We focused on the region spanning from the islands in the northern and central Moluccas, across New Guinea to the Bismarck and Solomon Archipelago and refer to it as “the New Guinea region”. We downloaded GenBank-deposited mitochondrial DNA (mtDNA) sequences of frog species native to this region and compared the taxonomic and geographic distribution of the most commonly sequenced loci (COI, cytb, 12S and two non-overlapping regions of 16S rRNA, here referred as “16Sa” and “16Sb”). In addition, we downloaded the data of Arida et al. (2021) from the Barcoding of Life Database (BOLD) in order to increase geographic sampling. A summary of the genetic dataset is presented in Table 1 and the full list of sequences is available in the Suppl. material 2. Sequences were retrieved from GenBank using the filter tool, specifying genera occurring in our study region (based on the distributional data of the International Union for Conservation of Nature, IUCN, https://www.iucnredlist.org) and targeted loci. These data were further filtered to remove species that do not occur in the study region. Coordinates for sampling localities were directly retrieved from publications by contacting the authors or were georeferenced with a ~ 10 km precision using Google Earth and Mapcarta (https://mapcarta.com/) when only locality names were available.

Given the unevenness of taxonomic and geographic sampling across loci, we delimited MOTUs using the 16Sb locus only. To expand our sampling, we produced 534 new 16Sb sequences (DNA extraction and sequencing protocol in Suppl. material 1: appendix S1) from specimens collected in western New Guinea during field work undertaken over the last 20 years, including 160 (out of the 167) specimens that were included in Arida et al. (2021). The sequences were aligned using MAFFT (version 7) (Katoh et al. 2019) following the E-INS-i method. MOTUs were delimited using three complementary single-locus methods: Automatic Barcode Gap Discovery (ABGD; Puillandre et al. (2012)); 2) single-rate Poisson Tree Processes (PTP; Zhang et al. (2013)); and 3) the Generalised Mixed Yule Coalescent approach (GMYC; Pons et al. (2006); Monaghan et al. (2009)); Briefly, 1) ABGD infers an intra- and interspecific threshold based on genetic distances, 2) PTP infers species boundaries from the number of substitutions on the branches of a Maximum-Likelihood phylogenetic tree and 3) GMYC uses an ultrametric tree to infer a temporal threshold delimiting species (each partitioning method is detailed in Suppl. material 1: appendix S2). For all methods, four species were used as outgroups: Heleophryne purcelli, Leiopelma hochstetteri, Pipa pipa and Rhinophrynus dorsalis. The final delimitation was based on a majority-rule consensus across the three partitions obtained (Dellicour and Flot 2018), i.e. defining as an MOTU any lineage supported by at least two of the three methods (Fig. 1; see the Suppl. material 2 for all delimitations’ results). We followed the nomenclature of Frost (2024) and assigned taxon names to MOTUs represented by type specimens or to those geographically closest to the type locality in case of conflicting initial identifications.

The robustness of the delimitations was further assessed for 38 MOTUs (i.e. 19 pairs) by searching for concordant patterns of acoustic differentiation (e.g. Fig. 1) using both published and newly acquired audio recordings (see Suppl. material 1: appendix S3 for data availability). These focal species complexes provided an overall evaluation of the robustness of the delimitation otherwise based on mtDNA only.

We investigated call variation following a note-centred approach (Köhler et al. 2017) with particular focus on: 1) note duration; 2) inter-note duration; 3) note repetition rate per second; 4) number of notes per call; 5) number of pulses per note; 6) amplitude modulation (i.e. increasing or decreasing amplitude within a note); 7) dominant frequency; and 8) note frequency modulation (absence, upward or downward). We conservatively considered slight variations in dominant frequency (7), note and inter-note duration (1 and 2) between allopatric MOTUs to be non-diagnostic, due to potential variation inherent to specimens’ body size and recording environment temperature (Gerhardt and Huber 2002). On the other hand, non-overlapping numbers of notes per call (4), note amplitude modulation (6) and note frequency modulation (8) were considered as supporting the delimitations regardless of whether the MOTUs occurred in sympatry or allopatry. All comparisons are summarised in Suppl. material 1: table S1 and individually presented in Suppl. material 1: figs S1–S6.

Figure 1. 

