Papua New Guinea terrestrial‐vertebrate richness: elevation matters most for all except reptiles

To examine species richness patterns in Papua New Guinea's terrestrial vertebrates and test for geographical congruence between the four classes, and between lizard and snake subgroups. To assess the environmental correlates of Papua New Guinean terrestrial‐vertebrate richness, and contrast effects of varying analytical resolution and correction for spatial autocorrelation. We predict congruence in the bird, mammal and to a lesser extent amphibian richness, with weak congruence or incongruence between reptiles and the other taxonomic groups. We further predict these patterns will stem from relative or in the case of reptiles dissimilar, correlative trends with environmental predictors such as elevation and temperature.


INTRODUCTION
Tropical regions hold the greatest species richness on the planet (Fine, 2015). Species richness patterns and drivers within tropical regions, however, remain poorly known (Tuomisto et al., 2014). Analyses at continental and global scales can mask regional correlations between species richness and environmental predictors and thus fail to identify the underlying causes of regional spatial-richness patterns (Qian & Ricklefs, 2008). Previous research demonstrated temperature to be an important determinant of species richness in temperate highlatitude regions, with primacy shifting to water availability in tropical and subtropical low-latitude regions (Hawkins et al., 2003). It was suggested that future studies should focus on assessing the interaction between these two richness predictors, as well as other key factors such as environmental heterogeneity (Stein et al., 2014). Furthermore, assessing the determinants of biogeographical patterns across multiple taxonomic groups provides an insight into how these trends vary with differing attributes, such as ecology and dispersal ability (Keith et al., 2012). A comprehensive understanding of the spatial overlap of species richness, or cross-taxon congruence, particularly within tropical regions, is crucial for efficient conservation planning and assessing the impacts of climate change (Grenyer et al., 2006;Qian & Ricklefs, 2008).
Broad-scale global analyses typically reveal similar species richness patterns across vertebrate taxa (Grenyer et al., 2006;Qian & Ricklefs, 2008). However, considerable variation in vertebrate congruence at continental and regional scales has also been observed, particularly between lizards and other taxa (Powney et al., 2010;Lewin et al., 2016). Cross-taxon congruence is expected when taxa respond similarly to environmental determinants or share large-scale diversification, colonization and extinction rates (Qian & Ricklefs, 2008). The spatial scale at which range map-based richness patterns are examined likely influences perceived patterns and drivers of richness (Rahbek, 2005;Hurlbert & Jetz, 2007;Field et al., 2009), especially within the tropics (Jetz et al., 2008). Despite the recognition of scale effects, fine-scale analyses and crosstaxon comparisons of richness determinants are uncommon (Belmaker & Jetz, 2011). While climatic determinants are strongest at coarse scales, processes such as biotic and abiotic filters regulating local community assemblages may provide better explanations for fine-grained richness (Whittaker et al., 2001;Belmaker & Jetz, 2011).
Attempts to explain broad-scale species richness gradients generally focus on predictors such as energy availability, environmental heterogeneity, study area and evolutionary time (Currie, 1991;Rahbek & Graves, 2001;Hawkins et al., 2003;Field et al., 2009). The energy-richness hypothesis (Wright, 1983) proposes that richness is limited by resource availability. Ambient energy is noted as a principle limiting factor for richness at high cold latitudes, with water availability a strong richness determinant at low warm latitudes (Hawkins et al., 2003). Although ambient energy and water availability may directly constrain richness, their interaction is purported to affect distribution patterns indirectly through plant productivity (Field et al., 2009). Plant productivity may influence animal-species richness through increased potential biomass, but it may also be consistent with the resource-speciation hypothesis in providing a greater diversity of resource types supporting more specialist species (Kaspari et al., 2000;Hurlbert, 2004). Tropical regions have the highest rates of net primary productivity (npp; Gillman et al., 2014); a factor thought to cause a trophic cascade of resource abundance, resulting in diversity peaks (Currie, 1991;O'Brien, 1998).
