# Novel detection of provenance in the illegal wildlife trade using elemental data

## Abstract

Despite being the fourth largest criminal market in the world, no forensic tools have been sufficiently developed to accurately determine the legal status of seized animals and their parts. Although legal trading is permissible for farmed or captive-bred animals, many animals are illegally removed from the wild and laundered by masquerading them as captive bred. Here we present high-resolution x-ray fluorescence (XRF) as a non-invasive and cost-effective tool for forensic classification. We tested the efficacy of this technique by using machine learning on a training set of zoo specimens and wild-caught individuals of short-beaked echidnas (Tachyglossus aculeatus), a small insectivorous monotreme in Australia. XRF outperformed stable isotope analysis (δ13C, δ15N), reducing overall classification error below 4%. XRF has the added advantage of providing samples every 200 μm on a single quill, enabling 100% classification accuracy by taking the consensus of votes per quill. This accurate and cost-effective forensic technique could provide a much needed in situ solution for combating the illegal laundering of wildlife, and conversely, assist with certification of legally bred animals.

## Introduction

The international illegal wildlife trade (IWT) is the fourth largest criminal market in the world, worth between USD\$7-23 billion annually1. Hundreds of millions of plants and animals are imported annually into the USA and Europe, used for medicine, food, pets, or fashion. The cost of IWT to biodiversity and species conservation is considerable2,3,4, as are costs to the individuals being caught and traded. Impacts occur during capture, both to the individuals being caught and to inhabitants of the wider ecosystem, and post-capture, including ongoing welfare harms to individuals5 and the potential spread of zoonotic and agricultural diseases6,7. While many animals are traded as body parts, the live pet trade is an increasing driver of biodiversity loss, particularly in the Asia-Pacific region7,8,9. Since 1973, the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES) has provided regulation of international trade of over 5,800 fauna species, covering endangered species, species subject to exploitation (e.g. exotic pets), and nationally protected species. Successful enforcement of the Convention is generally regarded as critical for combating IWT, although schemes for disincentivising trade and stimulating community behaviour change are also important10,11. However, enforcement falls to member countries and is typically carried out by customs officers. Problematically, although reports on wildlife seizures provide estimates on trade flows12, fraudulence is widespread, with much larger volumes seized than could possibly be legally obtained13.

Under CITES agreements, certified breeding farms are permissible for some species3,14,15,16. Breeding farms are approved to ostensibly reduce pressure from illegal poaching on wild populations and to assist in biodiversity conservation17. Their good intent is often thwarted by the establishment of unlicensed illegal breeding farms18 and the illegal sourcing of founder stock from the wild19. Furthermore, there are serious concerns that wildlife breeding farms are being used to launder illegally caught wildlife16,17,20,21. For example, at least 80% of green pythons (Morelia viridis) exported from Indonesia are suspected to be illegally wild-caught17. Likewise, illegal laundering of crocodile lizards (Shinisaurus crocodilurus), listed as endangered on the IUCN Red List, has been implicated in its decline in Vietnam due to unsustainable exploitation22. To combat this, much emphasis has been placed on tracking illegal trade networks23,24 and in developing forensic tools to determine the origins of species being traded22,25. However, stemming the illegal trade is challenged because existing forensic tools are either inaccurate, inaccessible, or too expensive, making it difficult to confidently determine specimen identity, geographic origin, or legal status.

Forensic tools have primarily focused on the determination of species identity, necessary to prevent the laundering of prohibited species under the name of similar but legally-tradable species26. Genetic techniques, like DNA metabarcoding, are suited to this application and have been used to successfully classify specimens26,27. A further focus is determining the geographic origins of specimens, particularly for species with large geographic ranges, like pangolin, reptiles, and birds13,25,28. Efforts to forensically test for geographic origin have centred on analysis of stable isotopes29,30, relying on intrinsic tissue signatures fractionated from diets during assimilation31,32,33. Keratinous tissue such as hair, feathers, scales, and nails, are metabolically inert following synthesis and so maintain an isotopic record reflecting the time and place where the tissue was synthesized, dependent upon temporal and spatial feeding habitats of different species and on tissue specific growth and moulting rates29. Importantly, isotopic signatures can be used to detect laundering in IWT by discriminating between captive-bred and wild-caught animals. Some success has been had with reptiles in Vietnam22 and with fish in China, Australia, Malaysia, and Indonesia34. However, both stable isotopes and genetic testing perform best for geographically isolated species with small home ranges or with restricted ecological niches22,25. Furthermore, both tools are expensive, require considerable sample preparation, and cannot be conducted in situ, limiting accessibility and versatility.

