Niche shifts and environmental non-equilibrium undermine the usefulness of ecological niche models for invasion risk assessments

Niche shifts and environmental non-equilibrium in invading alien species undermine niche-based predictions of alien species’ potential distributions and, consequently, their usefulness for invasion risk assessments. Here, we compared the realized climatic niches of four alien amphibian species (Hylarana erythraea, Rhinella marina, Hoplobatrachus rugulosus, and Kaloula pulchra) in their native and Philippine-invaded ranges to investigate niche changes that have unfolded during their invasion and, with this, assessed the extent of niche conservatism and environmental equilibrium. We investigated how niche changes affected reciprocal transferability of ecological niche models (ENMs) calibrated using data from the species’ native and Philippine-invaded ranges, and both ranges combined. We found varying levels of niche change across the species’ realized climatic niches in the Philippines: climatic niche shift for H. rugulosus; niche conservatism for R. marina and K. pulchra; environmental non-equilibrium in the Philippine-invaded range for all species; and environmental non-equilibrium in the native range or adaptive changes post-introduction for all species except H. erythraea. Niche changes undermined the reciprocal transferability of ENMs calibrated using native and Philippine-invaded range data. Our paper highlights the difficulty of predicting potential distributions given niche shifts and environmental non-equilibrium; we suggest calibrating ENMs with data from species’ combined native and invaded ranges, and to regularly reassess niche changes and recalibrate ENMs as species’ invasions progress.

We first quantified and compared the realized climatic niches of H. erythraea, R. marina, H. rugulosus, and K. pulchra in their native ranges (native niche) and Philippine-invaded ranges (Philippine niche) to investigate climatic niche changes that may have unfolded during their invasion, and to assess niche conservatism and environmental equilibrium. We did not include L. catesbeianus and E. planirostris because of the limited data on these species in the Philippines. We quantified realized climatic niche changes using both the COUE 47 and the n-dimensional hypervolume frameworks 49,50 . We described species climatic niches and environmental backgrounds using eight environmental variables representing a combination of means, extremes, and seasonality that are known to be ecologically relevant to amphibians and are not highly inter-correlated. We further tested for niche conservatism using niche equivalency and niche similarity tests 48,51 . We then investigated the implications of climatic niche changes for niche-based predictions of species' potential distributions by calibrating ENMs using data from the Philippine-invaded (Philippine ENMs) and native ranges (native ENMs), and both ranges combined (combined range ENMs). Recognizing that different ENM approaches can yield varying predictions of potential distributions despite being calibrated with the same data (e.g., presence, absence/pseudo-absence, and environmental variables), we quantified ENMs of the four alien species using eight statistical approaches. We then created ensemble models (EMs) for each set of ENMs and used these to predict the species' potential distributions in the Philippines and in their respective native ranges. Ensemble models identify the signals (i.e., high agreement among models) and filter the noises (i.e., high disagreement) from different ENMs, yielding a lower mean error than any of the constituent ENMs and their resulting predictions 56 . We related observed niche changes to aspects of the species' invasion history in the Philippines 52 and to predictions of potential distributions. We conclude with a discussion on the implications of our findings for invasion risk assessments.
We focused our niche shift analyses on the Philippines because we have a relatively robust species' occurrence dataset on the distribution of the alien amphibians in the Philippines 57 . Data in other invaded ranges of these species are limited, especially for H. erythraea, H. rugulosus, and K. pulchra (data on the invaded range of R. marina is similarly poor, with the exception of Australia 33 ). Including other invaded ranges in the analysis would therefore poorly represent the species' global invaded niches. Importantly, including data from additional jurisdictions in which these species are alien would unlikely change our conclusions, because these additional data would be swamped by the higher number of Philippine records 57 .