Environmental and topographic heterogeneity (i.e. elevational range) are purported to be among the most important factors influencing species richness (Stein et al., 2014), especially at medium to fine spatial scales (Field et al., 2009). Topographic heterogeneity is thought to increase the number of habitat types and resources available, which in turn increases the potential niche space allowing more species to co-exist (Stein et al., 2014). Topographically heterogeneous regions are thought to be particularly important for narrowly ranging endemics by acting as both cradles and museums of biodiversity (Fjelds a et al., 2012). The cradle hypothesis posits that increased range in topography and climate will create greater potential for speciation by isolation and divergent adaptation. According to the museum hypothesis, refuge potential in elevationally diverse regions is greater during periods of climatic fluctuation, increasing species persistence (Currie, 1991;Tews et al., 2004). Topography is suggested to influence richness most in regions experiencing the greatest climatic stability over the longest period of time, such as those in the tropics (Fine, 2015). Indeed, broad-scale studies have often highlighted the significance of tropical montane regions for species diversity (Orme et al., 2005;Cadena et al., 2011). At large spatial scales, species turnover in montane regions, caused by environmental heterogeneity and niche availability absent in homogenous landscapes, inflates species richness (McCain & Beck, 2016).
New Guinea is among the most biologically diverse regions on the planet (Brooks et al., 2006). The exceedingly rich vertebrate assemblage (c. 5% of the world's terrestrialvertebrate species on less than 0.2% of the Earths land surface) is thought to be the product of the island's large size (785,753 km 2 ), tropical climate, complex geological history and extensive mountain ranges (Allison, 2009;Shearman and Bryan, 2010). New Guinea has a remarkable degree of tectonic complexity (Baldwin et al., 2012), resulting in extensive mountain uplift (over 31% of its mainland area is above 1000 m, Allison, 2009). These extensive and often isolated mountain ranges produce high levels of range-restricted species. The climate of much of New Guinea is weakly seasonal, with temperature variation largely influenced by elevation, although the savanna areas have a distinct dry season and reduced precipitation overall (c. 1000-1500 mm annual rainfall; Allison, 2009). Roughly, two-thirds of the island is covered in broadleaf forests, mostly rain forest, which receive from 2000 to more than 10,000 mm of rainfall annually and lack a pronounced dry season. As the fauna on the eastern side of New Guinea, in the country of Papua New Guinea (PNG), has been more thoroughly surveyed (Heads, 2002), we restrict our study on the terrestrial vertebrates to this region. The many offshore islands that also comprise a significant portion of PNG are not included in our analyses, consequently our use of 'PNG' refers only to the mainland New Guinea portion of that country.
We assess species richness, cross-taxon congruence and environmental correlates of PNG's terrestrial vertebrates. To determine the effect of spatial scale on these relationships, we conduct analyses at three spatial resolutions. We predict that the aseasonal tropical climate will result in the highest terrestrial-vertebrate richness predominantly in montane regions. Sharp climatic clines coupled with decreasing connectivity at higher altitudes, are expected to promote species' elevational segregation (Janzen, 1967;Ghalambor et al., 2006) and increase the number of high-elevation range-restricted local endemics. This trend will be most evident in taxa with comparatively poor dispersal ability, such as amphibians. We further predict that PNG amphibian richness will be less constrained by water availability than observed in other, more water-limited, regions (Buckley & Jetz, 2007). While amphibians' physiological and ecological requirements for water (Feder & Burggren, 1992) will remain fundamentally the same, the extremely high degree of precipitation across most of PNG will reduce its level of constraint across the region. We predict that PNG bird and mammal species richness will be strongly positively correlated, exhibiting concordant trends to the environmental predictors due to the equivalent physiology and energetic requirements of endotherms (Grenyer et al., 2006). We predict that PNG reptiles will show the greatest degree of disparity in richness patterns. Reptiles are strongly constrained by temperature due to their physiology and they fundamentally differ from amphibians in their ability to control water loss with dry and scaly skin (Adolph & Porter, 1993). We predict reptile richness will peak in warm lowland regions, thus the ectothermic groups should display incongruent species richness patterns compared with congruent endothermic richness.

