Insights for protection of high species richness areas for the conservation of Mesoamerican endemic birds

To assess the representativeness values of Mesoamerican endemic birds within the current network of protected areas (PAs) to determine high‐priority and complementary conservation areas to maximize the long‐term protection of species.


| INTRODUC TI ON
Biodiversity conservation is critical, in part because the interaction of biotic communities with the physical habitat results in several ecosystem services that support human well-being (Cardinale et al., 2012). Paradoxically, species extinction rates due to human activities are currently 1,000 times higher than extinctions due to natural causes, and during this century, they are predicted to increase tenfold because of the accumulation of human impacts on natural ecosystems (De Vos, Joppa, Gittleman, Stephens, & Pimm, 2015). Developing methodologies that quickly provide accurate information for decision makers about what geographical areas must be prioritized is an urgent step towards reducing this extensive biodiversity loss and ensuring the provision of ecosystem services. This is particularly important in heavily threatened ecosystems that host high levels of species richness and endemism as well as agriculture and human settlement (e.g., Nori et al., 2016;Peters et al., 2019;Prieto-Torres, Nori, & Rojas-Soto, 2018;Strassburg et al., 2017).
Currently, twenty-five terrestrial biodiversity "hotspots" have been identified worldwide based on exceptional concentrations of species/habitat diversity. These sites encompass ~44% of the Earth's plant species and 35% of its vertebrates in just 1.4% of the land surface (Myers, Mittermeier, Mittermeier, Fonseca, & Kent, 2000).
Thus, they are the focus of many conservation programmes aiming to reduce the current rate of biodiversity loss (Cincotta, Wisnewski, & Engelman, 2000). For instance, ~38% of these hotspots are currently legally protected as parks and reserves that range from highly restrictive areas where all human activities are excluded to more inclusive management strategies involving local communities (see Schwartzman, Moreira, & Nepstad, 2000). Among these hotspots, Mesoamerica is considered among the most important high species richness sites for conservation across the Americas (Myers et al., 2000).
This biologically complex region extends from central Mexico to southern Panama and northern Colombia. It is mainly composed of highly fragmented tropical forest patches that vary in size and extent. Mesoamerica has wide topographic and climatic variability and a complex biogeographical history, all of which have promoted important biotic interchange events and extensive diversification in situ due to climatic changes and geological processes (Llorente, 1996;Prieto-Torres, Rojas-Soto, Santiago-Alarcón, Bonaccorso, & Navarro-Sigüenza, 2019;Ramamoorthy, Bye, Lot, & Fa, 1998;Ríos-Muñoz & Navarro-Sigüenza, 2012;Stehli & Webb, 1985). However, much this biodiversity remains unprotected. About ~72% of the Mesoamerican tropical forest ecosystems have already been converted to urban or agricultural uses (Bryant, Nielsen, & Tangley, 1997;Miller, Chang, & Johnson, 2001;Portillo-Quintero & Sánchez-Azofeifa, 2010;Weinzettel, Vačkář, & Medková, 2018). For most of the countries in this region, current protected areas (PAs) cover only a small proportion of the total surface area (less than 15%), which is far from the goal of 17% proposed in the Aichi targets (UNEP, 2010). Thus, generating a PA network that adequately represents the biodiversity in the long-term protection of these areas is an urgent task (Rodrigues et al., 2004;Venter et al., 2016).
From this perspective, different conservation planning schemes have been developed over the last decade (Ball, Possingham, & Watts, 2009;Ciarleglio, Wesley Barnes, & Sarkar, 2009;Moilanen et al., 2014;Sarkar & Illoldi-Rangel, 2010) promoting well-informed decisions to expand the PA network and contribute to the viability of long-term protection of biodiversity and ecosystem function (Watson, Grantham, Wilson, & Possingham, 2011). Generally, these approaches are based on the distribution of key biodiversity features (typically species distribution) and anthropic variables to identify the most important sites for conservation that are also compatible with sustained human development (Brum et al., 2017;Kukkala et al., 2016). However, in practice, it is difficult to compile this information comprehensively, assuring both