1. A reduction in forest area should result in a reduction of its number of species and, moreover, do so in a characteristic way according to the familiar species-area relationship. Brooks, Pimm & Collar (1997) applied this formula to the losses in forest area in the Philippines and Indonesia. Independently derived totals of the number of endemic bird species that are threatened with extinction broadly agree with these predicted losses. In some cases, however, predicted losses overestimate or underestimate the actual numbers of threatened species.
2. Within an island, the proportionate deforestation to date might be most extensive where there are many endemic species, or where there are few. To test this possibility, we obtained recent forest cover data for the region. We separated lowland (< 1000 m a.s.l.) from montane (> 1000 m a.s.l.) forest cover by overlaying topographic maps. From these data, we predict separately the numbers of montane and lowland endemic bird species likely to become extinct as a result of deforestation. We then compared these totals with the numbers considered threatened in the latest Red List.
3. Our predictions based on deforestation closely match the numbers of threatened endemic birds in the lowlands, but underestimate them in montane regions.
4. Our predictions based on deforestation underestimate the number of threatened montane mammal species even more seriously.
5. Lowland faunas of insular South-east Asia are under extreme threat because of massive deforestation. The region's montane faunas appear seriously threatened even by low levels of deforestation.
Tropical deforestation clears approximately 10% of the original area per decade (Whitmore 1997), with about half that area already gone (Myers 1994). This causes a large and predictable loss of species (Pimm et al. 1995). On-the-ground surveys and satellite imagery document the proportion of forest that has been lost. Given this information, we can apply the insight from MacArthur & Wilson's (1967) classic theory. It stated that small habitat islands should contain fewer species than in larger blocks of continuous habitat, just as small oceanic islands have fewer species than larger ones. Thus, as habitat is reduced, species should be lost in a characteristic way. Only species endemic to the area are at risk of global extinction, however, although many other species could be lost locally. We consider only an area's endemic species and predict what fraction of them is likely to become extinct (Pimm & Askins 1995) as habitat shrinks.
We have applied this formula to the birds of eastern North America (Pimm & Askins 1995), the Atlantic forests of South America (Brooks & Balmford 1996) and insular South-east Asia (Brooks et al. 1997). For the first case, the major forest losses were over a century ago and the predictions of extinctions match those observed. For the second and third cases, the predictions match the numbers of species considered threatened with ‘a high risk of extinction in the wild in the medium-term future’ in Birds to Watch 2 (Collar, Crosby & Stattersfield 1994), the source on which the latest Red Data Book of globally threatened species is based (Baillie & Groombridge 1996).
The third—and most detailed—case indicated some important discrepancies between observed and predicted numbers (Brooks et al. 1997). In the Philippines and on Sumatra our predictions based on deforestation underestimated the number of threatened endemic species. In the Lesser Sundas, Java and Sulawesi our predictions overestimated the number of threatened endemic species. Several hypotheses could explain why this should be the case.
First, Collar et al. (1994) may overestimate the number of threatened species (in the Philippines and on Sumatra) or underestimate it (in the Lesser Sundas, Java and Sulawesi). As our knowledge of South-east Asian birds improves, we may recognize the ‘responsible pessimism’ of Collar et al. (1994) as being overly cautious. That is, if one does not know much about a species, it is likely to be rare. Collar et al. might then conclude that it is vulnerable. Better knowledge may show it not to be.
These cases may be balanced by three possibilities that lead to the underestimation of the true number of threatened species. One possibility is that new species will be rare ones—the common ones have all been found. Another is that taxonomic revisions typically split off local populations of more widely ranging species. In both cases, these additions to the list will probably be of threatened species (Cuarón 1993). The last possibility is that science may have failed to describe some endemics before humans drove them to extinction, as happened extensively in Polynesia (Pimm, Moulton & Justice 1994).
The second hypothesis is that our predictions underestimate the number of threatened species at high levels of deforestation, when at least two factors will cause a greater loss of species from the remaining small forest fragments than expected (Laurance et al. 1997). Small fragments will typically be more isolated from each other than large ones (With 1997). Moreover, small fragments have large ratios of forest edge to forest interior and forest edges sometimes are poor habitats for forest interior specialists (Murcia 1995).
Some patterns support this possibility. The extensively cleared Philippine forests have more threatened species than one expects and the still mostly forested Borneo has fewer. Other patterns, however, counter it. Our predictions underestimate threat on the still heavily forested Sumatra and overestimate threat on the denuded islands of Java and the Lesser Sundas. In summary, the hypothesis cannot fully explain the discrepancies. Nonetheless, it begs our asking whether the details of the patterns of fragmentation might be useful in improving the formula's predictions.