Results
Varying levels of realized climatic niche change. Following the COUE framework 47 , we first quantified the occurrence densities of the native and Philippine niches in a weighted biplot of the first two principal components (PCs) of a Principal Components Analysis (PCA) calibrated using pooled environmental backgrounds in the species' native and Philippine-invaded ranges. The first two PCs captured ~71-76% of the variation in the environmental data. Supplementary Table S1 (online) shows the correlations between environmental variables and PCs. The topology of occurrence densities across the species' native and Philippine niches in PC biplots revealed: that the Philippine niche of H. erythraea is a subset of its native niche, whereas the other species showed partial overlap between native and Philippine niches 58 ; high to complete niche stability for all species at the intersection of the 75 th percentile of environmental backgrounds (I 75 ; 93-100%) and at the intersection of the complete environmental background (I 100 ; 91-100%); low to moderate niche unfilling for all species at I 75 (6-24%) and at I 100 (13-30%); and zero to low niche expansion for all species at I 75 (0-7%) and I 100 (0-9%) ( Fig. 1; Table 1). Rhinella marina and H. rugulosus showed niche expansion into non-analogous environmental space (i.e., climates found in the Philippine environmental background but not in the species' native range environmental background). Rhinella marina, H. rugulosus, and K. pulchra shifted their niche centroids to warmer and wetter climates in the Philippines (negligible centroid shift in H. erythraea; Fig. 1; see also Supplementary Figs. S1-S4). niche equivalency and similarity tests. Species' native and Philippine niches showed low niche overlap (Schoener's index of niche overlap [D]) (Table 1), with the lowest overlap in H. rugulosus (D = 0.10) and the highest in K. pulchra (D = 0.26). Comparing observed niche overlap values to null distributions (i.e., overlap values estimated from niches constructed by randomly re-allocating pooled occurrences to both niches) revealed that native and Philippine niches were more similar than expected by chance for R. marina and K. pulchra (P ≤ 0.01; Table 1; see Supplementary Fig. S5). Meanwhile, niches were more similar than randomly generated niches only for K. pulchra (P = 0.01 for both randomization methods). There was also a trend toward the niches of H. erythraea and R. marina being more similar than randomly generated niches for one of the two types of niche similarity tests (P = 0.06 for N ↔ P randomization test). For the case of H. rugulosus, niche equivalency was rejected, whereas the results of the two niche similarity tests were inconclusive. We further quantified climatic niche changes using four (for H. erythraea) to five (for R. marina, H. rugulosus, and K. pulchra) PCs to explore alternative PC biplots (see Supplementary Figs. S6-9 and Tables S2).
Dissimilar multidimensional hypervolumes of realized climatic niches. We compared species' native and Philippine niches using the n-dimensional hypervolume framework 49,50 . Here, we quantified the native and Philippine niches using multidimensional hypervolumes, derived from four (for H. erythraea) to five (for R. marina, H. rugulosus, and K. pulchra) PCs used in the COUE framework. Examining the topology and geometry of the hypervolumes of the native and Philippine niches in multidimensional environmental space revealed: low similarity at both the 75 th percentile probability boundary of hypervolumes (H 75 ; Jaccard Similarity Index J = 0.04-0.18) and entire probability boundary of hypervolumes (H 100 ; J = 0.11-0.24); low to moderate intersection at both H 75   Hoplobatrachus rugulosus (c), and Kaloula pulchra (d) as depicted by the biplot of the first two PCs. Grey areas represent overlap of the species' native and Philippine niches. The solid and dashed contour lines respectively represent the intersection of 75% (I 75 ) and 100% (I 100 ) of available environments in the native range (blue) and in the Philippine-invaded range (red). Solid arrows point to the direction of centroid shift from the species' native niches' occurrence density centroid to the that of Philippine niche. Dashed arrows represent the distance and direction between the positions of the centroids of the occurrence density of the environmental backgrounds in the native range and the Philippines along the PC biplots. The correlation circle shows the distribution of eight environmental variables along with the PC biplot: bio 1 = annual mean temperature; bio 4 = temperature seasonality; bio 5 = maximum temperature of warmest month; bio 6 = minimum temperature of coldest month; bio 12 = annual precipitation; bio 15 = precipitation seasonality; bio 16 = precipitation of wettest quarter; bio 17 = precipitation of driest quarter.  Table 1. Summary results of three niche change metrics (stability, unfilling, and expansion), overlap index (Schoener's D), and niche equivalency and similarity tests (P values). I 75 and I 100 represent respectively the intersection of 75% and 100% of available environments in the native and Philippine-invaded range. N ↔ P and N → P represent the randomization methods employed in the niche similarity test. . The filled circles represent the centroids of the hypervolumes of the native niche (blue) and Philippine niche (red). Opaque dots represent true species' occurrence records, whereas transparent dots represent random records derived from Gaussian kernel density estimation.
