Environmenral Biology of Fishes 26: 285-294. 1989. 0 1989 Kluwer Academic Publishers. Prinred in rhe Nerherlands. Grazing catfish, fishing birds, and attached algae in a Panamanian stream Mary E. Power', Tom L. Dudley: & Scott D. Cooper? ' Department of Biological Sciences, University of California, Santa Barbara, Santa Barbara, CA 93616, U.S.A. Department of Zoology, University of California, Berkeley, Berkeley, CA 94720, U.S.A. Received 18.7.1988 Accepted 3.1.1989 Key words: Predator avoidance, Spatial variation in herbivory. Tropical stream communities. Multi-trophic level interactions Synopsis In streams where algivorous fishes abound, striking variation of attached algae often develops along depth gradients. with bands of high standing crops in shallow water (<20cm) and sparse standing crops on deeper substrates. Experimental results from a stream in central Panama support the hypothesis that vertical variation in algal standing crops arises when grazing fishes avoid predators in shallow water by forgoing food resources that accumulate there. When 38 rocks bearing algae in a stream in central Panama were transferred from shallow (<30cm) to deeper (>20cm) water. algae were rapidly consumed by grazing catfish. Catfish were removed from three stream pools and left in place in three control pools. Ten days after catfish removal. algal standing crops in deep and shallow areas of removal pools were similar. while algal standing crops were higher in shallow than in deep areas of control pools. Catfish were exposed to fishing birds in open-topped enclosures. In one of three series of these pens, most catfish in shallow pens (10 and 20cm) disappeared after 14 days. while catfish in deeper pens (30 and 50cm) did not. Other groups of catfish which were caged 8 days showed differences in behavior depending on whether they had been fed or starved. After their release into their home pool. starved catfish spent more time feeding than did fed catfish. Despite their apparently increased hunger levels. starved catfish did not venture into shallow water to obtain algae. These results support the view that predator induced avoidance by grazers of certain areas can produce spatial pattern in the flora of flowing water communities. Introduction By restricting the foraging area of their prey. pred- ators can create spatial refugia for organisms on which their prey feed. Multi-trophic level effects of predators via their sublethal influences on prey behavior have been found in a growing number of freshwater habitats (Schlosser 1987. 1988. Mittel- bach 1984. Gilliam & Fraser 1987. Holomuzki 1986, Power 1987. Kerfoot & Sih 1987). Distribu- tions of attached algae often clearly indicate spatial variation in herbivory, which may arise from pred- ator avoidance by grazers. or from other factors (Hay 1981. 1984. Sammarco 1982. Power 1983. Power et al. 1985. Lubchenco 1978. and many others). -We investigated the role of predator avoidance by grazers in establishing vertical gra- dients in algal standing crops that are widespread in streams populated by grazing fishes. In tropical and temperate streams with high den- sities of grazing fishes. bands of algae often develop along shallow margins. but algal abundance drops 286 off steeply when depths exceed 15-25cm. This depth corresponds to the maximum depths at which wading and diving birds feed most frequently and effectively (Whitfield & Cyrus 1978, Boag 1982, Kramer et al. 1983, Power 1984a, 1987). Bands of algae along stream margins have been observed in streams in Panama (Power 1984a) and Trinidad (J.A. Endler, personal communication) inhabited by grazing catfishes (Loricariidae), and in streams in Oklahoma with dense populations of grazing minnows, Campostoma anomalum (Power & Matthews 1983). Here we report results from a series of simple experiments that tested three hy- potheses: H(1): Grazing by catfish limits algal depth distri- butions. Predictions: Algal abundance will de- crease as catfish densities increase with depth; in pools from which catfish are removed, algal den- sities will be more similar on deep and shallow substrates. H(2): Predation hazard, primarily from fishing birds, constrains catfish depth distributions. Pre- dictions: Catfish will avoid water shallower than 20 cm. where birds fish frequently and effective- ly. Catfish confined in shallow water will dis- appear faster than catfish confined in deeper water. H(3): Catfish will take risks if rewards are suffi- cient. Prediction: Catfish will graze in shallow water, where algal standing crops are high, when sufficiently hungry. much sunlight reaches the bed. Where the river runs through more resistant basalts, channel walls are steep and stable, forest grows to the edge of the stream, and canopies <25% open shade the bed. The Rio Frijoles