Third, there is a pronounced moisture gradient within the region. The northern and western islands tend to be wetter and the southern and eastern islands tend to be drier (Whitmore 1984). The endemics of the dry monsoon forests of the lowlands of Lesser Sundas and, to a lesser extent, eastern Java and southern Sulawesi, may be able to adapt to the secondary growth and scrub that deforestation leaves behind.
We present new data on the distribution of lowland and montane forest cover and endemism for both birds and mammals to explore these hypotheses.
Materials and methods
Our analysis requires three steps. First, we measure the distribution of remaining forest. Second, we predict the numbers of threatened species using the species-area relationship. Third, we compare these predictions to the independently assessed lists of threatened species.
Insular South-east Asia holds four archipelagos: the Philippines, the Greater and Lesser Sundas, and northern Wallacea (Fig. 1). The latter archipelago comprises the large island of Sulawesi and the smaller islands of the Moluccas. These, along with the Lesser Sundas, form the biogeographic province of Wallacea. We assume that historically all four archipelagos were forested (Collins et al. 1991), with predominantly moist forest in the north (where most deforestation is recent) and dry monsoon forest in the south (where much of the forest has been cleared for centuries). Brooks et al. (1997) gave a broad overview of the forest cover of the region.
We divide these four archipelagos based primarily on biogeography (e.g. Heaney 1986; ICBP 1992), combining, for example, Java with Bali, and Sulawesi with the Sula Islands. The bathymetry of the surrounding seas largely dictates this biogeography. It determined which islands were connected during periods of lower sea level throughout the Pleistocene (Heaney 1991). We measure deforestation at the scale of these island groups (Fig. 1). Brooks et al. (1997) differed in not combining Simeulue, Mentawai and Enggano as the West Sumatran Islands (Holmes 1994) and including the tiny island of Tanahjampeah with the rest of the Lesser Sundas (Dutson 1995). Here, we also shorten terminology from Brooks et al. (1997), simplifying Sangihe and Talaud into the Far Moluccas (Lazell 1992) and the Central Visayas into simply Cebu (Dutson, Magsalay & Timmins 1993).
There is no clear-cut boundary between lowland and montane forest (Whitmore 1984). On steep slopes, forest that is montane can reach as low as 500 m a.s.l. Conversely, plateaus can hold forest that is lowland in character at altitudes well over 1000 m a.s.l. One consequence is that the transition from lowland to montane forest is at lower altitudes on small islands. These islands necessarily have their steepest slopes at a lower altitude. Nevertheless, the contour line of 1000 m broadly marks the transition between forest zones. We used this contour to define lowland and montane forest in our data set, recognizing that considerable variability exists.
Predicting numbers of extinct and threatened species using the species-area relationship
The species-area relationship, S = cAz (where S = number of species, A = area and c and z are constants), estimates the numbers of species found in different-sized areas (Preston 1962). Simple rearrangement of the relationship gives a prediction of the number of species that will eventually be lost when a particular habitat is reduced in size. Dividing Ssurviving = cAsurvivingz by Soriginal = cAoriginalz gives Ssurviving/Soriginal = (Asurviving/Aoriginal)z. Because Sextinct = Soriginal – Ssurviving, we can derive Sextinct = Soriginal – [Soriginal(Asurviving/Aoriginal)z]. Rosenzweig (1995) reviewed a wide range of empirical and theoretical work which suggests that for island situations the appropriate value for the exponent, z, is that originally derived by Preston (1962) of ≈ 0·25. Appendix 1 provides values of Asurviving and Aoriginal (and hence Asurviving/Aoriginal) for the South-east Asian islands and island groups and for the lowland and montane portions of those islands. The data are expressed as percentages. Appendices 2 and 3 present the calculations of Sextinct that we derive from them.
Counting species: endemism, altitudinal distribution and conservation status
We did not include in the threatened totals any species for which Collar et al. (1994) do not list habitat destruction as a threat. Thus we exclude from these totals the 11 species in the region listed under criterion D (‘a very small population and/or a very small range’), as these are not threatened by current deforestation. We also excluded poorly known ‘Data Deficient’ species and those listed as ‘Near Threatened’ or ‘Conservation Dependent’.
We should also exclude any species threatened not by deforestation but by direct human exploitation (under criteria A1c and/or A2c), but no bird species in the region are listed solely under these criteria (Collar et al. 1994). Earlier, we (Brooks et al. 1997) did exclude five species of Wallacean parrots [Lorius garrulus (Linnaeus) 1758 and Cacatua alba (Müller) 1776 from Halmahera, L. domicella (Linnaeus) 1758 and C. moluccensis (Gmelin) 1788 from Seram, and C. sulphurea (Gmelin) 1788 from the Lesser Sundas and Sulawesi] that are primarily threatened by the cage-bird trade (Smiet 1985). In this analysis, we include these species, because all five are also under threat from habitat destruction (Collar et al. 1994). We did this for two reasons. The first was for consistency, because a number of other birds are threatened by direct exploitation as well as habitat loss (Collar et al. 1994). The second is because, realistically, threat from direct exploitation will increase as habitat loss increases (Diamond 1984).
