A new spin on a compositionalist predictive modelling framework for ...

Biological Conservation 160 (2013) 150–161

Contents lists available at SciVerse ScienceDirect

Biological Conservation journal homepage: www.elsevier.com/locate/biocon

A new spin on a compositionalist predictive modelling framework for conservation planning: A tropical case study in Ecuador Rubén G. Mateo a,⇑, Manuel de la Estrella b, Ángel M. Felicísimo c, Jesús Muñoz a,d,1, Antoine Guisan e,1 a

Real Jardín Botánico (RJB-CSIC), Plaza de Murillo 2, E-28014 Madrid, Spain Departamento de Botánica, Ecología y Fisiología Vegetal, Universidad de Córdoba, 14071 Córdoba, Spain c Escuela Politécnica, Universidad de Extremadura, E-10071 Cáceres, Spain d Universidad Tecnológica Indoamérica, Bolívar 20-35, Ambato, Ecuador e Department of Ecology & Evolution, University of Lausanne, 1015 Lausanne, Switzerland b

a r t i c l e

i n f o

Article history: Received 29 August 2012 Received in revised form 14 January 2013 Accepted 21 January 2013

Keywords: Community-level modelling Richness hotspots strategy Natural history collections Reserve network design Species richness patterns Tropical areas

a b s t r a c t Knowledge about spatial biodiversity patterns is a basic criterion for reserve network design. Although herbarium collections hold large quantities of information, the data are often scattered and cannot supply complete spatial coverage. Alternatively, herbarium data can be used to fit species distribution models and their predictions can be used to provide complete spatial coverage and derive species richness maps. Here, we build on previous effort to propose an improved compositionalist framework for using species distribution models to better inform conservation management. We illustrate the approach with models fitted with six different methods and combined using an ensemble approach for 408 plant species in a tropical and megadiverse country (Ecuador). As a complementary view to the traditional richness hotspots methodology, consisting of a simple stacking of species distribution maps, the compositionalist modelling approach used here combines separate predictions for different pools of species to identify areas of alternative suitability for conservation. Our results show that the compositionalist approach better captures the established protected areas than the traditional richness hotspots strategies and allows the identification of areas in Ecuador that would optimally complement the current protection network. Further studies should aim at refining the approach with more groups and additional species information. Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction A common strategy in conservation planning is the design of reserve networks with the final aim of preserving the most unique and biodiverse areas in situ (Possingham et al., 2006; Margules and Pressey, 2000; Myers et al., 2000; Prendergast et al., 1999). Conservation strategies vary, but the main trouble in biodiversity conservation is not really related to gaps in knowledge or technical complexities, but to budgetary difficulties: conservationist and governments cannot afford assisting all species under threat, especially for lack of funding in the most biodiverse countries (Bruner et al., 2004). Protected-area systems must ideally represent regional biodiversity and provide conditions to separate it from processes that threaten its persistence (Margules and Pressey, 2000). Conservation efforts should be based on biodiversity richness, density of unique species (i.e., endemics) and threat patterns, rather on ⇑ Corresponding author. Present address: Institute of Botany, University of Liège, B22 SartTilman, B-4000 Liège, Belgium. Tel.: +32 497648813. E-mail address: [email protected] (R.G. Mateo). 1 These authors contributed equally to this paper as senior authors. 0006-3207/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.biocon.2013.01.014

percentage of area officially included in protected-area systems (Rodrigues et al., 2004). In recognition of the need for more representative protected areas and the limited resources that government and conservationist can usually allocate to implement them, systematic approaches to conservation planning have been developed in recent years (Moilanen et al., 2009; Margules and Sarkar, 2007; Margules and Pressey, 2000; Pressey et al., 1993). Unfortunately, direct knowledge about the distribution of organisms is usually scarce. When spatial patterns of biodiversity are not accurately known, reserve designers and conservation managers must use indirect measures of biodiversity patterns, such as those derived from aerial photography and remote sensing sources, or other environmental indicators based on climate, topography, geology and soil attributes (Ferrier et al., 2007.; Margules and Pressey, 2000; Wilson et al., 2005). A complementary strategy is to use the data stored in natural history collections (Gaubert et al., 2006; Graham et al., 2004; Loiselle et al., 2003; Newbold, 2010), which are now available through public databases, such as TROPICOS (www.tropicos.org), or distributed search engines, such as the Global Biodiversity Information Facility network (GBIF, http://www.gbif.org/). However,

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

these databases also have disadvantages, such as defective spatial coverage, and spatial and taxonomic bias (Ferrier, 2002). A worldwide networking effort is currently under way to reduce the bias in these global biodiversity data (Bisby, 2000; Guralnick et al., 2007; Soberón and Peterson, 2004), but this effort will require many years before robust data sets of complementary biodiversity data will be available, especially in tropical zones (Cayuela et al., 2009). These sampling issues can lead to underestimate the true species distributions. In other words, an error of omission occurs because a species may inaccurately appear absent from a section of its real distribution because the area has not been sampled (Underwood et al., 2010). This defective spatial coverage of biodiversity data could be overcome, however, using species distribution models. Species distribution models (SDMs; Guisan and Zimmermann, 2000) have become widely used in conservation biology. They allow the expression of habitat suitability as a function of various ecologically meaningful environmental predictors across large spatial coverage. Predictive modelling is a growing discipline and many modelling techniques and approaches have been proposed (Elith and Leathwick, 2009). To cope with this wealth of options, a recent improvement to SDMs has been to ensemble models built with different initial data, modelling techniques or environmental change scenarios into a single, final prediction (Araújo and New, 2007; Grenouillet et al., 2011; Marmion et al., 2009). These improved species predictions – and associated uncertainty – can then be combined to make community predictions (Ferrier and Guisan, 2006). By summarising individual ensemble species distribution models (S-SDMs; Guisan and Rahbek, 2011), one can, for instance, predict potential species richness (Dubuis et al., 2011; Mateo et al., 2012), which can then be used to represent the distribution of richness hotspots in an area (Parviainen et al., 2009) and support the design of reserve networks (Araújo et al., 2004). Predicting richness hotspots is only one of many possible SDM applications in conservation, but other types of predictions at the species, community or landscape levels (e.g., predicted complementarity) are preferred or required to reach specific conservation goals (Whittaker et al., 2005), which may also use SDM predictions as inputs. However, SDMs are not free of errors and uncertainty (Carvalho et al., 2010; Underwood et al., 2010), because: (1) the distribution data on which they are based contain errors (see above); (2) they may not include all environmental, ecological and historical factors that affect species distributions (Guisan and Zimmermann, 2000); (3) there might be uncertainty in the environmental variables used to generate the SDMs, either through measurement errors or as a consequence of the resolution at which variables are mapped and (4) these environmental variables may be partially collinear (as likely here to some extent), potentially hampering full models’ predicting performance. These errors and uncertainties commonly give rise to an overestimation of species distributions (Segurado and Araújo, 2004). These commission errors – false positives – can have an untoward impact on conservation decisions process, because areas where species do not occur might be selected for conservation effort, resulting in both a waste of funds and an unrecognised failure to achieve a conservation target (Rondinini et al., 2006). Although hotspots have been used to refer to those areas rich in endemic species, the term has also been applied to areas of high diversity in species, endangerment and rarity (e.g. García, 2006; Godown and Peterson, 2000; Grand et al., 2004; Ortega-Huerta and Peterson, 2004; Rutledge et al., 2000). We refer here to hotspots as high richness areas, i.e. simple stacking of SDMs (or richness models), as most traditionally practiced in the scientific literature (e.g. García, 2006; McClean et al., 2005; Ortega-Huerta and Peterson, 2004; Pineda and Lobo, 2009; Urbina-Cardona and Flores-Villela, 2010). This traditional richness hotspots

