Biogeographical analyses to facilitate targeted conservation of orchid diversity hotspots in Costa Rica

We conduct a biogeographical assessment of orchids in a global biodiversity hotspot to explore their distribution and occurrences of local hotspots while identifying geographic attributes underpinning diversity patterns. We evaluate habitat characteristics associated with orchid diversity hotspots and make comparisons to other centres of orchid diversity to test for global trends. The ultimate goal was to identify an overall set of parameters that effectively characterize critical habitats to target in local and global orchid conservation efforts.

The Mesoamerican global biodiversity hotspot has been highlighted for many reasons, including its exceptionally large number of endemic plants (Myers et al., 2000). Research shows, however, that species richness is unevenly distributed within global hotspots, and local priorities need to be established (Cañadas et al., 2014;Harris, Jenkins, & Pimm, 2005;Médail & Quézel, 1997, 1999Murray-Smith et al., 2009). Studies also indicate there is little correspondence between hotspots of various taxonomic groups (Prendergast, Quinn, Lawton, Eversham, & Gibbons, 1993;Reid, 1998). Consequently, detailed analyses of specific taxonomic groups at finer geographic scales are necessary to identify conservation priorities within global diversity hotspots.
The overall objective of this study was to investigate the spatial distribution and geographic affinities of orchids in Costa Rica to identify underlying factors driving diversity patterns. We aimed to answer the following questions: 1) How are orchids distributed in Costa Rica and where do diversity hotspots occur? 2) What geographic variables most strongly influence the distribution of orchid hotspots? 3) Where are the most suitable habitats for potentially undocumented orchid diversity hotspots in Costa Rica?
We hypothesize that orchid diversity patterns are linked to specific geographic conditions and that suitable habitats and potentially undocumented hotspots exist outside of currently protected areas.
Results from this analysis will improve our understanding of fundamental biogeographical patterns in Costa Rica, Mesoamerica and other biodiversity hotspots. They will also support calls to identify potential Important Orchid Areas (IOAs), that is sites with exceptionally rich orchid floras and/or important habitats for their conservation (Seaton, Kendon, Pritchard, Puspitaningtyas, & Marks, 2013), much like Important Plant Areas (IPAs; Anderson, 2002;Foster et al., 2012). Ultimately, our results will advance conservation efforts by pinpointing specific targets for protecting an important component of global plant biodiversity.

| Study site
Costa Rica has an exceptional orchid flora that is well studied (Hammel-Lierheimer et al., 2003). The country is renowned for its commitment to orchid biology and conservation, and extensive resources for documenting orchid diversity and distributions are available (Blanco, Pupulin, & Warner, 2005;Bogarín, Pupulin, Arrocha, & Warner, 2013;Pupulin, 2007;Rivero, 1998). Consequently, the Costa Rican orchid flora represents a unique opportunity to evaluate biogeographical affinities of the family within a global diversity hotspot.

| Mapping orchid occurrences and richness hotspots
To map orchid distribution patterns in Costa Rica, we used collection records from the National Museum of Costa Rica (CR), the Lankester Botanical Garden (JBL) and the National Biodiversity Institute (INB) herbaria, which house the largest collections in the country (Thiers, 2015). We amassed distribution data from 12,316 records representing 1,320 orchid species. Geographic coordinates from each record were imported into ArcGIS 10.1 (Environmental Systems Research Institute, 2011) to create a point layer showing the distribution of orchids throughout the country. A network of 1-km 2 grid cells spanning the country was created, and the number of species in each cell was quantified. The resulting layer was used as the input feature class K E Y W O R D S biodiversity hotspot, Costa Rica, critical habitat, distribution mapping, orchids (Orchidaceae), plant conservation, random forest model, spatial ecology, species richness, weighted sum model in the hotspot analysis (Getis-Ord Gi*) tool to identify hotspots of orchid diversity, defined as grid cells with z-scores ≥ 1.96 and associated p ≤ 0.05 (Environmental Systems Research Institute, 2014; see Appendix S1 in Supporting Information for detailed methodology).