A number of future changes may affect the number of species likely to become extinct in the region. The discovery of new species or taxonomic reassessment of existing forms may increase the number of species considered threatened (Cuarón 1993). Only two definite new species—Aethopyga linaraborae (Kennedy, Gonzales & Miranda 1997) and Gymnocrex talautensis (Lambert 1998)—have been described in the region since the publication of Collar et al. (1994). The status of another new (nonforest) species, Lonchura pallidiventer (Restall 1996) is unclear, because it is known only from trade specimens, probably derived from Borneo. Three further new species from the Philippines—a woodcock Scolopax sp. (Harrap & Fisher 1994), a bleeding-heart Gallicolumba sp. (Klop, Curio & de Soye 1998) and a shortwing Brachypteryx sp. (Morris 1996)—have yet to be formally described. Nevertheless, the number of new species will probably be low.
Continuing deforestation across the region (Whitmore 1997) will certainly cause many more species to become threatened. In contrast, the very act of listing the region's species as threatened will hopefully act as a conservation catalyst (Collar 1996). While we recognize that in combination these factors will probably lengthen the list of threatened species, we conservatively restrict our analysis to species listed as threatened by Collar et al. (1994) alone. Appendix 2 summarizes the bird data.
We extend our analyses to mammals. There must be gaps in the mammal data, for even the distribution of North American mammals is incompletely known (Grayson & Livingston 1993). Heaney (1997) reports that at least 13 new species are currently being described from the Philippines alone. However, the region's mammal faunas are in broad agreement with general biogeographic patterns (Heaney 1986), suggesting that they are well enough known for us to make coarse predictions of the number of endemics that will become extinct following deforestation.
We compile a database of the region's endemic mammals. Our source was the latest taxonomy and distribution of the world's mammals, Wilson & Reeder (1993). All the endemic species for which Wilson & Reeder (1993) give habitat information are forest species. Furthermore, Heaney et al. (1989) found that forest mammals in the central Philippines tended to be endemic to particular islands, while species of deforested areas tended to have wide global ranges. Consequently, we assume that forest is the habitat for all endemic mammals of insular South-east Asia.
There are fewer data for altitudinal ranges for mammals than for birds. We again take this information from Wilson & Reeder (1993), which on occasion presents explicit ranges. More often we have to rely on qualitative information given by Wilson & Reeder (1993) or even, for the least-known species, simply whether or not the type locality is listed as a mountain. We use Musser (1987) for Sulawesi (who gives explicit altitudinal ranges for nonvolant mammals), and Payne, Francis & Phillipps (1985) for all Bornean endemics.
We take our list of threatened and extinct species from Baillie & Groombridge (1996). Six insular South-east Asian endemic mammals are listed as extinct, although numerous other species may have been lost (e.g. Terborgh 1974). Of the known extinctions, two are recent losses of fruit bats from the Western Visayas (Heaney & Heideman 1987). In contrast, three rodents from Flores (Musser 1981) and one from Timor (Glover 1986) are known only from fossil deposits 3550 years old. We include these species because a small possibility remains that they survive to the present (Wilson & Reeder 1993).
We exclude from consideration 17 taxa that Baillie & Groombridge (1996) deem to be species, but Wilson & Reeder (1993) do not. Twelve of these are ‘lumped’ by the latter, one was described too recently (Gonzales & Kennedy 1996) to be included, and four (‘Otomops johnstonei’, ‘Apomys gracilostris’, ‘Hipposideros madurae” and ‘H. sorensoni’) we have not been able to trace. We do not count 21 species as threatened by deforestation. Of these, 19 are endemic to one of the archipelagos, the other two are widespread within the region. These 21 are listed as under pressure from direct exploitation only (Red List criteria A1c and/or A2c). We also exclude a further 13 species listed as threatened by their naturally tiny ranges and/or populations (criteria D1 and/or D2).
As with birds, we recognize that Baillie & Groombridge (1996) will almost certainly underestimate likely species extinction as a result of the effect of taxonomic revision (e.g. Heaney 1997) and accelerating deforestation (e.g. Whitmore 1997). Nevertheless, we once again take the Red List (Baillie & Groombridge 1996) as our conservative, empirical estimate of which insular South-east Asian endemic mammals are likely to become extinct in the medium-term future. Appendix 3 summarizes the mammal data.