151

methodology is aimed to the conservation of the largest possible number of species in the smallest possible area (Kareiva and Marvier, 2003; Whittaker et al., 2005). However, using richness hotspots to set priorities is questionable when considering a broader range of objectives, such as preserving unique components of biodiversity (rarity, endemism), maintaining functioning ecosystems throughout the world or providing the greatest variety of distinct plant and animal lineages for future evolutionary breakthroughs (Kareiva and Marvier, 2003; Orme et al., 2005). For instance, some authors have already argued for the importance of studying biodiversity ‘‘coldspots’’ within the design of conservation networks (Bøhn and Amundsen, 2004; Kareiva and Marvier, 2003; Price, 2002). One such approach is the compositionalist approach (Callicott et al., 1999; Whittaker et al., 2005; Williams and Araújo, 2000), which considers different pools of species separately (similar biogeographic regions or level of endemism here, see material and methods). Each pool consists of species sharing characteristics that differ from other pools, and represents a different biodiversity value, eventually including the importance of each of these groups in a final map. In the richness hotspots strategy, the importance of these groups is masked by the total set of species. Therefore, the compositionalist approach potentially represents a complementary view to the richness hotspots approach at regional scale, i.e. a spatially more informative approach than for example taking into account information of different biogeographical areas or rates of endemism. For example, despite the extreme conditions of the Ecuadorian páramo, over 1500 species of vascular plants are estimated to be present (León-Yánez, 2000). This figure means that 10% of the Ecuadorian flora is present in 5% of the territory (Mena-Vásconez et al., 2001). Much endemism is restricted to Páramo formations due to their unique environmental conditions (Young et al., 2002). The richness hotspots approach would presumably predict maximum biodiversity richness in the Amazon basin, the two Andean slopes, and the Chocó biogeographic region. Therefore, this approach is unlikely to identify the unique plant formations from Páramo and will be unable to include information on endemic species. An approach allowing the consideration of individual species or pools of species is required for these tasks. The compositionalist approach, by allowing the representation of compositions of particular species and pools of species, should identify Páramos as optimal areas for conservation. Both the traditional richness hotspots and the compositionalist approaches require comprehensive species distribution maps, but these are rarely available in a spatially-explicit way, especially in tropical countries (Kareiva and Marvier, 2003). One possible way to overcome this issue is to use model predictions as a surrogate for real distribution data, but there are still few published examples of such model-based approaches that consider biogeographic specificities and endemism to design conservation areas. Here, we propose such a complementary compositionalist approach based on species distribution models (SDMs) to identify areas of maximised suitability for conservation (AMSC). We illustrate the approach with plants in Ecuador, with focus on the Coastal, Andean, Páramo, Amazonia regions, and endemic species. We take advantage of the most comprehensive plant database compiled so far for Ecuador, with 408 species from herbarium collections. We use advanced ensemble SDMs based on six modelling techniques for each individual species and subsequently stack them (according to Guisan and Rahbek, 2011) by region and endemism to generate compositionalist maps of species richness. Finally, we evaluate the conservation status of plant diversity in Ecuador using the potential richness maps derived from the simple stacking of all species predictions and the new compositionalist approach considering stacked groups of species. Ultimately, we show how

152

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

conservation recommendations can be drawn from the latter in the perspective of optimising the design of new protected areas. 2. Material and methods 2.1. Study area We chose Ecuador as study area for six reasons: (1) it is one of the most biologically diverse countries in the world (Jørgensen et al., 1992), and should therefore receive particular conservation attention. It host c. 2400 vertebrates and more than 16000 vascular plant species (Jørgensen and León-Yánez, 1999). Other authors (Balslev and Renner, 1989) estimated that the flora of Ecuador counts c. 20000 vascular plant species, of which more than 4000 are endemic to the country (Valencia et al., 2000). (2) The current tendency of global loss of biodiversity (Chapin et al., 2000; Hansen et al., 2010; Pimm et al., 1995) is also reflected in conservation problems within Ecuador, especially affecting the South-western lowlands (Best and Kessler, 1995), by a large number of factors, including agriculture, deforestation, mining, and climate change (Best and Kessler, 1995; Dodson and Gentry, 1991; Mecham, 2001; Ulloa Ulloa and Jørgensen, 2004), making the need for new tools for large-scale assessment particularly urgent. At the same time, Ecuador has no spatially explicit coverage of biodiversity data, thus rendering nature management and conservation decisions more difficult than in countries with better data coverage (for example European and North American countries). This combination of extremely high biodiversity richness, need of appropriate protection, and lack of biodiversity data to support conservation planning makes Ecuador an excellent area for testing new modelbased conservation approaches. (3) Few distribution modelling studies have been developed in tropical areas (Cayuela et al., 2009), although these areas often have a high biodiversity and are therefore in greater need of model-based conservation decision tools than temperate regions. (4) Ecuador has some of the best known plant diversity of all tropical countries (Loiselle et al., 2008), thus enabling ecological studies to build on previous taxonomic knowledge; (5) the vast ecological and orographic diversity of the country (Skov and Borchsenius, 1997) makes it an ideal study area for developing species distribution models. (6) Ecuador is a tropical country with thoroughly studied biodiversity and an extensive network of natural reserves (MAE, 2011) covering the ecosystems in the country fairly well (Lessmann, 2011; Sierra et al., 2002). We used the National System of Protected Areas (NSPA) of Ecuador as a reference to determine whether our new compositionalist approach is able to retrieve the areas included in the NSPA better than the commonly used richness hotspots approach. We are also interested in exploring which one of the two methods can improve the current NSPA by including underrepresented biogeographic elements or endemics and found that the compositionalist map does a better job. Considering scenarios of changing climate or land use was not within the scope of this study.

senting a fair representation of the proportions of plant habits in the Ecuadorian flora (cf. http://www.tropicos.org/ProjectAdvSearch.aspx?projectid=2); (3) they include a large proportion of species native to Ecuador; (4) they represent a complete environmental and geographical coverage of Ecuador (Fig. 1). We used the available records stored in TROPICOS database (Missouri Botanical Garden). Specialists from both the Missouri Botanical Garden (Saint Louis, USA) and Real Jardín Botánico (Madrid, Spain) checked all of the records, verifying the data location (latitude/longitude). When possible, all errors were corrected; otherwise, the record was removed. We then performed an exploratory data analysis and a multinomial linear regression on a dummy variable to detect outliers based on Mahalanobis distances with respect to the centroid of the data group, contrasted with a chisquare distribution (Filzmoser et al., 2005). Although outliers may represent sink populations or high dispersal capabilities (Pulliam, 2000), they are also indicative of georeferentiation or taxonomic identification mistakes. All potential outliers were analysed individually, and those not offering enough taxonomic and geographic reliability were discarded. Included collections were overlaid on a 0.0083° (1 km2 at the Equator) resolution grid. Where multiple collections of the same species occurred in the same pixel, a single presence was retained. Only species with at least 15 unique presences were modelled to avoid generating low-performance models (Mateo et al., 2010b; Papesß and Gaubert, 2007; Wisz et al., 2008). The final database included 17689 records for 408 species (Fig. 1). Each of the 408 species was first classified according to biogeographic area preference following an established classification of species per biogeographic area based on relevant literature for Ecuador (Jørgensen and León-Yánez, 1999). We recognised four major biogeographic areas to group species: (1) the coastal area (Coastal), (2) the Andean area (Andean), (3) the Amazonian (Amazonian), and 4) above 3000 m a.s.l., in the Páramo eco-region (Páramo). Of the 408 taxa included, 48 are endemic to Ecuador. Many other way to combine species for conservation prioritization exist, but it was not our aim to test them all, but to show the potential of species distribution models used in more refined way than a simple stacking of all species.

2.2. Species data We used 408 plant taxa for our study, comprising 110 genera and 19 families. Plants are the basic support of terrestrial ecosystems and dramatically influence the status of all other taxonomic groups, and the specific taxa included in this study were selected based on the following characteristics: (1) they have been extensively reviewed by specialists, guaranteeing reliable identifications; (2) they reflect a wide range of floristic relationships and include a wide range of life forms (118 epiphytes, 108 herbs, 23 lianas or vines, 1 parasite, 60 shrubs and subshrubs, and 98 trees, although strict assignment is sometimes difficult), roughly repre-

Fig. 1. Localisation of the collections used in this study.

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

2.3. Environmental predictors All nineteen bioclimatic variables from Worldclim 1.3 (Hijmans et al., 2005) were selected as potential predictor variables (1 km  1 km cell side). These bioclimatic variables are calculated using the monthly mean temperature and rainfall and using the digital elevation model as a covariate. These are: annual mean temperature, mean diurnal range, isothermality, temperature seasonality, maximum temperature of warmest month, minimum temperature of coldest month, annual temperature range, mean temperature of wettest quarter, mean temperature of driest quarter, mean temperature of warmest quarter, mean temperature of coldest quarter, annual precipitation, precipitation of wettest month, precipitation of driest month, precipitation seasonality, precipitation of wettest quarter, precipitation of driest quarter, precipitation of warmest quarter, and precipitation of coldest quarter. These variables represent environmental factors of potential biological relevance (Austin, 2007). As there were no a priori ecophysiological reason for selecting a subset of climatic variables as potential predictors of all 408 species, and assuming that managers would even less have this selection capacity, we therefore kept all variables in the analyses, therefore making it more straightforward for managers. Most of the modelling techniques used allow some variables selection, but not all, so we cannot exclude that some overfitting may still have occurred for species with few occurrences. Yet this should

153

result at worse in more conservative predictions for designing conservation areas, because overfitted models will lead to geographically more restricted predictions around the known localities of the species. 2.4. Modelling techniques We used six modelling techniques (Fig. 2) to represent the different options available for species distribution modelling: regression analysis, environmental envelopes, genetic algorithms, classification techniques, and maximum entropy. Boosted regression trees (BRT), logistic multiple regression (LMR), and multivariate adaptive regression splines (MARS) are discriminative techniques that need presence and absence data. Gower’s metric distance (GMD), and maximum entropy (MAXENT) are profile techniques that use presence-only data. Finally, genetic algorithm for rule-set prediction (GARP) is a mixed technique that uses a combination of profile and discriminative techniques. Herbarium collections provide us with presence-only data, which a priori prevent the use of discriminative techniques that require presences and absences. To overcome this issue, we generated pseudo-absences with the extension ‘‘Random Point Generator 1.28’’ (ArcView 3.2) and following the constraints in Mateo et al. (2010a): we generated approximately the same number of pseudo-absences as presences to avoid problems associated with unbalanced prevalence; (2) to collect information on the different

Fig. 2. Workflow of this study.