The lack of a validated, rigorous, and cost-effective forensic method to identify laundered wildlife is a substantial impediment to the enforcement of regulations and laws needed to protect threatened species and wild populations. Here we propose the use of a novel technique to overcome the inherent limitations of existing forensic approaches for specimens with wider geographical ranges and/or more diverse feeding strategies. High resolution x-ray fluorescence (XRF) is a non-destructive analysis tool that measures a wide range of elemental abundances present in samples35,36. High resolution XRF has found prominence in its application for elemental analysis of sediment cores and in dendrochronology to determine temporal and spatial variability of samples37,38, but can also be used to provide information on geographic location and on diet39. For example, elemental analysis has recently been used to determine country of origin for aquaculture shrimp40. Despite being readily available, non-destructive, and cost-effective, no previous studies have tested the efficacy of this method as a forensic tool for combating IWT.

As a test case, we chose to employ XRF to discriminate between captive-bred and wild-caught specimens of a species that is highly desired in the international pet trade and which has accessible wild populations and zoo collections, the short-beaked echidna (Tachyglossus aculeatus). Short-beaked echidnas are native to Australia, Indonesia, New Guinea, and Papua New Guinea, and can be legally traded from Indonesia if they are accompanied by a permit declaring them as captive-bred. However, echidnas are notoriously difficult to breed in captivity and the survival rate of captive born juveniles is poor. In the forty years preceding 2015, fewer than 20 are known to have been born in captivity and successfully raised in zoos in Australia. Despite this, 25 short-beaked echidnas were seized or traded as captive-bred on Indonesian trade routes between July 2011 and July 2012 alone. Question marks over the legality of seized specimens remain unanswered, stymying enforcement by the lack of evidence proving that specimens were wild-caught and prohibiting certification if they were in fact captive-bred. In this study we utilised both high resolution XRF and stable isotope analysis to classify samples taken from a reference set of echidnas of known origins (either captive-bred or wild-caught). Our goal was to provide an assessment of the merit of adopting elemental analysis as a forensic tool in IWT enforcement.

## Results

### Elements

As echidna quills differed in length, the number of XRF scans, spaced every 200 μm, ranged between 94 and 342. Quill lengths did not differ significantly between wild and captive echidnas (p = 0.4924) and the median number of samples per quill was 192 for both groups. Univariate testing of elemental differences between captive and wild echidna quills indicated that five of the 24 elements were significantly different, controlling for nesting within each quill (Table 1). In each of the five cases, elemental counts of Calcium (Ca), Magnesium (Mg), Nickel (Ni), Sulphur (S), and Zinc (Zn) were significantly lower in wild caught echidna quills. Variation in counts along each quill were marked, contributing to the lack of discriminatory power among many of the elements when analysed univariately (Fig. 1).

### Stable isotopes

Captive echidna quills had significantly higher values of δ13C (p = 0.001) and δ15N (p < 0.001) than wild echidnas (Welch Two Sample t-test). For δ13C, values for wild echidnas ranged from −24.2 to −18.56‰ ($$\bar{x}$$ = −22.46 ± 1.64‰) and from −23.28 to −13.98‰ ($$\bar{x}$$ = −18.29 ± 3.54‰) for captive echidnas). For δ15N, values for wild echidnas ranged from 4.20 to 7.16‰ ($$\bar{x}$$ = 5.32 ± 1.05‰) and from 7.20 to 11.62‰ ($$\bar{x}$$ = 9.59 ± 2.14‰) for captive echidnas. Isotopic signatures separated captive-bred quills from wild-caught quills well, except for captive quill 4 (Fig. 2). The quill from a wild-caught echidna that was subsequently in captivity for 5 months (W-C1) was closer in signature to wild echidnas than captive ones.