Species delimitation combining molecular and acoustic data, exemplified in an Oreophryne species complex. In this example, the species delimitation between O. roedeli and O. cf. roedeli 2 is confirmed by their difference in note repetition rate and mean number of pulses per note. The subtree was extracted from the ultrametric tree used for the GMYC analysis. Mean genetic p-distance (%) between lineages is indicated at the nodes. Dots above tree branches indicate posterior probabilities > 0.95. All detailed comparisons and specimen vouchers are presented in Suppl. material 1: appendix S3. Call data for O. cf. roedeli 1. and O. cf. roedeli 3 were unavailable.

We undertook additional analyses in order to better characterise potential geographic bias in the distribution of: 1) genetic sampling and 2) MOTUs across the fifteen subregions (Fig. 2) that are representative of the geographic and environmental diversity within our study region (justifications in Suppl. material 1: appendix S4). MOTU richness and endemism per subregion were compared to estimates of species richness derived from IUCN species-distributional data (IUCN, https://www.iucnredlist.org). These distributional data are based on minimum-convex polygons defined for each MOTU/species (i.e. linking all outermost occurrence points), regardless of the potential non-suitability of habitat within polygons. Richness was defined as the number of polygons intersecting each subregion or as simple presence in the case of MOTUs with fewer than two occurrences. Endemism was defined as the proportion of MOTUs/species uniquely found in each subregion. All maps were generated on QGIS v.3.20 (QGIS Development Team 2020).

Figure 2. 

Distribution of sampled localities for all compiled georeferenced mtDNA sequences (12S, 16Sa, 16Sb, COI, cytb) of amphibians across the focal area, with the number of sequences per subregion indicated by grey shading. The vertical white-dashed line represents the international border between the western half (Indonesia) and the eastern half (Papua New Guinea) of the island of New Guinea. Abbreviations: CRW, western central range; CRC, central central range; CRE, eastern central range.

The complete dataset of mitochondrial data comprised 2301 DNA sequences (1767 from online repositories and 534 newly generated) (Table 1, Fig. 2, Suppl. material 1: fig. S7), representing 1537 specimens (out of which 1380 were georeferenced) distributed across 226 unique sampling localities (i.e. within a 10 km radius) and including six families: Microhylidae, Hylidae, Ceratobatrachidae, Ranidae, Myobatrachidae and Dicroglossidae. When considering all families and online data only, more sequences were available for the 12S locus, but we chose to focus on the 16Sb locus for the rest of our study because it had the most comprehensive taxonomic and geographic sampling (Table 1; Suppl. material 1: fig. S7A). Combining GenBank and newly acquired sequences, 16Sb is the best-sampled locus for all families (35% of all sequences), followed by 12S (21%), 16Sa (20%) and cytb (16%). Microhylidae was the most-represented family with 1523 sequences, followed by Hylidae (353) and Ceratobatrachidae (298). It is noteworthy that more than half of the Microhylidae sequences corresponded to three genera only (Mantophryne, Hylophorbus and Choerophryne) that have been the focus of specific studies using 12S, 16S (a and b) and cytb (Oliver et al. 2013; Oliver et al. 2017; Ferreira et al. 2023).

The majority of the georeferenced sequences are from the eastern half of New Guinea (1177 sequences; Fig. 2), mainly from the Papuan Peninsula (367 sequences, 32 unique localities) and the eastern central range (CRE) (274 sequences, 37 unique localities). Conversely, in the western half of New Guinea and its surrounding islands (963 sequences), most sequences were from specimens collected in the Bird’s Neck and Bird’s Head subregions (393 and 312 sequences from 27 and 11 unique localities, respectively). Across subregions, microhylids are best represented in the dataset (175, 342, 259 and 145 sequences in the eastern central range, Papuan Peninsula, Bird’s Neck and Bird’s Head, respectively).

Most recognised species of Ceratobatrachidae and Ranidae in our study region are represented by molecular data (69% and 86% of 58 and 15 recognised species, respectively). By contrast, molecular data are available for only half of the recognised species in Microhylidae and Hylidae (Table 1). With the exception of Aphantophryne and Callulops, species-poor genera (sensu Frost (2024)) tend to have the highest taxonomic coverage: Mantophryne, Paedophryne, Barygenys and Hylophorbus (78–100%).