| Characterizing geographic attributes of orchid hotspots
To quantify geographic attributes associated with orchid hotspots in Costa Rica, we used GIS data layers for environmental and geopolitical variables characterizing the country (Ortiz-Malavasi, 2009; Table 1). Environmentally, we focused on abiotic and biotic variables that have known potential to influence orchid communities and their distribution. We considered biotemperature and precipitation because they have been shown to be limiting factors for several orchid species physiologically and ecologically (Crain & Tremblay, 2017;Liu et al., 2010;Vollering, Schuiteman, Vogel, Vugt, & Raes, 2015). Orchids can also be bound to particular forest levels (e.g. lowland or montane), terrain characteristics (i.e. slopes) and elevation ranges, each of which was included in our models, due to potential effects on microclimates, disturbance regimes and breeding systems (Jacquemyn, Micheneau, Roberts, & Pailler, 2005;Pupulin, 1998;Vollering et al., 2015;Whitman, Medler, Randriamanindry, & Rabakonandrianina, 2011).
The Holdridge life zone system incorporates measures of biotemperature, precipitation and evapotranspiration to predict climax vegetation patterns (Holdridge, 1947), and thus, we included this attribute as an additional predictor variable. Soil attributes were accounted for in our models because they can influence the distribution of terrestrial orchids and their mycorrhizal associates (e.g. McCormick et al., 2012;Phillips, Brown, Dixon, & Hopper, 2010). Soils may also influence the distribution of potential host trees, vegetation types and habitats that are suited to epiphytic orchids (Gentry & Dodson, 1987). The final environmental variables we considered were vegetation cover and land use/land cover, as orchids can be limited to specific vegetation types, such as primary forests (Pupulin, 1998), and may be differently adapted or susceptible to various land uses, for example, agriculture or logging (Sosa & Platas, 1998).
Geopolitically, we focused on two government jurisdiction levels (province and county, i.e. cantón) to determine whether regional and local jurisdictions differ in terms of orchid diversity patterns that rise from differences in conservation priorities at the regional versus local scales (Porras, 2010;Pressey, Mills, Weeks, & Day, 2013). We also focused on attributes of the protected areas in the country and considered potential differences in nine regional conservation areas, the different types of protected areas (e.g. national parks or wildlife areas) and the individual protected areas, which can differ in terms of regulations and management (Evans, 1999;Sistema Nacional de Áreas de Conservación (SINAC), 2012. The size of each protected area was the final geopolitical variable considered in our models, as it is a known correlate of orchid species richness estimates (Schödelbauerová, Roberts, & Kindlmann, 2009).

| Modelling geographic suitability for orchid hotspots
To identify locations in Costa Rica with the greatest potential to support undocumented hotspots of orchid diversity based on habitat suitability, we quantified the proportion of each variable value's distribution that was identified as an orchid diversity hotspot. Variable values with larger proportions of their distribution corresponding with hotspots were considered more characteristic of potentially suitable habitat (Appendix S2.2).
We evaluated the overall importance of each geographic attribute for explaining the distribution of orchid hotspots by running a series of nonparametric random forest regression models (Breiman, 2001;Cutler et al., 2007;Liaw & Wiener, 2002;Strobl, Hothorn, & Zeileis, 2009; Appendix S1). Random forest models have been increasingly utilized in ecological applications and can offer several improvements over more traditional models (Cutler et al., 2007).
We combined the habitat suitability measures and the importance estimates of each geographic attribute in a weighted sum model (e.g. Ananda & Herath, 2009;Phua & Minowa, 2005;Store & Jokimäki, 2003) to quantify the geographic suitability (Suitability i ) of each grid cell in our map (Appendix S1). We used the suitability values from the weighted sum model to generate a new map documenting locations most likely to support orchid diversity hotspots based on their geographic attributes.

| Orchid occurrences in Costa Rica
Our distribution map revealed that orchids have been documented in 2,680 1-km 2 grid cells (5.1% of the country; Figure 1). Cells with the greatest number of species (≥100 spp; n = 2) occurred in the Cordillera de Tilarán. Cells with 50 to 99 species (n = 9) were also documented in the Cordillera de Tilarán, as well as in the Cordillera de Guanacaste, in the Cordillera Central and in the northern parts of F I G U R E 1 Distribution of orchid diversity in Costa Rica based on number of species present in 1-km 2 grid cells. Larger markers indicate a greater abundance of orchid species in a given cell. Numbered arrows indicate areas where cells with the greatest number of orchid species are located: 1) Cordillera de Tilarán, 2) Cordillera de Guanacaste, 3) Cordillera Central, 4) northern Cordillera de Talamanca, and areas where orchids were documented less frequently: 5) Nicoya Peninsula, 6) Guanacaste Lowlands, 7) San Carlos Plains, 8) Caribbean Lowlands, 9) central Cordillera de Talamanca and 10) Golfito and Punta Burica region the Cordillera de Talamanca. Orchid species occurred less frequently on the Nicoya Peninsula and in the lowlands of Guanacaste in northwest Costa Rica, the San Carlos Plains in the far north, the Caribbean Lowlands, the central portion of the Cordillera de Talamanca and the extreme southern tip of the country between Golfito and Punta Burica, although inaccessibility and limitations to collecting may play roles in the latter two regions. Overall, the distribution map suggests that Costa Rican orchids occur primarily in centrally located mountainous regions of the country, but with notable exceptions.