### Classification

Machine learning using both elemental and isotopic data could accurately classify quills as either wild or captive (Table 2). Elemental data provided the greatest precision, where only 185 misclassification errors from 4,627 sampling positions across the 23 quills were predicted (3.99%), compared with 2 out of 23 misclassification errors predicted using isotopic data (8.69%). Prediction errors were significantly higher for captive quills than wild quills (t-value = −5.661, p < 0.001). At the quill level (Fig. 3), C11 had the highest predictive uncertainty with 27.5% of samples along the quill having more than 50% of trees supporting misclassification, while the next most similar quills with predictive uncertainty were W3 and W6 (with 11.3% and 9.0% of samples supporting misclassification respectively). By aggregating classification predictions from samples along each quill (Fig. 3), consensus among sample classifications predicted quill status 100% of the time. By comparison, classification accuracy based on isotopic data was lower, with one captive quill misclassified as wild (8.33%) and one wild quill was misclassified as captive (9.09%).

Ranking predictive importance using mean decrease in accuracy identified the top five ranked elements for distinguishing captive-bred from wild-caught echidnas as nickel (60.5), iron (57.9), calcium (56.6), chromium (43.4), and zinc (38.5). Partial dependencies of predictor variable influences illustrated that counts were indicative of captive quills when above 600 for nickel, above 200 for calcium, above 300 for chromium, and above 250 for zinc. An absence of iron or very low iron counts were indicative of captive-bred echidnas, while iron counts above 5,000 were indicative of captivity for some echidnas. For the isotopic random forest model, mean decrease in accuracy was higher for δ15N (22.9) than for δ13C (12.6). Partial dependencies of predictor variable influence illustrated that values above 8‰ δ15N and above −16‰ δ13C were indicative of captive echidna quills.

## Discussion

Our study demonstrated the potential utility of high-resolution x-ray fluorescence for distinguishing between captive and wild short-beaked echidnas with a high level of accuracy (Fig. 4). Scans every 200 μm for 24 elements correctly classified samples from echidna quills 95.9% of the time. Furthermore, by taking multiple scans per quill to reflect temporal variation in diet as each quill is developed, classification by consensus was 100% accurate. We validated our findings by confirming that δ13C and δ15N stable isotopes are useful discriminators between captive and wild individuals, with a classification accuracy of 91.3%.

The utility of 13C and 15N isotopic data for determining geographic origin is typically restricted to species with narrow dietary niches and restricted home ranges, like echidnas. Species with diverse diets and large home ranges would be expected to exhibit larger variability in δ13C and δ15N ratios, reducing the discriminatory power to differentiate between groups as potential overlap between diets may occur. This issue does not appear to apply to elemental data because of the large array of elements for which data are collected and the specific locational information that elemental data represent41. Rather than reducing classification accuracy, variation in elemental signatures drives predictive performance. High resolution elemental scans along the quill provide a comprehensive analysis of temporal changes that reflect both dietary changes and changes in metabolism during periods of torpor, medical treatment (administering of medicinal drugs to captive animals), sexual development stage, and pregnancy/lactation in females41. Considerable variability in elemental abundances were observed in captive echidna quills (Fig. 1), likely due to the large range of ingredient sources with potentially global origins that comprise the captive diet. In contrast, wild echidnas experience less elemental variation in their environment because they have relatively small home ranges and restricted dietary sources, particularly as their primary dietary sources, termites and ants, are highly localized and do not travel large distances. While zoo diets aim to meet nutritional needs of the animal42, we showed that the diets of the two groups differed both elementally and isotopically. Captive diets were enriched in δ15N, one to two trophic levels higher than wild diets, and were less depleted in δ13C, two to four trophic levels higher than their natural dietary source.

Laundered echidnas are likely to spend some time in captivity prior to being traded. Although no information is available on growth rates of quills, we showed that elemental signatures were robust to detecting wild-caught echidnas when individuals had been in captivity for at least 6 months. Testing of a wild-captive quill (W-C1) suggested that while the isotopic signature had potentially shifted towards the captive group (Fig. 2), elemental analysis clearly identified the quill sample as wild with only 4% of scan samples supporting misclassification. The benefit of using elemental analysis of keratin samples is that they are long-term records incorporating locational information over months or even years. This means that wild animals would need to be kept in captivity for long time periods before their elemental signature shifted to captive.