The number of “species” delimited by the different methods ranged from 263 (PTP) to nearly 400 (386 (ABGD), 371 (GMYC)), with a majority-rule consensus of 372 MOTUs in our focal area. This count was reduced to 369 with our integrative approach, comprising 179 species recognised by current taxonomy and 190 MOTUs that could not be assigned to any taxon (i.e., unidentified or undescribed candidate species). Microhylidae are dominant in our dataset in every measure: total number of sequences (Table 1), all 16Sb sequences (Fig. 3A), MOTU counts and proportion of candidate species (Fig. 3B).

Acoustic data were available for 38 MOTUs (Suppl. material 1: appendix S3, table S1), amongst which 30 were supported as distinct species, with 11 recognised and 19 (63%) unrecognised species. The mean p-distance between the MOTUs confirmed as valid species by acoustic data is 4.0% (1.0–13.0%) for Microhylidae and 2.7% (2.0–3.0%) for Ceratobatrachidae; limited bioacoustic data prevented comparisons between related MOTUs in other families.

Adding the 560 recognised species known to be resident in our study region to our 190 candidate species suggests that a minimum of 750 species inhabit the region. Most candidate species (123) originate from the western half of New Guinea where 215 species are currently recognised. Based on the ratio between the candidate species (190) and the recognised species included in our dataset (179; thus ratio = 1.1) (Fig. 3B), we can extrapolate the total number of species remaining to be described from the number of recognised species (560) if we assume: 1) that all candidate species are confirmed; 2) that we have evenly sampled total taxonomic diversity; and 3) that taxonomic gaps are evenly distributed across the region. Extrapolated to all groups, there would be a total of 1176 frog species across New Guinea and its neighbouring islands (Fig. 3C), of which 616 would be unrecognised at present (560*1.1 = 616). However, candidate species richness is unbalanced across families, being mostly driven by Microhylidae, with 1.5 times more candidate species than recognised species (97 vs. 68). The same extrapolation in this family alone would lead to 501 unrecognised species in addition to the current 334 recognised species.

The Bird’s Head and Bird’s Neck subregions contain the highest number of MOTUs (75 and 61 respectively; Fig. 3D), MOTUs assigned to recognised species (24 and 34, respectively), 16Sb sequences (223 and 175, Suppl. material 1: fig. S7) and are occupied by species of all families, except Dicroglossidae. By contrast, the western central range is represented by no 16Sb sequences at all, but one 12S, two 16Sa and three cytb sequences (Suppl. material 1: fig. S7A).

Figure 3. 

A) Ultrametric tree used for the GMYC analysis, based on all 16Sb sequences. Numbers in parentheses indicate the number of sequences. The detailed tree is presented in Suppl. material 1: fig. S8; B) Final number of recognised species and candidate species amongst MOTUs across families; C) Global number of recognised species, candidate species from this study and extrapolation of unrecognised species per family. The extrapolation is based on the proportion of putative candidate species in B (details in Results: Species delimitation); D) Total number of MOTUs per subregion (grey shading). Pie charts indicate the proportion of each family in the final MOTUs delimitation (colours presented in A), for the eight richest subregions. Abbreviations: Mic, Microhylidae; Hyl, Hylidae; Cer, Ceratobatrachidae; Ran, Ranidae; Myo, Myobatrachidae; Dic, Dicroglossidae.

Out of the 369 delimited MOTUs, 294 are represented by one or two occurrence points and 75 by three or more occurrence points. The evaluation of our taxonomic sampling across subregions, based on the richness of MOTUs assigned to recognised species vs. richness derived from IUCN data (Fig. 4A), revealed that the Papuan Peninsula and eastern central range were the most deficient, with 151 and 126 recognised species missing in the MOTU dataset, respectively (Fig. 4B). This deficit was less pronounced in the Solomon Islands, Seram and Halmahera (two, four and eight recognised species missing) and the Bird’s Neck and Bird’s Head (21 and 25 recognised species missing). When comparing species richness between IUCN data and all MOTUs regardless of their correspondence to recognised or candidate species, the molecular dataset missed species the most in eastern New Guinea (Fig. 4C). In contrast, the westernmost peninsula of New Guinea and the Island of Halmahera displayed levels of species diversity higher than those published by IUCN, with the Bird’s Head in particular having an excess of 18 MOTUs over the number of taxonomically known species (Fig. 4C).