154

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

ecological conditions in the study area, we defined a minimum distance of 30 km between pseudo-absences; (3) to avoid increasing the false negative rate, we defined a buffer (30 km in diameter) around each presence from which pseudo-absences were eliminated. The distance of 30 km was calculated based on entropy measures (see Mateo et al., 2010a). In the case of MAXENT, models were generated with the default settings, i.e. MAXENT effectively generated its own pseudo-absences. As pseudo-absences were generated fully randomly in both cases, we do not think that it can significantly affect the results, especially since models were averaged and comparing the different techniques was not our aim. Elith et al. (2006) provides the most comprehensive comparison of the different available techniques and software used in species distribution modelling. We generated MARS models using MARS 2.0 (www.salford-systems.com), running 30 models per species and varying the following parameters: (1) the maximum number of basic functions; (2) whether or not interactions were allowed between basic functions; and (3) inclusion of all of the 19 Worldclim variables or elimination of the mean annual temperature and mean annual precipitation to reduce multicollinearity. We used MARS 2.0 instead of the mda library (Elith et al., 2006) related to the findings in previous works (Muñoz and Felicísimo, 2004; Mateo et al., 2010a). The LRM models were generated using SPSS 14.0 with forward conditional stepwise LMR. The MAXENT models were generated in MAXENT 3.1 with the default settings (‘‘Auto features’’, auto-generation of background points, convergence = 10–5, maximum number of iterations = 500, regularisation value b = 10–4). BRUTO (BRT) models were generated in R (www.rproject.org) with the ‘‘mda’’ package and the parameters detailed in Elith et al. (2006). The Gower distance models (GMD) were performed in DIVA-GIS (Hijmans and Guarino, 2012). GARP Desktop 1.1.6 was used to generate 100 models per species; we selected the bottom 20% of models with the lowest extrinsic omission error, and of those, the ten models around the median of the commission error were used to generate the GARP ensemble model presented in the results section. All techniques except Gower and GARP included an internal variables reduction approach.

2.6. Ensemble models In recent years, ensemble approaches have been increasingly applied to ecological modelling to predict the spatial distributions of single species (e.g. Broennimann et al., 2007). It combines (often through averaging) predictions from several modelling techniques to avoid the need to make a subjective choice for a single technique (Segurado and Araújo, 2004; Austin, 2007) and to provide informative uncertainty estimates (Araújo and New, 2007; Marmion et al., 2009). In addition to improving the prediction accuracy of individual species, ensemble modelling approaches also allow deriving more robust biodiversity predictions when stacking multiple species distributions (Dubuis et al., 2011; Mateo et al., 2012). All predictions from the different models were standardised by rescaling them to the 0–1 scale using the equation:

Rescaled Value ¼ ðCell Value  MinimumÞ=ðMaximum  MinimumÞ and the ensemble prediction for each species was generated by averaging the individual rescaled predictions obtained with the six modelling techniques (Araújo and New, 2007; Marmion et al., 2009; Mateo et al., 2012). 2.7. Binary models The ensemble model was reclassified into maps of presence/absence (binary maps) using a threshold approach. Different approaches have been proposed for the threshold selection (Jiménez-Valverde and Lobo, 2007; Liu et al., 2005). Here, due to the presence-only data available, we used a threshold approach minimising the commission error (Fielding and Bell, 1997). The commission error is the error to predict the presence of a species when it has not been actually observed, i.e. the proportion of false positives. Here, we allowed a maximum commission error of 0.05. We converted original models to binary models using Arc Macro Language and ArcInfo 9.3 Workstation software. 2.8. Deriving areas of maximised suitability for conservation (AMSC)

2.5. Assessing predictive performance The predictive performance of SDMs should ideally be evaluated using a set of independent data. In our study, this procedure was not possible for most species due to the scant number of available collections that require using all available data to fit the model: 94% of the species stored in the TROPICOS data base for Ecuador show less than 20 occurrences and only 1% have more than 100 (Loiselle et al., 2008). This problem is common in studies carried out in tropical zones (Raven and Wilson, 1992; Feeley and Silman, 2011). Therefore, we assessed the predictive performance of individual SDMs using resubstitution, i.e., calculating ROC plots and the AUC statistic (Fielding and Bell, 1997) on the same data used to fit the model. This approach is thus nothing else than a different measure of model fit, still informative when no independent data can be left out for evaluation (Araújo and Guisan, 2006). Additionally, we used independent data for model validation in the case of species of the genus Anthurium. We created this independent dataset for validation using as presences the sites where collectors other than the specialist of the genus had collected the species being modelled and, as absences, randomly selected target-group absences (Mateo et al., 2010a). Finally, we evaluated the biological meaningfulness of the models by comparing the different potential richness maps using a new methodology based on testing the ability of stacked SDMs (S-SDM; Guisan and Rahbek, 2011) to reconstruct observed altitudinal patterns of biodiversity (Mateo et al., 2012).

We employed two SDM-based strategies (Fig. 2) to derive ‘areas of maximised suitability for conservation’ (AMSC) in Ecuador: (1) the traditional richness hotspots approach of simply stacking species binary predictions (H-SDM), and (2) the new compositionalist approach (C-SDM), as follows. Hotspots map (H-SDM), resulting from summing all individual binary SDMs (Guisan and Rahbek, 2011) as traditionally practiced (García, 2006; McClean et al., 2005; Ortega-Huerta and Peterson, 2004; Pineda and Lobo, 2009; Urbina-Cardona and Flores-Villela, 2010). It has been criticised by other authors for several reasons (Whittaker et al., 2005), including that it suggests that biodiverse areas excluded from the defined hotspots are irrelevant in conservation (Bates and Demos, 2001). Following this traditional approach, we obtained a species richness model for Ecuador (H-SDM map). We then selected the top class (upper quartile) with higher species richness as the ‘area of maximised suitability for conservation’ (AMSC). The compositionalist-map (C-SDMs) still considers H-SDMs, but it is based on a compositionalist approach (Whittaker et al., 2005) that considers the Coastal, Andean, Páramo, Amazonia, and Endemic species pools separately. To implement this approach, we (1) generated richness maps for each group, (2) reclassified each of these maps into five classes (based on equal intervals), (3) selected the top class (upper quartile) with higher species richness per map, and (4) summed the five upper quartile maps (Coastal, Andean, Páramo, Amazonia, and Endemic species) to obtain the predicted

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

AMSC (Fig. 2). This final map represents the highest areas of biodiversity based on trends for each pools of species (biogeographic region and the endemic species richness map in this case), ultimately providing a more informed representation of conservation priorities. In contrast to the hotspot approach, the compositionalist map considered the different pools of species separately, giving the relative importance of these groups in the final map. In the hotspot map, the importance of these groups is masked by attributing the same weight to all species in the set. 2.9. Vegetation and protected areas The resulting hotspot map (richness hotspots) and compositionalist-map (biogeographic approach) were compared to the network of National System of Protected Areas (MAE, 2011) and a vegetation map of Ecuador (Sierra et al., 1999). Thus it could be analysed which of the two strategies define better the network of established protected areas, and also to investigate the possibility of recommending the creation of new protected areas. 2.10. Final map for recommendation of new protected areas We assumed that areas with high degradation (anthropization and environmental degradation, like large cities) should be excluded from the network of protected areas, although this is something that ultimately must be carefully evaluated in the decision-

155

making process. We indicated such areas so overlaying the Compositionalist-map (C-SDMs) with the ‘‘Human Footprint’’ created by Sanderson et al. (2009), that expresses the relative human influence in every biome as a percentage. 3. Results 3.1. Reliability of species distribution models The methods that obtain the best predictive performance with both AUC analyses performed (resubstitution and independent evaluation for the genus Anthurium) were BRT, MARS, MAXENT and LMR (Table S1, online Supplementary material). All species distribution models were highly accurate (Mateo et al., 2010a) in regard to AUC resubstitution values (all > 0.95) and fairly accurate in regard to AUC independent evaluation for the genus Anthurium (all > 0.73). Additionally, stacking of the individual species models generated richness models that proved to be well correlated with the observed alpha diversity richness patterns along elevation, and also with the gamma diversity derived from the literature (Mateo et al., 2012). 3.2. Differences between compositionalist and hotspot approaches The compositionalist approach yielded predictions for the Coastal, Andean, Páramo, Amazonia, and Endemic species pools

Fig. 3. Potential species richness maps for 408 species included in this study: Amazonia (a), Andean (b), Coastal (c), Endemic (d) Páramo (e), and total (f) species in Ecuador. Geographical projection used: World Geodetic System 1984.