| Hotspots of orchid diversity in Costa Rica
The hotspot analysis identified 5,684 grid cells, representing 10.8% of the country, as significant hotspots of orchid diversity (p ≤ .05;

| Geographic attributes of orchid hotspots in Costa Rica
Linking geographic attributes to known orchid diversity hotspots highlighted the most frequently associated environmental and geo-   Geographically suitable areas were widespread, but the majority of the most suitable areas were in mountainous regions (Figure 4). proportions of cells predicted to be highly suitable for orchid diversity hotspots were not identified as such by our model based on collection records.

| Orchid distribution patterns
As a vital component of plant diversity in the Mesoamerican global biodiversity hotspot, it is unsurprising that Orchidaceae, one of the largest and most widespread plant families (Chase et al., 2015;Dressler, 2005), was well represented throughout Costa Rica ( Figure 1). Like other hotspots, Costa Rica is geographically heterogeneous, having unique habitats along two coasts and among four main mountain ranges (Burger, 1980;Janzen, 1983;Kirby, 2011

| Orchid diversity hotspots and Important Orchid Areas
While orchids occurred throughout most of the country, distinct hotspots were identifiable. This finding can facilitate the identification of potential IOAs (Seaton et al., 2013) following general criteria established for IPAs (Anderson, 2002) by highlighting sites with exceptionally rich orchid floras and/or important habitats for their conservation. From a conservation perspective, the distribution of orchid hotspots is encouraging since the majority corresponded with protected areas. Hotspots not corresponding with protected areas represent important conservation targets, however (Seaton, 2007), particularly the 50 cells that are among the richest hotspots.
A large proportion of unprotected hotspots are adjacent to existing F I G U R E 4 Distribution of geographically suitable areas for orchid diversity hotspots in Costa Rica based on geographic attributes, including environmental and geopolitical variables (Table 1). Suitability measures of all 1-km 2 grid cells (Suitability i ) were derived from a weighted sum model integrating the performance measures of individual variable values in each cell (a ij ) and the corresponding variable weights (w j ). Numbered arrows indicate areas with large clusters of the most suitable habitats: 1) central Costa Rica including Cordillera Central and northern Cordillera de Talamanca, 2) Turrubares area, 3) Puriscal area, 4) Cordillera de Tilarán, 5) Cordillera de Guanacaste and 6) southern Cordillera de Talamanca protected areas, and small expansions of these preserves could help protect them. The remaining hotspots could be prioritized as new protected areas are established. Because orchid hotspots frequently correspond with diversity centres for other species groups, protection of these locations would provide ancillary benefits for biodiversity conservation beyond orchids Anderson, Cherrington, Tremblay-Boyer, Flores, & Sempris, 2008;Kohlmann et al., 2010;Kohlmann, Solis, Elle, Soto, & Russo, 2007;Seaton, 2007).