variables in the Philippines to those in species' native ranges revealed vast extents of clamping of several environmental variables -predictions in these areas are therefore uncertain (see Supplementary Figs. S18-21). Ensemble models of native ENMs (native EMs) predicted low climatic suitability in many areas in the Philippines where the species are known to occur (Figs. 4-7). Projecting native EMs to the species' presence/ pseudo-absence dataset in the Philippine-invaded range (E inv ) produced AUC scores ranging from no better than random (H. rugulosus), failed (R. marina and K. pulchra), to poor (H. erythraea), whereas TSS scores ranged from no better than random (H. rugulosus) to slightly better than random ( When projected to the Philippines and the native range, EMs of combined range ENMs (combined range EMs) predicted high climatic suitability in many areas where the species are known to occur, as well as in areas in the Philippines where the species have not yet invaded . When projected to E inv and E nat , the combined-range EMs of all species had fair to excellent AUCs. Meanwhile, the combined-range EMs of most www.nature.com/scientificreports www.nature.com/scientificreports/ species had lower TSS when projected to E inv and E nat relative to E eval (Table 3; see Supplementary Figs. S14-17). Projecting values of environmental variables in the species' combined ranges to either their native range or the Philippines revealed negligible clamping of environmental variables.

Discussion
Our findings revealed varying levels of niche change across four alien amphibian species introduced to the Philippines: high niche stability and low to moderate niche unfilling across all species, and insignificant to substantial niche expansion in all species except H. erythraea. All species except H. erythraea occupied hotter and wetter climates in the Philippines. Niche overlap was low for all species. Despite the low overlap, niche equivalency tests revealed niche conservatism in R. marina and K. pulchra, and niche similarity tests revealed niche conservatism in K. pulchra and to a lesser extent in R. marina and H. erythraea. The native and invaded niches of H. rugulosus were not equivalent, although niche similarity tests were inconclusive. Nonetheless, niche expansion in analogous (i.e., climates found in both the Philippines and each species' native range) and non-analogous environmental space provide evidence of niche shift in H. rugulosus. Niche unfilling revealed environmental non-equilibrium in the Philippine-invaded range for all species, whereas niche expansion revealed environmental non-equilibrium in the native range and/or adaptation in the invaded range for all species except H. erythraea. Hypervolume centroid distances consistently supported a shift in the niche centroids of all species except H. erythraea. Meanwhile, hypervolume intersection metrics generally revealed substantial unique fractions of Philippine niches, which is overall inconsistent with the findings of the COUE niche change metrics. Consistent . Predicted potential distribution of Hylarana erythraea in its native range and in the Philippines (columns) projected by EMs calibrated using data from the Philippine-invaded range, native range, and combined ranges (rows). Hylarana erythraea is native to South Asia, mainland Southeast Asia, and parts of maritime Southeast Asia (Borneo, Indonesia [Sumatra, Java, Lombok, and Riau Islands]). White dots represent species occurrence records, which were thinned to improve visibility. Relative climatic suitability increases from cold to warm colours. The maps were created using QGIS Geographic Information System software (v. 3.14; http://qgis.osgeo.org) and projected using WGS 1984 Coordinate Reference System.  Table 2. Summary results of hypervolume similarity index and hypervolume distance and intersection metrics.  www.nature.com/scientificreports www.nature.com/scientificreports/ with the findings of the niche overlap tests based on Schoener's D, values of Jaccard hypervolume similarity index were consistently low for all species. Finally, we found that native and Philippine EMs showed poor reciprocal transferability -that is, climatic characteristics of the Philippine niche poorly predicted the potential distribution of the species in the native range, and vice versa. Combined range EMs produced relatively better predictions of potential distributions across both native and invaded ranges.
Our findings build on previous studies 32, 33 , contributing to our growing understanding of climatic niche changes during the invasion of alien species, in general, and alien amphibian species, in particular. Our study is the first to assess niche changes in H. erythraea, H. rugulosus, and K. pulchra. Meanwhile, our findings of niche conservatism in the Philippine-invaded range of R. marina is somewhat consistent with the findings of Li et al. 32 -that is, the realized climatic niche of R. marina is conserved in its Indomalayan invaded range (including the Philippines); our findings showed niche equivalency and to a lesser extent similarity in R. marina in its Philippine-invaded range (Table 1), whereas Li et al. 32 showed niche similarity (P = 0.02; niche equivalency test was not conducted) in its Indomalayan invaded range. The inconsistent results of the niche similarity test between our study and Li et al. 32 were likely driven by the differences in the scale and geographic extent of our analysis (Philippines only vs. Indomalayan realm, respectively). Our study also extends the work of Tingley et al. 33 , who examined niche shifts across the native vs. Australian-invaded range of R. marina.
