EVOLUTION INTERNATIONAL JOURNAL OF ORGANIC EVOLUTION PUBLISHED BY THE SOCIETY FOR THE STUDY OF EVOLUTION

Vol. 36 July, 1982 No.4

Evolution, 36(4), 1982, pp. 637-657

NICHE SHIFTS AND COMPETITION IN DARWIN'S FINCHES: GEOSPIZA CONIROSTRIS AND CONGENERS

B. R. GRANT AND P. R. GRANT Division of Biological Sciences, University of Michigan, Ann Arbor, Michigan 48109

Received February 5, 1981. Revised October 20, 1981

The idea that interspecific competition titative ecological data to the hypotheses, is an important process in structuring because quantitative data were not used communities stems largely from David to construct the hypotheses. In fact, Lack Lack's work with Darwin's Finches (Lack, had almost no ecological data (see Abbott 1940, 1945, 1947, 1969). Lack made eco­ et al., 1977). logical inferences about feeding niches In our initial studies we analyzed eco­ from analyses of bill sizes and shapes. He logical and morphological data from six listed several instances where the niches species on eight Galapagos islands (Abbott of coexisting species were different, and et al., 1977; Smith et al., 1978). To follow where the niche of an absent species ap­ this general approach, we selected for more peared to be occupied by one or more detailed study three situations which have species present. These examples he inter­ been heralded as especially clear illustra­ preted as evidence of competitive dis­ tions of competitive effects (e.g., Lack, placement and exclusion. 1969; Williamson, 1972; Arms and Camp, Lack's aim was to offer a coherent the­ 1979), the small size of Geospizafortis on oretical framework for understanding the Isla Daphne Major in the absence of G. adaptive radiation of Darwin's Finches, fuliginosa (Boag, 1981), mutually exclu­ not to test specific hypotheses. In contrast, sive distributions of G. fuliginosa and G. our aim is to test the hypothesis of inter­ difficilis (Schluter and Grant, unpubl.), specific competition. This is needed be­ and differences in bill size between pop­ cause the evidence in general for interspe­ ulations of G. conirostris. Here we report cific competition as an important factor in the results of the third study in which we determining niche relationships and com­ test the hypothesis of competition between munity patterns has been critically debat­ G. conirostris and congeneric species. ed recently (e.g., Connell, 1975, 1978, 1980; Wiens, 1977; Connor and Simber­ The Interspecific Competition loff, 1978, 1979; Simberloff, 1978; Strong Hypothesis Applied to et al., 1979; Abbott, 1980; Grant and Ab­ G. conirostris bott, 1980; Ricklefs and Travis, 1980; Ro­ Geospiza conirostris occurs on Isla Es­ tenberry and Wiens, 1980; Wiens and Ro­ panola (Hood) and its satellite Gardner in tenberry, 1980). We can return to Darwin's the south of the archipelago, and on Isla Finches to perform such tests without cir­ Genovesa (Tower) in the northeast (Fig. cularity by obtaining and applying quan- 1). It has been recorded also on Islas Pinta 637 638 B. R. GRANT AND P. R. GRANT

-DARWIN

Kilometers , WOLF

1·

~ PINTA

o MARCHENA Q GENOVESA

O·

AN SALVADOR (jBortolom' , .Seymour RABIDA· DoAHNEBALTRA

PLAZA SUR

SANTA CRUZ

LOS G • HERMANOS SANTA FE

1· SAN CRIST6BAL ISABELl.

On,low Enderby • OChomPion SANTA MARIA . Coldwell GARDNER 'Gordner ESPANOLA~

92· 91· 90·

FIG. 1. The Galapagos Islands.

(Abingdon), Wolf (Wenman) and Darwin nirostris, G. fortis and G. scandens, and (Culpepper), but has not been seen on these that the niche on Genovesa, in the pres­ three islands recently (Harris, 1973; Bow­ ence of magnirostris, is a composite of the man, 1979, and pers. comm.; D. Schluter, niches of absentfortis and scandens (Fig. pers. comm.), so these latter three islands 2). According to this argument, the inter­ will be ignored. island difference in the niche of conirostris Lack (1945, 1947) argued from the beak is attributable to just one variable identi­ shapes of conirostris on Espanola and fied, namely the competitive influence of Genovesa that the niche on Espanola is a magnirostris. A competitive displacement composite of the niches of absent G. mag- between conirostris and magnirostris is DARWIN'S FINCHES 639

8 presented in the form in which it was con­ o structed. In his book Lack (1947) first pre­

J: sented his facts, then followed them with I- Q" an interpretation. But Richards (1948) W pointed out that there was almost as much 0 6 ..... lID interpretation in the presentation of facts ..... as in the section devoted to interpretation: finch species: Geospizajortis (+) from I. Daphne, G. tions to be explained were the distribution scandens (.) from I. Daphne, G. magnirostris (0) from I. Genovesa and G. conirostris (0) from I. of species and inter-island differences in Genovesa (G) and I. Espanola (E). beak shape; his hypothesis was that dis­ tribution and morphology were causally influenced by interspecific competition for food; and his main assumption was that also implied. Lack (1969) also proposed the feeding niche of a population was re­ that conirostris has competitively exclud­ flected in the average beak characteristics. ed magnirostris, fortis and possibly scan­ The above assumption and the corollary dens from Espanola, and that fortis and that diet differences parallel beak differ­ scandens have been competitively exclud­ ences have been supported by the results ed from Genovesa by conirostris and/or of recent work with several species of Dar­ magnirostris. The three species which ap­ win's Finches (Abbott et al., 1977; Grant parently have been excluded by coniros­ et al., 1976; Smith et al., 1978; Grant and tris are widely distributed throughout the Grant, 1979; Grant and Grant, 1980a; archipelago on almost all the major is­ Grant, 1981). Therefore we might expect lands, including those nearest to Espanola that conirostris on Espanola, with mean and Genovesa. However magnirostris has beak characteristics intermediate between become extinct on Sta. Marla and S. Cris­ those of the absent magnirostris, fortis and tobal (see Fig. 1) since Darwin's visit to scandens (Fig. 2), has an intermediate the archipelago in 1835 (Sulloway, 1982). feeding niche position too. However niche breadth and overlap are functions of niche Tests of the Competition Hypothesis size and shape as well as position (Pianka, Testing the hypothesis is difficult for two 1969; Colwell and Futuyma, 1971; Col­ reasons. The first is that it deals with well, 1979). Therefore an intermediate events in the past. Historical processes feeding niche position does not necessarily cannot be tested directly in the way that mean that the feeding niche of conirostris contemporary processes can, except in the combines the niches of the three missing unlikely circumstances that the reactants species or that the conirostris niche is par­ are brought together under the same con­ ticularly broad. For this reason we regard ditions as prevailed in the past. Instead, Lack's statements about niche similarities, the hypothesis can be tested through its differences and breadths to be testable, i.e., consequences (predictions). The assump­ falsifiable predictions of the competition tions upon which it rests may also be hypothesis. For example, if we found that tested, and if they are shown to be wrong conirostris on Espanola did not feed in a they must be replaced or the hypothesis manner exhibited by magnirostris, fortis must be discarded in favor of another. and scandens elsewhere, we would con­ The second difficulty is that Lack's ar­ sider the competition hypothesis, as it is gument was neither entirely explicit nor applied to the absence of these three 640 B. R. GRANT AND P. R. GRANT species, to be falsified by the failure of the nectar (Grant and Grant, 1980a, 198Ob; prediction. Grant and Grant, 1981). Our random Itis also predicted from the competition sampling of seeds and fruits on the ground hypothesis that missing species would be and on bushes, method of measuring them, less efficient than the species present in and our systematic recording of feeding locating and dealing with the foods on a behavior and diets have been described in particular island if these missing species Abbott et al. (1977)and Smith et al. (1978). were to arrive. Lack (1969) used this ar­ The only difference is that the study grid gument when invoking the competitive on Espanola was changed from the area exclusion hypothesis to argue for the com­ in 1973 to one 500 m south that contained petitive superiority of conirostris on Gen­ Opuntia megasperma trees. We did this in ovesa and Espanola, but inferiority else­ order to have Opuntia in both study areas where. We consider this to be another on Espanola and Genovesa where feeding prediction of the hypothesis that can be observations were made. tested by comparing the feeding efficiency of different species on the same species of RESULTS seeds. There are alternative explanations to the .Prediction of Niche Expansion competition hypothesis for the absence of on I. Espanola fortis and scandens from both islands and To test the prediction of the competition magnirostris from Espanola. One is that hypothesis that the feeding niche of con­ their food niches are not present on these irostris on Espanola combines the feeding islands (Bowman, 1961). We will examine niche of magnirostris,jortis and scandens, this possibility when testing the niche ex­ we first ask if conirostris feeds on the same pansion predictions. Other explanations food items taken by magnirostris, fortis are that they have been unable to reach and scandens on other islands. the islands, and that they may have ar­ Table 2 lists the plant species present rived in small numbers but, not finding on Espanola that are used for food by G. mates of their own species, hybridized with conirostris there, and by finches elsewhere the resident birds. We will consider these in the archipelago. This table shows that possibilities in the Discussion. many of the seeds eaten by G. conirostris are eaten by magnirostris, fortis and scan­ METHODS dens on other islands. With the exception Food supply and feeding habits of the of Sida spinose and Alternanthera echino­ finches vary seasonally (Smith et al., 1978; cephala (Table 2), the seeds on Espanola Grant and Grant, 1980a). To make mean­ which are not taken by conirostris but ingful inter-island comparisons we restrict which are taken by other finch species on our attention to data collected in the other islands are rare, either for natural months June to August, i.e., the first half reasons or because the plants have been of the dry (non-breeding) season: I. Es­ largely removed by introduced goats as in panola, 8-25 August 1979, supplemented the case of Croton scouleri. Seeds of the with data collected 26 May-5 June 1973 two exceptional species are very small and (Abbott et al., 1977); I. Genovesa, 23 soft, they do not form an important part June-4 August 1979; I. Daphne (no con­ of the diets of fortis and scandens else­ irostris) 21 June-7 July 1975. We chose where, and they are not eaten by magni­ the early and middle dry season because rostris elsewhere. Therefore conirostris on beak adaptations are clearly seen then Espanola do feed on the same food items (Grant and Grant 1980a, Smith et al., taken by magnirostris,jortis and scandens 1978). Later in the dry season, i.e., from elsewhere. October and November onwards, Opun­ A more difficult question to answer is tia cactus comes into flower and all four whether a sufficient supply of seeds is species feed intensively on the pollen and available on Espanola for the subsistence DARWIN'S FINCHES 641