Table 1 compares the observed numbers of threatened and nonthreatened endemics to those numbers of endemics predicted to be threatened and not to be threatened on the basis of the extent of deforestation. As a specific example, there are 182 species of birds endemic to the Philippines. Of these, 71 are threatened with extinction, while 111 are not. Predicted numbers arise as follows. Luzon has 25% of its forests remaining (Appendix 1). There are 25 species endemic to the island (Appendix 2), and from the extent of deforestation, the species-area formula predicts that 7·3 should be threatened. (Predicted values need not be integer values.) Comparable predicted numbers of threatened species for Mindoro, the Western Visayas, Cebu, Mindanao, the Sulus and Palawan are 2·6, 4·4, 3·0, 9·8, 1·7 and 2·7, respectively, for a total of 31·5 species. There are 82 species found across the Philippines but not outside the archipelago (Appendix 2). Averaged over the archipelago, 25% of the forest remains (Appendix 1), so of these 82, we predict that 24·0 should be threatened. Combined, 55·5 species should be threatened, and thus 182 – 55·5 = 126·5 should not be threatened.
Table 1. Endemic birds of insular South-east Asian archipelagos. See text for data sources. We compare the numbers of species threatened (Collar et al. 1994) and not threatened that are endemic to each archipelago with the numbers of endemic species predicted by deforestation to become extinct and to survive. In each case our null hypothesis is that deforestation predicts the number of threatened species (Brooks et al. 1997). For each archipelago we give a χ2-value and its significance level for this comparison. All tests have 1 d.f. In cases where an expected value is less than 5, the sampling distribution of χ2 no longer approximates the chi-squared distribution, and so we apply Yates’ correction (Spiegel 1994). We indicate in parentheses the effect of excluding from the observed counts of threatened species the Wallacean parrots which are threatened by trapping as well as habitat destruction
The null hypothesis is that our predictions based on deforestation match known threat, and thus that observed and predicted numbers do not differ more than expected by chance. The χ2 for the comparison, 71 observed vs. 55·5 predicted and 111 observed and 126·5 predicted, is 6·2. The probability of this happening were the null hypothesis true is P = 0·02 and so we reject that hypothesis. Our predictions do not accurately match the numbers of threatened species.
The first four lines of Table 1 recap Brooks et al. (1997). The formula satisfactorily predicts the observed numbers of threatened species on the Greater Sunda and Northern Wallacea. It significantly underestimates them in the Philippines, and overestimates them in the Lesser Sundas.
The second and third analyses are for montane and lowland forests, respectively. An interesting pattern emerges. The species-area formula underestimates the threat to montane species in the Philippines and overestimates the threat to lowland species in the Lesser Sundas. The remaining six comparisons are not significantly different from what deforestation predicts. Non-significant results are in the majority. This means that our formula satisfactorily predicts the number of threatened species. There are two significant results, however, and it is these refutations of our model that we seek to understand.
Comparisons of the island-by-island predictions (Appendix 2), show that our species-area formula consistently underestimates the threat to montane species. In the Philippines, the observed numbers of threatened species exceed those predicted on four of the five islands that hold montane endemics. The exception is Palawan, which has only two montane endemics and 96% of its montane forest remaining. Neither species is threatened, nor would we expect either to be, given that so much of the forest remains. On all three of the Lesser Sundas island groups that have montane endemics, the observed numbers of threatened species exceed those predicted. In the Greater Sundas, Sumatra has six threatened montane species; our predictions suggest that only one should be. For Java, Bali and Borneo combined, our formula predicts 3·3 threatened species, whereas there are only two. For Northern Wallacea, three islands have one threatened species each; we predict that only one of these islands should have a threatened species.
For the lowlands, comparisons of the island-by-island predictions (Appendix 2), show no consistent deviation from what the species-area formula predicts, with the aforementioned exception of the Lesser Sundas. There are 20 independent contrasts of predicted and observed numbers of threatened species. Observed values are smaller than expected values for 7 of these, larger for 12, and there is one tie. On the Lesser Sundas, however, the overall pattern is for the predictions to overestimate threat consistently. This applies to each of the four island groups, and to species found throughout the Lesser Sundas, but nowhere else.
We separate the four lowland North Wallacean parrot species (C. sulphurea occurs in the Lesser Sundas as well as on Sulawesi) that we did not count as threatened in Brooks et al. (1997) within the totals of threatened species (Table 1). Our inclusion of these species further increases the degree to which our observed values match the predictions in North Wallacea.
Table 2 repeats the analyses for mammals. Once again, the deviations from the expectations from deforestation separate according to whether the species are montane or lowland. The observed numbers of threatened lowland species are satisfactorily predicted by the extent of the deforestation both within each archipelago (Table 2) and across the 20 independent island-by-island comparisons (Appendix 3). There are slightly more threatened mammals in the lowlands of the Lesser Sundas than we predict, rather than significantly fewer than predicted for birds.