The observed niche shifts could either be due to shifts in the species realized climatic niches in their invaded ranges 27 , to a shift in the species' fundamental niches (e.g., changes in their environmental tolerances) during the Figure 6. Predicted potential distribution of Hoplobatrachus rugulosus in its native range and in the Philippines (columns) projected by EMs calibrated using data from the Philippine-invaded range, native range, and combined ranges (rows). Hoplobatrachus rugulosus is native to East Asia (China and Taiwan) and mainland Southeast Asia. White dots represent species occurrence records, which were thinned to improve visibility. Relative climatic suitability increases from cold to warm colours. The maps were created using QGIS Geographic Information System software (v. 3.14; http://qgis.osgeo.org) and projected using WGS 1984 Coordinate Reference System. www.nature.com/scientificreports www.nature.com/scientificreports/ course of invasion 26,37 , or both. Realized climatic niche shift is in line with the notion that species occupy a subset of their fundamental niche in their native range -indicating environmental non-equilibrium -due to dispersal limitations and/or biotic interactions 59,60 . In contrast, fundamental niche shift may be caused by adaptive changes post-introduction 37 . In the case of R. marina, we detected expansion of the species' Philippine niche into analogous and non-analogous environments. Phenotypic changes have been observed over the course of the species' invasion in Australia 61,62 . However, Tingley et al. 33 , by combining ecophysiological and correlative models, found that R. marina fails to occupy a substantial fraction of its fundamental niche in its native South American range, plausibly due to biotic interactions with closely related species. More importantly, they showed that R. marina has successfully filled its fundamental niche in its Australian-invaded range, reflecting realized climatic niche shift. In our analysis, the captured environmental backgrounds in the native range (derived from biomes intersecting the species' range) are likely more environmentally restricted than the fundamental niche, and the detected non-analogous climates in the Philippines are likely within the species' fundamental niche. Meanwhile, for H. rugulosus, discriminating whether realized or fundamental niche shift explains the observed niche expansion in non-analogous environmental space would require mechanistic approaches based on experimental measurements of the species' fundamental niche 26,33,63 .
We extended our main analysis by quantifying niche changes in the COUE framework 47 using alternative biplots constructed from combinations of the first four to five PCs. Interestingly, alternative biplots revealed inconsistent support for niche equivalency and similarity, and in some cases, high variance in niche change metrics (see Supplementary Table S2). Nonetheless, we argue that the biplots of the first two PCs, as presented here, Figure 7. Predicted potential distribution for Kaloula pulchra in their native range and in the Philippines (columns) projected by EMs calibrated using data from the Philippine-invaded range, native range, and combined ranges (rows). Kaloula pulchra is native to Southern East Asia (China), eastern South Asia (Bangladesh and India), mainland Southeast Asia, and some parts of maritime Southeast Asia (Singapore and Indonesia [Sulawesi]). White dots represent species occurrence records, which were thinned to improve visibility. Relative climatic suitability increases from cold to warm colours. The maps were created using QGIS Geographic Information System software (v. 3.14; http://qgis.osgeo.org) and projected using WGS 1984 Coordinate Reference System. www.nature.com/scientificreports www.nature.com/scientificreports/ provide the most reliable results, since they explain the highest amount of variation in the environmental data (>71%). The fact that the COUE framework is limited to analysing niches in two dimensions at a time is a significant limitation. To overcome this limitation, we further quantified species' climatic niche changes using the n-dimensional hypervolume framework 49,50 . Using four to five PCs that collectively captured ~99% variance in the environmental data, our findings again revealed varying levels of niche changes among species, which were generally inconsistent with our findings derived using COUE framework's niche change metrics. This highlights the difficulty of quantifying niche changes, as well as testing for niche conservatism and shift, due to varied conceptual theories underpinning different methodological frameworks. Nonetheless, the n-dimensional hypervolume framework, as currently implemented, does not account for environmental availability in each range (and thus analogous and non-analogous environments) and does not give greater weight to PCs that explain more environmental variation. To this end, we base our conclusions regarding niche conservatism and environmental equilibrium primarily from the findings revealed by the COUE framework.
Should we anticipate future niche changes in the four alien amphibian species' Philippine niches? We found niche unfilling for all species, revealing environmental non-equilibrium in the species' Philippine niches. This supports the findings of Pili et al. 52 that all species' invasions in the Philippines are incomplete, and that the species continue to spread. Moreover, for all species, PC biplots revealed analogous environmental space that is   www.nature.com/scientificreports www.nature.com/scientificreports/ unoccupied in both native and Philippine-invaded ranges. If accessible (in geographic space) and climatically suitable, these areas of environmental space present opportunities for future niche expansion. For H. rugulosus, PC biplots revealed unoccupied non-analogous environmental space in the Philippines, which presents opportunities for the species to further shift its Philippine niche. This emphasizes the need for regular reassessments of niche conservatism/shift as a species' invasion progresses.