156

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

Fig. 4. Comparison between the National System of Protected Areas (NSPA) and Areas of Maximised Suitability for Conservation (AMSC) in Ecuador predicted by two different approaches: (a) considering separate Coastal, Andean, Páramo, Amazonia and Endemic species and applying a compositionalist approach, and (b) stacking (addition) all individual binary SDMs, in which we selected the top class (upper quartile) with higher species richness as the biodiversity hotspots in Ecuador (hotspot map). The hotspot map predicted 3436 km2 (7.39%) of protected areas that currently exist in Ecuador as AMSC, while the compositionalist map predicted 9666 km2 (20.79%).

separately (Fig. 3). We found areas of maximum suitability for conservation (AMSC) maps from the two approaches: compositionalist and richness hotspots. The traditional richness hotspot approach predicted 38567 km2 (13.5% of the area of Ecuador), while the compositionalist approach predicted 61408 km2 (21.37%). Differences between the predictions by these two approaches and the protected areas network in Ecuador are presented in Fig. 4. It shows that the compositionalist map (Fig. 4a) better captures the zones currently included within existing protected areas. The traditional richness hotspot approach (Fig. 4b) only predicted 3436 km2 (7.39%) of the official protected areas network as AMSC, while the compositionalist richness approach predicted 9666 km2 (20.79%). Table 1 shows how much of the total area of each vegetation type in Ecuador is included in the protected areas network and how much is predicted as highly suitable according to the two strategies. The compositionalist richness map predicted 13 vegetation types that are not predicted using the simple stacking richness hotspots approach, all of which have high ecological value, e.g., different Páramo or high Andean forests, that should be considered in conservation programs. For the purposes of conservation planning and assuming that the current network is optimal, the compositionalist richness approach renders more robust results, which are significantly different from the richness hotspots methodology (t = 2.756; P = 0.008). 3.3. Determining new priorities for conservation areas Areas that could be subject to conservation measures (high potential richness values and low Human Footprint values) are the Chocó remnants in the Northwest and the Cutucú and Condor ranges in the Southeast (Fig. 5). Table 2 shows that the compositionalist approach predicts larger areas of high conservation interest (AMSC) than the hotspot approach does, in addition to 12 vegetation formations not appearing in the latter. 4. Discussion 4.1. Compositionalist approach on conservation planning Complete knowledge of the biodiversity distribution for use in conservation planning is not available in most countries, especially

in the tropics. It is therefore important to develop appropriate methods to extend the limited information available to unknown or poorly explored areas. Among extrapolation approaches, the most used so far has been to generate species richness predictions and to base conservation strategies on richness hotspots derived from stacking individual models (Dubuis et al., 2011; Guisan and Rahbek, 2011), which we call the richness hotspots strategy. In the present study, we propose a new way to generate diversity richness information (Estrella et al., 2012). Our proposal is to consider not only the sum of individual species, but also information on the biogeographic patterns, endemicity, or rarity, that are crucial to propose conservation strategies, which is a more complex procedure. This compositionalist approach does not require a different input data or a greater investment than the richness hotspots strategy, but a careful analysis of all the information that the initial dataset at hand can provide. As we show, the compositionalist approach reproduces more accurately the NSPA areas than the hotspot strategy (7.39% vs. 20.79%, Fig. 4), and it is able to detect 13 vegetal formations that would not be included in the reserve solution by the richness hotspots strategy (Table 1), suggesting new conservation priorities for Ecuador (Table 2 and Fig. 5). The richness hotspots strategy is also flawed in overestimating the contribution of Amazonian species (Fig. 3a, 240 species), most likely because it is the richest area in Ecuador in term of species. Although it is a highly diverse area, many of the species growing there appear in other parts of the Amazonian basin. This explains why the richest area in endemic species in Ecuador is not captured by the richness hotspots strategy (Fig. 4b) but is better considered in our compositionalist approach (Figs. 3d and 4a). We recognise that our methodology has limitations and that some of the assumptions made in the study need to be tested more broadly in future works. In this regard, our study was exploratory and only illustrative of a new approach, not aimed to provide a final conservation solution. One limitation is that we conducted this study using ‘‘only’’ 408 plant species models grouped in ‘‘only’’ five pools (with each a richness map), although our results show that the compositionalist approach covers more surface of the network of protected areas, established with a full, larger range of data (plants, animals, etc.). Thus, in future studies, it would be interesting to test the effects of using different organisms, species pools and environmental variables in the compositionalist approach. Another limitation is the use of the current network of protected areas to evaluate the effectiveness of the compositionalist

157

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

Table 1 Vegetation formations in Ecuador according to Sierra et al. (1999), with indication of the total area in hectares for the country, percentage of the total area included in the National System of Protected Areas (NSPA), and the percentage of the total area of both approaches tested in this study included in the NSPA. The compositionalist map shows that there are 13 vegetation formations (bold) that would not be protected using the hotspot approach. One formation (Amazon cordillera montane humid shrub, underlined italic), currently not included in the NSPA, appears in both the compositionalist and hotspot approaches. Vegetation formations (Sierra et al., 1999) Amazon cordillera lower montane evergreen forests Amazon cordillera montane evergreen forests Amazon cordillera montane humid shrub Amazon cordillera upper montane humid shrub Amazon foothill evergreen forests Amazon lowland black-water inundated evergreen and palm forests Amazon lowland evergreen forests Amazon lowland lake graminoid vegetation Amazon lowland white-water inundated evergreen forests Coastal cordillera foothill evergreen forests Coastal cordillera lower montane evergreen forests Coastal cordillera lower montane evergreen forests Coastal cordillera montane cloud forests Coastal foothill deciduous forests Coastal foothill semideciduous forests Coastal lowland graminoid vegetation Coastal lowland deciduous forests Coastal lowland dry shrub Coastal lowland evergreen forests Coastal lowland inundated forests (guandal) Coastal lowland semideciduous forests Cushion páramo Dry páramo Dwarf mangroves Eastern Andes montane cloud forests Eastern Andes upper montane evergreen forests Espeletiapáramo Herbaceous páramo Mangroves Montane-lake graminoid vegetation North-eastern Andes lower montane evergreen forests Northern Andes montane dry shrub Northern Andes montane dry shrub Northern Andes montane humid shrub Savannah South-eastern Andes lower montane evergreen forests Southern Andes montane dry shrub Southern Andes montane humid shrub Southern Andes shrub páramo Super páramo Upper montane lake graminoid vegetation Western Andes lower montane evergreen forests Western Andes lower montane semideciduous forests Western Andes montane cloud forests Western Andes upper montane evergreen forests

Total (ha) in Ecuador 511,778 128,066 11243.56 8317 1,313,192 685,281 6,174,672 5919 492,734 400,528 1,130,277 15,561 61,689 62,831 453,352 24,229 1,234,265 307,551 3,140,291 2322 636,695 114,326 181,533 8229 894,631 928,644 53,824 1,162,081 265,864 842 328,481 212,830 26,465 480,485 231,641 346,747 295,165 112,636 52,721 7343 3093 539,707 187,219 940,418 601,551

approach. Using the current network cannot prove the superiority of an approach because, depending on how it was defined, it could be biased or inefficient at protecting biodiversity. What one can do instead is evaluate how different conservation prioritization approaches – e.g. based on SDMs – actually reflect the existing network, i.e. if they are rather supporting it or not from the perspective of the criteria used (here total and group species richness). Different reserve design approaches may produce different sets of solutions, each of which must be discussed with and among stakeholders, and does not constitute per se a final solution. In our study, we show that the compositionalist approach suggests new areas to be potentially protected, because they contain vegetation types that are not included in the present design (e.g. páramo formations, see Table 1). In this regard, we showed that in tropical countries, where the lack of adequate data is a recurrent limitation for taking informed conservation decisions, the compositionalist approach could represent a useful alternative or complementary strategy to the richness hotspots approach. Considering species’ biological requirements and distinct biogeographical components separately provides more information to support biodiversity prioritizations.