spatial and taxonomic representation (Carvalho, Brito, Pressey, Crespo, & Possingham, 2010). Unfortunately, information on the distribution for most species is incomplete, and when it exists, data are generally biased by site accessibility (Gaston & Rodrigues, 2003;Peterson, 2001). Thus, considering the immense efforts required to define maps of species' distributional ranges, the use of computational algorithms to generate species distribution models (SDMs) is an effective and widely accepted method to obtain accurate species distribution maps (Araújo et al., 2019;Peterson, 2001;Soberón & Peterson, 2005). This approach has been applied on a global scale in biogeography, macroecology and particularly in conservation planning (Araújo et al., 2019;Costa, Nogueira, Machado, & Colli, 2010;Hidasi-Neto et al., 2019;Nori, Villalobos, & Loyola, 2018;Prieto-Torres & Pinilla-Buitrago, 2017). In fact, combining SDMs and site-selection algorithms could provide more accurate data for reserve design in regions where protection of high-diversity areas at the lowest cost is urgent (e.g., Elith & Leathwick, 2009;Lessmann, Fajardo, Muñoz, & Bonaccorso, 2016;Pawar et al., 2007;Prieto-Torres et al., 2018).
Given that spatial and taxonomic representation of biodiversity is often poor at the regional level, species-level surrogates are often necessary in a conservation context to ensure that critical habitats and ecosystems within the region are not missed (Lessmann et al., 2016;Peralvo, Sierra, Young, & Ulloa, 2006). Birds, as a charismatic and well-known group of vertebrates, are good surrogates and defining areas that are important for bird conservation is an excellent first step to delineating areas that are important for conservation efforts more generally (Barnagaud et al., 2017;Kati et al., 2004). Besides, birds are important indicators of landscape conditions due to their strong vulnerability to environmental alterations (Chambers, 2008;Fahrig, 2003;Foley et al., 2005;Imbeau, Monkkonen, & Desrochers, 2001;Lawton et al., 1998;O'Connell, Jackson, & Brooks, 2000;Sekercioglu, 2006). In this sense, protecting birds is expected to provide benefits to other taxa (Gregory et al., 2005;Larsen, Bladt, Balmford, & Rahbek, 2012;Roberge & Angelstam, 2004). Therefore, birds have long attracted the attention of scientists, decision makers and non-governmental organizations to highlight and promote conservation policies and needs (e.g., Kujala, Burgman, & Moilanen, 2013;Prieto-Torres et al., 2018;Triviño, Kujala, Araújo, & Cabeza, 2018).
In this study, we focus on endemic bird species as indicators of overall diversity patterns across the region because Mesoamerica has high levels of endemism for birds (Eissermann & Avendaño, 2018;García-Moreno, Cortés, García-Deras, & Hernández-Baños, 2006;Navarro-Sigüenza & Sánchez-González, 2003;Peterson, Escalona-Segura, & Griffith, 1998;Peterson et al., 2003;Prieto-Torres et al., 2019;Sánchez-González, Morrone, & Navarro-Sigüenza, 2008;Sánchez-Ramos et al., 2018). Therefore, considering that endemic species reflect a unique history of the Earth and its biota, failure to protect them would result in major losses of unique species diversity for this highly threatened region and its ecosystems. In this context, using SDMs and conservation planning protocols (based on ZONATION software), we aim to (a) assess the current representativeness levels of Mesoamerican endemic bird species within existing PAs and (b) determine high-priority areas for conservation that complement the current PA network to maximize species representation and protection, in a way that considers the anthropic context. This information allows us to provide new and more accurate data on which areas require attention and therefore represents an important step to guide future establishment of new and efficient conservation areas across Mesoamerica. F I G U R E 1 Species richness distribution patterns of Mesoamerican endemic bird species (n = 182), showing the location of current protected areas according to the World Database of Protected Areas (UNEP & WCMC, 2019). A total of ~199,300 km 2 (i.e., 12.9%) of Mesoamerican terrestrial surface was covered by designed PAs (see Table 1). Birds (from left to right) in maps are Peucaea sumichrasti (NT); Campylorhynchus yucatanicus (NT); Piranga roseogularis (LC); and Icterus auratus (DD). The bird pictures were taken from Birds of the World's website (The Cornell Lab of Ornithology; Available in: https://birds ofthe world.org/bow/home)