Our findings do not conform to observed patterns linking niche unfilling with aspects of species' invasion history. Patterns in alien amphibian 32 , reptile 32 , and bird 31 invasions showed that residency time (i.e., the length of time since a species was first introduced) is inversely correlated with the magnitude of niche unfilling. Our findings regarding R. marina, which was introduced in the Philippines in 1934 52 , are somewhat conforming with this pattern. Meanwhile, the other species showed an inverse pattern; for example, H. rugulosus and K. pulchra, respectively introduced only in the 1990s and 2000s 52 , showed low levels of niche unfilling, whereas H. erythraea showed substantial niche unfilling, despite being first introduced in the Philippines for more than a century ago in the 1880s 52 .
Niche unfilling in birds 31 and in alien species in disjunct geographic areas (e.g., archipelagic systems) 33 have been linked to dispersal limitations (e.g., biogeographic barriers) 64 and propagule pressure (i.e., the absolute number of individuals involved in a human-mediated dispersal event and the number of discrete dispersal events) 59 . Similarly, the high niche unfilling in H. erythraea observed here may be due to its dependence on pathways characterized by low inter-island propagule pressure, especially during the early periods of its invasion. Hylarana erythraea is known to historically disperse inter-island primarily as a contaminant of commodities (agriculture, animals) (sensu Scalera et al. 65 ) 52 . The trade of these commodities (reflecting propagule pressure) is high intra-islands, especially on larger Philippine islands such as Luzon and Mindanao, where rapid and short-distance transport is possible and is critical for such perishable commodities. However, inter-island trade of these commodities is low, likely limiting the species dispersal throughout the archipelago. This low level of propagule pressure coincides with its spatio-temporal spread in the Philippines, in which it was first introduced and restricted among import-dependent small islands in the Central Philippines for more than a century, and only dispersed to the country's major islands of Luzon and Mindanao in the 1990s, where it underwent an increased rate of spread 52 . Meanwhile, high propagule pressure was likely an important contributor to the low levels of niche unfilling in R. marina, H. rugulosus, and K. pulchra. Rhinella marina was initially intentionally dispersed in many parts of the Philippines as a biological control agent, where it subsequently dispersed intra-islands primarily via natural dispersal. Moreover, the rapid inter-island dispersal of R. marina has been attributed to the Philippines' dependence on sea transport for trade and travel -this species has been observed to hitchhike on transportation vehicles (e.g., ships, boats, trucks) 52 . The use of multiple alternative introduction and dispersal pathways -providing alternative ways for intra-and inter-island dispersal -likely aided H. rugulosus and K. pulchra in negating effects of residency time and dispersal limitations, resulting in rapid dispersal throughout the country 52 .
Predictions of native EMs corroborated previous studies showing that ENMs based on the native range of species often underpredict suitable areas in the invaded range 26,29,[40][41][42] . Specifically, native EMs predicted low climatic suitability in many areas in the Philippines where the species occur. The observed poor predictive performance of native EMs was likely caused by the observed niche expansion and unfilling in the species Philippine niches. Interestingly, our findings, especially for H. erythraea, support previous studies 31, 35 showing increased predictive performance of native ENMs with decreasing niche unfilling and expansion in the invaded range. Meanwhile, Philippine EMs support previous studies showing ENMs calibrated using invaded range data may be able to capture niche shifts that occurred during invasion 26,29,[40][41][42] . However, due to environmental non-equilibrium often observed in alien species in their invaded ranges, ENMs calibrated using invaded range data may fail to capture parts of the species' climatic niche that is occupied in the native range but unoccupied in the invaded range. This leads to underprediction of the species' potential distribution in both the native and invaded ranges 22,23,25,41 , as observed in the predictions of the Philippine EMs. Overall, combined range EMs predicted high climatic suitability in areas in the Philippines where the species are known to occur, and predicted areas that are climatically suitable but are yet uninvaded by the species. Our findings corroborate those of previous studies in that combining data from the native and invaded ranges of an alien species, where available, can offset the predictive limitations of ENMs calibrated in only the native or invaded ranges 26,29,[40][41][42] . We recommend future researchers recalibrate combined range EMs as species' invasions progress, and when data from other ranges of the species are collected and/or made available.   www.nature.com/scientificreports www.nature.com/scientificreports/ niches. The observed niche changes, reflecting niche conservatism/shifts and environmental non-equilibrium, likely caused the observed poor reciprocal transferability of native and Philippine EMs and their constituent ENMs. Our findings support previous studies showing that ENMs calibrated with data from the combined native and invaded ranges, where available, will be most useful in informing invasion risk assessments 22,23,35,41,66 . In light of the implications of niche shifts and environmental non-equilibrium for ENM predictions, we suggest researchers and managers routinely incorporate quantification of niche changes into ecological niche modelling experiments 23,27,42,47,48 . In addition, the fact that the majority of alien species' invasions are incomplete 23,25,41,43 , such as the case of the four alien amphibian species studied here, emphasizes the need for researchers and managers to regularly reassess niche changes in alien species' invaded ranges as a species' invasion progress 33 . Consequentially, ENMs must be regularly recalibrated with updated data to reflect concurrent changes in the species' realized climatic niche in the invaded range 33,40,58 .