TABLE 1. Ground finches present on Islas Espanola, Daphne and Genovesa; the mean weights in grams and bill dimensions in millimeters (± one standard error) of adult males, with sample sizes in parentheses. * = measurements of live birds; all other measurements were taken from museum specimens by I. Abbott.

Upper mandible Upper mandible Lower mandible Island Finch species Weight length depth width Espanola G. conirostris 35.8 ± 0.8* (26) 15.0±0.1 (119) 6.2 ± 0.1 (120) 11.7 ± 0.1 (119) G. fuliginosa 13.8 ± 0.3* (7) 8.4 ± 0.1 (37) 3.6 ± 0.1 (37) 6.6 ± 0.1 (36) Daphne G. scandens 22.0 ± 0.4* (23) 14.9 ± 0.1* (22) 4.7 ± 0.1* (10) 8.9 ± 0.1* (23) Gcfortis 17.5 ± 0.4* (27) 10.3 ± 0.1 (31) 4.4 ± 0.1 (31) 8.2 ± 0.1 (31)

Genovesa G. magnirostris 36.0 ± 0.8* (26) 16.1±0.2 (38) 7.7 ± 0.2 (27) 15.9 ± 0.1 (38) G. conirostris 25.3 ± 0.4* (24) 14.1 ± 0.1 (64) 5.7 ± 0.1 (65) 9.8 ± 0.1 (65) G. difficilis 11.5 ± 0.1* (50) 9.1 ± 0.1 (83) 3.3 ± 0.0 (83) 6.2 ± 0.0 (81)

TABLE 2. Plant species present on Espanola that have seeds eaten by finches on Espanola and other islands in the archipelago. Percent coverage is given on the basis of 50 random quadrats at each site. (Depth X Hardness)! valuesfor seeds and percentage coverage of plants are takenfrom Abbott et al., 1977 and supplemented with our own data. Symbols: X = present on island; seeds eaten by M = G. magnirostris, C = G. conirostris on Espanola, Cg = G. conirostris on Genouesa, F = G. fortis, S = G. scandens, Ful = G. fuliginosa. Double underlining = ""'5%offeeding time spent on this food item at anyone time ofthe year.

Seeds eaten

Percent By finches By finches Plant species on Espanola (DH)'12of seeds occurrence on Espanola on other islands

Cordia lutea 13-14 34 C M Opuntia megasperma 10-11 X C Prosopis juliflora 10-11 50 C M Tribulus cistoides 9-10 X C M,F Bursera graveolens 4-6 X C M,F,S,~ -- - Ipomoea triloba 3-4 X C M, F, Cg Neptuna plena 3-4 X C M Cenchrus platyacantus 3-4 2 C M,E,S Rhynchosia minima 3-4 4 F Vallesia glabra 2-3 X C F Grabowskia boerhaavifolia 2-3 X F,S Lantana peduncularis 2-3 14 ~,Ful M,K,Cg Desmanthus virgatus 2-3 8 C Tribulus cistoides (unripe fruit on plant) 2-3 X C [,S Trianthema portulacastrum 1-2 16 C, Ful Chamaesyce recurva 1-2 2 F,Cg Croton scouleri 1-2 X M,F, Cg Abutilon depauperatum. 1-2 4 M Boerhaavia erecta 0-1 2 C, Ful F Cryptocarpus pyrformis 0-1 X Ful Aristida subpicata 0-1 X C, Ful F,~ Panicum fasiculatum 0-1 22 Q, Ful F,~ Sida spinose 0-1 18 F Altemanthra echinocephala 0-1 20 F,S 642 B. R. GRANT AND P. R. GRANT

TABLE 3. Plant species present on Genovesa that have seeds eaten by finches either on Genovesa or other islands in the archipelago. Symbols as in Table 2: D = G. difficilis.

Seeds eaten by

Percent Finches on Finches on Plant species on Genovesa (DR)'" occurrence Genovese other islands Cordia lutea 13-14 36 M C Tribulus cistoides 9-10 X C Opuntia helleri 7-8 8 M Bursera graveolens 4-6 58 M, Cg K,S,C Ipomea linearifolia 3-4 M, Cg F Cenchrus platyacanthus 3-4 X M EtS, C Lantana peduncularis 2-3 14 M,Cg,D EtS,~ H eliotropium angiospermum. 1-2 X D ~ Chamaesyce recurva 1-2 X Cg S Croton scouleri 1-2 44 M, Cg, D ~ Chamesyce amplexicaulis 0-1 4 Cg, D ~ S ida salviifolia 0-1 X D F Portulaca howelli 0-1 X K, S - Eragrostis cilianensis 0-1 16 Cg, .Q F

of populations of magnirostris, fortis and below). Therefore the abundant supply of scandens. The answer is clearest with Cordia lutea seeds and the sparse supply magnirostris. In the dry season on I. Gen­ of O. megasperma seeds on Espanola ovesa magnirostris feed almost entirely on should be sufficient to sustain a population the seeds of two species, Cordia lutea and of magnirostris in dry seasons. In addition Opunii« helleri (Smith et al., 1978; Grant there are other large hard seeds such as and Grant, 1980a). Cordia lutea occurs Prosopis juliflora which occur abundantly about as frequently on Espanola as on throughout Espanola (Table 2), and al­ Genovesa (Table 3). Opuntia helleri is though they are absent from Genovesa, more common on Genovesa than O. they constitute a major proportion of mag­ megasperma is on Espanola, and the seeds nirostris diets in the coastal areas of Isla of O. helleri are smaller and softer than Pinta (D. Schluter, pers. comm.). the seeds of O. megasperma (Tables 2 and Geospiza fortis subsists on a variety of 3). Opuntia helleri is distributed through­ small seeds in the dry season on I. Daphne out Genovesa at approximately the same prior to Opuntia flowering (Grant and frequency as in the study grid (Table 3), Grant, 198Ob). Those seeds, or close rel­ whereas on Espanola O. megasperma is atives, are present on Espanola (Table 2; patchily distributed. However it is doubt­ see also Abbott et al., 1977). In particu­ ful if the relative scarcity of Opuntia and larly dry years on Daphne some fortis add hardness of the seeds are factors respon­ the moderately hard Tribulus cistoides to sible for the absence of magnirostris on their diet (Grant and Grant, 1980b). Trib­ Espanola. First, in our five visits to Gen­ ulus cistoides is also present on Espanola. ovesa during the dry season Cordia seeds There is inevitable uncertainty about were more important in the diet of mag­ whether the Tribulus seeds are sufficiently nirostris than Opuntia seeds (Grant and common to sustain a population of fortis Grant, 1980a). Second, the large O. mega­ in the middle-late dry season in a dry year , sperma seeds are well within the cracking and whether Opuntia flowers are suffi­ range of magnirostris (Table 2, also see ciently common at the end of the dry sea- DARWIN'S FINCHES 643

TABLE 4. Proportion offoraging time spent on food items and feeding activities in the early dry season. Opuntia was not flowering in July and August on Espanola and Daphne. *Two individuals seen to crack O. helleri seeds.