Table 2. Endemic mammals of insular South-east Asian archipelagos. See text for data sources and Table 1 legend for details of the structure of the table and mechanics of the comparisons. We indicate in parentheses the effect of including in the observed counts of threatened species the 21 species which are threatened by hunting alone
There are only two montane endemics in the Lesser Sundas, so we cannot draw any conclusions. In the other three archipelagos, the predicted numbers of threatened species are significantly fewer than those observed. In the 16 independent contrasts of predicted and observed numbers (Appendix 3), observed numbers exceed predicted numbers 10 times, and there are 4 ties. All the ties are cases where the prediction is that no species should be threatened and none are.
We indicate in Table 2 the effect of including those mammals listed by Baillie & Groombridge (1996) as threatened by direct exploitation only. In nearly every case, this would increase the degree by which our number of threatened species exceed our predictions. For the few montane cases, their inclusion further increases the statistical significance of the already significant tests. For the lowlands of three archipelagos, the inferences change. With the inclusion of these species, deforestation significantly underestimates threat.
Across insular South-east Asia, for birds and mammals, we satisfactorily predict the numbers of threatened species, based on the extent of deforestation, with three key exceptions. In increasing order of importance:
1. If we include in our counts of threatened species those 21 mammal species which are threatened solely by direct human exploitation (Baillie & Groombridge 1996), then our predictions underestimate threat in nearly all lowland regions.
2. We consistently overestimate the degree of threat to birds in lowland habitats across the Lesser Sundas.
3. The predicted numbers of threatened species of birds and mammals derived from estimates of deforestation consistently underestimate the observed numbers in montane forests. What could explain these exceptions?
General problems of bias
One worry is that the Red List may not have been compiled independently of the extent of deforestation. As a hypothetical example, suppose that Collar et al. (1994) looked at the island of Cebu and, from their knowledge that only a few hectares of the island's forest remain, concluded (correctly) that all its endemic species would be on the verge of extinction. Such a process would guarantee a match between observed and predicted numbers of threatened species, when all the island's species are threatened. As we discussed in Brooks et al. (1997), the decisions of Collar et al. (1994) and Baillie & Groombridge (1996) to list species were made based on case-by-case review (with habitat loss within the species’ range only one of the factors considered). In any case, this process would not so readily obtain a match when not all the island's species should be threatened. How many species should they choose—and which ones? Nor does the hypothetical example explain the results where observed and predicted numbers diverge so consistently.
Another possibility is that observed numbers of threatened species might be higher than predicted because the species may be threatened by factors other than deforestation. These factors will tend to be correlated with habitat loss (Diamond 1984). It may be easier to hunt in forest fragments or from forest edges, for example. We try to control for these possibilities in three ways.
First, we excluded all species listed as threatened solely as a result of their being hunted, trapped or otherwise exploited. We discuss the consequences of including these below. We included the many species that are listed because of deforestation as well as direct exploitation (e.g. O’Brien & Kinnaird 1996). Our results would not have changed significantly had we excluded the five lowland Wallacean parrots that are threatened by deforestation as well as by trapping (Table 1). Second, we excluded all species threatened because of their tiny ranges alone, regardless of the extent of deforestation. Third, we checked that no species were listed as threatened based on projected future threat only (criterion A2): none are, although it is not impossible that future threat could also be included in the B criterion.
Underestimates of threat to lowland mammals
Baillie & Groombridge (1996) list 21 species of mammals as threatened by direct exploitation alone: seven Philippine endemic bats [Acerodon leucotis (Sanborn) 1950, Haplonycteris fischeri Lawrence 1939, Nyctimene rabori Heaney & Peterson 1984, Otopteropus cartilagonodus Kock 1969, Pteropus leucopterus Temminck 1853, P. pumilus Miller 1911 and Rhinolophus subrufus K. Andersen 1905] and one flying lemur [Cyanocephalus volans (Linnaeus) 1758]; the bat Dobsonia peroni (E. Geoffroy) 1810 from the Lesser Sundas; four bats (Nycteris javanica E. Geoffroy 1813, Pipistrellus cuprosus Hill & Francis 1984, Mormopterus doriae K. Andersen 1907 and Otomops formosus Chasen 1939) and three primates [Nasalis larvatus (Wurmb) 1787, Presbytis potenziani (Bonaparte) 1856 and Hylobates klossii (Miller) 1903] from the Greater Sundas; the bat Hipposideros inexpectatus Laurie & Hill 1954 and the palm-civet Macrogalidia musschenbroekii Schwarz 1910 from Sulawesi; the bat Acerodon humilis K. Andersen 1909 from the Far Moluccas; and two endemics widespread within insular South-east Asia, the primate Trachypithecus auratus (E. Geoffroy) 1812 and the rodent Haeromys pusillus (Thomas 1893). That two-thirds of these species are bats is not surprising. Bats are heavily hunted throughout the region (Heaney & Heideman 1987). But why should including these heavily hunted species cause deforestation to underestimate threat in the lowlands but not in the montane regions? The explanation is simply that people live in the flat lowland regions. Lowland mammals are threatened beyond what we expect as a result of deforestation simply because their habitat is accessible to their human predators.