Materials and Methods
Species occurrence and environmental data. Occurrence records of the four alien amphibian species in the Philippines and the species' native ranges were obtained from the Global Biodiversity Information Facility (GBIF) 67 , collections of local natural history institutions, scientific publications, expert observations, and field surveys conducted by the authors (see Supplementary Methods) 57 . The pooled dataset was cleaned manually so that only high-quality records were used in the analysis; records conforming to these set of conditions were retained: (1) georeferenced; (2) with year of record; (3) with "county" or "municipality" locality data (sensu Darwin Core Task Group 68 ); (4) inland coordinates and within the ascribed locality (up to the third administrative level [admin2] of the Global Administrative Areas v3.6 [https://gadm.org/]) -assessed in InfoXY v.2 tool of speciesLink (http://splink.cria.org.br/); (5) and inside the known native range of the species, as depicted by species range maps [69][70][71][72] . To minimize spatial sampling bias, occurrence records were thinned by subsampling records to a resolution of one record per 5 km 2 (2.5 arc-min) 73,74 . Initial visual assessment of the spatial distribution of the records of R. marina showed sampling bias across its native range, where records were denser in Central America relative to other parts of its range. Thus, we further reduced native-range records of R. marina to a resolution of 10 km 2 (5 arc-min) 33 . We initially planned to omit occurrence data outside the years 1970-2000, to align with the temporal reference of the environmental variables (described below) 75 . However, this would have resulted in the omission of a considerable proportion of collected occurrence data (~60-100%), especially data on recent range expansions of the species in the Philippines (K. pulchra was first recorded in the 2000s), and would inevitably negatively affect our analysis of realized climatic niche changes and our modelling of species' ecological niches. In addition, after our extensive data cleaning process, the final thinned dataset only contained data recorded from 1950 to present (Table 4). environmental variables. Extreme temperature and precipitation can negatively affect amphibian development (e.g., tadpole and egg developmental rates/morbidity), ecophysiology (e.g., locomotor performance, communication, sensory systems), and energy acquisition and allocation [76][77][78] . Temperature and precipitation seasonality alter growth, reproductive cycles, phenology, and prey-predator dynamics [79][80][81][82] . At a broad geographical scale, precipitation and temperature extremes and seasonality influence amphibian biogeography (population declines and extirpations, shifts in geographic distribution, species extinction) [83][84][85] . Thus, in quantifying realized climatic niche changes and ecological niche modelling, we used environmental variables representing a combination of means, extremes, and seasonality that are known to be ecologically relevant to amphibians and are not highly inter-correlated (Pearson's correlation coefficient |r | ≤ 0.7) 86 : annual mean temperature (bio 1), temperature seasonality (bio 4), maximum temperature of the warmest month (bio 5), minimum temperature of the coldest month (bio 6), annual precipitation (bio 12), precipitation seasonality (bio 15), precipitation of the wettest quarter (bio 16), and precipitation of the driest quarter (bio 17). Bioclimatic variables were obtained from WorldClim v.2, and represent averages of monthly minimum, mean, and maximum temperature and of precipitation for 1970-2000 75 with a spatial resolution of 5 km 2 (2.5 arc-minutes).