Species geographical range data
We compiled geographical range data for all 232 PNG native mainland reptile species using ArcGIS 10.0, following the taxonomy of Uetz & Ho sek (2015), but including three lizards (Cryptoblepharus novaeguineae, Gehyra baliola and Sphenomorphus anotus) considered valid by the IUCN Melanesian Reptile Working Group, 2014 (see Appendix S1 and S2 in Supporting Information). We obtained reptile range maps from the Bishop Museum Pacific Biological Survey Project and during the IUCN Melanesian Reptile Redlist Project (July, 2014). We converted all point-locality data to polygons using a 10 km 2 buffer (radius = 1.72 km), in accordance with IUCN mapping standards (http://www.amphibians.org/wp-content/uploads/2013/09/Red-List-Mapping-standards-ARLA-Jan2014-web-version.pdf). To assess trends in species richness within reptiles, we divided PNG reptiles into lizards (n = 150) and snakes (n = 68), omitting crocodiles (n = 2) and turtles (n = 12) due to their low species numbers. Following the taxonomy of Frost (2015), we obtained ranges for 217 of the 275 PNG amphibian species from the IUCN (2014) and digitized the remaining 58 maps ourselves (see Appendices S1 and S2). We obtained range maps for PNG's 635 breeding-bird species from BirdLife International and NatureServe (2014) and amended 64 of these according to the taxonomy of Pratt & Beehler (2014; see Appendix S1). We obtained range maps for 232 PNG mammal species, 219 were obtained from IUCN (2012) and 13 maps for recently described species were digitized by us (see Appendices S1 and S3).
Species richness for each terrestrial vertebrate group was obtained by collating range maps per group and combining them with three gridded layers of Papua New Guinea (see Appendix S4). Grid cells were either 25 9 25 km 2 (n = 620), 50 9 50 km 2 (n = 160) or 100 9 100 km 2 (n = 41). Equivalence in grid-cell area was ensured using a Behrmann equalarea projection; wherein grid cells areas were 625, 2500 or 10,000 km 2 . For convenience, we refer to these grid layers as 'quarter-degree grid', 'half-degree grid' and 'one-degree grid', as they approximate to these sizes. We omitted grid cells with < 50% land area to negate potential species-area relationships.

Environmental data
Three climatic predictors (temperature, precipitation and net primary productivity) and one topographic predictor (elevation) were used to derive eight environmental predictors (see Appendix S5). Temperature and precipitation data were obtained from the PNG Resource Information System (3rd Edition; Bryan & Shearman, 2008) and used to produce mean-annual and annual-range measures for both temperature (a proxy for ambient energy availability) and precipitation (a proxy for water availability). Net primary productivity (npp) data were obtained from Imhoff et al. (2004) and used as a proxy for rate of biomass production in primary producers (Hurlbert, 2004). Insufficient variance in npp heterogeneity precluded its use as a richness predictor. Elevational data were obtained from the PNG Resource Information System (3rd Edition; Bryan & Shearman, 2008) and used to create three topography predictors: mean elevation, elevational range (maximumminimum elevation per grid cell) and a 'mid-elevation' predictor (see below). The 'mid-elevation' model predicts that species richness has a unimodal relationship with elevation, thus within a given grid-cell richness will increase with the proportion of area at Papua New Guineas elevational mid-point. The function to simulate this follows the formula: where j represents the total gridded domain of PNG, and i a given cell within the grid. E i represents the mean elevation in cell i, and E jmid is half the difference between the maximum and minimum mean elevation recorded in grid j.

Simulation models
Richness patterns are typically investigated through curve-fitting analysis (Rahbek et al., 2007) and as such cannot model species ranges directly or generate quantitative predictions of species richness (Gotelli et al., 2009). Spreading-dye models (termed 'Geometric constraint models' in Jetz & Rahbek, 2001) were developed as an alternative method for assessing richness gradients in heterogeneous landscapes. These spatially explicit models can simulate speciation, dispersal and extinction of species in heterogeneous landscapes (Gotelli et al., 2009). Spreading-dye models reconstruct distributions by assigning cells to a species' range within the gridded domain. After the initial cell has been selected the range is allowed to expand to adjacent cells, with the cell number in each species' predicted range equivalent to its observed range. The initial grid cell and subsequent adjacent cell selection can either occur stochastically, with all grid cells equal in selection probability, or be weighted according to underlying environmental predictor values across the gridded domain (Storch et al., 2006;Rahbek et al., 2007;Szabo et al., 2009).