| Study area
The geographical range for this study was the Mesoamerican re-  (Escalante, Sánchez-Cordero, Morrone, & Linaje, 2007). This region encompasses all sub-tropical and tropical ecosystems (grouped into five biomes, over 60 vegetation type and 41 ecoregions) and is considered both a centre of origin and a corridor for terrestrial species (Olson et al., 2001;Jiménez & López, 2007). Thus, during the last 30 years, great efforts have been made to conserve representative samples of these ecosystems, resulting in more than 3,800 PAs (including National Parks and wilderness areas) throughout the region (Jiménez & López, 2007;IUCN & UNEP-WCMC, 2019).

| Species selection and occurrence records
We created a complete list of the permanent resident and endemic bird species inhabiting Mesoamerica, defined as species whose distributional range is limited only to the study area. This list was compiled from sources that offer information on the habitat characteristics for each species (e.g., Howell & Webb, 1995;Prieto-Torres et al., 2019;Stotz, Fitzpatrick, Parker, & Moskovits, 1996) Salvin & Godman, 1879-1904. Access number for downloaded GBIF records for each species is detailed in the Appendix S1.
Next, to identify problematic or imprecise species occurrences, we compared the spatial distribution of records obtained with the species ranges defined by the Neotropical Birds website (see details at https://neotr opical.birds.corne ll.edu) and removed all mismatched records. For cases where the geographical information of localities was dubious (e.g., likely data transcription errors), the lat-long coordinates were verified using ArcMap v.10 (ESRI, 2011) and Google Earth, and records located outside Mesoamerica and those with geographical information that could not be verified were eliminated.
We also removed points located within cities because these occurrences may not reflect the habitat requirements of species. These steps were important to identify problematic or imprecise species occurrence data with incorrect climate values because the choice of climate baseline and reduction of sampling bias affects model performance for each species (Boria, Olson, Goodman, & Anderson, 2014;Roubicek et al., 2010). Likewise, for this study we decided to exclude species with less than 15 independent occurrence records available because low sample size may affect model performance (Owens et al., 2013;Pearson, Raxworthy, Nakamura, & Peterson, 2007).
After discarding individual species models that were not statistically significant (see below), our full dataset contained 48,477 individual records of 180 endemic species--belonging to 12 orders, 34 families TA B L E 1 Current area of designated protected areas (PAs) and the additional area identified as priority conservation areas to increase coverage to match Aichi targets (17%) in Mesoamerica by country and 125 genera (see Appendix S1)--which we then used to build our models.

| Environmental data
Because building species distribution models relies on the environmental variables associated with occurrence points of the bird species, we gathered 19 bioclimatic variables summarizing aspects of precipitation and temperature for the Earth's surface from the layers of WorldClim 2.0 (Fick & Hijmans, 2017)  as implemented in the "corrplot" (Wei & Simko, 2017) and "usdm" (Naimi, 2017) libraries in R software (R-Core-Team, 2019). Detailed information about the set of environmental variables used for each species is shown in Appendix S1.