National System of Protected Areas (NSPA) (%)

Hotspot map (%) within NSPA

Compositionalist map (%) within NSPA

24.72 62.25 0.00

46.12 3.05

59.34 13.76

41.03 0.00 33.70 3.12 37.71 7.30 19.93 0.00 4.14 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.33 0.00 0.00 0.00 0.00 0.00 30.34 0.00 0.00 0.00 0.00 1.58 0.00 0.00 0.00 0.00 0.00 0.29 0.00 0.04 0.00

78.63 0.00 23.41 6.00 43.72 9.75 25.03 0.01 31.83 3.28 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 53.83 18.95 0.00 6.45 5.44 66.64 36.05 0.00 0.00 41.15 0.00 0.00 2.11 0.00 9.37 0.00 0.19 0.00 0.07 29.26 41.06 0.00 25.72 10.48

90.98 7.42 54.02 19.13 80.64 18.15 13.22 1.09 71.49 9.53 0.00 0.00 3.50 0.58 0.83 2.91 59.04 1.23 71.32 30.04 81.40 40.54 37.60 50.80 28.37 19.77 1.90 35.02 0.00 0.00 0.40 1.90 7.05 0.00 4.91 29.54 78.28 90.85 12.81 0.00 13.05 5.51

Quoting Whittaker et al. (2005): ‘‘the hotspot methodology is logical only if the exclusive goal of conservation is to protect the largest possible number of species in the smallest possible area’’. When conservation efforts differ from this objective, such as protecting the highest number of forest ecosystems, diversity of taxonomic groups, number of endemic species, or the highest number of endangered species, then the hotspot strategy could be complemented with a compositionalist approach that includes all of the components of biodiversity that we would like to preserve. The richness hotspots approach can help prioritize between suitable areas from the compositionalist approach by identifying those that are the richest in absolute. Our example of the compositionalist approach only included distinct biogeographic regions and endemic species, but other components could be added, such as endangered species or biotopes/ecosystems of high conservation value. However, it still proved superior in capturing the current conservation network. The richness hotspots approach tends to over-represent forest formations sheltering the areas of highest diversity in Ecuador, such as the Amazon basin, the two Andean slopes and the adjacent area of the biogeographic Chocó, missing other formations of great conservation interest, such as the Coast and Páramo

158

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

Fig. 5. Human Footprint within the AMSC predicted by the compositionalist map. This analysis attempted to measure the degree of anthropisation and thus the environmental degradation. These maps are useful for making recommendations on possible new protected areas (green and yellow colours) based on AMSC while prioritising less anthropomorphic areas. 1. Chocó. 2. Cóndor range. 3. Cutucú range. 4. Sumaco Napo-Galeras National Park. 5. Catacachi-Cayapas Ecological reserve. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

the compositionalist approach, which includes all of the components of biodiversity that we would like to preserve. In this case, a compositionalist strategy could be a good option for managers, identifying complementary spaces for conservation. There is no single best surrogate in the design of reserves, such as overall species richness or species richness of one indicator group, that indicate the best areas to conserve biodiversity in general. Accounting for various compositional aspects was shown to be important and should be practicable in most situations (Margules and Pressey, 2000). The compositionalist approach that we illustrated here is promising to develop new conservation priorities, such as looking for areas of maximised suitability for conservation (AMSC), and we therefore encourage its use alone or in complement to the commonly used richness hotspots or other approaches (e.g. complementarity). We are aware that our work was limited by the exclusive use of well-studied taxonomic groups, but it is promising that this first implementation of the compositionalist approach proved capable of uncovering the actual protection network despite this limitation. It would be interesting to continue developing this methodology using different indicator groups (such as liverworts, mosses, fungi, and animals) in other biogeographic areas or to include other variables in the analysis (e.g., socioeconomic variables). 4.2. Principal applications for conservation in Ecuador

regions (Figs. 3 and 4); also importantly, the richness hotspots approach cannot take endemicity into account. In summary, the hotspot strategy could be appropriate to recognise the most diverse areas that should receive attention and are important from a conservation point of view worldwide in terms of their biodiversity (Myers et al., 2000). When conservation efforts differ from this objective, such as protecting the highest number of forest ecosystems, diversity of taxonomic groups, number of endemic species, or the highest number of endangered species, then the hotspot strategy is likely to fail, and it should be complemented by

Recent studies revealed that Ecuador, as other tropical countries, is suffering a severe degradation of its natural ecosystems (Sáenz and Onofa, 2005), and one of the greatest deforestation rates in the Neotropics (Cuesta-Camacho et al., 2006), with more than 2000 species under threat (IUCN, 2011). In the comparison between protected areas and the Human Footprint (Fig. 5), we showed that the Ecuadorian National System of Protected Areas (NSPA) has effectively kept land transformation from trespassing within limits of protected areas. Currently, NSPA includes remnants of vegetation, far from towns and roads (low

Table 2 Efficiency of hotspot and compositionalist approaches in recognising declared reserves in Ecuador. The Areas of Maximised Suitability for Conservation (AMSC) extension, in hectares, outside the current National System of Protected Areas (NSPA) are indicated for each vegetation formation in Sierra et al. (1999). The compositionalist map includes 12 vegetation formations (bold) that do not appear as priorities using the hotspot approach. Vegetation formations (Sierra et al., 1999)

Hotspot map Area outside NSPA (ha)

Compositionalist map Area outside NSPA (ha)

Amazon cordillera lower montane evergreen forests Amazon cordillera montane evergreen forests Amazon cordillera montane humid shrub Amazon foothill evergreen forests Amazon lowland black-water inundated evergreen and palm forests Amazon lowland evergreen forests Amazon lowland lake graminoid vegetation Amazon lowland white-water inundated evergreen forests Coastal cordillera lower montane evergreen forests Coastal foothill evergreen forests Coastal lowland evergreen forests Cushion páramo Dry páramo Eastern Andes upper montane evergreen forests Espeletiapáramo Herbaceous páramo Northeastern Andes lower montane evergreen forests Northern Andes montane dry shrub Northern Andes montane humid shrub Southeastern Andes lower montane evergreen forests Super páramo Upper montane lake graminoid vegetation Western Andes lower montane evergreen forests Western Andes montane cloud forests Western Andes montane cloud forests Western Andes upper montane evergreen forests

187,773 3811 4613 392,009 16,506 2,175,055 101 84,156 0 45,678 0 0 0 0 0 0 74,143 0 0 5465 0 0 1273 105 6129 0

233,249 14,107 8841 275,798 24,616 2,490,684 242 104,864 59 285,226 153 24,626 22,167 27,424 13,871 293,976 83,889 218 10,129 31,231 5 165 169,563 191,346 32,845 57,695

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

Human Footprint values, green and yellow colours in Fig. 5), subject to less immediate threats than neighbouring areas that have usually been heavily transformed by humans (high Human Footprint values, orange and red colours in Fig. 5), such as the Cotacachi-Cayapas Ecological Reserve, the only large preserve in the Chocó area, or the Sumaco Napo-Galeras National Park (Fig. 5). These two areas harbour intact vegetation patches in a matrix of strongly modified landscapes. These modified landscapes have high potential richness values (Fig. 5) so that, in the past, these areas were intact vegetation zones, connecting present protected areas with each other. Further areas that should be subject to conservation measures according to our analyses are the Chocó remnants in the Northwest and the Cutucú and Condor ranges in the Southeast (Fig. 5). The humid forest of the tropical Andes region is one of the most diverse areas in the world (Fjeldsa et al., 2005; Rahbek and Graves, 2001) and thus a top priority for conservation (Mittermeier et al., 1998; Olson and Dinerstien, 1998). Our results show that the most diverse area of Ecuador lies in the Chocó bioregion (NW Ecuador). This area has long been recognised as a biodiversity hotspot (Myers, 1988), harbouring the richest bird diversity in the world (Terborgh and Winter, 1982) and the largest plant diversity of occidental Ecuador (Gentry, 1982, 1986). The next highest areas in diversity richness are both Andean slopes – a global conservation priority (Mittermeier et al., 1998; Olson and Dinerstien, 1998) – and the Amazon basin. Although 20% of the country is under some type of protection, not all of the biogeographic regions are represented in the NSPA. Specifically, the Chocó bioregion is the worst represented, and Coastal, South Andean and South Amazonian formations are also poorly represented (Lessmann, 2011; Sierra et al., 2002; Cuesta-Camacho et al., 2006). The compositionalist approach is able to find most of these under-represented areas in the Chocó or the Andean slopes, and also the Cutucú and Condor ranges in the Southeast (green and yellow areas in Fig. 5), demonstrating its advantage over the richness hotspots approach, which totally fails for the Chocó, and severely for the Andean slopes. According to the Human Footprint (Sanderson et al., 2009), the most threatened zones inside AMSC correspond to the areas where oil industry is prominent (Fig. 5), although not all conservation problems come directly from oil extraction. The opening of roads and the free transportation to the local indigenous population has increased the market of bush meat (Suárez et al., 2009), which must be currently considered as a severe threatens to Yasuní extraordinary diversity (Bass et al., 2010). One formation (Amazon cordillera montane humid shrub, underlined italic in Table 1), currently not included in the NSPA, appears in both the compositionalist and hotspot approaches like a formation that should be protected. Table 2 shows that the compositionalist approach predicts larger areas of high conservation interest (AMSC) than the richness hotspots approach does, in addition to 12 vegetation formations (two of them typical vegetation of the Coast, five for the Páramos and 5 of the Andes). It is important to mention that formations with fewer taxa will be underestimated by the latter in conservation programs, irrespective of the importance, endemicity or rarity of the taxa involved. Formations with fewer species have higher chances of summing up less species in the richness hotspots approach than the compositionalist approach, where summations are done separately for each formation. For example, Páramo is poorer in absolute number of species than formations in the Andean slopes. It definitively will sum less species in the richness hotspots approach than formations in both slopes, and its conservation interest would be underestimated, even although in absolute terms, especially in the context of global change, such interest could be greater (Young et al., 2002). Other formations that would be underestimated are, no doubt, the dry coastal formations.