When looking at the degree of endemism per subregion, distributional data for both MOTUs and IUCN range maps suggested higher levels of endemism in archipelagos (e.g. Bismarck and Solomon Isl.) and in the eastern half of New Guinea (Suppl. material 1: fig. S9). However, estimates of endemism rates for MOTUs were much more homogeneous across subregions and higher on average (85%) than those obtained for species in the IUCN dataset (35%).

Figure 4. 

A) Regional species richness based on IUCN data; B) Number of missing recognised species in the MOTU dataset based on the difference between the count of MOTUs assigned to recognised species and IUCN data presented in A; C) Difference in species richness, between the total MOTUs dataset and IUCN data. Negative values indicate a species-deficit in the DNA-based dataset, as compared to IUCN data.

Although DNA sequencing has been increasingly accessible, with the rate of publication of new sequences rocketing over the last two decades (Smith 2016), the New Guinea region still remains poorly represented in DNA sequence repositories. Our georeferenced dataset highlighted that most (81%) of the mtDNA sequences from New Guinean amphibians published before this study (GenBank) originate from the eastern half of the island, in the country of Papua New Guinea (Fig. 2). This sampling bias is also apparent in other taxa (e.g. ants; Kass et al. (2022)). We partially filled this gap by providing newly acquired sequences from specimens recently collected in western New Guinea and we encourage others to pursue efforts to document species diversity in this part of the island by combining field surveys and molecular identification protocols (Arida et al. 2021).

Despite relatively intensive frog sampling in the eastern half of New Guinea, where 75% of all recognised species from our study region occur, 49% of these are not yet represented by molecular data (Table 1). Missing data are most prevalent in Microhylidae and Hylidae and especially in species having narrow ranges (Suppl. material 1: table S3). Studies focusing on specific genera have already started to fill these gaps (e.g. Oliver et al. (2017); Ferreira et al. (2023)) and work on the Hylidae is in progress (S.C. Donnellan, pers. comm.).

Using 16S mtDNA, we delimited 369 MOTUs in the study region and a significant fraction of these (179) are named and considered taxonomically valid. Overall, the ABGD and GMYC delimitations were congruent with one another and mainly drove our final consensus (Suppl. material 1: table S4), while PTP tended to be more conservative, except for Microhylidae. A similar pattern between delimitation methods was also found by Arida et al. (2021), who analysed variation at the COI locus in a subset of samples included in our study (Suppl. material 1: table S5). The MOTU consensus in Arida et al. (2021) only partially overlaps with ours (Suppl. material 1: table S5), highlighting the benefits of integrating additional data (e.g. bioacoustics, morphometrics) to define robust candidate species hypotheses (Padial et al. 2010).

Thirty of the 38 MOTUs, for which acoustic data were available, displayed notable differences in call features and we consider 19 of these to represent confirmed candidate species. MOTUs supported by acoustic data display mean p-distances at the 16S locus of 2.75–4% (minimum of 1% between Asterophrys pullifer and Asterophrys cf. pullifer), broadly overlapping with the 3–5% distances that have been advocated as useful thresholds to flag “candidate species” in amphibians (Vences et al. 2005; Fouquet et al. 2007). Hence, a 3% threshold would be relatively conservative with our 16S dataset, since 69, 56 and 26 pairs of MOTUs in Microhylidae, Hylidae and Ceratobatrachidae display p-distances below 3% (including eight that were supported by acoustic data). Yet, because mtDNA-based approaches can lead to either species-oversplitting or lumping (Sukumaran and Knowles 2017; Dellicour and Flot 2018), the true extent of unrecognised diversity remains uncertain in the absence of bioacoustic data and the lack of morphological data prevents further characterisation that could strengthen the candidate species hypotheses. Nevertheless, because Microhylidae and Ceratobatrachidae are direct-developing frogs that often have narrow distributions and climatic niches (Oliver et al. 2022), we consider it likely that a very large portion of allopatric MOTUs in these groups correspond to distinct species. Conversely, we advise caution regarding Hylidae, for which species diversity remains ambiguous in our dataset due to sampling gaps and because numerous species complexes have been documented (e.g. Sulaeman et al. (2021); Günther et al. (2023)).