| Species distribution models
To construct the potential distributional area models for each bird species, we used MaxEnt 3.4.1 (Phillips, Anderson, & Schapire, 2006), which uses the maximum entropy principle to calculate the most likely distribution of focal species as a function of occurrence localities and environmental variables (Elith et al., 2011). Although other computer programs are also available for modelling species' distribution ranges, we decided to use MaxEnt because it has been proven to perform better when only presence data are available (Elith et al., 2011), as is our case. This software produces robust models with ≥15 occurrence points are available for each species (Elith et al., 2011;Wisz et al., 2008).
On the other hand, given that SDMs must consider historical factors affecting species' distributions, we used specific areas for model calibration for each species, known as the accessible area or M (Barve et al., 2011;Soberón & Peterson, 2005). For each species, a mask or GIS polygon delimiting this calibration area was established based on the intersection of occurrence records with the WWF Terrestrial Ecoregions (Olson et al., 2001) and the Biogeographical Provinces of the Neotropical region (Morrone, 2014). In effect, we assumed that this defined region has been explored by each species (i.e., reached by dispersal from existing populations) and thus represents both the species' tolerance limits as well as historical and ecological barriers to dispersal (such as rivers or valleys) across the Mesoamerican region.
All models were run with no extrapolation to avoid artificial projections from extreme values of ecological variables (Elith et al., 2011;Owens et al., 2013). Other MaxEnt parameters were set to default. We used the bootstrap resampling option from MaxEnt to calibrate the habitat suitability models of each species, which randomly resampled 75% of the occurrence data (training points) 100 times to generate the models (i.e., replicates), while using the remaining 25% of the dataset (testing points) to assess the model's accuracy by computing the area under receiver operating characteristic curves (AUC; Elith et al., 2006;Fielding & Bell, 1997). Then, we retained for subsequent analyses only the model that represented the mean environmental suitability value for each species. We converted continuous cloglog habitat suitability probability outputs for each species (Phillips et al., 2006) into binary presence-absence maps by setting the decision threshold to "10th percentile training presence." We used this threshold criterion to minimize over-predictions in our final binary maps, allowing better recovery of species' distributional areas (Liu, White, & Newell, 2013). Finally, model performance was evaluated by calculating the commission and omission error values (Anderson, Lew, & Peterson, 2003) and the partial ROC curve test (Peterson, Papes, & Soberón, 2008)

| Conservation prioritization
ZONATION 4.0.0b (Moilanen et al., 2005) was used to determine high-priority areas for the conservation of endemic bird species across Mesoamerica. This software establishes a hierarchical prioritization of areas of the study region, allowing the identification of key sites for the conservation of species and areas for an optimally balanced expansion of an existing reserve network. This is based on biodiversity features (here, bird species distribution) and different "penalization" variables (here, anthropic pressures) for each pixel (Di Minin, Veach, Lehtomäki, Montesino Pouzols, & Moilanen, 2014;Moilanen, 2007;Moilanen et al., 2005Moilanen et al., , 2014. The way the "loss of conservation" value is aggregated across features within a pixel depends on so-called "cell-removal rules" . Here, we decided to run our analysis implementing two different removal rules: core area zonation (CAZ) and additive benefit function (ABF).
The most important difference between these rules is that ABF assigns higher importance to cells with many features and retains a higher average proportion of features (i.e., prioritizes high species richness), while the CAZ prioritizes areas containing rare and/ Given that most bird species cannot be adequately protected in highly modified areas (Pimm et al., 2014) because human influence tends to diminish habitat quality, and therefore, the potential for conservation, it was important to prevent the software from assigning high conservation values to highly modified areas. To do this, we assigned negative weight values to pixels with >50% cover loss and extremely disturbed landscapes in a reclassified land cover map (Defourny et al., 2016) and to pixels with high human influence in the Global Terrestrial Human Footprint map (Venter et al., 2016;WCS & CIESIN, 2005). By assigning negative weights to these pixels, the sum of the positive (i.e., the summary of biodiversity features) and negative weighted was zero, allowing a balanced solution for prioritization (Faleiro, Machado, & Loyola, 2013;Moilanen et al., 2011).
Both ABF and CAZ prioritizations were run with the "edge removal" function activated and BLP (Boundary Length Penalty distribution smoothing) set to 0.5. This function forces the program to remove cells from the defined edges-to-area ratio of remaining landscape, increasing the connectivity of priority and protected areas in the landscape . ZONATION's warp factor was set at the default (warp factor = 10). All variables had a spatial resolution of 0.008333° (~1 km 2 ) and were cropped to the study area (from 7° to 22°N and from −102° to −97°W; see Figure 1).
After running the prioritization analyses, we plotted performance curves for both analyses to quantify the proportion of the original occurrences retained for each biodiversity feature, at each top fraction of the landscape chosen for conservation Moilanen et al., 2014). We generated two performance curves, one for all species and one for only threatened species (CR, EN, VU). This allowed us to determine the representativeness of the current PA network and the priority areas reaching 17% of the available territory, as proposed in the Aichi targets (UNEP, 2010). Finally, to determine the relative importance of current PAs within Mesoamerica, we repeated these prioritization analyses but did not include the shape file of PA features as a hierarchical mask (see above). Graphical results of this last step are provided as Appendix S1.