159

Our study delivers two important results regarding Ecuador conservation policy. First, the compositionalist approach proved more efficient in reproducing the spatial pattern of established protected areas. That outcomes show that following the richness hotspots approach in reserve design will overlook regionally important areas of diversity and that the compositionalist-map does a better job of protecting this diversity than would be accounted for in the richness hotspots approach. Second, whichever method is used, some important areas of potential conservation interest are currently not included in any wildlife reserve in Ecuador, which is in accordance with other recent studies (Lessmann, 2011; Sierra et al., 2002; Cuesta-Camacho et al., 2006). Ecuador authorities are aware of this situation, and thus established as a goal for the 2009–2013 national strategic planning to increase a 5% the area currently under protection (SENPLADES, 2009); the present exercise can thus be envisaged as a timely help in directing conservation actions to complement the current reserve network. The results obtained in this study represent a complementary view to the information available to conservation managers in Ecuador. Yet, conservation areas proposed in this work are not absolute and require extensive analysis and collection of additional information. There is an urgent need to fill important gaps of information (e.g. information on endemic and endangered species) in tropical areas (Cayuela et al., 2009). To fill these information gaps will facilitate the development of a feedback system for the analysis of conservation priorities. Using distribution models has obvious advantages and significant challenges. But certainly more data on the distribution of species in these areas are necessary to advance conservation efforts. 4.3. Reliability of species distribution models Model reliability depends on data quantity and quality as well as species ecological characteristics (Mateo et al., 2010b). In the case of narrowly distributed species, reliable models can be generated from data sets with few presences (McPherson and Jetz, 2007; Pearson et al., 2007). In the case of widely distributed species, model accuracy and reliability can be reduced due to different factors (Mateo et al., 2010b). For example, presence data used to generate the model could represent regional niche variation or only part of the ecological heterogeneity of the species. Therefore, the minimum number of presences needed to train the model for a restricted species, that can be really low (Pearson et al., 2007), cannot be extrapolated without considering species’ biology and biogeographic range (Bean et al., 2012). From the perspective of the climatic variables, our use of all 19 Worldclim predictors was not optimal due to collinearity issues, and this may have affected some models. Yet, it would be equally difficult for managers intending to use these models to select an optimal subset of these predictors. A future study should propose a standard procedure to reduce the number of collinear variables in SDMs for optimal use by managers. Finally, SDMs do not consider some aspects that are certainly crucial to the current species’ realised distributions, such as historical barriers (Guisan and Zimmermann, 2000; Soberón and Peterson, 2005). In this study, we are interested in present species richness patterns, and local commission errors (i.e. beyond barriers) are unlikely to change significantly the areas of maximised suitability for conservation we identify, based on many species. Moreover, it would be difficult to assess what constitute a barrier for the 408 species used in this study, as this is highly depending on the dispersal strategy of each species (see Engler and Guisan, 2009), and these data were (and are still) not available for such analysis. Without such knowledge, any choice of barrier would be highly subjective. However, as this represents an interesting perspective, it would be worth adding in future studies.

160

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161

4.4. Conclusions From our study, we can conclude that: (1) using a modelling approach based on spatial predictions of individual species separated in distinct biogeographic pools allowed to recover the current network of protected areas in Ecuador, which was originally designed independently from the present study. (2) The new compositionalist approach provides a complementary view to the traditional richness hotspots approach. (3) Our predictions, regardless of method, allowed the identification of additional areas in Ecuador that are absent from the current protection network. For Ecuador, further studies should aim at refining the approach with more taxonomic groups, especially animals, and also with additional geographic information, especially in the coastal areas, as well as species characteristics, such as rarity. This increase of information would help redefining the current reserve network, and would inform about new areas which conservation would maximise the economic, social and political investment dedicated to their protection (Cuesta-Camacho et al., 2006; Lessmann, 2011). Acknowledgments We thank the ‘Fundación BBVA’ for its financial support and the Missouri Botanical Garden for generously providing the data on the plant distribution. We especially thank the collectors and specialists who made this work possible. M.d.l.E. was funded by a Spanish Juan de la Cierva Grant JCI-2009-05243 of the Ministry of Science and Technology; he also thanks J.A. Devesa for his support. Grant CGL2009-09530-BOS of the Ministry of Science and Technology of Spain currently supports J.M. We thank anonymous referees and the Editor, for their comments, which allowed us to improve the manuscript greatly. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.biocon.2013.01.014. References Araújo, M.B., Guisan, A., 2006. Five (or so) challenges for species distribution modelling. J. Biogeogr. 33, 1677–1688. Araújo, M.B., New, M., 2007. Ensemble forecasting of species distributions. Trends Ecol. Evol. 22, 42–47. Araújo, M.B., Cabeza, M., Thuiller, W., Hannah, L., Williams, P.H., 2004. Would climate change drive species out of reserves? An assessment of existing reserveselection methods. Glob. Change Biol. 10, 1618–1626. Austin, M., 2007. Species distribution models and ecological theory: a critical assessment and some possible new approaches. Ecol. Model. 200, 1–19. Balslev, H., Renner, S.S., 1989. Diversity of East Ecuadorian lowland forests. In: Holm-Nielsen, L.B., Nielsen, I.C., Balslev, H. (Eds.), Tropical Forests: Botanical Dynamics, Speciation, and Diversity. Academic Press, London, pp. 287–295. Bass, M.S., Finer, M., Jenkins, C.N., Kreft, H., Cisneros-Heredia, D.F., McCracken, S.F., Pitman, N.C.A., English, P.H., Swing, K., Villa, G., Fiore, A.D., Voigt, C.C., Kunz, T.H., 2010. Global conservation significance of Ecuador’s Yasuní National Park. PLoS ONE 5, e8767. Bates, J.M., Demos, T.C., 2001. Do we need to devalue Amazonia and other large tropical forests? Divers. Distrib. 7, 249–255. Bean, W.T., Stafford, R., Brashares, J.S., 2012. The effects of small sample size and sample bias on threshold selection and accuracy assessment of species distribution models. Ecography 35, 250–258. Best, B.J., Kessler, M., 1995. Biodiversity and Conservation in Tumbesian Ecuador and Peru. BirdLife International, Cambridge. Bøhn, T., Amundsen, P.-A., 2004. Ecological interactions and evolution: forgotten parts of biodiversity? Bioscience 54, 804–805. Bisby, F.A., 2000. The quiet revolution: biodiversity informatics and the internet. Science 289, 2309–2312. Broennimann, O., Treier, U.A., Müller-Schärer, H., Thuiller, W., Peterson, A.T., Guisan, A., 2007. Evidence of climatic niche shift during biological invasion. Ecol. Lett. 10, 701–709. Bruner, A.G., Gullison, R.E., Balmford, A., 2004. Financial costs and shortfalls of managing and expanding protected-area systems in developing countries. Bioscience 54, 1119–1126. Callicott, J.B., Crowder, L.B., Mumford, K., 1999. Current normative concepts in conservation. Conserv. Biol. 13, 22–35.