Our findings are significant as they provide strong support for x-ray fluorescence as a new forensic tool for combating the global illegal wildlife trade. It should be noted that our study was limited in both sample size and spatial extent, while our captive echidnas were from the same source and fed the same diet. Further research is needed to ascertain detection accuracy among a more diverse set of samples, as well as applying the technique to other highly traded species with keratinous tissues such as birds (feathers), reptiles (scutes, nails, sloughs), and mammals (nails, scales, quills). An advantage of this technique is that it is non-invasive: samples can be taken from animals with minimal handling, and sometimes no handling, while repeat scans can be taken from samples to reduce sensitivity to outliers while still maintaining the ability to examine phenological patterns or post-treatment differences. Here we employed high resolution XRF in a laboratory setting to detect differences between captive and wild echidna quills, however, the method has great potential as an in situ forensic tool as hand-held multi-elemental XRF equipment are available. While hand-held devices typically scan for only a handful of elements at a time, machine learning provides a robust technique for obtaining sufficient predictive accuracy when combined with high resolution training sets. Our research team are currently measuring elemental abundances using hand-held devices to distinguish between captive and wild animals to provide an economical field-based tool for use in places like markets and farms and during border control where trading of animals frequently occurs.

## Methods

### Sample collection and preparation

Quills from captive short-beaked echidnas were sourced from Taronga Zoo, Taronga Conservation Society Australia (Sydney, Australia). Quills from wild echidnas were sourced from those presented for veterinary care or post-mortem examination at the Taronga Wildlife Hospital at Taronga Zoo and from the Wildlife Veterinary Department of Mogo Zoo (Mogo, Australia). A total of 24 quills were sampled from 22 echidnas. Twelve captive samples were collected from adult residents at Taronga Zoo for between 22 months and 20 years ($$\bar{x}$$ = 10.5 years). All captive echidnas were fed the same diet. Eleven wild quill samples were collected from ten individuals presented for veterinary care. One wild echidna was sampled twice after 1 and 7 days in care. Echidnas classified as wild had quills sampled at between 0 and 26 days in care ($$\bar{x}$$ = 4 days). These echidnas had been recovered or rescued from a range of locations within 290 km of Sydney, Australia. One echidna was classified as wild-captive as it was sampled 188 days after being kept for veterinary care. Quills were cleaned of any surface contaminants using a lint free cloth and deionized water and stored in paper envelopes at room temperature. A sample of Taronga Zoo captive echidna diet (Vetafarm, Wagga Wagga, 2700, NSW, Australia) was oven dried for 48 hours at 60 °C, homogenised using a mortar and pestle, and ground to a fine powder for stable isotope analysis.

All methods were carried out in accordance with approvals authorised by the Taronga Conservation Society Animal Ethics Committee (3a/02/15, 4a/02/18 and R16B229).

### Elemental and SIA analyses

Elemental analysis was performed at the Australian Nuclear Science and Technology Organisation (ANSTO) Lucas Heights facility using an ITRAX micro X-ray fluorescence (μXRF) core scanner. Whole quills were mounted length-ways along the ITRAX scanning platform and scanned at a resolution of 200 μm. Quills were aligned using a laser beam to ensure the scanning path (4 mm wide) covered all quills. A Molybdenum target tube was used with a dwell time of 10 seconds at settings of 30 kV and 55 mA. Elemental abundance data was collected for 24 elements (Fig. 1). Following ITRAX scanning the same quills were then ground using a ball mill (Retsch Mixer Mill 200) to create a homogenized powder for each quill. Samples were weighed into tin caps and stable carbon (13C) and nitrogen (15N) isotopes were analysed at the Bioanalytical Mass Spectrometry Facility (BMSF) at the Mark Wainwright Analytical Centre (MWAC) University of New South Wales. A Delta V Advantage Isotope Ratio Mass Spectrometer and Flash 2000 Organic Elemental Analyzer fitted with a Conflo 4 was used. A sample of captive diet from Taronga Zoo was also analysed for 13C and 15N for comparison with captive quill results. Data are available upon request from the authors.