August 1979 July 1975 July 1975 July 1979 July 1979 Espanola Daphne Daphne Genovesa Genovesa G. conirostris G. fortis G. scandens G. magnirostris G. conirostris Small seeds on ground .291 .464 .490 .087 .197 Seeds (DH)! 0-1 .272 .039 .012 1-2 .002 .004 .0lD .029 2-3 .064 .034 .191 4--6 .023 .001 .006 7-14 .295 .861 .011* Tribulus mericarp fragments + .142 .057 Opuntia flower .025 Opuntia aril .040 .034 .191 .045 .495 Opuntia pad ripping and probing for arthropods .280 .020 .005 .036 Opuntia spine base .001 .001 .009 Bark ripping .027 .157 Gleaning .005 .039 Sulidae egg .006 .004 .001 Total seconds 26,135 20,266 5,285 12,255 19,416

son. But given the large range of alter­ tatively and flower production and seed native foods for fortis on Espanola, we set are low, G. scandens feeds mainly on conclude that a population of fortis could diptera larvae extracted from rotting be sustained there. Opuntia pads in the dry season (D. Schlu­ Geospiza scandens feeds on small seeds ter, pers. comm.). On Isla Champion off and cactus products. In years such as 1975, Floreana, Opuntia megasperma var. when there is abundant rainfall during the megasperma is abundant, and like O. wet season, small seeds are plentiful in the megasperma var. orientalis on Espanola, dry season and these are taken from plants the aril is fibrous, but the seeds are even and from the ground (Table 4). Geospiza harder. In August 1980 we recorded G. scandens then turns to Opuntia pollen and scandens feeding for proportionately 56.8% nectar when flowers start blooming in Oc­ of its foraging time on beetle larvae and tober (Smith et al., 1978; Grant and Grant, pupae from rotting Opuntia pads (see also 1980b). In the dry season in dry years on Table 4 for data from Daphne). We found Daphne it feeds largely on Opuntia echios many beetle larvae and pupae in Opuntia seeds and arils; for example, the propor­ pads on Espanola. tion of time spent foraging on these items Not one of the three species, magniros­ was .900 in August of 1980, a relatively tris, fortis or scandens, is dependent upon dry year. On Espanola the large and hard one or more food types that are lacking on seeds of Opuntia megasperma are certainly Espanola, and several of their diet items beyond the cracking power of scandens, are present. The evidence contradicts the and the aril is fibrous and less rewarding hypothesis that the three species are ab­ than the succulent arils of Opuntia seeds sent because suitable foods are lacking. elsewhere. Despite these observations, a scandens population could be self-sustain­ The Prediction of Niche Expansion ing on Espanola through a diet of small on I. Genovesa seeds and insect larvae taken from Opun­ To test the prediction that the feeding tia pads, because on Isla Marchena, where niche of conirostris on Genovesa com­ Opuntia helleri reproduces mainly vege- bines the feeding niche oi fortis and scan- 644 B. R. GRANT AND P. R. GRANT

dens we follow the same argument con­ We have found that the range and cerning missing species on Espanola. That abundance of food plants present on Es­ is, we first ask if Genovesa conirostris feed panola and Genovesa should be sufficient on the same food items as missing fortis to support populations of fortis and scan­ and scandens do on other islands. Table dens on both islands and magnirostris on 3 lists the plant species on Genovesa and Espanola. We now ask if conirostris feed­ the diets of present and absent finch ing niches combine the feeding niches of species. It shows that many food items or the missing species and answer this by else close relatives present on Genovesa comparing the feeding niches of coniros­ which are consumed by conirostris are tris with the feeding niches of the missing eaten by fortis and scandens on other is­ species on other islands in a similarity lands. analysis. The question of whether there is suffi­ cient food on Genovesa to support a pop­ Comparison ofAllopatric ulation of fortis and scandens can be an­ Feeding Niches swered with a similar argument to the one Table 4 shows the proportion of feeding presented above concerning the absence of time spent on various food items in July fortis and scandens on Espanola. Plants and August. Several parts of Opuntia as are widely distributed throughout Gen­ well as seeds are a source of food for the ovesa (Grant and Grant, 1980a) at ap­ finches, particularly scandens and coni­ proximately the same frequency as in the rostris. Therefore Opuntia feeding must study grid (Table 3). Many small seeds be considered, together with seed con­ present and common on Genovesa are sumption, in a comparison of feeding consumed by fortis and scandens else­ niches. Seeds have been grouped into where (Table 3). Although O. helleri seeds depth-hardness classes for comparative are hard and beyond the cracking powers purposes, except that the mericarps (fruit of fortis and scandens, the aril is fleshy parts) of Tribulus cistoides have been con­ and succulent and similar to the aril of O. sidered separately because of the unique echios which is eaten by fortis and scan­ problems they pose to birds (Grant, 1981a). dens on Daphne (Table 4). Opuntia helleri The result is a classification of proportion­ is abundant and flowers bloom profusely al feeding in 14 categories. Grouping seeds from November to May (Grant and Grant, into a few size categories could bias the 1980a) and would provide pollen and nec­ results in favor of greater similarity. We tar for both species. Rotting Opuntia pads believe that bias is minimal because many contain diptera larvae. Thus a diet of small of the seeds within one category are from seeds, Opuntia arils and pollen would be the same plant species on different islands. available for fortis on Genovesa. This is Where they are not, all the seed species similar to the diet of fortis on Daphne. If within the category are eaten by both scandens can subsist on Marchena and species being compared on at least one is­ Champion on dry season diets of small land in the archipelago. For example, seed seeds and arthropod larvae taken from size-hardness category 7-14 includes Opuntia pads, they could surely survive Opuntia helleri and Cordia lutea on Gen­ on Genovesa where in addition to a supply ovesa, and Opuntia megasperma, Prosopis of small seeds Opuntia is common and julifiora and Cordia lutea on Espanola. would provide arils, diptera larvae, pollen Geospiza magnirostris eats O. helleri and and nectar. Therefore many food items C. lutea seeds on Genovesa and Prosopis present on Genovesa are consumed by for­ julifiora on Pinta. Geospiza conirostris tis and scandens elsewhere, and there are takes all three species on Espanola. no known essential constituents missing. To compare feeding niches we use the So there is no reason to supposefortis and index of similarity 1-0.5 (!, IPH - P2i I) scandens are absent from Genovesa be­ where PH and P2i are the proportions of cause their foods are not present. foraging time spent by the two species on DARWIN'S FINCHES 645

TABLE 5. Similarities in the foraging of pairs of Geospizaspecies observed over8-day periods.

Espanola G. conirostris v Genovesa G. conirostris SIM = .31 v Daphne G. fortis SIM = .38 v Daphne G. scandens SIM = .40 v Genovesa G. magnirostris SIM = .42 v Espanola G. juliginosa SIM = .37 v G. magnirostris and conirostris SIM = .53 v G. scandens and fortis SIM = .41 v G. magnirostris, fortis and scandens SIM = .65 Genovesa G. conirostris v Genovesa G. magnirostris SIM=.14 v Daphne G. fortis SIM = .27 v Daphne G. scandens SIM = .42 v G.fortis + scandens SIM = .30