Overestimates of threat in lowland monsoon forest
The lowland birds of the monsoon forests of Java and the Lesser Sundas are less threatened than the extent of deforestation would predict. These islands, because much of their forest has been cleared for centuries, may have lost lowland endemics before they were known to science (van Balen 1994).
One of these missing endemics could have been Argusianus bipunctatus (Wood) 1871, known only from a portion of a male primary feather described in 1871 (Davison 1983; Sibley & Monroe 1990), although the validity of this taxon is in doubt (Parkes 1992). Delacour (1951) noticed that Java is the most obvious gap in the Sundaic range of the congeneric A. argus (Linnaeus) 1766, and suggested that A. bipunctatus had replaced the species there. Davison (1983) argued that because Javan ornithology dates back ‘much earlier than the discovery date of A. bipunctatus’ the species was more probably from Tioman, off peninsular Malaysia. However, there is no reason that the feather could not have been obtained on Java well before 1871. Another missing endemic could have been Gallirallus sharpei (Büttikofer) 1895 known from a single specimen purchased in Amsterdam in 1865, and which Olson (1986) argued must have derived from the Sundaic region. By comparing the lowland avifaunas of Java with Borneo and Sumatra, van Balen (1994) further speculates that trogons, malkohas and frogmouths have also been lost. Only archaeological exploration will be able to provide further evidence for this ‘missing species’ hypothesis (Pimm et al. 1994).
Alternatively, our results could support a second hypothesis that the birds endemic to dry monsoon forest are more tolerant of deforestation than are moist forest species. This remains to be tested elsewhere and it does have broader relevance; for example, the New World dry forests are the ‘most endangered tropical ecosystem’ (Janzen 1988). Interestingly, only 17% of the region's threatened birds are dry forest species, compared to the 57% that are moist forest species (Collar, Wege & Long 1997). The proportions of total dry and total moist forest species considered threatened are not known, but the data do suggest that dry forest species may be under proportionately less threat. A similar example comes from the largely deciduous forests of the Western Ghats of India. Despite the fact that these ‘have been much degraded by human activities’, none of their 19 endemic bird species are considered threatened (Gaston & Zacharias 1996). Their birds are ‘largely opportunistic habitat users’ (Daniels 1996), able to adapt to, or at least forage in, coffee plantations (Shahabuddin 1997). These patterns should not be interpreted as suggesting that dry forest species are unharmed by deforestation.
These two hypotheses are not necessarily exclusive. Monsoon forests are more easily settled by people than are rainforests, and so the drier southern islands of the region have been inhabited for a longer time than the rainforest islands (Collins et al. 1991). Ancient deforestation in the lowlands of Java and the Lesser Sundas could have provided an ‘extinction filter’ (Balmford 1996), driving the most vulnerable species to extinction while the survivors adapted to the remaining degraded habitats.
For mammals, deforestation does not overestimate threat in the Lesser Sundas. This may be a result of inclusion in the Red List of the Lesser Sundas’ four (probably) long-extinct species (Musser 1981; Glover 1986). This would be consistent with the ‘missing species’ hypothesis. At least three more extinct murids are known from Timor but await formal description (Glover 1986). Alternatively, we might expect mammals, being poor dispersers (Brown 1971), to be less tolerant than birds of the fragmentation of any kind of forest. If this is the case, then we should not be surprised to find that lowland deforestation overall predicts threat to the region's mammals very well.
Underestimates of threat in montane forest
Why are more montane species threatened than we expect from the loss of habitat? There are several hypotheses. First, what if the underestimate of the level of threat based on deforestation is really a consistent overestimate of the level of threat by the Red List? Perhaps montane endemics, with naturally tiny ranges and populations, are perceived to be more threatened than they actually are. In our analysis, we assumed that the listing process rigidly followed the IUCN criteria for threatened status (Baillie & Groombridge 1996). All species threatened only by naturally tiny ranges or populations should have been listed under criterion D—which we exclude. If such species have been perceived as rarer than they are—perhaps because of the difficulties of carrying out fieldwork in montane forest—the Red List may overestimate threat to montane endemics. Only further detailed fieldwork will answer this possibility.
Second, our species-area formula based on deforestation may underestimate the threats in montane forest because fragmentation can threaten species over and above the loss of area alone (Laurance et al. 1997). As a simple product of their rugged topography, might remaining montane forests have proportionately more forest edge than those surviving lowland forests? This hypothesis is straightforward to test if we consider that as levels of fragmentation increase, the ratio of edge: area will increase as well (Krummel et al. 1987). For the Philippines forests, Dawning, Iverson & Brown (1993) found that forest patches with a high edge to area ratio in 1934 tended to be small and likely to be deforested subsequently.