| Underlying geography of orchid hotspots
Analysing geographic attributes associated with orchid diversity hotspots enabled us to identify the most important environmental and geopolitical features supporting orchid diversity in Costa Rica.
These geographic attributes were comparable to those highlighted in studies of other locations and species groups.
These settings provide a variety of environmental conditions, and habitat heterogeneity is commonly cited as an important driver of diversity (Crain & White, 2013;Zhang, Yan, et al., 2015). Studies on orchid distribution and diversification in Central and South America theorize that recent geologic events, including rapid growth of mountain ranges, volcanic activity and glaciation at high elevations, are primary drivers of orchid evolution and speciation (Dodson, 2003;Kirby, 2011). Our results showing peak diversity levels in mountainous areas supported this assertion. Studies on other tropical plant and animal communities have produced similar findings (Gentry & Dodson, 1987;Kluge, Kessler, & Dunn, 2006;Kohlmann et al., 2007Kohlmann et al., , 2010Krömer et al., 2005;McCain, 2004). There were exceptions, however, as some hotspots occurred at coastal lowland sites (Figure 2), suggesting various local geographic attributes can support orchid diversity and must be accounted for in conservation planning.
Precipitation levels and temperature ranges associated with orchid hotspots in Costa Rica were comparable to those characterizing hotspots documented elsewhere (Acharya et al., 2011;Zhang, Chen, et al., 2015). Large varieties of epiphytic plant groups achieve maximum diversity levels under similar conditions (Gentry & Dodson, 1987). It is suggested that optimal water-energy dynamics obtained under cool moist conditions promote high biodiversity levels and our results fit this claim (O'Brien, 2006). Furthermore, the cool moist conditions associated with hotspots in Costa Rica are similar to those promoting inflorescence and flower development in some orchids (Blanchard & Runkle, 2006), thus providing another possible explanation for their abundance at those sites.
Soil characteristics were not as strongly linked with the distribution of orchid hotspots, which may be related to the primarily epiphytic habit of most orchids, but interesting patterns still emerged.
Orchid hotspots were generally found in moist, weathered, mineral soils that are relatively infertile. While some evidence indicates that epiphyte diversity is generally higher on fertile soils (Gentry & Dodson, 1987), our results did not support this since hotspots were uncommon or absent in the most fertile soil types in the country, for example, Mollisols, Alfisols and Entisols. Soil characteristics might affect the distribution of potential host trees and mycorrhizal fungi, however (Gentry & Dodson, 1987;Phillips et al., 2010), and these factors might explain the connection between orchid hotspots and soil types in Costa Rica. The fact that soil order and suborder were significant variables in our regression model suggests that soil characteristics are important attributes related to the distribution of orchids and should be considered when evaluating prospective habitats.
Concerning land use/land cover types as well as vegetation cover in Costa Rica, our results support documented trends from other locations and organismal groups. Orchid hotspots most frequently coincided with natural primary forests, whose value to orchid richness and overall biodiversity has been repeatedly recognized (Barthlott, Schmit-Neuerburg, Nieder, & Engwald, 2001;Gibson et al., 2011;Hundera et al., 2013). Nevertheless, a number of hotspots corresponded with less pristine habitats, including coffee plantations. This is not entirely surprising, however, as cool, humid, middle-elevation sites that are prime habitats for many orchids are also ideal for coffee production (Moguel & Toledo, 1999;Perfecto, Rice, Greenberg, & Van der Voort, 1996). Some studies even show that populations of orchids, epiphytes and other species can be well maintained in coffee plantations (Moguel & Toledo, 1999;Solis-Montero, Flores-Palacios, & Cruz-Angón, 2005;Sosa & Platas, 1998;Williams-Linera, Sosa, & Platas, 1995). Still, other studies show that diversity of orchids and other species in coffee plantations is variable and largely dependent on specific management strategies; intensively, farmed areas are generally less diverse (Hundera et al., 2013;Philpott et al., 2008). Consequently, research on coffee plantations and orchid diversity in Costa Rica is warranted to determine whether richness levels are a result of specific management strategies maintaining crucial properties of natural forests.
Geopolitically, our results support findings of other biodiversity studies in Costa Rica (see Janzen, 1983;Nadkarni & Wheelwright, 2000). In most cases, the highest diversity levels were associated with existing protected areas, particularly within the Central Volcanic and Arenal-Tilarán Conservation Areas, which encompass large portions of diverse habitats in the Cordillera de Tilarán and the Cordillera Central. Several national parks exist in these regions, and of the various protected area types in Costa Rica, national parks support the largest proportion of orchid hotspots. Although this pattern was somewhat expected since the national park system in Costa Rica includes many of the oldest and largest protected areas with strict conservation regulations (Brüggemann, 1997), it highlights the importance of these parks. Large proportions of protected zones were also occupied by orchid hotspots, and they could be targeted for increased attention and regulation concerning orchid conservation.
While correspondence between hotspots and protected areas highlights the significance of conservation efforts in Costa Rica, it may also be indicative of species and habitat losses outside these safe havens. The majority of orchid hotspots occurred in provinces with the highest population densities, which may point to high threat levels for hotspots outside protected areas (Rosero-Bixby & Palloni, 1998;Sánchez-Azofeifa, Harriss, & Skole, 2001). Consequently, unprotected hotspots near encroaching development and high population centres need attention from conservation planners. Expanding existing reserves while establishing smaller ones in strategic areas would be viable options for protecting additional orchid diversity centres.