### Statistical analyses

Differences in elemental signatures between captive-bred and wild-caught echidnas were tested by constructing univariate generalised mixed models in the ‘lme4’ package in R43. Negative binomial models were used to account for overdispersed data, while quill identity was used as a random variable to account for nesting of scans. We then used the tree-based machine learning algorithm random forest44 to determine whether quill samples from echidnas could be classified using elemental composition and δ13C and δ15N isotope ratios, implemented in the ‘randomForest’ package in R. Random forest is a machine learning technique that constructs an ensemble of unpruned classification trees by bootstrapping samples of training data and random predictor variables for tree induction44. Classification is made by aggregating (majority vote or averaging) the predictions of the ensemble. Two sets of random forest models were built for elemental and stable isotope data separately, running the default number of 500 trees and four variables at each split for elemental data and one variable for isotopic data. Variable importance was determined by ranking the mean decrease in accuracy for each predictor variable and by examining partial dependency plots (Supplementary Information).

### Ethics

This work was carried out under Animal Care and Ethics approvals from Taronga Conservation Society: AEC 4a/02/18 (previously prior to Feb 2018 it was 3a/02/15). Title: Collection of opportunistic samples for researchers from live animals during veterinary procedures or routine husbandry procedures. And Taronga Conservation Society Internal Sampling Licence Approval number R16B229. Title: Tracing the trade: using keratinous tissue to combat illegal wildlife trade.

## Data Availability

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

## References

1. 1.

Rosen, G. E. & Smith, K. F. Summarizing the evidence on the international trade in illegal wildlife. Ecohealth 7, 24–32, https://doi.org/10.1007/s10393-010-0317-y (2010).

2. 2.

Blundell, A. G. & Mascia, M. B. Discrepancies in reported levels of international wildlife trade. Conserv. Biol. 19, 2020–2025, https://doi.org/10.1111/j.1523-1739.2005.00253.x (2005).

3. 3.

United Nations Office on Drugs and Crime. Wildlife and forest crime analytic toolkit. Report No. Revised Edition, 212 (2012).

4. 4.

García-Díaz, P., Ross, J. V., Ayres, C. & Cassey, P. Understanding the biological invasion risk posed by the global wildlife trade: propagule pressure drives the introduction and establishment of Nearctic turtles. Glob. Chang. Biol. 21, 1078–1091, https://doi.org/10.1111/gcb.12790 (2015).

5. 5.

Baker, S. E. et al. Rough trade: Animal welfare in the global wildlife trade. Bioscience 63, 928–938, https://doi.org/10.1525/bio.2013.63.12.6 (2013).

6. 6.

Smith, K. M. et al. Summarizing US wildlife trade with an eye toward assessing the risk of infectious disease introduction. Ecohealth 14, 29–39, https://doi.org/10.1007/s10393-017-1211-7 (2017).

7. 7.

Bush, E. R., Baker, S. E. & Macdonald, D. W. Global trade in exotic pets 2006–2012. Conserv. Biol. 28, 663–676, https://doi.org/10.1111/cobi.12240 (2014).

8. 8.

Wu, J. & Sadovy de Mitcheson, Y. Humphead (Napoleon) wrasse Cheilinus undulatus trade into and through Hong Kong. (TRAFFIC, 2016).

9. 9.

Auliya, M. et al. Trade in live reptiles, its impact on wild populations, and the role of the European market. Biol. Conserv. 204, 103–119, https://doi.org/10.1016/j.biocon.2016.05.017 (2016).

10. 10.

Biggs, D. et al. Developing a theory of change for a community-based response to illegal wildlife trade. Conserv Biol 31, 5–12, https://doi.org/10.1111/cobi.12796 (2017).

11. 11.

Cooney, R. et al. From poachers to protectors: Engaging local communities in solutions to illegal wildlife trade. Conservation Letters 10, 367–374, https://doi.org/10.1111/conl.12294 (2017).

12. 12.

Shepherd, C. R., Connelly, E., Hywood, L. & Cassey, P. Taking a stand against illegal wildlife trade: the Zimbabwean approach to pangolin conservation. Oryx 51, 280–285, https://doi.org/10.1017/S0030605316000119 (2017).

