INTRODUCTION
From 1995 to 2000, human population growth occurred in virtually all global hotspots and, although human disturbance can occur in the absence of human settlements, thus it is likely that habitats will continue to be converted and species will continue to become threatened or extinct (Cincotta et al. Reference Cincotta, Wisnewski and Engelman2000). The Atlantic Forest biome is a global biodiversity hotspot, with high conversion rates and high plant endemism (Myers et al. Reference Myers, Mittermeier, Mittermeier, Fonseca and Kent2000), and home to approximately 70% of Brazil's 169 million people (Jacobsen Reference Jacobsen, Galindo-Leal and Câmara2003). Species extinction is likely to be high in this geographically restricted region with high species endemism, heavy habitat loss and a rapidly increasing human population (Laurance Reference Laurance2006, Reference Laurance2009). For the Atlantic forest region, there are some studies related to conservation planning, such as population viability analysis of mammals (for example, marsupials: Brito & Grelle Reference Brito and Grelle2004; and primates: Brito & Grelle Reference Brito and Grelle2006; Oliveira et al. Reference Oliveira, Hankerson, Dietz and Raboy2010), and a spatial conservation prioritization with all species of a mammal group (Pinto & Grelle Reference Pinto and Grelle2009). There are also other major conservation initiatives by non-governmental organizations (NGOs) and governmental agencies (Tabarelli et al. Reference Tabarelli, Pinto, Silva, Hirota and Bedê2005). This biome urgently requires further investigation of current conservation methods and the establishment of priority areas for conservation.
The use of human population density in reserve selection procedures is based on the assumption that an increase in density represents a threat to conservation (Luck Reference Luck2007). For example, human population is a strong predictor of extinction risk, especially for large-bodied mammals (Cardillo et al. Reference Cardillo, Mace, Jones, Bielby, Bininda-Emonds, Sechrest, Orme and Purvis2005). Human population size (HPS) may also be an indicator of conflict between conservation and development (Balmford et al. Reference Balmford, Moore, Brooks, Burgess, Hansen, Williams and Rahbek2001). There are another two implicit and related reasons to include human population in the reserve selection approach: habitat conversion is directly related to HPS (Laurance et al. Reference Laurance, Albernaz, Schroth, Fearnside, Bergen, Venticinque and Costa2002), and this variable is also a surrogate for monetary costs (Luck et al. Reference Luck, Ricketts, Daily and Imhoff2004; Rangel et al. Reference Rangel, Bini, Diniz-Filho, Pinto, Carvalho and Bastos2007). Highly populated areas generally have no or very few habitat remnants. The opposite is not necessarily true. It is likely that there is some remaining habitat for conservation in counties with low human populations, but habitat conversion caused by humans can also occur in areas of low human density (Cincotta et al. Reference Cincotta, Wisnewski and Engelman2000), for example in large estates used for farming or cattle raising. Correlations between human population and species richness are almost always positive (Balmford et al. Reference Balmford, Moore, Brooks, Burgess, Hansen, Williams and Rahbek2001; Chown et al. Reference Chown, van Rensburg, Gaston, Rodrigues and van Jaarsveld2003; Luck et al. Reference Luck, Ricketts, Daily and Imhoff2004; Vázquez & Gaston Reference Vázquez and Gaston2006), but this does not mean that conservation conflicts will necessarily exist; there may be biological reasons for this correlation (for example Luck et al. Reference Luck, Smallbone, McDonald and Duffy2010) or it may be an artefact of sampling bias. Several researchers tried to identify networks of priority areas with low human population using the principle of complementarity in Africa (Balmford et al. Reference Balmford, Moore, Brooks, Burgess, Hansen, Williams and Rahbek2001), the Cerrado of Brazil (Diniz-Filho et al. Reference Diniz-Filho, Bini, Pinto, Rangel, Carvalho and Bastos2006; Pinto et al. Reference Pinto, Mathias, Blamires, Diniz-Filho and Bini2007), Europe (Araújo et al. Reference Araújo, Williams and Turner2002) and Australia and North America (Luck et al. Reference Luck, Ricketts, Daily and Imhoff2004). As well as species richness, the beta biodiversity pattern is strictly linked to complementarity.