Microhylidae , Hylidae and Ceratobatrachidae together include 93% of all our MOTUs and 91% of the MOTUs assigned to recognised species. This proportion mirrors the proportion of taxonomically known species from these families (94%; Frost (2024)) and supports the idea that our sampling is representative of actual species diversity in the region. Following this assumption and assuming that the ratio between candidate and recognised species within our MOTU dataset is representative of the whole region (see Results), we obtained an estimate of 1,200 frog species in the New Guinea region. This estimate is subject to uncertainty and could well be an overestimate due to the nature of mtDNA (cf. paragraph above) and also because unsurveyed or poorly surveyed areas could be home to higher proportions of previously-known species (i.e. low beta diversity). On the other hand, our estimate could also be an underestimate if our dataset contains false negatives (species undelimited by mtDNA) or there is overlooked high beta diversity, as suggested by recent work in eastern New Guinea (Kraus 2021; Dahl et al. 2024).

Although the validity of our estimate of frog species richness across all of the New Guinea region remains to be confirmed, we argue that a two-fold increase in species diversity seems reasonable in the western half of New Guinea where most of our candidate species occur (Fig. 5). However, major uncertainties persist in areas where data remain substantially lacking, such as the western portion of the central range, which is thought to be highly biodiverse (Fig. 4A). This part of the island is the least sampled of all New Guinea both for molecular data and for most taxa, except perhaps some arthropod groups and birds (Toussaint et al. 2021; Kennedy et al. 2022; Letsch et al. 2023).

We delimited a total of 190 candidate species, of which 123 occur in western New Guinea (mainly in the Bird’s Head and Bird’s Neck subregions; Fig. 5). Amongst those, only six overlap with the ~ 200 candidate species identified by Oliver et al. (2022) from the New Guinea region. This gap derives from the differing sources of data used: mtDNA data in this study vs. expert surveys in Oliver et al. (2022). We can further highlight the geographic unbalance across datasets, i.e. most candidate species of Oliver et al. (2022) are from eastern New Guinea and many candidates from western New Guinea were not considered. These two largely non-overlapping datasets illustrate the scale of the Linnean shortfall for frogs in this region; while estimating a total of 700 species, Oliver et al. (2022) did not include at least 117 species identified here. Combining these two candidate species datasets, both incomplete and patchy in different ways, generates a minimum estimate of 800 species (~ 240 currently undescribed).

Figure 5. 

Examples of candidate species from the Bird’s Head and Bird’s Neck subregions (West Papua, Indonesia). The full list of candidate species is presented in the Suppl. material 2.

Due to large geographic and taxon sampling gaps for the 16Sb locus in several regions on New Guinea (e.g. Microhylidae in the central range, Hylidae in southern New Guinea and Ceratobatrachidae in the central portion of northern New Guinea), assessments of species distribution and endemism are necessarily limited. The Papuan Peninsula stands out as being the most sampled subregion when considering all mtDNA sequences (Fig. 1) and the most species-rich according to the IUCN (Fig. 4A), but one of the least sampled and least diverse in our MOTUs dataset (Fig. 3D and Suppl. material 1: fig. S7). This discrepancy derives from the positive correlation between the number of available 16Sb sequences per subregion and the number of delimited MOTUs (Figs 3D, 4B, 7B). How the correlation between species richness and geographic area varies across the study region remains to be tested with more data. However, we suspect that most areas of New Guinea show high levels of species richness, as suggested by the similar areal extents and species diversity of: 1) the Papuan Peninsula (85,000 km²) where 162 species are resident according to the IUCN and 2) the combined Bird’s Head and Bird’s Neck (90,000 km²) that are our best 16Sb-sampled subregions and have 110 delimited MOTUs in our analyses. Such diversity levels in western New Guinea are in line with recent studies that uncovered high levels of undocumented diversity in numerous taxa in this part of the island (Kadarusman et al. 2012; Cámara-Leret et al. 2020; Natusch et al. 2020; Arida et al. 2021; Milá et al. 2021; Ferreira et al. 2023; Oliver et al. 2025).

Our regional endemism rates for MOTU’s may be overestimates for some regions due to sparse sampling of species that may be more widely distributed. This is particularly true for Hylidae (also Ranidae, Dicroglossidae and Myobatrachidae) that tend to have wider ranges than microhylids and may occur throughout unsampled areas. This contrast in distribution patterns is probably tightly linked to reproductive features and habitat, i.e. between species breeding in waterbodies, which are abundant in the lowlands and the direct-developing Microhylidae that can breed virtually anywhere humidity is sufficient (Duellman et al. 1999; Bickford 2004). Endemism is generally high for amphibians in tropical mountains that act as habitat islands, which ultimately favours biological diversification (Anthelme et al. 2014; Stein et al. 2014). This pattern has been reported for frogs in eastern New Guinea (Oliver et al. 2017; Tallowin et al. 2017; Dahl et al. 2024) and undoubtedly occurs in western New Guinea, which harbours similar topographic heterogeneity (Oliver et al. 2022; Ferreira et al. 2023).