food items classified in Table 4. This is the niche of conspecific conirostris popu­ appropriate for discrete classes of food lation on Genovesa, or to any pairwise items not measured along a common scale combinations of niches of these species (Whittaker, 1960). We do not have com­ (Table 5). Furthermore Espanola coniros­ parable estimates of nectar, pollen, aril, tris has the greatest niche breadth (Table arthropod and seed density; therefore we 6). are unable to use Hurlbert's (1978) method The competition hypothesis also pre­ of comparing diet similarity in relation to dicts that the feeding niche of Genovesa food availability. Table 5 gives the simi­ conirostris is more similar to the compos­ larities based on the combined observa­ ite niche offortis and scandens than to the tions of eight successive days. There are niches of these two considered separately no confidence limit estimates on the sim­ or to the niche of the conirostris popula­ ilarity values, but the values are not close tion on Espanola. This prediction is not to 0 and 1 where sampling distortions are fully realized (Table 5). The niche of Gen­ likely to be greatest (Ricklefs and Lau, ovesa conirostris is most similar to the 1980). Figure 3 shows cumulative similar­ niche of scandens considered alone. The ities over a period of eight successive days. composite niche of scandens and fortis, The stability of the similarities after about the niche of fortis alone and the niche of five days suggests that our results would Espanola conirostris are all about equally not have been very different if we had ex­ similar to the niche of Genovesa coniros­ tended the period of observation. iris. Reference to Table 4 will show the Espanola conirostris is closest in feed­ reasons for these results. Genovesa coni­ ing habits to magnirostris (similari­ rostris is similar to scandens because both ty = .42). This results from both species exploit Opuntia, but conirostris spends taking large and hard seeds. The differ­ more time ripping the bark off Bursera ence is due to Espanola conirostris also and Croton branches to get at arthropods, taking small seeds and ripping open mainly termites, and less time feeding on Opuntia pads to feed on dipteran larvae small seeds on the ground than scandens and pupae. In this respect Espanola con­ does. The similarity between conirostris irostris resemblefortis and scandens. And and fortis is lower because fortis exploits in fact, in agreement with the prediction Opuntia parts less and small seeds on the of the competition hypothesis, the feeding ground more than both scandens and con­ niche of Espanola conirostris is more sim­ irostris. ilar to an unweighted composite of mag­ Thus the evidence shows that the con­ nirostris, fortis and scandens niches on irostris niche on Genovesa is similar to the other islands than it is to the niches of scandens niche but does not support the these three species treated separately, to prediction that it is similar to a composite 646 B. R. GRANT AND P. R. GRANT

TABLE 6. Measurements offeeding diversity lf!, Pl2 ESPANOLA and niche breadth HlHmar of Geospizaspecies. These are alternative measures of the variation in feeding .8 characteristics of each species. Pi is the proportion of feeding time spent on the ithfood item. H = - Ipl lOg10Pi and Hmar = lOglOS where S = the number of .6 food items taken. Data are from Table 4.

II) w Feeding Niche diversity hreadth I- ·4 ex Espanola conirostris 3.96 0.90 < Daphnefortis 3.20 0.63 ... Genovesa conirostris 3.19 0.68 ·2 Daphne scandens 3.14 0.61 ~ Genovesa magnirostris 1.33 0.32 II)

C> on I. Genovesa predicts a low similarity Z in feeding niches between these two 0 GENOVESA w species. It also predicts that the two con­ w .8 irostris populations differ in their feeding u, niches in such a way that magnirostris feeds on items taken by conirostris on Es­ w > ·6 panola but not by conirostris on Geno­ I- ...... vesa. < -. Our results (Table 5) support these pre­ ... ·4 ~ /.-.~·\ l------­. dictions. There is a particularly low sim­ ~ • /'" "'\ •.J.-/.'...... <,e __e_e ilarity between magnirostris and coniros­ ~ .. &,., ...- ..- .. _ii_ tris on Genovesa, lower than any other U ·2 .".",0 0 __ 0 pair of species compared. The similarity /0 -0_0 between the two conirostris populations is also low, but when Espafiola conirostris 2345678 is compared to a composite of Genovesa conirostris and magnirostris the similarity DAYS rises to .53. Reference to Table 4 shows FIG. 3. Similarities in feeding niches between the reason for these results. The low sim­ species, calculated from observations made on day 1, day 1 + 2, day 1 + 2 + - + n over 8-day pe­ ilarity between magnirostris and Geno­ riods. Similarities are between G. conirostris on I. vesa conirostris is due to magnirostris Espanola (above) or I. Genovesa (below) and the fol­ feeding almost entirely on large-hard seeds, lowing: G. magnirostris (0), G. fortis (.A.), G. scan­ whereas conirostris exploits Opuntia parts dens (.), G. conirostris on 1. Genovesa (0), G.fortis and arthropods. The two populations of and G. scandens combined (e), G. magnirostris and G. conirostris combined (.6) and G. magnirostris, G. conirostris differ in four ways. The first is fortis and G. scandens combined (+). G. conirostris a discrete difference. Espanola conirostris on 1. Espanola is most similar to G. magnirostris, G. feeds on large hard seeds, Genovesa con­ fortis and G. scandens combined, but G. conirostris irostris does not. The three other differ­ on 1. Genovesa is most similar to G. scandens alone. ences are ones of degree. Opuntia helleri arils on Genovesa are succulent and abun­ fortis and scandens niche. Note too in Ta­ dant, and are taken more frequently than ble 6 that the niche of Genovesa coniros­ are the comparatively dry and fibrous arils tris is not particularly broad or diverse. of O. megasperma on Espanola. Also on Genovesa the tip-biting, twisting, tearing Comparison ofSympatric motion employed to strip bark off trees to Feeding Niches reach arthropods is more highly devel­ The hypothesis of a competitive dis­ oped, and occurs more frequently, than on placement of magnirostris and conirostris Espanola. This is a habit employed fre- DARWIN'S FINCHES 647

TABLE 7. Measurements in millimeters of random, cracked and rejected Cordia lutea seeds on I. Espanola and I. Genouesa.

Length Depth Length X depth

Island Sample N x ± SE N x ± SE N x ± SE

Espaiiola Random 106 10.3 ± 0.1 106 6.1 ± 0.1 106 63.2 ± 1.4 Cracked by G. conirostris 108 8.8 ± 0.1 108 5.3 ± 0.1 108 46.5 ± 1.0 Genovesa Random 59 10.4 ± 0.2 59 6.8 ± 0.1 59 71.5 ± 1.9 Cracked by G. magnirostris 54 9.6 ± 0.2 54 6.3 ± 0.1 54 59.9 ± 1.9 Rejected by G. magnirostris 22 9.7 ± 0.3 22 6.3 ± 0.2 22 61.6 ± 2.8 quently by a tree finch with a similar beak all stones seen to be picked up and reject­ size and shape, Camarhynchus psittacula, ed without cracking by magnirostris. Ta­ on other islands such as Pinta, Marchena ble 7 gives the measurements, and Table and Santa Cruz (unpubl. observ.). Finally 8 gives the results of statistical compari­ Opuntia pad ripping was rare on Geno­ sons. vesa in July yet quite common on Espa­ Cordia stones are larger on Genovesa nola in August. However it was common than on Espanola, but on both islands on Genovesa in November, (Grant and birds tend to select the small stones. Ge­ Grant, 1979; Grant and Grant, 1980a). In ospiza magnirostris rejected stones of the this instance, the difference between the same size as the ones they cracked, there­ niches of the two populations may be a fore the basis of rejection in this small little exaggerated because niches change sample must have been something other seasonally and our observations on the two than size. Many of the stones seen to be islands were not made at exactly the same rejected had emergence holes of a bruchid time of the year. beetle, Amblycerus galapagoensis so this was possibly the reason for rejection. Food Selection and Feeding Efficiency Eighty-five percent of the random sample To test the prediction that missing of Cordia stones on Genovesa had larvae species would be less efficient than species of this beetle; the larvae are eaten by mag­ present in dealing with foods, we concen­ nirostris (and also by conirostris on Es­ trated on the feeding of adults of Espanola panola). The rejected stones without conirostris and Genovesa magnirostris emergence holes were not particularly upon a common food type, the seeds of large. Therefore the difference in size be­ Cordia lutea. Cordia lutea seeds were tween stones cracked and stones at large chosen because the stones or pits that con­ reflects a selection by the finches that does tain them (Wiggins and Porter, 1971) are not involve handling. In this case size is the largest and hardest of the fruits and probably the cue used. seeds taken both by conirostris on Espa­ How would magnirostris perform on nola and magnirostris on Genovesa. They Espanola? The average size (length X are a major dry season food item in the depth) of Cordia stones cracked by mag­ diets of both species and for this reason, nirostris on Genovesa is not significantly and because they are large and hard, they different from the average size of stones are most likely to reveal any difference in on Espanola (t I 58 = 1.10, P > .1). There­ feeding efficiency between the two species fore magnirostris would probably not be (Grant et al., 1976). We collected and size-selective on Espanola. In contrast to measured random samples of stones from this result, if Espanola conirostris were the ground, the remains of all cracked present on Genovesa it would be able to stones found which were intact enough to crack only the smallest Cordia stones, as permit measurement and, on Genovesa, indicated by a comparison of a random 648 B. R. GRANT AND P. R. GRANT

TABLE 8. Differences between samples of Cordia lutea tested by t tests. One to three asterisks indicate statistical significance «.05, <'01 and <'001, respectively). Log-transformed data were used to test for a difference between random and cracked seeds on Espanola (length x depth) because variances were unequal.