The order Primates is the mammal group with the highest number of globally threatened species (Grelle et al. Reference Grelle, Paglia, Silva, Rocha, Bergallo, Alves and Van Sluys2006) and, in Brazil, not all species are represented in reserves of strict protection (Pinto & Grelle Reference Pinto and Grelle2009). Selecting priority conservation areas with lower human population densities is possible and desirable because the majority of the factors associated with primate species threat in Atlantic forest are related to human population, such as habitat loss, hunting for food, sport or pets, mining, illegal palm-harvesting, roads, power lines, predation from domestic pets and exotic species (see for example Machado et al. Reference Machado, Drummond and Paglia2008; IUCN 2010).
Prioritization exercises are often done using grids organized in hexagons or squares (see Reyers Reference Reyers2004; Pinto et al. Reference Pinto, Diniz-Filho, Bini, Blamires and Rangel2008; Tognelli et al. Reference Tognelli, Arellano and Marquet2008), but geopolitical units may also be relevant, since this is the scale at which administrative decisions relevant to conservation are made in practice (Hunter & Hutchinson Reference Hunter and Hutchinson1994; Ando et al. Reference Ando, Camm, Polasky and Solow1998). Indeed, Hunter and Hutchinson (Reference Hunter and Hutchinson1994) referred to the extent of study, but in Brazil, using counties as spatial units to separate grains in reserve selection procedures is pertinent since this is the smallest geopolitical unit at which this kind of administrative, and consequently conservation, decision is generally made.
In this study, we evaluate the possibility of minimizing conservation conflicts as function of HPS by applying a systematic conservation approach (Margules & Sarkar Reference Margules and Sarkar2007) to identify priority areas that represent endemic primate species in the Atlantic forest biome. We used the Brazilian counties as units of analysis within this biome, as these are the smallest administrative units where the political decisions are taken and HPS is estimated. Using the results we obtained and other findings of knowledge bias (Dennis & Thomas Reference Dennis and Thomas2000; Dennis et al. Reference Dennis, Sparks and Hardy1999), we also investigated the hypothesis that HPS in the set of counties with available data is greater than expected by chance alone.
METHODS
We obtained the map for the Brazilian counties from the Brazilian Agency of Geography and Statistics (IBGE; http://www.ibge.gov.br) and superimposed it onto the Atlantic forest biome (Fundação SOS Mata Atlântica 2009; Fig. 1). If a county's area of Atlantic forest biome was < 25% of its total area, it was excluded from the set. The remaining 3130 counties with Atlantic forest coverage ≥ 25% of their total area were considered the units of analysis in our study. We organized the presence/absence data for the 19 primate species endemic to the Atlantic Forest biome (see Pinto & Grelle Reference Pinto and Grelle2009) per county in a binary matrix. Species richness was calculated by summing the species present in each county.
Figure 1 Distribution of the Atlantic forest biome among the Brazilian states. CE = Ceará, RN = Rio Grande do Norte, PB = Paraíba, PE = Pernambuco, AL = Alagoas, SE = Sergipe, PI = Piauí, BA = Bahia, GO = Goiás, MG = Minas Gerais, ES = Espírito Santo, RJ = Rio de Janeiro, MS = Mato Grosso do Sul, SP = São Paulo, PR = Paraná, SC = Santa Catarina and RS = Rio Grande do Sul.
Human population data were obtained from the IBGE population census for the year 2007, and these were associated with the county map. We log-transformed these data to normalize distribution. We analysed the linear correlation between HPS and species richness. Autocorrelation in data may inflate Type I errors (see Legendre et al. Reference Legendre, Dale, Fortin, Gurevitch, Hohn and Myers2002; Diniz-Filho et al. Reference Diniz-Filho, Bini and Hawkins2003); we therefore used the corrected degrees of freedom using the Dutilleul (Reference Dutilleul1993) method in the test of significance for correlation, implemented through SAM software (Rangel et al. Reference Rangel, Bini and Diniz-Filho2006, Reference Rangel, Bini and Diniz-Filho2010).