To simulate species richness according to our predictors, we used spreading-dye models at a quarter-, half-and onedegree grid scale (Jetz & Rahbek, 2001;Storch et al., 2006). For each terrestrial vertebrate group, we developed nine species richness simulation models, eight simulation models derived from environmental predictors and one null model. The null model specifies a single geographical origin for each species, and allows each range to spread cohesively and randomly within the gridded domain (Gotelli et al., 2009). Simulations were modified to assess the role of environmental gradients on species richness by weighting cells according to the predictor values (Storch et al., 2006;Rahbek et al., 2007;Szabo et al., 2009). The gridded predictors: mean elevation, elevational range, mean annual npp, mean annual temperature, annual temperature range, mean annual precipitation and annual precipitation range were rescaled to between 0 and 1, using minimum and maximum values for each predictor. This ensured that a given proportional change resulted in an equivalent change in cell selection probability across all environmental predictors . The relative probability of initial cell and subsequent adjacent cell selection was determined by variation in the environmental predictor values. Thus, environmental determinism per cell ranged from completely random (i = 0) to strongly deterministic (i = 1). Simulated terrestrial-vertebrate species richness, derived from the nine predictors at three spatial resolutions, was carried out 999 times for each group, and the gridded mean species richness calculated from all models.

Statistical analysis
All statistical analyses were carried out in R 3.3.0 (R Development Core Team, 2016). Cross-taxon congruence among PNG terrestrial-vertebrate group richness was initially assessed using Pearson's correlation coefficient; however, non-independence in the spatial data was detected using Moran's I (see Appendix S6). Spatial autocorrelation arises when neighbouring locality values in spatial data are more similar than expected by chance (Legendre, 1993), leading to increased type-1 error rates and incorrect parameter estimates (Lennon, 2000). To account for spatial autocorrelation, we used Dutilleul's modified t-test (Dutilleul, 1993), provided in the package 'SpatialPack' (Osorio & Vallejos, 2014).
To identify the strongest environmental correlates of each PNG terrestrial vertebrate group, we regressed observed species richness against richness simulated by the eight environmental predictors and the single stochastic predictor. We initially conducted ordinary least squares (OLS) regression, then accounted for spatial autocorrelation in model residuals by using simultaneous autoregressive (SAR) models (Kissling & Carl, 2008), provided in the package 'spdep' (Bivand & Piras, 2015). The spatial structure was specified as the eight nearest neighbours to each cell, and neighbour cells were weighted according to the row-standardized coding scheme. We assessed the relative explanatory power in the OLS and SAR models using Akaike's information criterion (AIC c ).

Species richness
PNG bird and mammal richness have generally concordant patterns at the quarter-degree grid scale, whereas amphibian and reptile richness patterns are distinct from these and from each other (Fig. 1). Amphibian richness (13-55 species per grid cell, mean 27 AE 8 SD; Fig. 1) peaks within the Central Highlands and north-western Bewani and Torricelli mountain ranges, with less pronounced peaks on the south-eastern Owen Stanley mountain range. Lowest amphibian richness occurs in the south-western Trans-Fly, north-western Sepik, and coastal lowland areas. Bird richness reveals a strong affinity to the Papuan peninsula, with greatest richness bordering the Owen Stanley mountain ranges and lowest within the interior of the Central Highlands, south-western Trans-Fly and north-western Sepik lowlands (133-411 species per grid cell, mean: 289 AE 47; Fig. 1). Mammal richness peaks trace the Central and southern Owen Stanley Mountains periphery, with lowest richness in the Trans-Fly and Sepik lowlands (35-118 species per grid cell, mean: 76 AE 21). Reptile richness peaks occur within the lowlands of the Trans-Fly, Huon Gulf and south-central Papuan Peninsula, with richness decreasing towards the major mountain range interiors (10-86 species per grid cell, mean 53 AE 15; Fig. 1). Dividing PNG reptiles into lizards (n = 150) and snakes (n = 68) reveals strikingly different richness peaks (Fig. 1). Lizard richness peaks in northern lowland areas along the Huon Gulf and north-west West Sepik Province, whereas snake richness peaks in the lowlands of the Trans-Fly and south-central Papuan Peninsula.