Length Depth Length x depth

Island Sample df. df· df· df· Espanola Random vs. cracked 188 7.00*** 141 3.21** 21.2 9.99*** by G. conirostris Genovesa Random vs. cracked 111 3.81 *** 111 3.43*** 81 5.79*** by G. magnirostris Genovesa Random vs. rejected 79 2.41* 79 2.51* 79 2.84** by G. magnirostris Genovesa Cracked by G. magni- 74 0.35 74 0.08 74 0.30 rostris vs. rejected by G. magnirostris Espanola vs. Genovesa Random 163 0.48 163 5.67*** 163 3.44*** Espanola vs. Genovesa Cracked 160 4.31 *** 89 4.57*** 160 7.97*** sample of stones from Genovesa and the it would have slightly higher energetic costs sample cracked by conirostris on Espa­ through being slightly larger. But this dis­ nola (t160 = 7.97, P < .001). This is further advantage would probably be more than illustrated in Figure 4. Itis not surprising, offset by two feeding advantages; it would in view of this result, that the small-billed presumably handle the small Cordia stones form of conirostris which does occur on slightly faster than conirostris, and it Genovesa (Fig. 2) does not crack stones of would be able to deal quickly with the Cordia lutea. large Cordia stones which conirostris has Not only do magnirostris on Genovesa difficulty in cracking or else avoids. Small take larger Cordia stones than conirostris differences in metabolism would only be on Espanola, they take much less time on important if, in addition, magnirostris average to crack them (13.6 ± 2.6 s, were much poorer at finding Cordia stones N = 36)than conirostris on Espanola (24.2 than conirostris, in which case energetic ± 3.3 s, N = 45; t79 = 3.25, P < .001). costs of searching would be much higher However the total handling time on Cor­ in magnirostris than conirostris. dia stones is not significantly different Figure 5 shows that the advantage pos­ (P > .1) between conirostris (89.7 ± 10.3 sessed by magnirostris over conirostris in s, N = 45) and magnirostris (71.6 ± 6.5 cracking Cordia stones is held, although s, N = 44). Total handling time refers to to a lesser extent, with smaller and softer the period of time between picking up a seeds. In this figure, the average time each stone and finishing feeding on the seed or species takes to crack a seed is seen to rise beetle larva. We did not include searching with an increase in seed size and hardness. time and we did not detect any difference Unfortunately the number of points for in the time taken to extract seeds and lar­ each finch species is too small for analysis vae. Overall the total efficiency of Cordia by regression, and pairwise comparisons seed exploitation is approximately the same of finch species are limited by the small in the two species, in terms of energy in­ number of seeds of a given type eaten by take per unit of energy expended. two species. The main point we wish to A very small difference in metabolism make is that there is no detectable reversal between the species is suggested by the of the relative efficiencies of magnirostris small difference in body weights in Table and conirostris towards the lower end of 1. Thus if magnirostris was on Espanola the seed size-hardness scale. A graph of DARWIN'S FINCHES 649

20 GENOVESA. "

10

V) a:: w co ~ :::> ES PANOl A. e /\ GENOVESA. B Z 20 ~ 10 I• \.,~o \ " 0 0 0/0 0_0 '0.../\ / / \ 0 a-a ...... 0-0" '0 25 50 75 100 125 25 50 75 100 125 CORDIA STONES lENGTH x DEPTH FIG. 4. Frequency distributions of the sizes of Cordia lutea stones measured in two dimensions in millimeters. On I. Espanola the stones sampled at random (0) are compared with those cracked by G. conirostris (...) in A, and with the superimposed distribution of seeds cracked by G. magnirostris (.) on I. Genovesa in B. On I. Genovesa the stones sampled at random (0) are compared with those cracked by G. magniros,,"s (.) in A and with those cracked by G. conirostris (...) on I. Espanola in B.

seed handling times plotted against seed 60 size-hardness gives the same results. Figure 5 also shows that conirostris has w :l': 45 a feeding efficiency advantage over fortis ;: (without taking metabolic requirements o Z into account). Thus the prediction of the ;;: 30 v competition hypothesis that conirostris has -c a feeding efficiency advantage over fortis v'" Z is found to be correct, but the prediction « 15 that conirostris has a feeding efficiency w :l': II.. -~, ~, advantage over magnirostris was not sup­ ported by our results; rather magnirostris 3 6 9 12 15 appears to have an advantage over Es­ panola conirostris. A necessary qualifica­ SEED SIZE- HARDNESS tion is that if immigrant magnirostris were FIG. S. The average time in seconds to crack seeds and fruits of several plant species as a function immatures, they might be at a feeding ef­ of the size and hardness of those seeds as indexed by ficiency disadvantage in relation to adult the square root of the depth in millimeters x the conirostris, because on Genovesa imma­ hardness in kgf. The finch species are G. magniros­ ture magnirostris take significantly longer iris from I. Genovesa (0), G. scandens from I. Daphne (.), G. fortis from I. Daphne (e) and from to crack Cordia stones (36.9 ± 8.15, I. Santa Cruz (@) where it is larger (Grantet al., 1975), N = 4) than do adult magnirostris and G. conirostris from I. Espanola (0) and from I. (t38 = 3.83, P < .001). Genovesa (@). Curves have been fitted by eye. 650 B. R. GRANT AND P. R. GRANT

DISCUSSION seed availability can produce annual shifts in feeding niches (Grant and Grant, 198Ob), The interspecific competition hypothe­ our results may be partly dependent on the sis has been supported by several but not relatively plentiful' food supply in the years all of our observations. Our ecological data of our study. We have tried to eliminate support Lack's hypotheses that fortis and this bias as much as possible by restricting scandens have been competitively exclud­ our comparisons to years when there was ed from Espanola by conirostris; that comparable rainfall in the preceding wet scandens has been competitively excluded season on all study islands. Rainfall in 1975 from Genovesa by conirostris; and that was above average on Daphne (Grant and competitive character displacement be­ Boag, 1980), and in 1979 about average tween conirostris and magnirostris could on Genovesa and Espanola. But even in have occurred on Genovesa. The evidence years of substantial rainfall, food supply does not support the view that fortis has limits population sizes during the dry sea­ been excluded by conirostris from Geno­ son (Smith et al., 1978). In dry years this vesa, or that magnirostris has been ex­ limiting influence is especially strong (Boag cluded by conirostris from Espanola. We and Grant, 1981). It would be preferable will first discuss the main assumptions to make inter-island comparisons in a dry upon which these conclusions rest, then year, but the only data we have from a consider alternative explanations for our dry year comes from Daphne (Grant and results and finish by pointing out two im­ Grant, 1980b; Boag, 1981; Boag and plications of our study. Grant, 1981). We have assumed that an equal weight­ A fourth assumption is that other po­ ing of niches is appropriate for the com­ tential biotic factors not considered, such parison of conirostris niches with the as disease and predation, can be ignored. combined niches of missing species. An Very little is known about disease, and unequal weighting may be more realistic, nothing is known to suggest that it would perhaps based upon degree of bill size and be an important factor on one island and shape similarities, but this would not alter not on another. The three main predators the results strongly. Second, we have as­ are the Galapagos hawk Buteo galapa­ sumed that the results are not substan­ goensis (deVries, 1976), and the short-eared tially affected by the particular allopatric owl Asio fiammeus (Grant et al., 1975; populations used in comparisons with Grant and Grant, 1980a) and mocking­ conirostris. This assumption is especially birds Nesomimus spp. (Downhower, 1978; important in our use of magnirostris data, Grant and Grant, 1980a). The owl is pres­ since those data come from an island where ent on all our study islands, mockingbirds one population of conirostris is present. are present on both Espanola and Gen­ Our magnirostris data from Genovesa in ovesa, but the hawk is present on only July (this study) and November (Grant and Espanola. From this it could be argued Grant, 1980a) are similar to magnirostris that magnirostris is predator-excluded data from 1. Pinta and I. Marchena (D. from Espanola, but this would be incon­ Schluter, pers. comm.) and I. Santa Cruz sistent with the coexistence of magniros­ (Smith et al., 1978) in showing that large tris and hawks on other islands (Lack, and hard seeds predominate in dry season 1947; Harris, 1973), and would presum­ diets. Beak sizes of magnirostris on these ably require a preferential predation upon islands are similar (see Lack, 1947). magnirostris for reasons of its slightly Therefore we believe this assumption is larger size or its rarity. approximately correct. A fifth assumption is that the inferred Third, we have assumed that the results competition is with the specified species are not dependent upon which year we and not with others. This may be incor­ studied the finches. Since year-to-year rect. Since Lack's (1969) last treatment of variation in rainfall, plant production and the subject, MacArthur (1972) and DARWIN'S FINCHES 651

MacArthur et al. (1972) have drawn at­ modern post-disturbance conditions. tention to the competitive influence of However, the argument that bills of con­ many species upon one, and introduced temporary conirostris are adapted to past the term diffuse competition to describe conditions is difficult to sustain in the face it. In the present context magnirostris , of evidence of rapid morphological change fortis and scandens may be absent from in a population of fortis under selection Espanola for reasons of competition with for large bill size (Boag and Grant, 1981). conirostris, fuliginosa, Nesomimus mac­ If the altered vegetation resulted in a donaldi (mockingbird) and Zenaida gal­ change in conirostris phenotype it prob­ apagoensis (dove). Likewise the absence of ably occurred in the last century: we have scandens and fortis from Genovesa may been unable to detect a change this cen­ be due to the combined influence of mag­ tury in comparisons between specimens nirostris, conirostris, difficilis, N eso­ collected in 1905-06 and modern ones. mimus parvulus and Zenaida galapago­ Even though the competition hypothe­ ensis or some combination of these species sis has been largely supported by the re­ (e.g., conirostris and magnirostris: Lack, sults of the tests and its assumptions are 1969). All eat seeds and Opuntia parts, reasonable, other hypotheses may be just and all show a divergence in diets from as good or better. We now consider three wet season to dry season which suggests other hypotheses. an avoidance of competition (Grant and The first alternative is that species are Grant, 1980a). The remaining land bird absent from an island because their food species on these islands, Certhidea oliva­ niches are absent. For example, Bowman cea, Dendroica petechia and Myiarchus (1961 p. 286) wrote "the nature of the magnirostris (Espanola only), can be ig­ vegetation on Hood Island [Espanola] is nored because they have diets of arthro­ such that possibly there are too few hard­ pods and are therefore not likely to com­ coated seeds to support a population of G. pete with the others. magnirostris, and too few Opuntia cactus The data in Tables 2 and 3, combined plants to support G. scandens." Presum­ with data in Abbott et al. (1977), Smith et ably this explanation can be extended to al. (1978) and Grant and Grant (1980a), account for the absence of fortis as well. make the possibility of diffuse competition The hypothesis was tested in our study plausible, because they show that some and found to be incorrect. These species foods taken frequently by fortis elsewhere are not absent from Espanola because of are taken frequently by fuliginosa on Es­ the absence of an appropriate food supply, panola and by difficilis on Genovesa; Lack nor are fortis and scandens absent from (1969) ignored these two species in his Genovesa for that reason. analyses. The hypothesis of diffuse com­ The food supply hypothesis has been in­ petition is extremely difficult to test (Ab­ voked by Bowman (1961) to explain dif­ bott and Grant, 1976), nevertheless we ferences in feeding niches a) between co­ must acknowledge that it could succeed in existing species, such as conirostris and explaining the absence of fortis from Gen­ magnirostris on Genovesa, and b) be­ ovesa where the hypothesis of competi­ tween populations of the same species on tion with conirostris fails. different islands, such as conirostris on Finally we have assumed that changes Genovesa and Espanola. It is succinctly in the vegetation of Espanola caused by stated as follows: "Anatomical differences goats introduced in the last century have between closely related species on anyone not created a bias in our comparisons with island are best thought of as biological ad­ the undisturbed islands of Daphne and justments which have been evolved when Genovesa. Such a bias could arise if cur­ the forms were in isolation. These adjust­ rent conirostris morphology and feeding ments could have prevented 'competition' were adapted more to past pre-distur­ from occurring between the forms when bance conditions on Espanola than to subsequently they came together on the 652 B. R. GRANT AND P. R. GRANT same island" (Bowman, 1961 p. 296). test generally supported the competition Therefore, conirostris populations differ hypothesis, and we have no reason to turn in beak morphology and diets because they to the operationally more complex food have become adapted to the different food hypothesis to explain morphology and dis­ supplies the islands possess, and coexist­ tribution of the finches. ing conirostris and magnirostris popula­ A second explanation for the absence of tions on Genovesa differ because their ad­ fortis and scandens from both islands and aptations occurred originally on different magnirostris from Espanola is that they islands. have not been able to reach these islands. This hypothesis, like the competition These two islands may be relatively in­ hypothesis, rests on the assumption that accessible to birds by virtue of their po­ diet and beak shape are correlated, hence sitions in relation to neighboring islands feeding niches are correctly inferred from (see Fig. 1) and prevailing winds (Power, beak morphology. Bowman (1961) pre­ 1975). This explanation for species ab­ sented empirical evidence to support this sences has been repeatedly rejected on the assumption as it applies to differences be­ grounds that inter-island wanderings are tween coexisting species, but not as it ap­ quite frequent (Lack, 1947, 1969; Bow­ plies to inter-island differences within man, 1961; Harris, 1973). For example, species. The food supply hypothesis has 15fortis individuals were collected on Es­ the additional merit of constituting a null panola in 1906, and 1. Abbott (pers. alternative (no interaction) to the compe­ comm.) observed one in June 1973. One tition hypothesis (Abbott et al., 1977; individual has been collected on Geno­ Strong et al., 1979; Grant and Abbott, vesa, and we observed two there in July 1980). It has two serious problems, how­ 1979 (see also Yang and Patton, 1981). We ever, that make it useful only to the extent also netted an adult male scandens on that other hypotheses fail. Genovesa at the same time and one indi­ The first problem is that it is doubtful vidual has been collected 20 miles from if there are enough islands in the archi­ the nearest island (Gifford, 1919). Geospi­ pelago for all sympatric populations to za magnirostris also make inter-island have evolved their adaptations for food movements (Grant et al., 1975). gathering in isolation from each other, i.e., A third hypothesis for the absence of allopatrically. This makes the hypothesis breeding populations of species on Espa­ unreasonable in its more general, all-em­ nola and Genovesa is that the few indi­ bracing form, but it could still be correct viduals who reach the islands and survive in specific instances. The second problem to the breeding season fail to find conspe­ is that it explains the absence of a species cific mates and hybridize with members of from an island by the absence of the ap­ one of the resident species. We refer to this propriate food supply, yet it allows for ad­ as the immigrant-hybridization hypothe­ aptation of new colonists to a local food sis. For example, it is plausible that small supply which differs from the supply on numbers of magnirostris hybridizing with the island of origin. With such compre­ conirostris have been partially responsible hensiveness it is close to being unfalsifia­ for the more magnirostris-like features in ble. The dividing line between conditions the Espanola conirostris bill. We do not that preclude establishment of colonists know if hybrid offspring would be viable, and conditions that allow establishment but viability is likely because two magni­ and persistence long enough for adapta­ rostris x conirostris pairs produced via­ tion to take place needs to be specified, ble offspring on Genovesa in 1980 and 1981 but it would be extremely difficult to do (unpubl. observ.). A parallel situation oc­ so operationally (see also Grant, 1975). curs on Daphne. Geospiza fuliginosa has Because of these difficulties we placed repeatedly immigrated since our study be­ greater emphasis in this paper on testing gan there in 1973. It has not established a the competition hypothesis. Results of the breeding population but a few individuals DARWIN'S FINCHES 653

have hybridized with the resident larger nola and more than 20 on Champion (un­ species G. fortis (Boag, 1981; Grant and publ. data) in contrast to a range of 2 to Price, 1981). Geospizafortis are unusually 7 for the other species (Grant and Grant, small on this island. 1981). The species is present on only the Hybridization might also occur between three southern islands and their satellites. two species if one becomes rare: the few Populations of Opuntia were decimated in individuals of the rare species then court this century and the last one by human heterospecifics in the virtual absence of destruction of natural habitat on Santa conspecifics. An unbalanced sex ratio in Marfa and S. Cristobal, and by introduced the common species (d. Boag and Grant, goats on all three islands. Nowadays, O. 1981) could facilitate the acceptance of megasperma only grows in large stands on heterospecific mates. We refer to this al­ the goat-free islets of Champion and ternative as the resident-hybridization hy­ Gardner off Santa Maria. Thus direct and pothesis. It may be invoked to explain the indirect effects of human activity probably absence of magnirostris from Espanola by caused the demise of the magnirostris supposing that magnirostris was once populations, although we have no detailed present and common, became rare, hy­ knowledge of why they were not sustained bridized with conirostris and disappeared by alternative foods such as the seeds of as a species. Cordia lutea and Prosopis julifiora. Some support for the hypothesis is pro­ The changes occurring on the southern vided by the particularly large form of islands in both habitat and finch popula­ magnirostris on the neighboring islands of tions in the last 150 years have been so Santa Maria and S. Cristobal. It was once poorly documented that most of these ideas common but became extinct on both is­ about past hybridization will remain as lands sometime after Darwin's visit in 1835 speculation. The hypothesis will be sup­ (Steadman, 1981; Sulloway, 1982). A few ported, however, if fossils of magnirostris museum specimens of intermediate size and of intermediates with conirostris are from these two islands (e.g., see Sulloway, found on Espanola; fossils of magnirostris 1982) suggest that magnirostris and fortis have recently been found on Santa Marfa hybridized during the period of decline of (Steadman, 1981). The immigrant-hybrid­ the magnirostris populations. By the time ization hypothesis can be investigated di­ Espanola was first visited by collectors at rectly, at least in principle. However, the end of the nineteenth century, mag­ magnirostris was not recorded on Espa­ nirostris was already extinct (or exceed­ nola by collectors at the turn of the cen­ ingly rare) on the other two islands. If it tury (e.g., Gifford, 1919), orin recent field had been present earlier on Espanola as studies (Abbott et al., 1977; Downhower, well, it would have been larger than con­ 1978; this study). The role of hybridization irostris in bill dimensions by much more in the evolution and distribution of the than 15%, which is the minimum differ­ finches remains an open question. ence between currently coexisting species We conclude by making a general point in the archipelago (Grant, 1981b). Even that emerges from this discussion. Biogeo­ the smaller form of magnirostris on Santa graphic analyses have made an important Cruz is larger than conirostris in bill depth contribution already by showing how by more than 15%. widespread competitive effects might be Why did the two, possible three, pop­ (Lack, 1947, 1976; Diamond, 1972, 1975; ulations of magnirostris go extinct? The MacArthur, 1972; Brown, 1975; for criti­ large bill may have been adapted to crack­ cism see Connor and Simberloff, 1978, ing the particularly large and hard seeds 1979; Simberloff, 1978; Strong et al., 1979), of Opuntia megasperma. This species of but we believe that because interspecific Opuntia has the largest and hardest seeds competition is an ecological process, eco­ of any Opuntia species in the archipelago; logical studies are required to test the hy­ the size-hardness index is 10.9 on Espa- pothesis. The present study has provided 654 B. R. GRANT AND P. R. GRANT

some ecological tests but it has also iden­ seeds were rare and conirostris were able tified areas of ignorance that deserve at­ to subsist on arthropods found in Opuntia tention not only among Darwin's Finches pads and beneath bark. Extinctions are but in all similar studies of island com­ not likely to be observed, but they can be munities. inferred from extrapolations of observed The first is the colonization of islands fluctuations. To understand the absence of or, to be more general, the entry of new species on islands and the ecological and species into a community. In the case of morphological differences between those Darwin's Finches, most inter-island wan­ present, there is no real substitute for di­ derers are first-year birds. As many as 100 rect studies of food availability and use by or more may arrive on an island following coexisting species over a long period of time one breeding season, and invariably all (Wiens, 1977; Smith et al., 1978; Abbott, leave (or die) before or during the next 1980; Grant and Grant, 198Ob). breeding season (Grant et al., 1975; un­ publ. observ. on I. Daphne). They do not SUMMARY have sub-normal weights, and with few This study tested predictions of the exceptions they do not exhibit reproduc­ competition hypothesis which Lack (1945, tive behavior if they stay up to the begin­ 1969) applied to Darwin's Finches on the ning of the breeding season. Factors other Galapagos islands in order to explain their than interspecific competition for food are adaptive radiation. Lack suggested that likely to influence the decisions of individ­ Geospiza conirostris has competitively ex­ uals to stay or to leave. These factors pre­ cluded G. fortis and G. scandens from I. sumably include those which stimulate re­ Espanola and I. Genovesa and G. mag­ productive activity, and the possible nirostris from I. Espanola; and further that tendencies of birds to return to their natal G. conirostris populations exhibit com­ area or nearby to breed. Therefore ease of petitive displacement (Genovesa), com­ dispersal from one island to another, which petitive release (Espanola), or both. We seems to be indicated by the known im­ tested four predictions of this hypothesis migrations, should not be equated with using quantitative ecological data and ob­ ease of establishing a breeding population tained mixed results. (Grant, 1977). In particular, the absence (1) The prediction that Espanola coni­ of magnirostris from Espanola may have rostris has a broad feeding niche which is more to do with conditions suitable for more similar to a combined magnirostris­ starting a breeding population than with fortis-scandens niche than to the niches of dispersal to the island, food supply or these species considered separately was competition (see also Abbott, 1978, 1980). upheld. The niche of Espanola conirostris The second area of ignorance is the pro­ is unusually broad. cess of competition over long periods of (2) The prediction that Genovesa coni­ time. We have shown how annual varia­ rostris had a feeding niche similar to a tion in rainfall alters plant production and combined fortis-scandens niche was not produces substantial changes in feeding upheld. The niche of Genovesa conirostris niches between years (Grant and Grant, is more similar to the niche of scandens 1980b). We have also seen large changes alone. between years in the numerical propor­ (3) The prediction that conirostris and tions of conirostris and magnirostris on magnirostris have dissimilar niches on Genovesa and fortis and scandens on Genovesa, and that the difference lies in Daphne (Grant and Grant, 198Ob). There magnirostris taking food items exploited may have been a fluctuating competitive by conirostris on Espanola but not on balance between conirostris and magni­ Genovesa, was verified. rostris on Espanola, leading to the extinc­ (4) The prediction that conirostris on tion of the latter perhaps during a series Espanola was more efficient than magni­ of dry years when Cordia and Prosopis rostris at dealing with foods on that island DARWIN'S FINCHES 655 was not upheld, although conirostris on ship was determined by the toss of a sex­ both islands were more efficient than for­ biased coin. tis and scandens. LITERATURE CITED We conclude that interspecific compe­ ABBOTT, I. 1978. Factors determining the number tition has influenced the distribution and of land bird species on islands around South­ morphology of Darwin's Finches, but our Western Australia. Oecologia 33:221-233. results do not support the hypothesis of ---. 1980. Theories dealing with the ecology of competitive exclusion in two particular land birds on islands. Adv. Eco!. Res. 11:329-371. cases; the exclusion of fortis from Gen­ ABBOTT, I., L. K. ABBOTT, ANDP. R. GRANT. 1977. Comparative ecology of Galapagos Ground ovesa and magnirostris from Espanola Finches (Geospiza Gould): evaluation of the im­ by conirostris. We cannot explain the ab­ portance of floristic diversity and interspecific sence of fortis and magnirostris from the competition. Eco!. Monogr. 47:151-184. two islands by the lack of suitable foods ABBOTT, I., AND P. R. GRANT. 1976. Non-equi­ or by the failure of the species to reach librial bird faunas on islands. Amer. Natur. 100:507-528. these islands. Geospiza fortis may be ab­ ARMS, K., AND P. S. CAMP. 1979. Biology. Holt, sent because of a combined (diffuse) com­ Rinehart and Winston, N.Y. petitive effect of magnirostris, conirostris BOAG, P. T. 1981. Morphological variation in the and difficilis. A breeding population of Darwin Finches (Geospizinae) of Daphne Major Island, Galapagos. Ph.D. Thesis. McGill Univ. magnirostris may be lacking on Espanola BOAG, P. T., AND P. R. GRANT. 1981. Intense because whenever magnirostris arrive, in natural selection on a population of Darwin's small numbers, they die, depart or hy­ Finches (Geospizinae) in the Galapagos. Science bridize with conirostris. The only condi­ 214:82-85. BOWMAN, R. I. 1961. Morphological differentia­ tion under which competitive exclusion is tion and adaptation in the Galapagos finches. likely is the immigration of immature Univ. Calif. Pub!. Zoo!' 58:1-302. magnirostris, because immature birds are ---. 1979. Adaptive morphology of song dia­ likely to be less efficient at dealing with lects in Darwin's Finches. J. Ornithol. hard seeds than adult resident conirostris. 120:353-389. BROWN, J. H. 1975. Geographical ecology of des­ Two areas of ignorance in need of em­ ert rodents, p. 315-341. In M. L. Cody and J. pirical investigation are discussed. The first M. Diamond (eds.), .Ecology and Evolution of is the set of conditions that determine the Communities. Harvard Univ. Press, Cambridge. establishment of a breeding population on COLWELL, R. K. 1979. Toward a unified approach an island; food supply and competition are to the study of species diversity, p. 75-91. In J. F. Grassle, G. P. Patil, W. Smith, and C. Taillie not the sole factors. The second is the ex­ (eds.), Ecological Diversity in Theory and Prac­ tent of competition for food in a tempo­ tice. International Co-operative Pub!. House, rally fluctuating environment. Fairland, Maryland. COLWELL, R. K., AND D. J. FUTUYMA. 1971. On the measurement of niche breadth and overlap. ACKNOWLEDGMENTS Ecology 52:567-576. CONNELL, J. H. 1975. Some mechanisms produc­ This research carried out with the per­ ing structure in natural communities: a model and mission of the Direccion General de De­ evidence from field experiments, p. 460-490. In M. L. Cody and J. M. Diamond (eds.), Ecology sarrollo Forestal, Quito, Ecuador, the and Evolution of Communities. Harvard Univ. Charles Darwin Foundation and the Press, Cambridge. Charles Darwin Research Station. It was ---. 1978. Diversity in tropical rain forests and supported by grants from NRC Canada coral reefs. Science 199:1302-1310. (A2920) and NSF (DEB79-21119). We 1980. Diversity and the coevolution of competitors, or the ghost of competition past. Oi­ thank P. T. Boag, K. T. Grant and N. kos 35:131-138. Grant for field assistance, I. Abbott for CONNOR, E. F., AND D. S. SIMBERLOFF. 1978. permission to use his measurements of Species number and compositional similarity of birds, and I. Abbott, J Myers, T. D. Price, the Galapagos flora and avifauna. Eco!. Monogr. T. Rotenberry, D. Schluter, N. M. 48:219-248. J J ---. 1979. The assembly of species communi­ Smith and a referee for valuable com­ ties: chance or competition? Ecology ments on a manuscript. Order of author- 60: 1132-1140. 656 B. R GRANT AND P. R GRANT