13. 13.

Heinrich, S. et al. Where did all the pangolins go? International CITES trade in pangolin species. Global Ecology and Conservation 8, 241–253, https://doi.org/10.1016/j.gecco.2016.09.007 (2016).

14. 14.

D’Cruze, N. et al. A star attraction: The illegal trade in Indian Star Tortoises. Nature Conservation 13, 1 (2015).

15. 15.

Ogden, R. & Mailley, J. A review of wildlife forensic science and laboratory capacity to support the implementation and enforcement of CITES. (UNODC, 2016).

16. 16.

Janssen, J. & Chng, S. C. L. Biological parameters used in setting captive-breeding quotas for Indonesia’s breeding facilities. Conserv. Biol. 32, 18–25, https://doi.org/10.1111/cobi.12978 (2018).

17. 17.

Lyons, J. A. & Natusch, D. J. D. Wildlife laundering through breeding farms: Illegal harvest, population declines and a means of regulating the trade of green pythons (Morelia viridis) from Indonesia. Biol. Conserv. 144, 3073–3081, https://doi.org/10.1016/j.biocon.2011.10.002 (2011).

18. 18.

Shi, H., Fan, Z., Yin, F. & Yuan, Z. New data on the trade and captive breeding of turtles in Guangxi Province, South China. Asiat. Herpetol. Res. 10, 126–128 (2004).

19. 19.

Brooks, E. G. E., Roberton, S. I. & Bell, D. J. The conservation impact of commercial wildlife farming of porcupines in Vietnam. Biol. Conserv. 143, 2808–2814, https://doi.org/10.1016/j.biocon.2010.07.030 (2010).

20. 20.

Nijman, V. & Shepherd, C. R. Trade in tigers and other wild cats in Mong La and Tachilek, Myanmar – A tale of two border towns. Biol. Conserv. 182, 1–7, https://doi.org/10.1016/j.biocon.2014.10.031 (2015).

21. 21.

Nijman, V. & Shepherd, C. R. Wildlife trade from ASEAN to the EU: Issues with the trade in captive-bred reptiles from Indonesia. (TRAFFIC Europe Report for the European Commission, 2009).

22. 22.

van Schingen, M. et al. Can isotope markers differentiate between wild and captive reptile populations? A case study based on crocodile lizards (Shinisaurus crocodilurus) from Vietnam. Global Ecology and Conservation 6, 232–241, https://doi.org/10.1016/j.gecco.2016.03.004 (2016).

23. 23.

Patel, N. G. et al. Quantitative methods of identifying the key nodes in the illegal wildlife trade network. Proc. Natl. Acad. Sci. USA 112, 7948–7953, https://doi.org/10.1073/pnas.1500862112 (2015).

24. 24.

van Uhm, D. P. The illegal wildlife trade: Inside the world of poachers, smugglers and traders. (Springer, 2016).

25. 25.

Natusch, D. J. D. et al. Serpent’s source: Determining the source and geographic origin of traded python skins using isotopic and elemental markers. Biol. Conserv. 209, 406–414, https://doi.org/10.1016/j.biocon.2017.02.042 (2017).

26. 26.

Luczon, A. U., Ong, P. S., Quilang, J. P. & Fontanilla, I. K. C. Determining species identity from confiscated pangolin remains using DNA barcoding. Mitochondrial DNA Part B 1, 763–766, https://doi.org/10.1080/23802359.2016.1238752 (2016).

27. 27.

Mwale, M. et al. Forensic application of DNA barcoding for identification of illegally traded African pangolin scales. Genome 60, 272–284, https://doi.org/10.1139/gen-2016-0144 (2017).

28. 28.

Gaubert, P. et al. The complete phylogeny of pangolins: scaling up resources for the molecular tracing of the most trafficked mammals on Earth. J. Hered., https://doi.org/10.1093/jhered/esx097 (2017).

29. 29.

Hobson, K. A. Tracing origins and migration of wildlife using stable isotopes: a review. Oecologia 120, 314–326 (1999).

30. 30.