Cross-taxon congruence
Cross-taxon congruence between PNG amphibian, bird and mammal species richness is positive irrespective of spatial autocorrelation or resolution (Table 1). Mammal richness exhibits the strongest degree of congruence with amphibian and bird richness, across the three spatial resolutions (Pearson's r = 0.61-0.87, P < 0.01). In contrast, once corrected for spatial autocorrelation, reptile richness is not significantly correlated with amphibian, bird or mammal richness at any of the spatial resolutions. Lizard and snake richness largely exhibit no correlation with the other three terrestrial vertebrate groups. Lizard and bird richness are positively correlated at the quarter-degree resolution (Pearson's r = 0.39, P < 0.01); however, this relationship becomes insignificant at the coarser resolutions. A positive correlation between lizard and snake richness, also diminishes with decreasing spatial resolution (Pearson's r = 0.69 [P < 0.01] at the quarter-degree, 0.54 [P < 0.01] at the half-degree and non-significant at the one-degree resolution).

Simulated environmental predictors
Our analyses reveal that elevation provides the best predictor of PNG terrestrial-vertebrate richness (Table 2 and see Appendix S7). Correcting for spatial autocorrelation affected the variance explained by each predictor model. This altered the best-fit predictor models for amphibian and mammal richness at the finest resolutions and for birds at the two coarser resolutions. Spatial resolution similarly affects the best-fit predictor models, primarily between the quarter-and half degree for all terrestrial vertebrates except reptiles. PNG reptile richness exhibits a distinct, negative correlation with mean elevation at quarter-degree (OLS r 2 = 0.48, P < 0.01; SAR Naglekerke r 2 = 0.71, P < 0.01) and half-degree resolutions (OLS r 2 = 0.25, P < 0.01; SAR Naglekerke r 2 = 0.49, P < 0.01). At the quarter-degree resolution, PNG bird richness is strongly correlated with the mid-elevation richness model (OLS r 2 = 0.20, P < 0.01; SAR Naglekerke r 2 = 0.63, P < 0.01). Variance in amphibian richness is best explained by temperature range (OLS r 2 = 0.42, P < 0.01; SAR Naglekerke r 2 = 0.80, P < 0.01), and mammal richness by elevational range (OLS r 2 = 0.76, P < 0.01; SAR Naglekerke r 2 = 0.88, P < 0.01). Further, elevational range best explains the variance in amphibian, bird and mammal richness at both the half-and one-degree resolution, following correction for spatial autocorrelation. The stochastic spreading-dye and net-primary-productivity richness models consistently provide the worst predictors of PNG terrestrial-vertebrate species richness.