From the GIS data, we calculated the length of forest–nonforest edge (i.e. excluding the length of the forest–sea and lowland forest–montane forest edges) for montane and lowland habitats on all islands (Appendix 1). The edge-to-area ratio is scale dependent, so we calculated the value of edge divided by square root of area. We then compared these values using a two-tailed Wilcoxon rank test. The edge to square root of area was significantly greater in lowland than in montane forest across all of the islands (t = 3·92, 19 d.f., P < 0·01). The results reject the hypothesis that montane forest is more fragmented than lowland forest.
The third and fourth hypotheses suggest that there are systematic differences between the ranges of montane and lowland endemic species. Hypothesis three is that montane endemics, unlike lowland endemics, might have generally complementary ranges; for example, each mountain range might have its unique species. The larger the area sampled, the more mountain ranges would be included, and so the more species. Species numbers would increase in direct proportion to the area and so the species to area relationship would have a slope of unity. If this were the case, the destruction of forests would reduce the number of species in direct proportion to the area lost. Lose half the forest and lose 50% of the species—a considerably greater loss than the 15% loss predicted for a species-to-area relationship with an exponent of 0·25.
We can test hypothesis three in two ways: directly, by looking at whether the distributions of montane endemics are indeed complementary; and indirectly, by examining the slope of the species-to-area relationships.
Within the mountains of Mindoro, Palawan, Halmahera, Buru, Seram, Sumba and Timor the distributions overlap almost completely. On Luzon, Mindanao, Java and Flores, the distributions are nested within those of the most widely occurring species. No distributions are complementary to each other. On Sumatra, with 14 montane endemics, one species pair is complementary between the North and South Barisan range, Lophura hoogerwerfi (Chasen) 1939 and L. inornata (Salvadori) 1879. These forms may actually be a single species (Mackinnon & Phillipps 1993). On Borneo, with 28 montane endemics, only one species, Oriolus hosii Sharpe 1892, does not occur on the largest mountain, Mt Kinabalu, being replaced there (and elsewhere in the Greater Sundas) by the widespread O. cruentus (Wagler) 1827 (Mackinnon & Phillipps 1993). On Sulawesi, with 29 montane endemics, there is a single complementary species pair, Cyornis sanfordi Stresemann 1931 (endemic to the Minahassa Peninsula) and C. hoevelli (Meyer) 1903 (elsewhere on the island) and one species endemic to the isolated Lompobattang massif, Ficedula bonthaina (Hartert) 1896 (Fraser & Henson 1996). The distributions of montane endemics are perfectly complementary in only one case, the Western Visayas, where the three species have nonoverlapping ranges (Dickinson et al. 1991): Ptilinopus arcanus Ripley & Rabor 1955 on Mt Canlaon, northern Negros; Stachyris latistriata on Mt Baloy, Panay; and S. nigrorum Rand & Rabor 1952 on Mt Talinis, southern Negros.
In summary, very few of the single-island montane endemic birds of the region have complementary distributions. These observations reject the possibility that a high level of complementary montane endemism accounts for the disproportionately high threat that we observe to montane endemics.
Figure 2 plots the numbers of montane and lowland endemics (Tables 1 and 2) against their potential ranges (Appendix 1). These are endemics-area relationships (Harte & Kinzig 1997). Were the slopes of the montane species steeper than for the lowlands, this would both suggest greater complementarity in montane ranges and a greater susceptibility to deforestation. The lowlands of Sumatra, Java and Borneo have far fewer endemics than one would expect from their size. This is not surprising: these islands had Pleistocene connections to mainland South-east Asia (Heaney 1986). Excluding them, the log-log fits of species to area are nearly identical for montane and lowland birds and similar for montane and lowland mammals.
In summary, the endemic area relationships also reject a difference in the patterns of the ranges of montane and lowland species that would explain the greater levels of threat suffered by the former.
The fourth hypothesis is that species with small ranges may be disproportionately threatened by a given proportionate forest loss. This could be a perception, rather than reality—the consequence of classifying any species with a very small range as being threatened, even in the absence of direct evidence. Such perceptions may have a high probability of becoming reality, of course, because habitat destruction can quickly drive such species to extinction.
The disproportionate threat could also be a real effect for one of several reasons. Other things being equal, a temporary reduction in numbers, say 50%, is no threat to a population numbering in the millions, but it would be a matter of concern to a population numbering in the hundreds, and would be fatal to a population of two individuals.
Other things are often not equal, making life even harder for species with small ranges. Typically such species are less dense within their ranges than are more widespread species (Brown 1984). Less abundant species are more vulnerable than common species (Pimm 1991). The argument against this possibility is that correlation of range size and density is, at best, one with much variation. Across this region, many montane species are extremely common within their tiny ranges, maybe as a result of density compensation (MacArthur, Diamond & Karr 1972) in their depauperate communities (Terborgh 1977). A good example is Stachyris nigrorum, which we found to be the second most common bird on Mt Talinis (in South Negros in the Western Visayas), the single mountain from which it is known (Brooks et al. 1992). Only fieldwork to determine accurate population densities for the insular South-east Asian endemic species will resolve this problem.
The critical questions ask: Are species at greater risk than expected because they have small ranges? If so, are montane species at greater risk because a greater proportion of them have smaller ranges?
The numbers are too small to break down these results by archipelago. Across the four archipelagos, there are 183 montane endemic birds (Appendix 2), of which 141 occur only on one island, and 42 occur on more than one island within each archipelago. Of the more widespread 42 species, only four are threatened and deforestation predicts this number almost exactly (3·7). It is montane species found only on single islands that are disproportionately threatened: 34 are threatened, compared to the prediction of only 10·4. For mammals (Appendix 3), there are only seven widespread montane endemics, one is threatened, which matches the prediction. Again, the threatened species are endemics found on single islands.
These numbers are too few to draw strong conclusions. Nonetheless, they suggest that it is not the fact of being montane that makes a species unusually threatened. Being montane usually means being restricted to one island, and, by our definition, the montane forests of that island. Such species have typically the smallest ranges of those we consider. We do not have estimates of the ranges of all the species, so we cannot make the obvious comparison to those lowland species that have very small ranges.
This systematic failure of the predictions of threat when applied to species with the smallest range sizes matches our earlier conclusion (Brooks et al. 1997). We found then that endemics restricted to single islands were more threatened, compared to predictions, than species found throughout an archipelago. These, in turn, are more threatened, compared to predictions, than species found throughout the region.
The biodiversity of Indonesia and the Philippines is under great threat. Montane species are disproportionately threatened, because, we assert, they have the smallest ranges. This finding has precedents. Renjifo et al. (1997) recently highlighted the disproportionate threat faced by montane avifaunas in the northern Neotropics. Within insular South-east Asia, the discovery of Aethopyga linaraborae by Kennedy, Gonzales & Miranda (1997) has attracted attention to the highland birds of Mindanao. Crucially, the lowlands of insular South-east Asia hold by far the greater absolute numbers of threatened species. These lowlands should therefore remain key elements of conservation strategy, supplemented by montane areas.
Our analysis indicates that threat to insular South-east Asian mammals, especially the montane endemics, is underestimated by the extent of deforestation. This is well-documented (Oliver 1994; Heaney 1997). It is also not surprising. We would expect poor dispersers like mammals to be particularly susceptible to habitat loss (Brown 1971). Furthermore, while 11% of the world's birds are threatened, 25% of mammals are threatened (Baillie & Groombridge 1996). The pattern results from there being so many mammals with small, montane distributions. This has implications for conservation strategy. Hague et al. (1986) showed that distribution patterns across the Philippines are similar across higher taxa and hence indicated that conservation planning could be based on any one taxon (but see van Jaarsveld et al. 1998). Our results suggest that it is important to compare patterns of threat as well as patterns of distribution to choose a given higher taxon (e.g. birds or mammals) as a baseline for conservation strategy.
Finally, estimates of the total size of the ‘sixth extinction’ usually make arguments based on projections of tropical forest loss, the known fraction of species found in tropical forests, and the species to area relationship. One such argument runs ‘if we protect 5% of the forests in national parks, then we will likely save 50% of the 80% of all species, the fraction that is found within those forests.’ Our results show that this argument is flawed. Saving 5% of tropical forests will not save as many as 50% of the original forest species if those parks are small ones.
Thanks to: C. Billington, S. Blyth, R. Luxmoore, J. Paine and M. Spalding for help analysing forest cover data; G. Anderson, W. Brown, E. Curio, B. Drewes, S. Echternacht, S. Henson, R. Kays, S. Lazell, M. McKinney, T. O’Brien and G. Shahabuddin for providing literature; S. Albon, A. Balmford, N. Collar, S. Echternacht, M. McKinney, L. Manne, K. Orvis, D. Vázquez, C. Wilder, two anonymous referees, and everyone credited by Brooks et al. (1997) for insight and comments; and T. Fisher, P. Magsalay and everyone credited by Brooks & Dutson (1996), Butchart et al. (1996) and Evans et al. (1993) for their help in the field. This study was funded jointly by a Pew Charitable Trust Scholarship in Conservation and the Environment to S.L.P. and by the World Conservation Monitoring Centre.
Table 4. Endemic birds of South-east Asian islands. See text for data sources. Single-island values are the sums of those for all islands within an archipelago. Totals for each archipelago are the sum of the single-island values and those for species endemic to but widespread within the archipelago. These are the values that were used in Table 1.
Table 5. Endemic mammals of South-east Asian islands. See text for data sources. We obtained overall values for each archipelago in the same way as for birds (Appendix 2) and used these values in Table 2.