supports dense populations of three species of algae-grazing loricariid catfish: Rineloricaria uracantha Kner & Steindachner, An- cisrrus spinosus (Eigenmann & Eigenmann) and Hyposfornus plecostomus (Linnaeus). A fourth species, Chaetostoma fischen' Steindachner, is less common and occurs primarily where deep, fast flowing habitat is available (Power 1981, J.D. McPhail, D.L. Kramer, G.E.E. Moodie. unpub- lished data). Armored catfish in the Rio Frijoles graze almost entirely on periphyton on hard substrates: bed- rock, boulders, cobbles, and large pieces of sub- merged wood. They are the only major grazers of attached algae in deeper stream habitats (Power 1981), where invertebrate grazers appear to be lim- ited by high densities of predatory fishes. During an intensive 28 months field study from 1978 to 1980. algal standing crops were very scant on sub- strates that were deeper than 20cm, but relatively large standing crops of algae, including filamentous bluegreens, occurred in shallower habitats along stream margins (Power 1984a). In March of 1985, when we carried out experiments reported here, algal standing crops were also generally higher in shallow water. Methods and results Study site The Rio Frijoles is a clear stream that flows over marine sedimentary rocks and metamorphic basalts in the Parque Nacional Soberania of central Panama (9O9' N, 79"44' W). Baseflow discharge va- nes from 0.5 m3 s-' in the rainy season to 0.1 m3 s-' in the dry season (Power 1981). Most of the stream's watershed lies within mature second growth forest; only short reaches of some head- waters lie within land that is presently cleared. In reaches underlain by soft sedimentary marine de- posits. the stream maintains a wide floodplain. for- est canopy is >25% open (Power 1984b), and Algal transfers. We tested the hypotheses that al- gae in shallow water were unpalatable. or that they accumulated too rapidly to be controlled by graz- ers, by transferring 38 algal-covered cobbles from shallow to deep water. Grazing catfish removed all conspicuous algae from these substrates within hours, and sometimes within minutes, showing that the food was acceptable, and that catfish could easily have depleted periphyton in shallow water had they fed there. Effect of depth on disappearance of loricariids from open-fopped pens. The most common piscivorous 287 birds sighted along the Rio Frijoles were mature and immature little blue heron, Egrefta caerufea, green kingfishers. Chloroceryle americana, and night-fishing rufescent tiger herons, Tigrisoma lin- eatus. Birds observed fishing during the 1978-80 study (Power 1984b) and the 1985 study reported here generally foraged in water <20 cm deep (in 27 out of 28 observations where depth of fishing was measured). To examine the vulnerability of armored catfish at various depths to predators, we enclosed fish in open-topped pens that were similar in area (1 m') and substrate (gravel <45 mm median diameter), but were set in 10,20,30 and 50cm of water. Pens were made of black plastic screen (6.4mm mesh) supported by four 2.54cm diameter PVC pipes. The tops of pen walls projected about 10 cm above the water surface. Four pens (one at each depth) were installed in each of three reaches of stream where we had previously seen little blue herons fishing. We stocked each pen with size-matched groups of the two most common species of lor- icariids: Rineloricaria uracantha, a thin cryptic spe- cies abundant in deeper riffles and in heads, tails and margins of pools, and Ancistrw spinosw, the most common species in deeper stream pools (Fig. 1). After twelve days, catfish remaining in pens were counted. P In one of the series of four pens, most R. uracantha had disappeared from pens in 10 and 20 cm of wa- ter. but the majority remained in the 30 and 50cm deep pens (Fig. 2). Roughly half of the A. spinosw were also missing from the 10 and 20cm pens. and most were still present in pens that were 30 and 50cm deep (Fig. 3). We did not witness predation. but saw both little blue herons and green king- fishers within one meter of the shallow cages in this series .of pens on several occasions. At the other two stream reaches, all catfish re- mained in all the pens when experiments were ter- minated after twelve days of exposure. We do not have sufficient information to evaluate whether site differences or simply the short duration of the experiment accounted for the lack of predation at these other two reaches. Fig. 1. Ancistm spinosus and Rineloricaria uracantha, the two most common loricariid catfish in the Rio Frijoles. Effect of loricariid removal on attached algae in pools. Three pairs of pools. one sunlit, one half shaded, and one darkly shaded, were selected. Paired pools were physically similar and close or adjacent along the reach (Table 1). Loricariids were counted in pools by a snorkeller (MEP) who 1 Mar 10 Mar 22 20 cm 20 cm c .