Optimization routines were used to select networks to represent each species at least once with a minimum total number of sites. These routines use a total cost function that should be minimized (Ball et al. Reference Ball, Possingham, Watts, Moilanen, Wilson and Possingham2009); it increases if any species is not represented and if additional counties are selected. The total cost function also increases as more populated counties are selected (an individual cost related to HPS was given to each county in the second approach; see below). For this purpose we used the simulated annealing algorithm (Andelman et al. Reference Andelman, Ball, Davis and Stoms1999; Possingham et al. Reference Possingham, Ball, Andelman, Ferson and Burgman2000, Reference Possingham, Wilson, Andelman, Groom, Meffe and Carroll2006), which starts with a random set of reserves and, at each iteration, randomly swaps sites in and out of that set, measuring the change in the total cost function. Only changes that diminished cost were retained. In the cost function, the non representation of any species increased the cost, as well as the inclusion of any site.
Two approaches were used to select reserve networks. In the first (referred to as ‘without human constraint’ hereafter), all the counties were considered to have the same cost. In the conflict minimization approach (‘human constraint’ hereafter), the cost of counties varied according to the HPS. The counties' costs needed to be rescaled to values between 0 and 1, to values comparable to the cost 1 of not representing any species, once these were used in the same total cost function. A linear function was then adjusted such that the smallest ln-human population value (6.69) was 0.4, and the largest ln-human population value (16.20) was 1. To obtain the other counties' cost values we used the function obtained (Fig. 2):
Figure 2 Values attributed as costs to the counties obtained as a linear function of ln-human population size. We provide a detailed explanation of cost in the Methods section.
We could not attribute a zero value to the smallest ln-HPS because the county would then have had no cost, and would enter a network more easily.
We used the software MARXAN (Ball & Possingham Reference Ball and Possingham2000; Possingham et al. Reference Possingham, Ball, Andelman, Ferson and Burgman2000, Reference Possingham, Wilson, Andelman, Groom, Meffe and Carroll2006; Game & Grantham Reference Game and Grantham2008), which incorporates the simulated annealing algorithm, in the reserve selection procedures. We ran it 300 times with 10 000 000 iterations for each approach. In the approach without human constraint and with no costs attributed to the counties, most of the solutions represented all primate species with the minimum number of counties, but some did not; therefore we used the first 100 networks that did represent all the primate species with the minimum number of counties to map the irreplaceability of the counties. The frequency of each county in the various optimized networks indicates its relative importance for complementarity solutions, a simple measure of irreplaceability of that county (Meir et al. Reference Meir, Andelman and Possingham2004; see also Ferrier et al. Reference Ferrier, Pressey and Barrett2000 for a more complex approach). The highest degree of irreplaceability is when a county occurs in each of the 100 representation solutions.
The HPS in all the networks for both approaches was calculated by summing the human population of the counties within each network. These values were compared using a null distribution composed of the HPS in 100 000 random networks with the same number of counties of the networks obtained with the simulated annealing algorithm. We used RRS (Randomization Reserve Selection) software (Rangel et al. Reference Rangel, Pinto, Diniz-Filho and Bini2004) to select the random networks and calculate the HPS in these networks. The area of forest remnants in the counties that composed the network minimizing HPS was obtained from Fundação SOS Mata Atlântica (2009). The results obtained led us to investigate whether the counties which had some presence data had a total HPS greater than expected by chance alone. In order to test this using a null model, we selected 332 random counties from the 3130 counties in 10 000 simulations.
RESULTS
Occurrences of the primate species endemic to the Atlantic forest were distributed along 332 counties from the 3130 counties with an area of Atlantic forest biome composing ≥ 25 of their total area (Fig. 3). The five counties that had the highest number of species were: Belmonte and Una (Bahia state) in the most north-easterly coast region, Linhares and Santa Teresa (Espírito Santo state) in the east, and Teófilo Otoni and Timóteo (Minas Gerais state) in the interior. The species with the greatest number of occurrences was Cebus nigritus (114 counties). Three species occurred in only three counties, namely Callicebus coimbrai, Cebus flavius and Leontopithecus caissara. C. coimbrai is endangered, and C. flavius and L. caissara are critically endangered (IUCN 2010). Other species are also endangered, but had greater numbers of occurrences.
Figure 3 Spatial pattern of endemic primate richness per county in the Atlantic forest biome, in number of species (for state names, see Fig. 1).
There was a positive correlation between HPS and primate species richness in Atlantic forest (r = 0.239, p < 0.001, corrected degrees of freedom = 843.1), indicating possible conservation conflicts. The 100 best networks selected to represent the primate species without human constraint where all counties had the same cost contained nine counties. Only one county, Una (Bahia), occurred in all networks and it contained five primate species (Fig. 4). The other counties can be replaced by one another in different networks.