Quantifying niche changes. We quantified the native and Philippine niches of the four alien amphibian species using the COUE framework 47 and the n-dimensional hypervolume framework 49,50 . We then tested for niche equivalency and similarity between the native and Philippine niches of the species 48,51 . Finally, we assessed for niche conservatism and environmental equilibrium based on findings of the methodological frameworks and niche equivalency and similarity tests. coUe Approach. Using the COUE framework 47 , we quantified the species' native and Philippine niches in weighted PC biplots and decomposed climatic changes in species' niche using a unified set of metrics 27,47 . We analyzed this using the ecospat package 87 in R v.3.6 88 . We first transformed each species' global environmental space based on the eight environmental variables described above, onto a biplot defined by the first two PCs of a PCA. We calibrated the PCA using pooled environmental backgrounds from the species' native and Philippine ranges. We defined ecologically relevant environmental backgrounds following the suggestions of Guisan et al. 47 . In the native range, environmental backgrounds include all the biomes 89 inhabited by the species (i.e., biomes that are intersected by the species' native range [69][70][71][72] ). In the Philippines, environmental backgrounds included the whole of the Philippines. The first two PCs captured ~71-76% of the variation in the environmental data. The correlations of environmental variables with the PCs are shown in Supplementary Table S1 (online). We then divided the species' global environmental space, as depicted in a PC biplot, into a grid consisting of 100 ×100 cells, bounded by the minimum and maximum values in the environmental background. Lastly, we projected the scores of the species occurrence records onto the gridded environmental space and grouped the scores per grid cell. We applied a Gaussian kernel density function with a standard bandwidth to estimate the smoothed density of occurrences in each cell of the gridded environmental space 90 . We further quantified the species realized climatic niche using alternative biplots from combinations of the first four (for H. erythraea) to five (for R. marina, H. rugulosus, and K. pulchra) PCs (see Supplementary Figs. S6-9).
We visually examined the niches' occurrence densities in gridded environmental space, as depicted in a PC biplot, and categorized their topological interaction into five possible patterns 58 : (1) complete to near-complete overlap of the two niches; (2) the Philippine niche is a subset of the native niche; (3) native niche is a subset of the Philippine niche; (4) partial overlap between the two niches; or (5) completely disjunct niches. We decomposed niche changes in analogous environmental space into niche stability, niche expansion, and niche unfilling 27,47 . We calculated these indices in I 75 (to remove marginal climates resulting from kernel smoothing) and I 100 of environments available in each range. These indices are not confounded by non-analogous climates, making measured expansion indicative of changes in the realized niche 27 . Thus, we also visually assessed niche unfilling and expansion in non-analogous environmental space.
Niche overlap, equivalency, and similarity. We quantified the overlap between each species' native and Philippine niches using Schoener's index D of niche overlap 91 , which estimates the overall similarity between the two niches over the global environmental space 51 . Schoener's D ranges from zero (no overlap) to one (complete overlap) 51,91] . We quantified niche conservatism by testing estimates of D under two alternative hypotheses, representing two extremes across a spectrum of niche conservatism: (1) niche equivalency which determines whether realized niches of two entities in two geographical ranges are conserved in the strictest sense (i.e., whether observed niche overlap is effectively indistinguishable when randomly re-allocating pooled occurrences of both niches between them), and (2) niche similarity, which estimates whether the realized niche occupied in one range is more similar to the niche occupied in the other range than randomly generated niches 48,51 For each species, we tested the hypothesis of niche equivalency by simulating two niches based on randomly permuted pooled occurrence records of the native and Philippine niches (the simulated niches maintain the same number of occurrence records as that observed between the native and Philippine niches). The overlap between the two simulated niches was then estimated using Schoener's D. This process was repeated 1,000 times to generate a null distribution of niche overlap values of simulated randomly permuted niches and to confidently reject/ accept the hypothesis of niche conservatism. We inferred niche conservatism if the observed niche overlap was greater than 95% of null distribution values (using the "greater" test of ecospat.niche.equivalency.test function in the ecospat package, which is a one-tailed test). In contrast, observed values outside the density of 95% of null distributions suggest a significant difference in niches 48,51 .