- co Standard length (cm) Fig. -7. Numbers of R. uracanrha stocked on March 10 and remaining on 'March 22 in open-topped enclosures at four water depths. 4 288 A ncisirus I Mar 10 1 Mar 22 I 20 cm E 6- 1 *Ocm `I 1 3- C .- n 30cm I 30 cm * 31- go f 61 ot ql I 53 ,50cm I 3-6 6-15 0 3-6 6-15 Standard length (cm) Fig. 3. Numbers of A. spinosrcs stocked on March 10 and re- maining on March 21 in open-topped enclosures at four water depths. recorded numbers and the estimated lengths of each species. One pool from each pair was selected by coin toss for loricariid removal. Fish were netted by snorkellers using aquarium net frames fitted with monofilament gill net, a capture technique that only minimally disturbed the habitat. Over a 6d period, we removed sufficient numbers of ior- icariids from these pools to depress their numbers (Table 2). In removal pools, the numbers of cat- fish, particularly of R. urucunthu, that were re- moved often exceeded those previously counted. This may be due to immigration of fish to pools during the removal period. as pools were adjoined by deep riffles which provided habitat for small loricariids and passage for all size classes. In addi- tion. resident R. uracantha may have been over- looked during counts. as they are extremely cryptic when quiet and half-buried in sand. The smallest of the four species. R. uracantha has lower grazing rates (as estimated by area covered per individual per time) than the larger loricariid species (Power 1984b). Before removal, and 10 and 17d after removal, algae were collected from substrates above and below the 20 cm depth contour in each pool. With an aspirator fitted with a glass intake nozzle. we simultaneously scraped and aspirated algae from known areas. using a flexible template pressed to the substrate. Algae were preserved in 1-3'6 Lu- gol's solution, and later examined under 3OOx and Table 1. Charactenstics of paired pools in loricariid removal experiments. Pool Length (m) Max. width (m) Area* (m') Max. depth (cm) Canopy (% open) Substrate Otter Scat (control) Pyralid (removal) Lower Polistes (control) Upper Polistes (removal) Cecropia (control) Blue Clay (removal) 10-25 upstream basalt 39.0 9.4 288 100 <10 downstream bedrock 22.1 4.9 85 97 <10 basalt bedrock 32.0 5.2 131 74 25-50 bedrocb cobbles 22.9 4.2 76 76 3-50 bedrock cobbles 12.7 4.2 42 45 > 50 cobbles clay 9.7 5.5 42 55 >50 cobbles Clay ~~ o Plan view ares estimated by assuming pools are elliptical in shape 289 weighed in the laboratory. Preserved samples were dried to constant weight at 80°C. weighed. com- busted 1 h at 51OoC, and re-weighed for determina- tion of ash-free dry weight. Unfortunately, most algal samples were lost in transit between Panama and California. Too few samples were available to compare algal standing crops in the two sunny pools on any date, or in the other pools before loricariid removals, or 17d af- terwards. Sufficient samples were available to com- pare algal biomass in the dark and half-shaded pools 10 d after loricariid removals. Algal standing crops in control pools were 4-20 times higher in shallow than in deep areas. whereas algal standing crops in removal pools were similar in shallow and deep areas (Table 3). Using the mean algal biomass for deep and shallow areas of each of the four pools as replicates (two depths x four pools = eight val- ues). we performed a two-way ANOVA examining the effects of treatment and depth. and their inter- action. on algal standing crops. Main effects of treatment and depth on algal biomass were not significant (F= 2.4 and 0.2. p= 0.19 and 0.70. df = 1 and 1. respectively). The interaction of depth and treatment. however. could account for a significant amount of the variation in algal biomass (F = 8.8. p = 0.04, df = 1). These results suggest that differences in algal standing crop between shallow and deep areas of pools were reduced by the removal of loricariids. Effect of food deprivation on loricariid behavior Table 2. Effect of loricariid removal on loricariid numbers in pools and space use. To examine the effect of hunger on the tendency of loricariids to forage in shallow water, we manipulated hunger levels of catfish from a single large pool with a dense catfish stock. Before the manipulation, the pool was mapped, and areas of hard bedrock substrate, and soft grav- el, sand and detrital substrates were measured. In addition, we measured the area of stream bed un- der four depth intervals (Table 4). Markers were placed on the stream bed to delineate depth con- tours so that the loricariids occupying various sub- strates and depth intervals could be counted during bankside observations. Sitting on the bank. one of us (MEP) observed catfish