| Geographic suitability and orchid conservation applications
Our geographic models suggest environmental and geopolitical variables are valuable predictors of orchid diversity in Costa Rica.
Habitat suitability estimates accurately coincided with the majority of known hotspots. Thus, our modelling approaches offer a useful framework for evaluating suitable habitats for orchid hotspots. with findings from other areas that have been evaluated (Acharya et al., 2011;Liu et al., 2015;Müller et al., 2003;Vollering et al., 2015;Zhang, Chen, et al., 2015), and it appears that models incorporating similar geographic attributes will be fitting for predicting distributions of orchid diversity hotspots in other areas.
Identifying connections between geographic attributes and orchid hotspots will ultimately benefit conservation efforts in many ways.
Geographic suitability models will help optimize efforts to identify and protect undocumented hotspots. Geographically suitable locations not yet identified as hotspots could be targeted for floristic surveys and analysis. If seemingly ideal habitats do not support hotspots, it may point to other factors limiting orchid diversity, such as inadequate pollinator populations or mycorrhizal associates, that must be addressed (Reiter et al., 2016). In either case, habitat suitability mapping should help streamline orchid conservation efforts.
Understanding links between orchid hotspots and geographic characteristics should also prove useful for implementing translocation or reintroduction projects. Geographically suitable areas may offer important alternatives for mitigating habitat losses or disturbances in other parts of orchid species' ranges (Crain & Tremblay, 2014;Reiter et al., 2016;Seaton et al., 2013). Early identification of suitable areas may also be useful for assisted migration efforts for combating range shifts due to global climate change (Anderson, Cherrington, Tremblay-Boyer, et al., 2008;Crain & Tremblay, 2017;Liu et al., 2010;Reiter et al., 2016;Seaton, Hu, Perner, & Pritchard, 2010). With the aid of habitat suitability estimates, these conservation measures are more likely to succeed because appropriately chosen translocation sites will reflect orchids' natural habitats and may be more likely to support pollinators or mycorrhizal symbionts (Reiter et al., 2016).

| Caveats and considerations for future research
While our study presents a useful framework for analysing biogeographical underpinnings of orchid diversity and the results should provide valuable information for advancing the development of systematic targeted conservation plans in Costa Rica, there are several caveats to consider when interpreting geographic analyses of biodiversity over larger regional scales. It is important to note that while diversity hotspots may represent efficient targets for conservation efforts, not all orchid species will be distributed in these locales (Kareiva & Marvier, 2003). Orchids occurring outside of hotspots, particularly rare species, must not be neglected as they are likely to have unique attributes and/or ecological significance in their own settings. Additionally, our analysis makes use of orchid diversity estimates within each individual grid cell to identify hotspots for conservation targets, but does not explore beta diversity across the collective network of hotspots, which will be an important consideration for conservation decision-makers and a worthwhile extension of this study (Socolar, Gilroy, Kunin, & Edwards, 2016). Furthermore, our analysis does not include an assessment of phylogenetic diversity or conservation status (e.g. rarity or threat) of orchids, which will be important considerations for providing optimal benefits for preserving various components of overall orchid diversity within the hotspot network (Crain & Tremblay, 2014;Swenson, 2011). A final caveat of biodiversity research using GIS to look at species and habitat distributions over larger spatial scales is that species' distributions and actual on-site habitat conditions do not necessarily equate with model predictions, particularly at very small scales. Accordingly, systematic assessments of species' statuses and actual habitat conditions within individual hotspots should be conducted through targeted field surveys and/or detailed, remotely sensed data to confirm the suitability and significance of individual hotspots identified in our analyses.

| CON CLUS IONS
Understanding the biogeography of orchids in Costa Rica is a crucial step towards the conservation of this important component of biodiversity. The world's orchid flora is incredibly unique and widespread, and hotspots warrant significant attention from scientists, conservation agencies and local stakeholders. By analysing geographic attributes associated with orchid hotspots, including environmental and geopolitical variables, critical factors related to their distribution can be identified and explored to understand ecological relationships, identify suitable habitats and document conservation targets. Our findings highlight the importance of analysing the complex geographic patterns and ecological underpinnings of orchid diversity hotspots and will help researchers and conservation practitioners take necessary steps towards preserving orchids in the most important centres of biological diversity.

ACK N OWLED G EM ENTS
We sincerely thank our families and friends for their ongoing support. We gratefully recognize the staff from the Costa Rican National Herbarium, the Lankester Botanical Garden (LBG) herbarium and the National Biodiversity Institute (INBio) herbarium, for providing access to their orchid databases. We also thank the research staff at LBG, without whom this research would not be possible. We convey our special thanks to Carlomagno Soto and staff at the La Selva Biological Station for technical and logistic support. We also acknowledge the editors and reviewers who provided comments that greatly strengthened the overall manuscript. This project was funded in part by the Organization for Tropical Studies, the Department of Biology at the University of Puerto Rico (UPR) and the Decanato de Estudios Graduados e Investigación at UPR.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from