DISCUSSION
This study provides a comprehensive assessment of the biogeographical patterns of species richness for terrestrial vertebrates in Papua New Guinea, a remarkably diverse but understudied region. Our analyses reveal congruent richness patterns in all Papua New Guinea terrestrial vertebrates except for reptiles and the constituent lizard and snake groups. Our findings are generally consistent with the crosstaxon congruence observed in most groups at the global scale (Grenyer et al., 2006;Qian & Ricklefs, 2008), and reptiles being outliers within regions such as Australia (Powney et al., 2010), and Africa (Lewin et al., 2016). As in previous studies (Grenyer et al., 2006;Belmaker & Jetz, 2011) spatial resolution influenced the degree of congruence among richness patterns for PNG amphibians, birds and mammals, but had no effect on reptile incongruence. These congruent versus contrasting patterns are evidently linked to the underlying abiotic and biotic factors governing each species' distribution, a trend increasingly apparent at finer spatial scales. Climatic variables are frequently considered to provide the strongest predictors of broad-scale species richness (Hawkins et al., 2003;Field et al., 2009), while other studies have suggested topographic heterogeneity measures may strongly regulate species richness gradients, especially at small spatial scales (Kerr & Packer, 1997;Whittaker et al., 2001;Ricklefs, 2006;Rahbek et al., 2007;Stein et al., 2014). Belmaker & Jetz (2011) noted that measures of global amphibian, bird and mammal richness patterns display contrasting relationships with climate, but similar relationships with topographical heterogeneity. The primacy of topographical heterogeneity in predicting species richness is particularly evident in high-energy regions such as the tropics (Orme et al., 2005;Grenyer et al., 2006;Davies et al., 2007;Kreft & Jetz, 2007;Thomas et al., 2008;Qian, 2010). Our analyses reveal that elevational range strongly correlates with PNG amphibian, bird and mammal richness at coarser spatial resolutions. Furthermore, in accordance with Belmaker & Jetz (2011), we show that all terrestrial vertebrates, except reptiles, display similar topographical-heterogeneity-richness slopes (Table 2). We thus provide evidence for the existence of elevationalbased mechanisms promoting species richness through high levels of persistence and speciation in PNG amphibian, bird and mammal groups. In contrast, reptile richness is strongly constrained by elevation, with peaks in the warmer lowlands, suggesting that these are central regions for reptile diversification and persistence.
Our analyses reveal PNG reptile species richness is strongly constrained by elevation and, as predicted, is positively correlated with temperature (Whittaker et al., 2007;Qian, 2010), probably through direct effects on life history traits . Adaptation to survive in water-limited regions has enabled reptiles to persist in the southern Trans-Fly and Port Moresby savanna areas. Richness peaks in these areas likely reflects the accumulation of numerous Australian reptile species, largely wide-ranging snakes, during the land connections of the last series of glaciations. Except for reptiles, vertebrate species richness in savanna is generally lower than in closed-canopy wet forest, suggesting decreased competition and predation pressure may also enable a greater diversity of reptiles to exist in these areas (Sweet & Pianka, Table 1 Cross-taxon congruence in Papua New Guinea (PNG) terrestrial-vertebrate species richness at the quarter-, half-and onedegree spatial resolution. Cross-taxon congruence using Pearson's r values above the diagonal, with P-values, corrected and uncorrected in parenthesis, below the diagonal. Boldface represents significant Pearson's r values, corrected for spatial autocorrelation. Regular font represents significant Pearson's r values, uncorrected for spatial autocorrelation. Italics script represents insignificant Pearson's r values. Degrees of freedom equate to 618 at the quarterdegree, 118 at the half-degree, and 39 at the one-degree spatial resolution. . The northern reptile richness peaks, dominated by lizards, likely represent important transition zones between once-geographically distinct assemblages and thus contain a high degree of phylogenetic diversity. Although elevation per se is unlikely to cause species richness gradients, topographic heterogeneity likely acts as a surrogate for precise local climatic conditions, such as temperature and precipitation, purportedly strong determinants of species richness (Qian, 2010). Interplay among factors such as slope and aspect will increase the range of climatic conditions in geographically proximate mountains, thus increasing diversity in taxa with slightly different climatic envelopes. Climatic stratification along tropical elevational gradients purportedly limits the elevational extent to which any particular species' physiological profile can function (Janzen, 1967;Ghalambor et al., 2006). Turnover in elevational plant communities led to elevational range being commonly used as a proxy for habitat heterogeneity, with this inferred increased habitat diversity purported to allow for increased sampling of habitat specialists (Rahbek & Graves, 2001;Davies et al., 2007; but see Hortal et al., 2013). It is likely that these processes have been highly influential in promoting amphibian, bird and mammal diversity in PNG's topographically complex regions. In contrast, reptile richness may be limited along mountain slopes by factors such as restricted access to solar radiation caused by increased humidity, vegetation and cloud cover (McCain, 2010). Climatic predictors such as temperature, precipitation and primary productivity, are frequently reported as strong correlates of broad-scale vertebrate richness patterns (Jetz & Rahbek, 2002; but see Adler et al., 2011). However, these predictors consistently had lower effects sizes on PNG terrestrial-vertebrate richness, compared with the topographic predictors (Table 2). In particular, npp performed poorly across all taxonomic groups and analytical resolutions. This indicates that these climatic predictors do not constrain species richness to the same degree as in higher latitude regions. The stochastic spreading-dye model, essentially equivalent to a mid-domain effect which predicts higher richness at the centre of an island or continent (Gotelli et al., 2009), also performed poorly for all taxa, except for reptiles where it was reversed (reptiles being rich at coastal lowland areas), probably reflecting the effects of temperature. Cross-scale variation in predictor primacy and the effect of spatial autocorrelation, highlight the need for careful examination of broad-scale biogeographical data. Failure to do so can affect the study inferences, as illustrated in this study (Table 1 & 2).