Figure 4 Irreplaceability pattern obtained using the frequency of occurrence of the counties in the 100 best networks with nine counties from the first approach; all the counties have the same cost (for state names, see Fig. 1).
In the ‘with human constraint’ approach, one minimal network with nine counties representing all primate species was obtained and this contained the minimum possible human population (Fig. 5). The other networks had more counties or a higher total human population. The counties selected were Passo de Camaragibe, Santo Amaro das Brotas, Una, Machacalis, Monte Belo, Santa Teresa, Silva Jardim, Cananéia and Gália (Table 1).
Figure 5 Single network with nine counties selected in the ‘with human constraint’ approach, each county has a cost directly related to its human population size (for state names, see Fig. 1).
Table 1 Name, state, human population size, vegetation cover and strict use reserves in the nine counties selected in the human constraint approach. PN = Parque Nacional (National Park); RB = Reserva Biológica (Biological Reserve); PE = Parque Estadual (State Park); EE = Estação Ecológica (Ecological Station); RVS = Refúgio de Vida Silvestre (Wildlife Refuge). AL = Alagoas; SE = Sergipe; BA = Bahia; MG = Minas Gerais; ES = Espírito Santo; RJ = Rio de Janeiro; SP = São Paulo.
HPS was greater in counties selected in the ‘without human constraint’ approach, as compared to the ‘with human constraint’ approach. The networks selected without human constraint (Fig. 4) had an average ln-HPS = 13.26 (SD = 0.79; maximum = 15.71; minimum = 12.09), and this was not greater than expected by chance alone (p = 0.11). The ln-HPS in the network selected with human constraint was 11.78, which was not smaller than the average of the null distribution (p = 0.27). The average ln-human population in 100 000 random networks was 12.30 (SD = 0.79; maximum = 16.91; minimum = 10.12). A network devised using human population constraint had fewer people compared to the networks selected without human constraint constraint (Fig. 6).
Figure 6 ln-human population in 100 000 random networks with nine counties. ln-human population in the 100 networks selected to conserve primates (the point represents the mean, the right and left limits of the box show the standard deviation, and the right and left limits of the line represent the minimum and maximum). The vertical arrow shows the ln-human population in the network selected to conserve primates while minimizing the human population.
The total ln-HPS in the 332 counties that had some kind of endemic primate species occurrence data was 17.54. Only one of the 10 000 sets of 332 random counties (p < 0.01) had a ln-HPS as high as 17.54.
DISCUSSION
Several studies have defined priority areas for conservation in South America for different groups of organisms (Fjeldså Reference Fjeldså2000; Cavieres et al. Reference Cavieres, Arroyo, Posadas, Marticorena, Matthei, Rodríguez, Squeo and Arancio2002; Thiollay Reference Thiollay2002; Arzamendia & Giraudo Reference Arzamendia and Giraudo2004; Diniz-Filho et al. Reference Diniz-Filho, Bini, Vieira, Souza, Bastos, Brandão and Oliveira2004, Reference Diniz-Filho, Bini, Pinto, Rangel, Carvalho and Bastos2006, Reference Diniz-Filho, Bini, Pinto, Rangel, Carvalho, Vieira and Bastos2007; Tognelli Reference Tognelli2005; O'Dea et al. Reference O'Dea, Araújo and Whittaker2006; Pinto et al. Reference Pinto, Mathias, Blamires, Diniz-Filho and Bini2007, Reference Pinto, Diniz-Filho, Bini, Blamires and Rangel2008) and identified conservation priorities in the entire Neotropical region (Wege & Long Reference Wege and Long1995; Loyola et al. Reference Loyola, Becker, Kubota, Haddad, Fonseca and Lewinsohn2008a, Reference Loyola, Oliveira, Diniz-Filho and Lewinsohnb, Reference Loyola, Kubota, Fonseca and Lewinsohn2009) or part of it (Galván & Vázquez Reference Galván and Vázquez2008). Only one study has used point locality data for primates evaluating the efficiency of the reserve network in 0.25° × 0.25° grid cells (Pinto & Grelle Reference Pinto and Grelle2009). Our study is the first to use counties as the unit of analysis including human population in the reserve selection process in South America, although counties have been the focus of studies elsewhere (Abbitt et al. Reference Abbitt, Scott and Wilcove2000).
