Using citizen science records from iNaturalist to document geographical range outliers in Australian skinks

Data collection

iNaturalist observation data for skinks in Australia were downloaded directly from iNaturalist at 9:37 pm (AEST) on 21 July 2023 (iNaturalist community 2023), by using the following query:

‘Quality grade’ refers to the grade given to observations on iNaturalist to determine their use in research and ‘Identifications’ refers to the level of agreement in community identification on an observation. The ‘place_id’ denotes Australia, and the ‘taxon_id’ denotes the family Scincidae. This provided 45,443 records. Those records that had not been identified to the species level (n = 6041) were excluded, giving 39,402 records in total.

Three range map sources were used to provide a broad baseline for the accepted current consensus on known species distributions. IUCN Red List range maps were downloaded from iucnredlist.org at 5:00 pm (AEST) on 1 August 2023 (IUCN 2023b). Range maps from the Global Assessment of Reptile Distributions (Hereafter GARD) were primarily v1.7, the latest publicly available version of the database released in 2022 (Roll et al. 2017; Caetano et al. 2022). However, they also included range maps from the upcoming v2.0. The third source was the sixth edition of A Complete Guide to Reptiles of Australia (Wilson and Swan 2021), the most recent version of the comprehensive and well-regarded Australian reptile field guide.

A number of species varied taxonomically among iNaturalist and the various range map sources. In the case of contentious recent taxon merges or splits, for example, the convergence of six previously separate Ctenotus species to Ctenotus inornatus (Rabosky et al. 2014), species were merged and, if needed, range maps were combined. This is primarily because iNaturalist users may vary on their opinion on the taxon change and old records may not yet be updated, so individuals will be identified inconsistently. This creates a large number of complex outliers from the smaller ranges of more divided groupings. Range maps were also combined for species that are physically indistinguishable, for example, Carlia crypta and Carlia rubigularis (Singhal et al. 2018), as identification was performed through photo records.

iNaturalist observations were compared with range maps from the IUCN and GARD using QGIS (QGIS Development Team 2023) and visually compared with range maps from the field guide. All visual outliers (n = 1153) were examined. These outliers were documented, along with whether they fell inside or outside each of the three range maps and their distance to the relevant digital range map/s, unless they met the following seven outlier exclusion criteria (n = 770):

Individual skinks were then identified to species level as best as possible from the images provided. If it was found that an individual was misidentified on iNaturalist, the observation was discarded unless it was still outside of the range of its actual species, in which case it was reclassified correctly. In a number of cases, the individual could not be identified down to species level with the images provided. These observations were discarded.

Further, while some observations did not explicitly meet any exclusion criteria, they were eliminated for a combination of factors that made them less reliable, primarily a long time between observation and posting. Many observations were uploaded to iNaturalist a number of years after the photo was taken, and although records were not eliminated on the basis of this alone, they were treated with suspicion and eliminated if, for example, the environment pictured did not match its stated location, or if the location accuracy was also extremely low.

The remaining outlier records (n = 656) were examined to determine whether they were truly novel range outliers, or whether they had been noted elsewhere in the literature. If an observation was not outside of all map sources (GARD, IUCN, and field guide), it was not regarded as novel. If any documentation of the species occurring in the area could be found elsewhere, it was not regarded as novel. Observations fairly close to a species’ range were not regarded as novel; this cut-off was placed at 30 km, including location accuracy in the measurement. However, this could be lower for highly range-restricted species or higher for species with broad ranges, especially when in arid zones. It could also be lower if the observation was isolated in a high-traffic area such as a large city, where a translocation is more likely. Novel records (n = 111) were then further scrutinised and an estimation of the likely source of the outlier/s was made as translocation (n = 54), extension (n = 18), or undetermined (n = 39). Non-novel outliers were examined to illuminate differences and limitations among the various range maps used in this study.

Statistical analysis

All analyses were performed using R statistical software (R Core Team 2023), by using the tidyverse, car, and emmeans R packages (Fox and Weisberg 2019; Wickham et al. 2019; Lenth 2024). The proportion of each subfamily of Australian skinks (Egerniinae, Eugongylinae, Sphenomorphinae) with iNaturalist records was compared with χ² goodness-of-fit tests. The mean number of iNaturalist records of the three subfamilies was compared using a single-factor ANOVA and a post hoc Tukey’s test. For those species with iNaturalist records, the proportion of each subfamily with outliers and with novel outliers were also compared with χ² goodness-of-fit tests.

The mean range-size data of Australian skink species obtained from GARD (Roll et al. 2017; Caetano et al. 2022) was compared between species with and without iNaturalist records. For those species with iNaturalist records, mean log-transformed range size was also compared between those with and without both all range outliers and novel range outliers. These comparisons used a Welch test.

Representation of Australian skinks on iNaturalist

Of the 324 skink species documented in Australia on iNaturalist, five (Eutropis multifasciata, Plestiodon fasciatus, Plestiodon gilberti, Plestiodon laticeps and Sphenomorphus incognitus) were not native Australian skink species. These observations were all either captive, misidentifications, or not accompanied by photo records.

319 out of a total of ~462 native Australian species had iNaturalist records. Most of these species had comparatively few records: 38 (11.9%) had a single record, and 146 (45.8%) had fewer than ten. Only 60 species (18.8%) had over 100 records, and eight species (1.7%) had over 1000, with Eulamprus quoyii (n = 3464) having the most records (Fig. 1).

Fig. 1.

Distribution of the number of iNaturalist observations among native Australian skink species. Bars are 20 units wide. Species with over 1000 observations are labelled.

The three subfamilies of Australian skinks were represented unequally on iNaturalist.

Egerniinae species were most likely to be represented (94% of species with iNaturalist records), followed by Eugongylinae (79.4%) then Sphenomorphinae (59.1%). (χ² = 33.94; d.f. = 2; P < 0.0001) (Fig. 2a). Egerniinae species also had the most records (mean = 269.7; s.d. = 629.7), again followed by Eugongylinae (mean = 105.3; s.d. = 307.4) and then Sphenomorphinae (mean = 42.3; s.d. = 237.3) (single-factor ANOVA with log(x + 1)-transformed data: F = 38.8; P < 0.0001) (Fig. 2b). Tukey’s post hoc analysis (Supplementary Table S1) showed a significant difference between the number of records recorded on iNaturalist among all three subfamilies of Australian skinks.

Fig. 2.

(a) The proportion of species with iNaturalist records among each subfamily of Australian skinks, and (b) the mean number of iNaturalist observations within each subfamily of Australian skinks. Error bars show standard error.

The known range size of a species also had a significant effect on how likely it was to be recorded on iNaturalist. Range sizes were compared for 448 native Australian skink species. Of these, 307 species had iNaturalist observations. Those species with observations on iNaturalist had significantly larger range sizes (mean = 581,068 km²; s.d. = 1,016,827) than did those without (mean = 114,438 km²; s.d. = 284,751) (Welch t-test using log-transformed range sizes: t(d.f.) = −8.44; P < 0.0001).

Geographical range outliers

Of the 319 native Australian species documented on iNaturalist, 82 (25.7%) were found to have at least one geographical range outlier, and 33 (10.3%) had at least one novel range outlier (Table 1). A full record of all outliers can be found in the supplementary material (Table S2).

Subfamilies were again not equally represented (Fig. 3). Among those 319 species on iNaturalist, Egerniinae had the highest proportion of both species with outliers (38.3%) and novel outliers (17%), which was not significantly different from Eugongylinae species (35.2% outliers; 15.7% novel outliers) (outliers: χ² = 0.14; d.f. = 1; P = 0.71) (novel outliers: χ² = 0.14; d.f. = 1; P = 0.7); however, these were significantly higher than the proportions for Sphenomorphinae species (16% outliers; 4.9% novel outliers) (outliers: χ² = 16.82, d.f. = 2, P = 0.0002) (novel outliers: χ² = 10.69, d.f. = 2, P = 0.005).

Fig. 3.

The proportion of species among the three subfamilies of Australian skinks with and without (a) outliers, and (b) novel outliers on iNaturalist.

Range size had a significant effect on how likely the species on iNaturalist were to have outliers, but only if they were novel (Fig. 4). There was no significant difference in range size between species with range outliers (mean = 708,114 km²; s.d. = 1,138,222) and without range outliers (mean = 535,175 km²; s.d. = 967,873) (Welch t-test using log-transformed range sizes: t(d.f.) = −1.2251; P = 0.2) (Fig. 4a). Conversely, there was a significant difference in range size between species with novel range outliers (mean = 940,764 km²; s.d. = 1,358,900) and without novel range outliers (mean = 538,06 km²; s.d. = 962,159) (Welch t-test using log-transformed range sizes: t(d.f.) = −3.6532; P = 0.0006) (Fig. 4b).

Fig. 4.

Average range size of Australian skink species with iNaturalist observations (a) with and without range outliers and (b) with and without novel range outliers. Error bars show standard errors.

The majority of species had very few individual outliers, with 71.9% of those with outliers and 84.8% of those with novel outliers having fewer than five. Many species had a high proportion of non-novel outliers, most dramatically Tiliqua rugosa, which had 205 outliers, only one of which was novel (Tables 1, S2). These non-novel outliers were usually due to differences between range maps (Fig. 5). The IUCN range maps were generally more restrictive, with 292 observations falling outside IUCN maps but covered by GARD maps (204 of these were due to Tiliqua rugosa, with 88 being due to other species). By comparison, 54 observations fell outside GARD range maps, but were within IUCN maps. In total, 38 observations fell outside both digital range maps while still being covered by the field guide maps. The remaining 160 non-novel outliers fell outside of all range maps but could be accounted for in other scientific literature. These observations were often on islands where the species is well-documented. For example, 42 Carlia longipes observations were on Fitzroy, Green and Lizard islands, which were not covered by range maps despite the species being known to inhabit them (Sprackland et al. 2004). However, some were on the mainland. Cryptoblepharus pannosus had 36 non-novel observations not covered by the three range map sources used as a baseline for this study (Table S2). One in Murray Bridge, South Australia (SA), too close to its known range to be considered novel, and a widespread group in Western Victoria (VIC). Those outliers in Western VIC were primarily in National and state parks, particularly the Grampians, Little Desert and Mount-Arapiles-Tooan. The species is known to be naturally occurring in this area (Robertson and Coventry 2019).

Fig. 5.

The number of observations of Australian skinks on iNaturalist that lay outside each digital range map, as well as in or outside the range maps of Wilson and Swan (2021)’s A Complete Guide to Reptiles of Australia.

Fifty-four observations that potentially attributed to translocations were spread across 12 species, with the majority of records being of Carlia sexdentata. This species had 38 outliers, which were all novel (Table 1), 37 of which fell within the city limits of Darwin, Northern Territory (NT), ~350 km from its nearest known range in Arnhem Land, NT, with the earliest record from 2010. The other outlier was some distance from Darwin, in Katherine, NT (Fig. 6), being recorded in 2021. Lampropholis delicata had a total of 29 outliers, 16 of which were novel and three of which could be certainly attributed to translocation (Table 1). These observations were of wild living individuals on the grounds of Adelaide Zoo. Of the 10 other species that showed evidence of translocation (Table 1), the majority had a single isolated record, whereas Cryptoblepharus pulcher had two isolated observations in VIC and SA, and Pseudemoia spenceri had two observations in VIC 40 km apart (Table 1).

Fig. 6.

Map showing the locations of Carlia sexdentata range outliers on iNaturalist in relation to IUCN and GARD range maps (range maps were identical) in green. Outliers are represented by blue markers.

Representation of Australian skinks on iNaturalist

Significant patterns in iNaturalist representation were found among both species range size and subfamily for Australian skinks. The skink species represented on iNaturalist tended to have much larger range sizes, which suggests that iNaturalist is a less successful tool for detecting geographic outliers of range-restricted species, which are already being seldomly recorded in general. As the popularity of iNaturalist continues to increase, the number of observations will continue to grow and hopefully provide data for more range-restricted species, as well as those species that occur in locations with limited human accessibility. Species among different subfamilies varied consistently both in the likelihood that they would be represented on iNaturalist, and in the number of iNaturalist observations. Egerniinae, the most highly represented despite being the least speciose, is a group also known as the ‘social skinks’ (Gardner et al. 2008), which are generally large and charismatic animals, including the genera Egernia and Tiliqua, which are often kept as pets and included a number of captive observations on iNaturalist. These characteristics seem to make it much more likely for individuals of these species to be recorded than those of other subfamilies. In contrast, Sphenomorphinae is the largest subfamily, yet the least represented. Sphenomorphinae includes the two most speciose genera of Australian skinks, namely, Ctenotus, with ~107 species, and Lerista, with ~97. These vast majority of these species are desert-dwelling (Rabosky et al. 2014; Wilson and Swan 2021) and, hence, further away from major coastal human population centres in Australia, which may explain why Sphenomorphinae generally does not have a large number of records on iNaturalist. Sphenomorphinae also has a large number of legless or near-legless species. These species tend to be more cryptic than limbed ones as they are often small burrowers (Camaiti et al. 2021, 2022), further reducing the chance for observation by iNaturalist users in this subfamily.

Both range size and subfamily had an impact on whether a species documented on iNaturalist had range outliers, although it was less distinct than overall presence on iNaturalist. Species with range outliers had significantly larger ranges than those without, but only if those outliers were novel. Whereas we were able to find outliers regardless of range size, this suggests that our method was more successful in finding novel outliers, and potentially undocumented populations, for those skink species on iNaturalist with larger range sizes. Sphenomorphinae species had a much lower probability of having both outliers in general and novel outliers than did the other subfamilies, likely because the number of records for these species tended to be much lower. However, although species within the subfamily Eugongylinae had significantly fewer records on iNaturalist than did those within Egerniinae, there was no significant difference in the probability that outliers would be found, regardless of novelty. This suggests that the number of records a species has on iNaturalist is not the only factor that determines how effective this method will be at locating range outliers, and that even species with fewer overall records may still have range outliers if they are, for example, well suited to accidental translocation or their current known range is inaccurately documented.