Haché, S., Hobson, K. A., Villard, M. A. & Bayne, E. M. Assigning birds to geographic origin using feather hydrogen isotope ratios (δ2H): importance of year, age, and habitat. Can. J. Zool. 90, 722–728, https://doi.org/10.1139/z2012-039 (2012).

31. 31.

Carolan, J. V., Mazumder, D., Dimovski, C., Diocares, R. & Twining, J. Biokinetics and discrimination factors for 13C and 15N in the omnivorous freshwater crustacean. Cherax destructor. Mar. Freshwater Res. 63, 878–886 (2012).

32. 32.

Mazumder, D., Johansen, M. P., Fry, B. & Davis, E. Muscle and carapace tissue–diet isotope discrimination factors for the freshwater crayfish Cherax destructor. Mar. Freshwater Res. 69, 56–65, https://doi.org/10.1071/MF16360 (2018).

33. 33.

Fry, B. Stable Isotope Ecology. (Springer, 2006).

34. 34.

Gopi, K., Mazumder, D., Saintilan, N., Yusoff, F. M. & Sammut, J. Stable isotope analysis as a tool for determining seafood provenance. J. Environ. Biol. (2018).

35. 35.

Croudace, I. & Rothwell, G. Micro-XRF sediment core scanners: Important new tools for the environmental and earth sciences. Spectroscopy Europe 22, 6 (2010).

36. 36.

Croudace, I. W., Rindby, A. & Rothwell, G. R. ITRAX: description and evaluation of a new multi-function X-ray core scanner. Geological Society, London, Special Publications 267, 51–63, https://doi.org/10.1144/GSL.SP.2006.267.01.04 (2006).

37. 37.

Thomson, J., Croudace, I. W. & Rothwell, R. G. A geochemical application of the ITRAX scanner to a sediment core containing eastern Mediterranean sapropel units. Geological Society, London, Special Publications 267, 65–77, https://doi.org/10.1144/GSL.SP.2006.267.01.05 (2006).

38. 38.

Grudd, H. Tornetrask tree-ring width and density AD 500-2004: a test of climatic sensitivity and a new 1500-year reconstruction of north Fennoscandian summers. Clim. Dyn. 31, 843–857 (2008).

39. 39.

Hobson, K. A. & Norris, R. D. In Terrestrial Ecology Vol. 2 Terrestrial Ecology Series (eds Hobson, K. A. & Wassenaar, L. I.) 1–19 (Elsevier, 2008).

40. 40.

Li, L. et al. Assessment of elemental profiling for distinguishing geographic origin of aquacultured shrimp from India, Thailand and Vietnam. Food Control 80, 162–169, https://doi.org/10.1016/j.foodcont.2017.04.046 (2017).

41. 41.

Iyengar, G. V. Elemental analysis of biological systems. Volume 1. Biomedical, environmental, compositional, and methodological aspects of trace elements. (Taylor and Francis, CRC Press, 1989).

42. 42.

Moore, B. D., Marsh, K. J., Wallis, I. R. & Foley, W. J. Taught by animals: how understanding diet selection leads to better zoo diets. Int. Zoo Yearbook 39, 43–61, https://doi.org/10.1111/j.1748-1090.2005.tb00004.x (2005).

43. 43.

R: A language and environment for statistical computing (R: Foundation for Statistical Computing, Vienna, Austria, 2017).

44. 44.

Breiman, L. Random forests. Machine learning 45, 5–32 (2001).

## Acknowledgements

The authors wish to acknowledge the following people: S Chng and C. Beastall (TRAFFIC), D. Chong and Dr. T. Allcock, veterinary and nursing staff (Taronga Wildlife Hospital), Dr. R. Spindler (Taronga Conservation Society Australia), and the Centre for Ecosystem Science, UNSW.

## Author information

Research lead by K.B., research concept developed by K.B., P.M., L.T. and M.S. Sample collection by P.H., L.T. and M.S. Sample analysis by K.B. D.M. and P.G., statistical analysis by D.R. and K.B. Manuscript written by K.B., D.R., P.M., L.T., M.S. and D.M. Figures prepared by D.R. and K.B.

Correspondence to Kate J. Brandis.

## Ethics declarations

### Competing Interests

The authors declare no competing interests.

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions