benthic invertebrates are low. When habitable area in Black Brows Beck was greater than about 70%, the mean percentage difference from the Ricker curve was less than ±20% for both species (Fig. 5), and there was probably little direct competition for food or space. However, as habitable area decreased progressively below 70%, departure from the Ricker curve increased markedly for both species, probably because of a change in environmental conditions that favoured the bullheads but not the trout. There was a reduction in flow and hence drift food for the young trout. This would have little effect on the nocturnally foraging bullheads, and the reduction of flow would probably facilitate the capture of their invertebrate food. Freshwater shrimps Gammarus pulex L. increased in abundance during droughts and were the preferred food of both species. This shift in favour of bullheads would continue while habitable area was being reduced. During severe droughts, trout mortality was very high in smaller streams neighbouring Black Brows Beck and trout were absent when the habitable area was in the range 15–20%. Both the width and depth of the streams decreased, and flow often ceased. They were reduced to a series of small pools; temperatures were too high and oxygen levels too low for trout to survive (Elliott 2000). However, bullheads were still present, probably because of their higher temperature tolerance and wider thermal limits for feeding when compared with trout (Elliott & Elliott 1995). Their numbers did not decrease markedly until habitable area was in the range 10–15%, with corresponding decreases in width and depth. These severe conditions never occurred in Black Brows Beck, and there was always some flow, probably because the stream originates from underground springs.

The evidence for direct competition between bullheads and juvenile salmonids is equivocal. Bullheads introduced into a river in northern Finland sometime before 1979 had no significant effects on juvenile salmon densities (Pihlaja et al. 1998). Gabler, Amundsen & Herfindal (2001) examined the diets of the two species in the same river and concluded that there was little evidence for competition between them. Similarly, bullheads were found to have no detectable effects on densities of accompanying fish species, including juvenile salmon and trout, in the River Avon in southern England (Carter et al. 2004). However, in experimental studies in artificial streams, the downstream drifting of newly emerged trout fry increased in the presence of bullheads, and predation by bullheads on the fry was high for the first few days after their emergence from the gravel nest (Gaudin & Heland 1984; Bardonnet & Heland 1994; Gaudin & Caillere 2000). Significant differences in early life-history and reproductive traits between brown trout living above and below a waterfall, as well as lower densities of trout below the waterfall, were associated with the presence of the alpine or Siberian bullhead Cottus poecilopus Heckel below the waterfall (Olsen & Vøllestad 2001, 2003). However, there was no evidence for direct competition between the two species. The diet of the alpine bullhead was almost identical to that of salmon parr in a river in northern Norway and it was concluded that the two species were competing for food (Gabler & Amundsen 1999). However, as noted earlier, direct competition will occur only when food is scarce, and there was no evidence for this in the Norwegian river. Finally, it should be noted that there is little evidence to support the persistent claim that bullheads eat the eggs of salmonids; erroneous conclusions probably arose from careless examination of ripe female bullheads, distended with their large orange eggs (Mills & Mann 1983).

The present study has shown that there was no simple direct relationship between bullhead and trout densities in Black Brows Beck. The relationship between the two species was more complex and was influenced by at least one external factor, namely changes in habitable area. This case study shows that, during a major perturbation, the relationship between the densities of two species can change markedly in favour of the less numerous species. The competitive coexistence between the two species is therefore a dynamic process that changes through time with periodic changes in the environment.

Acknowledgements

This work was financed by the Natural Environment Research Council through a grant to the Freshwater Biological Association.

Ancillary