About 74% of the species in this study are considered threatened (IUCN 2010). Two of the nine counties selected had <15% vegetation cover, and neither of these contained any strict reserve (Table 1). Of the other counties, five had partial strict use reserve protection. The local knowledge to guarantee that the selected areas have available native areas to be conserved is not available, but the areas most converted and that have the greatest potential for conservation conflict have not entered the reserve networks we identified. Point locality occurrence data are considered a conservative approach because they minimize commission errors when compared to studies that use geographical distributions (O'Dea et al. Reference O'Dea, Araújo and Whittaker2006; Rondinini et al. Reference Rondinini, Wilson, Boitani, Grantham and Possingham2006; Pinto & Grelle Reference Pinto and Grelle2009).
The average HPS in networks found using reserve selection without human constraint was high but no greater than in random networks with the same number of counties. Minimum networks that represented all the endemic primate species had large HPS. The present results also show that it is possible to minimize human conservation conflict in the Atlantic forest. Human population in the single network with human constraint (with costs) was lower than that in the networks selected without constraints, albeit not lower than those in random networks. This also corroborates that the occurrence data that are available derive from highly populated areas. Although there are less densely populated counties, insufficient occurrence data exist for these. Of the 3130 counties in the Atlantic forest, 332 had some kind of endemic primate species occurrence data, and there were more people in the counties for which there were data than expected by chance alone. Surveys may thus be directly related to areas with more people or factors associated with this, such as taxonomists' home ranges or study areas, proximities to work centres and roads (Dennis et al. Reference Dennis, Sparks and Hardy1999; Dennis & Thomas Reference Dennis and Thomas2000, Kadmon et al. Reference Kadmon, Farber and Danin2004). There is insufficient research data for areas with low human populations.
The size of the species' range may also have influenced the results (Pinto et al. Reference Pinto, Mathias, Blamires, Diniz-Filho and Bini2007). There is no way to avoid conservation conflicts for species with very small geographic ranges if these ranges are restricted to areas with high human population density, although this may not be a problem in this study because we used occurrence point data instead of species geographic distributions. Species are artificially restricted when occurrence data are used, but such data assure the species will be represented in the networks selected.
Field surveys are essential to minimize the Wallacean shortfall, a concept associated with low knowledge of species' distributions (Lomolino Reference Lomolino, Lomolino and Heaney2004; Whittaker et al. Reference Whittaker, Araújo, Paul, Ladle, Watson and Willis2005), but conservation planning is an urgent need that cannot wait for the solution. The prioritization of areas for species conservation needs to be implemented using information already available, even if the selection is biased for some taxonomic groups and incomplete. For example, in the present study, the fact that data were biased did not mean they were wrong; rather that primate occurrence points were more frequent in those areas for which the data were biased. These data should be used for conservation planning, but this does not eliminate the need for additional inventories. Updating of strategies must follow the data actualization. Improvements can also be made in reserve selection procedures, including other variables such as vegetation cover and current land costs.
The present study shows that it is possible to minimize the HPS in the reserve selection procedure, but conservation conflicts in the Atlantic Forest will continue given this is the Brazil's most populous region and data availability seems linked to high human population. Our study is an initial exploration of this issue at a broad scale following the framework of conservation biogeography (Whittaker et al. Reference Whittaker, Araújo, Paul, Ladle, Watson and Willis2005) in the Atlantic Forest biome. Such broad-scale approaches can provide overall guidelines for downscaled conservation strategies. Furthermore the maintenance of the few natural areas remaining in this biome is needful.
ACKNOWLEDGEMENTS
Míriam Plaza Pinto was supported by a FAPERJ doctoral scholarship and a CNPq pos-doctoral scholarship, and Carlos Eduardo Viveiros Grelle by a CNPq productivity fellowship. This study was supported by grants from CNPq and FAPERJ (Jovem Cientista do Estado). We thank Fabiano Melo, Helena Bergallo, José Alexandre Diniz-Fillho, Rafael Loyola, Mariana M. Vale and an anonymous reviewer for reading an earlier version of this manuscript.