Non-novel outliers and differences in range map sources

Non-novel outliers for Australian skinks were most commonly only outside IUCN Red List ranges, which suggests that IUCN ranges are more conservative than are those of the other range maps used. This is perhaps not surprising given that the vast majority of IUCN Red List range maps for Australian skink species were created in 2017 (Tingley et al. 2019), whereas GARD continually updates their range maps according to the latest data and expert opinion (Caetano et al. 2022), and the sixth edition of the field guide was released in 2021 (Wilson and Swan 2021). Thus, more recently discovered populations are less likely to be featured in IUCN maps. The potentially lower accuracy in IUCN Red List range maps is a concern because these maps are used to assist conservation decisions, as well as being very commonly used in broader ecological research (Herkt et al. 2017; Hughes et al. 2021). These results provide additional indication that the range maps produced for the Red List are not the most accurate available for Australian skinks. Further, both digital range map sources often had issues with adhering to certain coastlines, particularly that of south-western Tasmania, likely owing to faults in mapping software or approaches. Islands where the species is well known were also often not included. This suggests that digital mapping can often be less accurate along complex coastlines owing to the characteristics of mapping software (Visconti et al. 2013). Map creators should use caution when describing coastal and island-dwelling species. However, the need for this is likely to be reduced as mapping software becomes more accurate over time.

Non-novel outliers that were not recorded on any of the three range map sources act to highlight our gaps in knowledge regarding distribution. Many of these outliers were part of island populations, which were likely to be less well documented on digital range maps because of the aforementioned characteristics of mapping software. The field guide is likely to have excluded these island populations because of size constraints, because range maps in the guide were physically very small and, so, unable to accurately represent offshore islands. Cryptoblepharus pannosus was an anomaly in that it had a large number of non-novel outliers in western Victoria, an area well-populated by humans where the species is known to occur (Robertson and Coventry 2019). It is unknown why the species was not recorded here in any of the map sources used primarily in this study, but this shows that unusual oversights can occur in these broad efforts to map species distributions.

Novel outliers and their potential origins

Limitations of the method

Although there were large numbers of novel outliers, which strongly suggest established populations in some species, most often species had very few outliers. Because of the nature of iNaturalist, it is unfortunately only useful as a source of presence data and does not provide any absence data (Mesaglio and Callaghan 2021). This means that these isolated outliers may actually be part of established populations, but we have no way of knowing with the data available on iNaturalist. Additionally, our method focused only on observations that were visually outside the range of the species they are listed as on iNaturalist. However, for an observation to be listed as a certain species it must be identified as such, and identifications on iNaturalist rely on the time and skill of users as well as the precision of any artificial intelligence employed to make an initial identification. Whereas this may be less of an issue for some taxa, many skinks can be very difficult to identify (Wilson and Swan 2021). This project had to eliminate 6041 observations of skinks because they were not identified to the species level. Even those that were identified to a species level included a number of misidentifications. It is quite possible that other range outliers exist on iNaturalist, but have simply been misidentified as a different species for which they are within range (Gorleri and Areta 2022). This method cannot account for such occurrences; however, doing so would involve examining every single observation for a taxonomic group so as to accurately identify them, which would massively increase the workload required.

A key feature of iNaturalist is the ability to obscure the location of a given record. This can be performed manually by users, usually either for personal reasons or for the protection of rare animals, which are possibly targeted by poachers for the pet trade, but it is also performed automatically if a species is threatened. There is no way to petition iNaturalist for unobscured records to use in scientific research; so, each user must be contacted to access exact location data (Contreras-Díaz et al. 2023; Farquhar et al. 2024). This can greatly affect the usefulness of iNaturalist for research purposes, particularly for the most vulnerable species, and reduces the effectiveness of this method, in particular because it is difficult to make assumptions on the origin of range outliers without fairly exact location data. iNaturalist records can provide evidence of previously undocumented populations, but we must be cautious because of the potential inaccuracy of citizen science records. Field studies must be performed to fully confirm a species’ presence in an area, as well as to determine their likely origin and their impact on local ecology. Any conclusions about novel populations found in this study should be taken as a suggestion for further research.

Conclusions

Our project identified novel range outliers in Australian skinks, including an established translocated population and evidence of a number of likely accidental translocations in a variety of species. Understanding species distributions is vital to conservation and management efforts, and our results can hopefully provide a basis for further study into these geographical range outliers so as to confirm their presence and determine their origins with more certainty, as well as how they may be affecting local ecosystems. Our method was more successful for certain taxa and species than for others, and its effectiveness may vary geographically, given that iNaturalist users are not uniformally distributed in space. However, the number of species represented on iNaturalist continues to increase with the number of observers, and this method could already be successful for many species around the world, particularly for groups such as reptiles and amphibians, which are both likely to be Data Deficient and to have undergone accidental human-assisted dispersal in recent years.