Carvalho, S.B., Brito, J.C., Pressey, R.L., Crespo, E., Possingham, H.P., 2010. Simulating the effects of using different types of species distribution data in reserve selection. Biol. Conserv. 143, 426–438. Cayuela, L., Golicher, D.J., Newton, A.C., Kolb, M., Alburquerque, F.S.D., Arets, E.J.M.M., Alkemade, J.R.M., Pérez, A.M., 2009. Species distribution modeling in the tropics: problems, potentialities, and the role of biological data for effective species conservation. Trop. Conserv. Sci. 2, 319–352. Chapin, F.S., Zavaleta, E.S., Eviner, V.T., Naylor, R.L., Vitousek, P.M., Reynolds, H.L., Hooper, D.U., Lavorel, S., Sala, O.E., Hobbie, S.E., Mack, M.C., Díaz, S., 2000. Consequences of changing biodiversity. Nature 405, 234–242. Cuesta-Camacho, F., Peralvo, M., Ganzenmüller, A., Sáenz, M., Novoa, J., Riofrío, G., Beltrán, K., 2006. Identificación de vacíos y prioridades de conservación para la biodiversidad terrestre en el Ecuador continental. EcoCiencia, Quito. Dodson, C.H., Gentry, A.H., 1991. Biological extinction in Western Ecuador. Ann. Mo. Bot. Gard. 78, 273–295. Dubuis, A., Pottier, J., Rion, V., Pellissier, L., Theurillat, J.-P., Guisan, A., 2011. Predicting spatial patterns of plant species richness: a comparison of direct macroecological and species stacking approaches. Diver. Distrib. 17, 1122–1131. Elith, J., Leathwick, J., 2009. Species distribution models: ecological explanation and prediction across space and time. Ann. Rev. Ecol. Evol. Syst. 40, 677–697. Elith, J., Graham, C.H., Anderson, R.P., Dudík, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberón, J., Williams, S., Wisz, M.S., Zimmermann, N.E., 2006. Novel methods improve prediction of species’ distributions from occurrence data. Ecography 29, 129– 151. Engler, R., Guisan, A., 2009. MIGCLIM: predicting plant distribution and dispersal in a changing climate. Divers. Distrib. 15, 590–601. Estrella, M., Mateo, R.G., Wieringa, J.J., Mackinder, B., Muñoz, J., 2012. Legume diversity patterns in West Central Africa: influence of species biology on distribution models. PLoS ONE 7 (7), e41526. Feeley, K.J., Silman, M.R., 2011. The data void in modeling current and future distributions of tropical species. Glob. Change Biol. 17, 626–630. Ferrier, S., 2002. Mapping spatial pattern in biodiversity for regional conservation planning: where to from here? Syst. Biol. 51, 331–363. Ferrier, S., Guisan, A., 2006. Spatial modelling of biodiversity at the community level. J. Appl. Ecol. 43, 393–404. Ferrier, S., Manion, G., Elith, J., Richardson, K., 2007. Using generalized dissimilarity modelling to analyse and predict patterns of beta diversity in regional biodiversity assessment. Divers. Distrib. 13, 252–264. Fielding, A.H., Bell, J.F., 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ. Conserv. 24, 38–49. Filzmoser, P., Garrett, R.G., Reimann, C., 2005. Multivariate outlier detection in exploration geochemistry. Comput. Geosci. 31, 579–587. Fjeldsa, J., Álvarez, M.D., Lazcano, J.M., León, B., 2005. Illicit crops and armed conflict as constraints on biodiversity conservation in the Andes Region. Ambio 34, 205–211. García, A., 2006. Using ecological niche modelling to identify diversity hotspots for the herpetofauna of Pacific lowlands and adjacent interior valleys of Mexico. Biol. Conserv. 130, 25–46. Gaubert, P., Papes, M., Peterson, A.T., 2006. Natural history collections and the conservation of poorly known taxa: ecological niche modeling in central African rainforest genets (Genetta spp.). Biol. Conserv. 130, 106–117. Gentry, A.H., 1982. Phytogeographic patterns in northwest South America and southern Central America as evidence for a Chocó Refugium. In: Prance, G. (Ed.), Biological Diversification in the Tropics. Colombia Univ. Press, New York, pp. 112–136. Gentry, A.H., 1986. Species richness and floristic composition of the Choco region plant communities. Caldasia 15, 71–79. Godown, M.E., Peterson, A.T., 2000. Preliminary distributional analysis of US endangered bird species. Biodivers. Conserv. 9, 1313–1322. Graham, C.H., Ron, S.R., Santos, J.C., Schneider, C.J., Moritz, C., 2004. Integrating phylogenetics and environmental niche models to explore speciation mechanisms in Dendrobatid frogs. Evol. Psychiat. 58, 1781–1793. Grand, J., Buonaccorsi, J., Cushman, S.A., Griffin, C.R., Neel, M.C., 2004. A multiscale landscape approach to predicting bird and moth rarity hotspots in a threatened pitch pine-scrub oak community. Conserv. Biol. 18, 1063–1077. Grenouillet, G., Buisson, L., Casajus, N., Lek, S., 2011. Ensemble modelling of species distribution: the effects of geographical and environmental ranges. Ecography 34, 9–17. Guisan, A., Rahbek, C., 2011. SESAM – a new framework integrating macroecological and species distribution models for predicting spatio-temporal patterns of species assemblages. J. Biogeogr. 38, 1433–1444. Guisan, A., Zimmermann, N.E., 2000. Predictive habitat distribution models in ecology. Ecol. Model. 135, 147–186. Guralnick, P., Hill, A.W., Lane, M., 2007. Towards a collaborative, global infrastructure for biodiversity assessment. Ecol. Lett. 10, 663–672. Hansen, M.C., Stehman, S.V., Potapov, P.V., 2010. Quantification of global gross forest cover loss. Proc. Natl. Acad. Sci. 107, 8650–8655. Hijmans, R.J., Guarino, L., Mathur, P., 2012. DIVA-GIS Version 7.5. Manual. (accessed 02.02.12). Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high resolution interpolated climate surfaces for global land areas. Int. J. Climatol. 25, 1965–1978.

R.G. Mateo et al. / Biological Conservation 160 (2013) 150–161 IUCN, 2011. The IUCN Red List of Threatened Species. Version 2011.1. (retrieved 01.05.11). Jiménez-Valverde, A., Lobo, J.M., 2007. Threshold criteria for conversion of probability of species presence to either-or presence–absence. Acta Oecol. 31, 361–369. Jørgensen, P.M., León-Yánez, S., 1999. Catalogue of the Vascular Plants of Ecuador. Jørgensen, P.M., Ulloa, C.U., Pedersen, H.B., Luteyn, J.L., 1992. The Quito Herbarium (QCA): 100,000 important collections from Ecuador. Taxon 41, 51–56. Kareiva, P., Marvier, M., 2003. Conserving biodiversity coldspot. Am. Sci. 91, 344– 351. León-Yánez, S., 2000. La flora de los páramos ecuatorianos. In: La biodiversidad de los páramos. Serie Páramo 7. GTP/AbyaYala, Quito, Ecuador, pp. 5–21. Lessmann, J., 2011. Como complementar las áreas protegidas del Ecuador continental. Qué proteger y a qué coste. Universidad Internacional Menéndez Pelayo, Madrid, Spain, p. 47. Liu, C., Berry, P.M., Dawson, T.P., Pearson, R.G., 2005. Selecting thresholds of occurrence in the prediction of species distributions. Ecography 28, 385–393. Loiselle, B., Howell, C.A., Graham, C.H., Goerck, J.M., Brooks, T., Smith, K.G., Williams, P.H., 2003. Avoiding pitfalls of using species distributions models in conservation planning. Conserv. Biol. 17, 1591–1600. Loiselle, B.A., Jørgensen, P.M., Consiglio, T., Jiménez, I., Blake, J.G., Lohmann, L.G., Montiel, O.M., 2008. Predicting species distributions from herbarium collections: does climate bias in collection sampling influence model outcomes? J. Biogeogr. 35, 105–116. MAE, 2011. Ministerio del Ambiente de Ecuador: Subsistema Patrimonio de Áreas Naturales del Estado (PANE) de Ecuador. (retrieved 01.12.11). Margules, C.R., Pressey, R.L., 2000. Systematic conservation planning. Nature 405, 243–252. Margules, C., Sarkar, S., 2007. Systematic Conservation Planning. Cambridge University Press, Cambridge, UK. Marmion, M., Parviainen, M., Luoto, M., Heikkinen, R.K., Thuiller, W., 2009. Evaluation of consensus methods in predictive species distribution modelling. Divers. Distrib. 15, 59–69. Mateo, R.G., Croat, T.B., Felicísimo, Á.M., Muñoz, J., 2010a. Profile or group discriminative techniques? Generating reliable species distribution models using pseudo-absences and target-group absences from natural history collections. Divers. Distrib. 16, 84–94. Mateo, R.G., Felicísimo, Á.M., Muñoz, J., 2010b. Effects of the number of presences on reliability and stability of MARS species distribution models: the importance of regional niche variation and ecological heterogeneity. J. Veg. Sci. 21, 908– 922. Mateo, R.G., Felicísimo, A.M., Pottier, J., Guisan, A., Muñoz, J., 2012. Do stacked species distributions models reflect altitudinal diversity patterns? PLoS ONE 7, e32586. McClean, C.J., Lovett, J.C., Küper, W., Hannah, L., Sommer, J.H., Barthlott, W., Termansen, M., Smith, G.F., Tokumine, S., Taplin, J.R.D., 2005. African plant diversity and climate change. Ann. Mo. Bot. Gard. 92, 135–152. McPherson, J.M., Jetz, W., 2007. Effects of species’ ecology on the accuracy of distribution models. Ecography 30, 135–151. Mecham, J., 2001. Causes and Consequences of Deforestation in Ecuador. . Mena-Vásconez, P., Medina, G., Hofstede, R., 2001. Los Páramos del Ecuador. Abya Yala/Proyecto Páramo, Quito, Ecuador. Mittermeier, R.A., Myers, N., Thompsen, J.B., Fonseca, G.A.B., 1998. Biodiversity hotspots and major tropical wilderness areas: approaches to conserving to setting conservation priorities. Conserv. Biol. 12, 516–520. Moilanen, A., Wilson, K.A., Possingham, H.P., 2009. Spatial conservation prioritization. Quantitative methods and computational tools. Oxford University Press, Oxford. Muñoz, J., Felicísimo, A.M., 2004. A comparison between some statistical methods commonly used in predictive modeling. J. Veg. Sci. 15, 285–292. Myers, N., 1988. Threatened biotas: ‘‘Hotspots’’ in tropical forests. Environmentalist 8, 1–20. Myers, N., Mittermeler, R.A., Mittermeler, C.G., Fonseca, G.A.B., Kent, J., 2000. Biodiversity hotspots for conservation priorities. Nature 403, 853–858. Newbold, T., 2010. Applications and limitations of museum data for conservation and ecology, with particular attention to species distribution models. Prog. Phys. Geogr. 34, 3–22. Olson, D.M., Dinerstien, E., 1998. The global 2000: a representation approach to conserving the Earth’s most biologically valuable ecoregions. Conserv. Biol. 12, 502–515. Orme, C.D.L., Davies, R.G., Burgess, M., Eigenbrod, F., Pickup, N., Olson, V.A., Webster, A.J., Ding, T.-S., Rasmussen, P.C., Ridgely, R.S., Stattersfield, A.J., Bennett, P.M., Blackburn, T.M., Gaston, K.J., Owens, I.P.F., 2005. Global hotspots of species richness are not congruent with endemism or threat. Nature 436, 1016–1019. Ortega-Huerta, M.A., Peterson, A.T., 2004. Modelling spatial patterns of biodiversity for conservation prioritization in North-eastern Mexico. Divers. Distrib. 10, 39– 54. Papesß, M., Gaubert, P., 2007. Modelling ecological niches from low numbers of occurrences: assessment of the conservation status of poorly known viverrids (Mammalia, Carnivora) across two continents. Divers. Distrib. 13, 890–902. Parviainen, M., Marmion, M., Luoto, M., Thuiller, W., Heikkinen, R.K., 2009. Using summed individual species models and state-of-the-art modelling techniques to identify threatened plant species hotspots. Biol. Conserv. 142, 2501–2509.