Journal of
Environmental predictor effect sizes, in both the OLS and SAR models, for PNG terrestrial vertebrates were lowest in reptiles at the one-degree resolution (Table 2). This indicates that extraneous factors, such as differing affinities to geological terranes, may be influencing reptile richness patterns in PNG. New Guinea's complex geotectonic history and major mountain-building events are purported to be a crucial factor shaping the regions biotic diversity (Georges et al., 2014;Toussaint et al., 2014). In particular, environmental change during orogenesis is likely to have driven diversification in New Guinea (Toussaint et al., 2014). This may explain the high alpha-and beta-diversity of amphibians, birds and mammals in the montane regions (see Appendix S8). Onceisolated terranes that are now accreted along New Guinea's northern coast have also been important allopatric speciation centres, especially for amphibians and reptiles.
We suspect that comparable environmental conditions and biotic diversity across New Guinea will result in equivalent cross-taxon congruence and environmental-richness correlates to those observed by us in PNG. Although the New Guinean biota remains poorly known compared to many other parts of the world, the overall patterns of biotic distribution and richness, particularly within Papua New Guinea, are reasonably well known and are unlikely to change appreciably as new species are discovered and named (Allison & Tallowin, 2015). We minimized distributional errors in PNG species by using expert-drawn range maps, combining both observation and speculative data. Expert-drawn maps are less susceptible to underestimation of true range size compared with point-observation data and do not grossly overestimate a species' range and elevational limits, as seen in extent-ofoccurrence range maps (Hurlbert & Jetz, 2007;Rocchini et al., 2011). Over-estimation of species ranges is particularly influential in narrow ranging species found in low tropical latitudes (Jetz et al., 2008).
The congruent spatial-richness patterns identified among PNG amphibian, bird and mammal species, and the incongruent reptile richness patterns, have important implications for regional conservation. Protected areas targeted towards high species richness in a single taxon are likely to provide coverage for a range of terrestrial-vertebrate species. However, such protected areas would fail to capture areas of high reptile richness. Disparity in reptile species richness, caused by their divergent environmental requirements, suggest that reptile-specific conservation approaches will provide the most effective method for preserving their regional centres of diversity. As reptile diversification centres correspond to important biogeographical regions, such as the Trans-Fly and Port Moresby savanna, conservation efforts targeting reptile hotspots will also have the added benefit of preserving these unique regions. Effective conservation strategies in PNG will ultimately require additional assessment of congruence patterns for range-restricted and threatened-species combined with a geographical understanding of the major threats to terrestrial vertebrates (Shearman & Bryan, 2010). However, these assessments will greatly benefit from further research addressing the taxonomic and distribution shortfalls particularly evident in groups, such as amphibians (Allison, 2009), and reptiles (Meiri, 2016). By revealing the spatial species richness patterns in PNG's terrestrial vertebrates and identifying their strongest environmental correlates, we provide baseline information, which may contribute to conservation planning in PNG.