Estimated and recognised levels of regional frog species richness on New Guinea (e.g. 100–150 species in the Bird’s Head and the Papuan Peninsula) are similar to comparably sized areas in Amazonia (eastern Guiana Shield) (Vacher et al. 2020) and Madagascar (Central plateau) (Brown et al. 2016). However, the diversification of the Amazonian and Malagasy amphibian fauna has a long history throughout the Cenozoic, whereas terrestrial ecosystems of New Guinea may only date back to the early Oligocene (e.g. Jønsson et al. (2011)). These differences of context are clearly exemplified when comparing the crown age of Microhylidae from these regions; ~ 60 vs. 20 Ma (Feng et al. 2017; Hime et al. 2021; Portik et al. 2023). The New Guinean clade of Microhylidae (at least 14 genera and more than 300 species; Frost (2024)) would then have rapidly diversified during the island’s recent orogenic development and adapted to fossorial, terrestrial, scansorial and arboreal habitats (Rivera et al. 2017; Hill et al. 2023), a pattern typical of an adaptive radiation (Moen et al. 2021).

The Maximum-Likelihood and ultrametric phylogenetic trees obtained from the analysis of 580 bp of 16S mtDNA are mostly congruent and revealed numerous lineages from western New Guinea that are new to science (Suppl. material 1: fig. S8). Since our phylogenetic inferences are based on a single mtDNA locus only, we refrain from discussing node age estimates or interrelationships further than for the most external branches. We note, however, that our tree topologies are largely congruent with recent phylogenetic hypotheses that focused on Ranidae and Ceratobatrachidae (Brown et al. 2015; Chan et al. 2020). Interestingly, the addition of new Cornufer lineages revealed overlooked diversity in western New Guinea that could be very important for understanding the evolutionary history of that group (Suppl. material 1: fig. S8).

Most incongruences between our trees and previously published phylogenies occur within Hylidae and especially within Litoria (sensu Frost (2024)). Most discrepancies concern weakly supported clades that are also unresolved in literature (Portik et al. 2023). On the other hand, microhylid phylogenetic relationships are difficult to evaluate herein due to insufficient sequence data and limited species sampling. This echoes the long-standing challenge of resolving microhylid phylogenetic relationships in the region (Rivera et al. 2017; Hill et al. 2022).

New Guinea’s forests are still relatively extensive compared to many other tropical areas (Vancutsem et al. 2021) and their amphibian faunas remains substantially intact (Oliver et al. 2022). The region has been considered a “wilderness area” (Mittermeier et al. 2003); consequently, it has received little conservation attention compared to other regions that face much more extensive and imminent threats (e.g. Madagascar; Ralimanana et al. (2022)). Yet, a quarter of recognised amphibians from the New Guinea region still awaits an IUCN conservation status assessment and might already be subject to some threats (Oliver et al. 2022). In addition, we showed here that the region hosts an important and undocumented cryptic amphibian diversity, a large portion of which may consist of species confined to restricted geographical areas and, therefore, likely to be more sensitive to both land-use and climate changes. At a minimum, this indicates that the conservation status of many species will need to be re-assessed as distributions become better known, as highlighted by Kraus (2021) for the south-easternmost part of New Guinea and its archipelagos. More generally, there is a crucial need to hasten formal species descriptions in the New Guinea region (best achieved using integrative approaches), in order to facilitate conservation while also providing a more robust foundation for future research. The significant extent of the Linnean and Wallacean shortfalls of this region should also emphasise the importance of taking early conservation measures for amphibians to mitigate the risks of destructive land-use change (Cordier et al. 2021) and potential biotic invasions (e.g. chytrid fungus; Bower et al. (2019)) and to address the vulnerability and resilience of locally endemic upland species to threats from climate change (Lenoir et al. 2020). This rationale also applies to other tropical regions that are subject to significant amphibian Linnean shortfalls (e.g. Neotropics and Madagascar) (Vacher et al. 2020; Carné and Vieites 2024) and already face massive threats to biodiversity (Luedtke et al. 2023).