DEVRIES, TJ. 1976. Prey selection and hunting Isla Daphne Major, Galapagos. Oecologia methods of the Galapagos Hawk, Buteo gala­ 19:239-257. pagoensis. Le Gerfaut 66:3-42. HARRIS, M. P. 1973. The Galapagos avifauna. DIAMOND, ]. M. 1972. Avifaunal equilibria and Condor 75:265-278. species turnover rates on the Channel Islands of HURLBERT, S. H. 1978. The measurement of niche California. Proc. Nat. Acad. Sci. USA 64:57-63. overlap and some relatives. Ecology 59:67-77. 1975. Assembly of species communities, LACK, D. 1940. Evolution of the Galapagos finch­ p. 342-444. In M. L. Cody and ]. M. Diamond es. Nature 146:324-327. (eds.), Ecology and Evolution of Communities. ---. 1945. The Galapagos finches (Geospizin­ Harvard Univ. Press, Cambridge. ae): a study in variation. Occas. Pap. Calif. Acad. DoWNHOWER, ]. F. 1978. Observations on the Sci. 21:1-159. nesting of the small ground finch Geospiza fuli­ ---. 1947. Darwin's Finches. University Press, ginosa and the large cactus ground finch G. con­ Cambridge. irostris on Espanola, Galapagos. Ibis 120:340-347. 1969. Subspecies and sympatry in Dar­ GIFFORD, K W. 1919. Field notes on the land win's Finches. Evolution 23:252-263. birds of the Galapagos Islands and of Cocos Is­ 1976. Island Biology, Illustrated by the land, Costa Rica. Expedition of the California Land Birds of Jamaica. Univ. Calif. Press, Academy of Sciences to the Galapagos Islands, Berkeley and Los Angeles. 1905-1906. xm. Proc. Calif. Acad. Sci., 4th Ser. MAcARTHUR, R H. 1972. Geographical Ecology. 2, Pt. 2:189-258. Harper and Rowe, N.Y. GRANT', B. R, AND P. R. GRANT. 1979. Darwin's MACARTHUR, R H., ]. M. DIAMOND, AND ]. R finches: population variation and sympatric spe­ Karr. 1972. Density compensation in island ciation. Proc. Nat. Acad. Sci. USA 76:2359-2363. faunas. Ecology 53:330-342. ---. 1981. Exploitation of Opuntia cactus by PIANKA, KR 1969. Sympatry of desert lizards birds on the Galapagos. Oecologia 49:179-187. (Ctenotus) in Western Australia. Ecology GRANT, P. R 1975. The classical case of character 50:1013-1030. displacement. EvoI. BioI. 8:237-337. POWER, D. M. 1975. Similarity among avifaunas ---. 1977. Review of island biology, illustrated of the Galapagos Islands. Ecology 56:616-626. by the land birds of Jamaica. David Lack, 1976. RICHARDS, O. W. 1948. Species formation on is­ Berkeley and Los Angeles, Univ. Calif. Press. lands. Review of D. Lack (1947). Darwin'sfinch­ Bird-Banding 48:296-300. es, Cambridge Univ. Press. ]. Anim. EcoI. ---. 1981a. The feeding of Darwin's Finches 17:83-84. on Tribulus cistoides (L.) seeds. Anim. Behav. RICKLEFS, R K, AND M. LAU. 1980. Bias and 29:785-793. dispersion of overlap indices: results of some ---. 1981b. The role of interspecific competi­ Monte Carlo simulations. Ecology 61:1019-1024. tion in the adaptive radiation of Darwin's Finch­ RICKLEFS, R K, AND]. TRAVIS. 1980. A mor­ es. In A. Levinton and R I. Bowman (eds.), phological approach to the study of avian com­ Patterns of Evolution in Galapagos Organisms. munity organization. Auk 97:321-338. Special Publication, Amer. Asso. Adv. Sci., Pa­ ROTENBERRY, ]. T., AND J. A. WIENS. 1980. cific Division. Habitat structure, patchiness, and avian com­ GRANT, P. R, AND I. ABBOTT. 1980. Interspecific munities in North American steppe vegetation: a competition, island biogeography and null hy­ multivariate analysis. Ecology 61:1228-1250. potheses. Evolution 34:332-341. SIMBERLOFF, D. S. 1978. Using island biogeo­ GRANT, P. R, AND P. T. BOAG. 1980. Rainfall on graphic distributions to determine if colonization the Galapagos and the demography of Darwin's is stochastic. Amer. Natur. 112:713-726. Finches. Auk 97:227-244. SMITH, ]. N. M., P. R GRANT, B. R. GRANT, I. ]. GRANT, P. R, AND B. R GRANT. 1980a. The ABBOTT, AND L. K. ABBOTT. 1978. Seasonal breeding and feeding characteristics of Darwin's variation in feeding habits of Darwin's Ground Finches on Isla Genovesa, Galapagos. EcoI. Finches. Ecology 59:1137-1150. Monogr. 50:381-410. STEADMAN, D. W. 1981. Vertebrate fossils in lava --_. 198Ob. Annual variation in finch numbers, tubes in the Galapagos Islands. Proc. Eighth In­ foraging and food supply on Isla Daphne Major, ternat. Congr. Speleology, Vo!. 1:549-550. Galapagos. Oecologia 46:55-62. GRANT, P. R, B. R GRANT, ]. N. M. SMITH, I. ]. STRONG, D. R, JR., L. A. SZVSKA, AND D. S. Sim­ ABBOTT, AND L. K. ABBOTT. 1976. Darwin's berloff. 1979. Tests of community-wide charc­ finches: population variation and natural selec­ ter displacement against null hypotheses. Evo­ tion. Proc. Nat. Acad. Sci. USA 73:257-261. lution 33:897-913. GRANT, P. R, AND T. D. PRICE. 1981. Population SULLOWAY, F.]. 1982. The Beagle Collections of variation in continuously varying traits as an eco­ Darwin's Finches (Geospizinae). Bull. British logical genetics problem. Amer. Zoo!' 21:00-00. Museum Nat. Hist., ZooI. Ser. GRANT, P. R,]. N. H. SMITH, B. R GRANT, I. ]. WHITTAKER, R H. 1960. Vegetation of the Sis­ ABBOTT, AND L. K. ABBOTT. 1975. Finch kiyou Mountains, Oregon and California. EcoI. numbers, owl predation and plant dispersal on Monogr. 30:279-338. DARWIN'S FINCHES 657

WIENS, J. A. 1977. On competition and variable WILLIAMSON, M. 1972. The Analysis of Biological environments. Amer. Sci. 65:590-597. Populations. Edward Arnold, London. WIENS, J. A., AND J. ROTENBERRY. 1980. Pat­ YANG, S. Y., AND J. L. PATTON. 1981. Genic terns of morphology and ecology in grassland and variability and differentiation in Galapagos scrubsteppe bird populations. Ecol. Monogr. finches. Auk 98:230-242. 50:287-308. WIGGINS, I. L., AND D. M. PORTER. 1971. Flora Corresponding Editor: R. E. Ricklefs of the Galapagos Islands. Stanford Univ. Press, Palo Alto.