We tested for niche similarity by simulating two niches based on occurrence densities with randomly placed centroids in the environmental background using two randomization tests: (1) centroids randomly placed in both the native and Philippine range (N ↔ P) and (2) centroid randomly placed in the Philippine-invaded range only (N → P) 48,51 . The simulated niches maintain the same number of occurrence records as that observed between the native and Philippine niches. This process was repeated 1,000 times, randomly shifting the centroid of simulated occurrence densities in each repeat, to generate a null distribution of niche overlap values of simulated random niches and to confidently reject/accept the hypothesis of niche conservatism. We inferred niche conservatism if the observed overlap value was greater than 95% of null distribution values (using the "greater" test of ecospat. niche.similarity.test function in the ecospat package) 48,51 . n-dimensional hypervolume framework. We analysed the multidimensional hypervolumes of species' native and Philippine niches using the hypervolume package 92 in R v.3.6 88 . Using the Gaussian kernel density estimation method, we generated multidimensional hypervolumes of species native and Philippine niches from homogeneously distributed random records. These random records were derived from the Gaussian kernel density estimates of the distributions of species' occurrence records in multidimensional environmental space comprising of the same four (for H. erythraea) to five (for R. marina, H. rugulosus, and K. pulchra) PCs used in the COUE framework, which collectively capture ~99% of variation in environmental data. The bandwidth for Gaussian kernel density estimates was estimated separately for each species niche using the Silverman bandwidth estimator (see Supplementary Table S3) 89 . The hypervolumes of the species' native and Philippine niches were compared using a similarity index (Jaccard) and niche changes were decomposed using hypervolume distance (minimum distance, centroid distance) and intersection (volume of the intersection, the unique fraction of hypervolumes) metrics 93 . We calculated the hypervolume metrics and similarity index in H 75 (to remove marginal climates resulting from kernel smoothing) and in H 100 . ecological niche modelling. We modelled the realized climatic niches of the four alien amphibian spe-  101 . We calibrated ENMs with the same eight environmental variables used to quantify the realized climatic niche changes and occurrence records from the species' (1) Philippine-invaded range, (2) native range, and (3) combined ranges. We tested for environmental variable clamping to identify locations where values of variables were outside the range used for calibrating ENMs -predictions in these areas are uncertain. This was done by projecting values of environmental variables from one range to another (e.g., the Philippines to species' native range). Barbet-Massin et al. 102 laid-out optimal settings in generating and weighting pseudo-absences for different statistical techniques used in ENM. Here, we generated and weighted pseudo-absences to yield good to optimal ENM performance across the eight statistical techniques 102 . We generated pseudo-absence data by randomly sampling 10,000 (2020) 10:7972 | https://doi.org/10.1038/s41598-020-64568-2 www.nature.com/scientificreports www.nature.com/scientificreports/ background points from areas where the species has no presence records and across all the biomes 89 inhabited by the species (same environmental backgrounds used in COUE framework). For each statistical approach, we equally weighted the presences and pseudo-absences used to calibrate the ENMs by setting a neutral (0.5) prevalence. We conducted the analysis using the biomod2 package 103,104 in R v.3.6 88 . We ran all ENMs with the default parameters of the statistical techniques.
Prior to modelling, we randomly split each presence/pseudo-absence dataset (datasets for native, Philippine, and combined range ENMs) into two parts: 70% for training and 30% for evaluation (E eval ). By assigning a specific subset for evaluation, the EMs (discussed below) and its constituent ENMs were evaluated using the same data, ensuring a fair evaluation between the performances of EMs compared to ENMs 104 . For each species, we further sub-sampled the training data subset randomly to 70% training and 30% testing. This was repeated 10 times to account for uncertainty due to random subset selection, resulting in 10 random sub-samples of training-testing data. We modelled each sub-sample of training-testing data using the eight statistical approaches described above, generating 80 Philippine ENMs, 80 native ENMs, and 80 combined ENMs per species.
To evaluate predictive performance, we projected the ENMs to their respective E eval . To evaluate reciprocal transferability, we projected the Philippine ENMs to the entire presence/pseudo-absence dataset in the native range (E nat ), native ENMs to the entire presence/pseudo-absence dataset in the Philippine-invaded range (E inv ), and combined ENMs to both E nat and E inv . We then computed for the AUC 105 and the TSS 106 . We interpreted the AUC values based on Swets et al. 107 , where values >0.90 = excellent, >0.80-0.90 = good, >0.70-0.80 =fair, >0.60-0.70 = poor, and >0.50-0.60 = fail. Meanwhile, TSS values range from -1 to 1, where 1 indicates perfect agreement and values zero or less indicate performance no better than random 106 .
For each species, we combined each set of ENMs (80 Philippine ENMs; 80 native ENMs; 80 combined ENMs) to build a consensus ensemble model where each ENM was weighted by its TSS scores (i.e., better performing models contribute more to the EM) 56,104 . We then evaluated the EMs using the evaluation datasets used in evaluating their constituent ENMs (E eval and E nat for Philippine EMs; E eval and E inv for native EMs; E eval , E inv , and E nat for combined range EMs). To predict species' potential distributions, we projected species' EMs to the Philippines and their respective native ranges.

Data availability
All data was gathered from publicly available sources and are available in the online registry of the Global Biodiversity Information Facility (HerpWatch Pilipinas, Inc.; 57 see Supplementary Methods).