grazing over a gridded bedrock platform marked off into squares 0.5 m on an edge. At 10-minute intervals from 1100 to 1140 h. the numbers and estimated lengths of vari- ous species and size classes of loricariids on the bedrock platform, and on areas of soft substrate, were recorded. After these baseline observations. we caged two groups of closely size-matched loricariids captured from the mapped pool. Fifteen R. uracanrha (SL range = 6:-105 mm), 7 A. spinosus (50-!23 mm) and 1- H. piecosromus (55-71mm) were placed in each cage. Catfish were enclosed in large 1 m3 roof- ed cages (mesh size = 6.4mrn). One group was held without food: the other was fed on attached algae transferred from stream margins and on canned green beans. After the first three days of captivity, captive loricariids readily consumed both foods. Fed and starved groups were marked with ~ Pool Loncanids counted Total No Loncanids removed Loncanids counted Total Yo (before removal) m-2 (after removal) m-z March 7-12 R' A" H' C' RAHCRAHC Otter Scat 3 65 6 200.070 00 0 2 7 0 6 !5 0.05 Pyralid 22 11 0 3 36 0.42 25 12 3 2 8 3 1 2 11 0.17 Lower Polistes 24 7 2 0 33 0.25 0 0 0 0 43 7 3 0 52 11.40 UpperPolistes 26 7 2 0 35 0.46 34 Y 6 0 12 3 0 0 !5 '1.20 5 11,:2 Cecropia 20 3 1 4 27 0.64 0 0 0 0 19 3 1 - '6 11.62 Blue Clay 22 13 1 1 37 0.88 55 13 1 2 31 !0 * R = Rineioricaria. A = Ancrsirus. H = Hyposiomus. C = Chaerosroma 290 differently colored plastic beads attached with stainless-steel wire through the dorsal armor imme- diately in front of the first dorsal spine. Mer 8 d, both groups were released into their home pool (cages were opened, and fish allowed to swim out). Before catfish were caged, they tended to active- ly feed on hard substrates, and to rest on soft sub- strates where they were less conspicuous (Rinefor- icarza) (Fig. 4) or under bedrock ledges, out of sight (Anchus, Hyposfomus). After experimental ma- nipulation of hunger, a greater proportion of the starved individuals sighted were on hard substrates and were feeding, relative to fed catfish or catfish observed before caging (Fig. 4). Fewer experimen- tally fed catfish were observed on platforms be- cause replete Ancistrus and Hypostomus rested un- der ledges (out of sight). Fed and starved individu- als were sighted during snorkelling censuses, show- ing that they had not emigrated from the pool. On Tuble 3. Effect of loricariid removal on periphyton biomass in pools. Pool Periphyton biomass (mg afdw cm-') <20cm depth >20cm depth mean SE (n) mean SE (n) Otter Scat (control) 4.0 1.8 (3) Pyralid (removal) 1.1 0.2 (3) Lower Polistes (control) 2.4 0.8 (4) Upper Polistes (removal) 1.4 0.6 (4) 0.2 0.1 (3) 1.2 0.6 (4) 0.6 0.5 (3) 0.8 0.3 (3) Tuble 4. Area of hard and soft substrates in the hunger mani- pulation pool. Depth (cm) Hard substrate Softmbstrate Total (m') (m7 (m9 <10 3.98 6.09 10.07 1&20 4.23 8.14 12.37 214 7.50 17.17 24.67 >a 1.62 14.92 19.54 Total 20.33 46.32 66.65 Hard Soft Before i.EJ!.L 20 0 c Fed Starved 20 r Fig. 4. Positions and behavior of loricariids before they were caged, and on the day of their release, after they were starved or fed in cages in their home pool. 'Hard substrate' is a bedrock platform; 'soft substrate' is gravel. sand. or detritus. each of 13 scan samples (Altmann 1974) made on the day of release, more starved than fed catfish were sighted on platforms, and more were feeding (Table 5). Larger proportions of starved than fed catfish, and of starved than unmarked catfish. were feeding when sighted during the 13 scan samples. Unmarked catfish, which had not been caged. were free to feed on natural substrates, and therefore were presumably less hungry than were caged fish that had been deprived of food. Of the uncaged and fed loricariids sighted on the bedrock platform, <26% were feeding, whereas over 75% of the starved fish that were observed were feeding (Ta- ble 5). One potential source of dependence in these data is the influence of a fish's behavior during one scan on its own or another fish's behavior during the next scan. This potential dependence is re- duced by the short duration of the behavioral states (resting, which usually lasted <30s, and feeding, 29 1 which usually lasted ~20s) relative to the 10min interval between scan samples. Another source of dependence could be time dependent changes in the probability of an individual's resting or feeding, for example, if fish satiated. However, there were no trends in the proportions of sighted starved or fed fish that were feeding or resting over the two hour observation period. Because these sources of dependence seemed minor, we performed median tests (Zar 1984) on proportions of fish sighted feed- ing vs. resting. The results (fed vs. starved: chi square= 6.04, pcO.025, df= 1: uncaged vs. starved: chi square= 7.58, p