| RE SULTS
Our species distribution models showed highly significant AUC ratios from the partial ROC test (ranging from 1.15 to 1.99, p < .05) and low omission errors (mean of 16.4 ± 9.9 [i.e., 7.3 ± 9.9 occurrence points]), indicating that the models were statistically better than random. Thus, the species distribution models were considered accurate under these performance diagnostics. Overall, our species models showed spatial distributional ranges from 1,584 to 555,300 km 2 (mean of 99,448 ± 120,109 km 2 ). We observed that 25.0% of species had small distributional ranges within the region, 50.0% had intermediate range sizes, and 25.0% had large distributional ranges ( Figure 1). According to the IUCN (see Appendix S1) only 20 of these species are classified as threatened (EN and VU), 10 as NT, one as DD and 149 species as LC. In addition, species richness patterns for endemic birds across Mesoamerica tended to be highest in areas that are considered boundaries between highly biodiverse ecosystems, such as tropical dry forests and cloud forests throughout Mexico, Guatemala, Costa Rica and Panama (Figure 1). In contrast, low species richness values were found along the coast and the Caribbean slope. Overall, we observed no significant differences (p > .05) for species richness values between areas within (16.6 ± 14.9 spp.) and outside (16.5 ± 12.5 spp.) the existing PA network.
Currently, a total of ~165,000 km 2 (i.e., 13.1%) of the surface of the region is covered by PAs (Table 1, Figure 1). This level of protection across regions represents, on average, 19.3% of the distribution area of the endemic bird species analysed here (Figure 2a)