161

Pearson, R.G., Raxworthy, C.J., Nakamura, M., Townsend Peterson, A., 2007. Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar. J. Biogeogr. 34, 102–117. Pimm, S.L., Russel, G.J., Gittleman, J.L., Brooks, T.M., 1995. The future of biodiversity. Science 269, 347–350. Pineda, E., Lobo, J.M., 2009. Assessing the accuracy of species distribution models to predict amphibian species richness patterns. J. Anim. Ecol. 78, 182–190. Possingham, H.P., Wilson, K.A., Andelman, S.J., Vynne, C.H., 2006. Protected areas. Goals, limitations, and design. In: Groom, M.J., Meffe, G.K., Carroll, C.R. (Eds.), Principles of Conservation Biology. Sinaeur Associates, pp. 507–549. Prendergast, J.R., Quinn, R.M., Lawton, J.H., 1999. The gaps between theory and practice in selecting nature reserves. Conserv. Biol. 13, 484–492. Pressey, R.L., Humphries, C.J., Margules, C.R., Vane-Wright, R.I., Williams, P.H., 1993. Beyond opportunism: key principles for systematic reserve selection. Trends Ecol. Evol. 8, 124–128. Price, A.R.G., 2002. Simultaneous ‘hotspots’ and ‘coldspots’ of marine biodiversity and implications for global conservation. Mar. Ecol. Prog. Ser. 241, 23–27. Pulliam, R.H., 2000. On the relationship between niche and distribution. Ecol. Lett. 3, 349–361. Rahbek, C., Graves, G.R., 2001. Multiscale assessment of patterns of avian species richness. Proc. Natl. Acad. Sci. 98, 4534–4539. Raven, P.H., Wilson, E., 1992. A fifty-year plan for biodiversity surveys. Science 258, 1099–1100. Rodrigues, A.S.L., Andelman, S.J., Bakarr, M.I., Boitani, L., Brooks, T.M., Cowling, R.M., Fishpool, L.D.C., da Fonseca, G.A.B., Gaston, K.J., Hoffmann, M., Long, J.S., Marquet, P.A., Pilgrim, J.D., Pressey, R.L., Schipper, J., Sechrest, W., Stuart, S.N., Underhill, L.G., Waller, R.W., Watts, M.E.J., Yan, X., 2004. Effectiveness of the global protected area network in representing species diversity. Nature 428, 640–643. Rondinini, C., Wilson, K.A., Boitani, L., Grantham, H., Possingham, H.P., 2006. Tradeoffs of different types of species occurrence data for use in systematic conservation planning. Ecol. Lett. 9, 1136–1145. Rutledge, D.T., Lepcyzk, C.A., Xie, J., Liu, J., 2000. Spatiotemporal dynamics of endangered species hotspots in the United States. Conserv. Biol. 15, 475–487. Sáenz, M., Onofa, Á., 2005. Preguntas clave: Reporte de los ecosistemas terrestres ecuatorianos. Indicadores de Biodiversidad para Uso Nacional (Proyecto BINU). Ecuador. Ministerio del Ambiente del Ecuador, Fundación EcoCiencia, Quito. Sanderson, E.W., Jaiteh, M., Levy, M.A., Redford, K.H., Wannebo, A.V., Woolmer, G., 2009. The human footprint and the last of the wild. Bioscience 52, 891–904. Segurado, P., Araújo, M.B., 2004. An evaluation of methods for modelling species distributions. J. Biogeogr. 31, 1555–1568. SENPLADES, 2009. Plan Nacional para el Buen Vivir 2009–2013: Construyendo un Estado Plurinacional e Intercultural. República de Ecuador. (accessed 15.12.09). Sierra, R., Cerón, C., Palacios, W., Valencia, R., 1999. Propuesta Preliminar de un Sistema de Clasificación de Vegetación para el Ecuador Continental. Proyecto INEFAN/ GEF-BIRF y EcoCiencia, Quito, Ecuador. Sierra, R., Campos, F., Chamberlin, J., 2002. Assessing biodiversity conservation priorities: ecosystem risk and representativeness in continental Ecuador. Landsc. Urban Plann. 59, 95–110. Skov, F., Borchsenius, F., 1997. Predicting plant species distribution patterns using simple climatic parameters: a case study of Ecuadorian palms. Ecography 20, 347–355. Soberón, J., Peterson, A.T., 2004. Biodiversity informatics: managing and applying primary biodiversity data. Philos. Trans. Roy. Soc. Lond. Ser. B Biol. Sci. 359, 689–698. Soberón, J., Peterson, A.T., 2005. Interpretation of models of fundamental ecological niches and species’ distributional areas. Biodivers. Inform. 2, 1–10. Suárez, E., Morales, M., Cueva, R., Bucheli, V.U., Zapata-Ríos, G., Toral, E., Torres, J., Prado, W., Olalla, J.V., 2009. Oil industry, wild meat trade and roads: indirect effects of oil extraction activities in a protected area in north-eastern Ecuador. Anim. Conserv. 12, 364–373. Terborgh, J., Winter, B., 1982. Evolutionary circumstances of species with small ranges. In: Prance, G.T. (Ed.), Biological Diversification in the Tropics. Columbia University Press, New York, pp. 587–600. Ulloa Ulloa, C., Jørgensen, P.M., 2004. Árboles y arbustos de los Andes del Ecuador. . Underwood, J.G., D’Agrosa, C., Gerber, L.R., 2010. Identifying conservation areas on the basis of alternative distribution data sets. Conserv. Biol. 24, 162–170. Urbina-Cardona, J.N., Flores-Villela, O., 2010. Ecological-niche modeling and prioritization of conservation-area networks for Mexican Herpetofauna. Conserv. Biol. 24, 1031–1041. Valencia, R., Pitman, N., León-Yánez, S., Jørgensen, P.M., 2000. Libro rojo de las plantas endémicas del Ecuador. Herbario QCA, Pontificada Universidad Católlica del Ecudor, Quito. Whittaker, R.J., Araújo, M.B., Jepson, P., Ladle, R.J., Watson, J.E.M., Willis, K.J., 2005. Conservation biogeography: assessment and prospect. Divers. Distrib. 11, 3–23. Williams, P.H., Araújo, M.B., 2000. Integrating species and ecosystem monitoring for identifying conservation priorities. Eur. Conserv. 4, 17–18. Wilson, K.A., Westphal, M.I., Possingham, H.P., Elith, J., 2005. Sensitivity of conservation planning to different approaches to using predicted species distribution data. Biol. Conserv. 122, 99–112. Wisz, M.S., Hijmans, R.J., Li, J., Peterson, A.T., Graham, C.H., Guisan, A., 2008. Effects of sample size on the performance of species distribution models. Divers. Distrib. 14, 763–773. Young, K.R., Ulloa, C., Luteyn, J.L., Knapp, S., 2002. Plant evolution and endemism in Andean South America: an introduction. Bot. Rev. 68, 4–21.