| D ISCUSS I ON
Our spatial conservation prioritization analyses showed that the current PA network is poorly representative of the distributional ranges of endemic bird species in Mesoamerica and does not efficiently cover the conservation needs. Using this macroecological approach, we identified sites that are important for species conservation through several practical methods, such as aggregation methods, uncertainty analysis, species prioritization and replacement cost analysis for current or proposed reserves . This is important because current and future conser-  (Tables 1 and 2). Firstly, according to the prioritization analyses (whether due to rarity or species richness), PA area needs to be increased to include the identified priority areas to accurately include their biodiversity. This is consistent with previous studies that have highlighted the need for additional PAs to avoid in major losses of species diversity throughout the region in the long term  Stattersfield, 1998). The priority conservation areas found here provide insights into where to focus future conservation expansion efforts to accomplish a representative and connected PA network. In fact, we showed that it is possible to greatly improve the efficiency of Mesoamerican PAs (e.g., increasing the species representativeness levels by more than 65%) by strategically expanding the current network by only 3.9%. This efficiency is important given ongoing deforestation and limited funding for conservation (e.g., Pouzols et al., 2014;Pringle, 2017;Wallace, Barborak, & MacFarland, 2003).
The picture is particularly alarming in Costa Rica, where our results suggest that an extension of more than 50% of the current PAs surface is needed, as well as in Panama, with at least an additional 18% of PA surfaces required (Table 1). TA B L E 2 Current area (in km 2 and percentage) of the current Protected Areas (PAs) and the complementary conservation areas estimated to increase coverage to match the 17% Aichi target throughout Mesoamerica by terrestrial ecosystem Mesoamerican biodiversity cannot be protected in reserves alone, since they are too isolated, too expensive to manage and too controversial in a region where poverty alleviation remains a more immediate priority than conservation. For instance, most of the birds occurring in Mesoamerica are widely distributed across the region, and most endemic species' ranges include at least two countries (e.g., Howell & Webb, 1995;Prieto-Torres et al., 2019;Sánchez-Ramos et al., 2018;Stotz et al., 1996). Thus, we argue for the implementation of trans-boundary policy collaborations for future conservation initiatives . This is particularly important considering that while most countries are behind the Aichi target connectivity element (Saura et al., 2018;Torres-Morales et al., 2019), in particular cases such as El Salvador and Honduras (Figures 1 and 2), only about half of the area currently under protection is effectively connected (Komar, 2002). Maintaining connectivity is particularly important and challenging in Mesoamerica because of the region's altitudinal and latitudinal gradients, which act as natural barriers to species movement and can increase the vulnerability of biodiversity to climate change and agricultural expansion (Harvey et al., 2008).
The priority conservation areas identified here could play a key role in halting biodiversity loss while acting as corridors to allow gene flow and migration between PAs (Dulloo et al., 2008). Likewise, future studies should address the contribution of the priority conservation areas defined here to the connectivity requirements and spatial movements for taxa we did not analyse, such as migratory species.
Fortunately, although each country maintains its own ministries of the environment, they all participate in the Central American System of Protected Areas (SICAP) formed in 1992, which has allowed the development of programmes such as the Mesoamerican Biological Corridor (DeClerck et al., 2010;Hilty, Chester, & Cross, 2012;Miller et al., 2001). This conservation programme seeks to apply the Convention on Biological Diversity's ecosystem approach to support conservation initiatives that are strongly linked to sustainable rural livelihoods, while simultaneously integrating regional scale PA connectivity. This is particularly important because it is an ongoing coordinated effort among countries to reach the Aichi targets, which has allowed some Mesoamerican PAs to progress through a variety of regional, national and internationally recognized reserves. In fact, Nevertheless, considering that conservation prioritization often takes place at smaller scales (Wallace et al., 2003), we also argue that additional land is not the only requirement to meet a given conservation goal. While studies like this one provide essential scientifically based information on a coarse scale, conservation actions can only be executed through the joint action of academia, NGOs, local communities and policymakers (Nori et al., 2016;Prieto-Torres et al., 2018). Undoubtedly, biodiversity conservation in human-modified landscapes of Mesoamerica cannot be effectively advanced if it cannot be defined and measured. Thus, the implementation of interdisciplinary and complementary programmes (including vegetation restoration) are crucial to ensure conservation in the region (Whitbeck, 2004;DeClerck et al., 2010;Galloway et al., 2005;Hansen et al., 2008;Janzen, 2000;Portillo-Quintero & Sánchez-Azofeifa, 2010 and future potential species ranges could represent a less costly and more effective strategy for guiding conservation decision-making to maximizing the long-term protection of biota (Hannah et al., 2007;Prieto-Torres et al., 2016;Triviño et al., 2018).
Of concern is the conservation of forest-dependent species that are unable to persist in an agricultural matrix, even when there is significant on-farm tree cover (e.g., Nori et al., 2013). Approaches evaluating conservation status for species in human-modified landscapes, in both spatial and temporal terms, are essential for shedding light on the ecological mechanisms underlying the persistence of wild biodiversity in those areas (Donovan & Strong, 2003;Nori et al., 2013) and the critical roles that species play in local ecosystems (Gardner et al., 2009). The identification of conservation areas for birds that are endemic, threatened or both, as well as areas with a high con- Therefore, large-scale studies would be an important step to guide the establishment of new conservation areas that are efficient for the entire Mesoamerican region.

ACK N OWLED G EM ENTS
Thanks to the curators of the multiple institutions worldwide who have supplied data, access to collections and invaluable logistic support. We also thank Javier Nori to provide the shapefile layer We thank Lynna Kiere, Luis A. Sánchez-González, and two anonymous reviewers for helpful comments on previous versions of the manuscripts.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13153.

DATA AVA I L A B I L I T Y S TAT E M E N T
The authors confirm that the data supporting the findings of this study are available within the article [and/or]