Open data for biogeography research of the genus Metrosideros across the south-central Pacific region

Introduction

The south-central Pacific region, defined here as the combination of the islands of Micronesia, Melanesia, and Polynesia, covers a vast proportion of the planet. The south-central Pacific region has distinct biophysical characteristics with high levels of endemism, including almost all species of the widespread genus Metrosideros (Myrtaceae) (Wright et al. 2021). The genus was thought to have originated in New Zealand, but based on earlier conclusive fossils may have originated in Tasmania in the Oligo-Miocene where it is now extinct (Tarran et al. 2016, 2017). From Australia, it may have dispersed across the Pacific through New Zealand and New Guinea where it is the major component of many Indigenous island floras. One such example is Metrosideros polymorpha (‘ōhi‘a lehua), which is dominant across many Hawai‘ian ecosystems (Mueller-Dombois and Fosberg 1998) where it is a key component of plant communities of all ages (Mueller-Dombois 1992). Unfortunately, this species is highly susceptible to rapid ‘ōhi‘a death (ROD), a disease caused by two recently introduced Ceratocystis fungal pathogens, Ceratocystis huliohia and Ceratocystis lukuohia. The origins of these species are unclear but they reside in Asia-Australian and Latin American Ceratocystis clades, respectively (Barnes et al. 2018). Many species in the genus are also susceptible to myrtle rust (caused by Austropuccinia psidii), another invasive fungal pathogen native to South America (Coutinho et al. 1998). Information on the distributions of susceptible species is critical for management of these diseases. For example, distributions of endemic Metrosideros across New Zealand have been used to inform the management response to myrtle rust (McCarthy et al. 2021). At present, ROD is only known from Hawai‘i, but there is a risk it may spread across susceptible Metrosideros in the Pacific. Given evidence that other Metrosideros species such as Metrosideros excelsa from New Zealand could also be susceptible to ROD (Luiz et al. 2022), information about the pan-Pacific distributions of Metrosideros species is critical for developing surveillance and management plans that could prevent wider impacts from potential ROD spread across the Pacific region.

Geographic information system (GIS) data on the locations of islands within the south-central Pacific region is of critical foundational importance for mapping plant distributions, as islands represent the ultimate limit of distributional extent of terrestrial species such as those in the genus Metrosideros. Unfortunately, the south-central Pacific region is cartographically disadvantaged as the convention of using the Greenwich Meridian as a default prime meridian results in some Pacific islands being artificially split. Also, there are numerous small islands within the south-central Pacific region, meaning that automated mapping from satellite imagery can be technically difficult, and creating the potential for differences in mapping scale to have pronounced effects on the inclusion or exclusion of small islands by different data creators. Therefore, appropriate GIS island location data needs to be created to support any mapping exercise in the south-central Pacific region.

For species occurrence data, the Global Biodiversity Information Facility (GBIF, https://www.gbif.org) system for aggregating worldwide biodiversity databases (Edwards et al. 2000) has become the de facto source of open access species occurrence data for biogeographical analyses. While GBIF aggregates occurrence data from a wide range of providers, there are other potential sources of data that should be considered. This is important as many Pacific nations and Metrosideros species are not well represented by GBIF data (GBIF 2024). Therefore, while GBIF provides an excellent starting point, we are conscious that there are other potentially important sources of open access occurrence data such as those from initiatives like the sPlotOpen project (Sabatini et al. 2021) that could inform a clearer picture of Metrosideros distributions in the south-central Pacific region. Investing effort in integrating these various open access datasets would provide a useful resource from which other projects that require occurrence records can be developed. In this regard, our focus on open access data is of critical importance such that the resulting data sources can be widely used.

Therefore, our objective was to create two open data products: (1) GIS data for the outlines of islands in the south-central Pacific region with relevant attribute information and in a suitable map projection with improved geometries; and (2) occurrence data of Metrosideros species compiled from several sources with unified and consistent taxonomy. Our hope is that creating this data within an open science framework will enable the broader research community to advance knowledge about Metrosideros distributions in support of issues such as those currently created by pathogenic fungi including Ceratocystis.

Materials and methods

Results

Occurrence data

No taxonomic harmonisation was required with all databases providing species names consistent with the accepted nomenclature (Table 1). We found that some species were commonly mis-identified across a wide taxonomic spectrum by either the iNaturalist contributors or the iNaturalist computer vision species identification algorithm (e.g. mis-identification of Melaleuca as Metrosideros polymorpha in California https://www.inaturalist.org/observations/9036079, or as M. excelsa in Portugal https://www.inaturalist.org/observations/82143976). Many of these mis-identifications had regional biases. For example, the related species M. excelsa, Metrosideros kermadecensis, Metrosideros collina, and M. polymorpha were frequently confused in cultivation. Metrosideros polymorpha appears to be very rarely cultivated outside of its native range of Hawai‘i, but was heavily over-reported in cultivation, especially in California where the correct identifications were usually M. excelsa and M. collina. Conversely, M. collina had been heavily under-reported in cultivation in places such as California, Spain, Portugal, New Zealand, and Australia where it had typically been mis-identified as M. excelsa and M. polymorpha.

Across all occurrence data sources, 14,726 Metrosideros observations were identified and included in our dataset. However, the taxonomic and geographic distribution of these occurrences was extremely skewed. Only eight island groups had any occurrences, and New Zealand and Hawai‘i alone contributed 98% of the occurrences, with the remaining island groups having few occurrences (Fig. 1), and 23 groups having no occurrences. Similarly, endemic species from New Zealand and Hawai‘i dominated the species occurrences, with only 27 of the targeted 54 Metrosideros species represented in the compiled dataset (Fig. 2).

Fig. 1.

Metrosideros species occurrence counts by island group. Note the log scale on the x-axis.

Fig. 2.

Metrosideros species occurrence counts. Note the log scale on the x-axis.

In terms of data sources, 99% of the occurrences came from GBIF. However, when broken down geographically, the additional plot data was concentrated specifically in Hawai‘i, the Lord Howe Island Group, and in New Caledonia. For New Caledonia in particular, the sPlotOpen data provided a good proportion of the available data (Fig. 3). When broken down by species, GBIF data dominates again, but with some species data primary or sole data source coming from HAVPlot and sPlotOpen (Fig. 4).

Fig. 3.

Metrosideros species occurrence proportions for each island group by data source.

Fig. 4.

Metrosideros species occurrence proportions by data source.

Regarding BIEN, we found that this dataset would only provide 33 additional occurrences, but without location uncertainty information. As the data was licenced under a CC-BY-NC-ND licence that precludes resharing the data in a modified form, no BIEN data was included in our compiled dataset.

Discussion

Compiling appropriate GIS data of the islands of the south-central Pacific was more challenging than expected as there were an array of options each with varying pros and cons that had to be reviewed. On balance, the use of OpenStreetMap was preferable primarily due to its permissive licencing model. The value of such reliable GIS data was immediately evident as it was used as part of our data cleaning process to identify species occurrences within marine locations in which Metrosideros species should occur.

We were less successful in compiling reliable data on all Metrosideros species’ occurrence. Unfortunately, we discovered that there was extremely limited data availability for some regions and species within GBIF, which is the world’s biggest aggregator of open access occurence data. Some Pacific islands had only tens of occurrences for Metrosideros species known to occur there. Similarly, some Metrosideros species had only tens of occurrences throughout the whole Pacific region. Given the expected distributions of many of these species (Table 1), this highlights a significant gap in some of the openly available Metrosideros occurrence data within the south-central Pacific region, likely in part due to restricted distributions and inaccessibility to rare species. What was interesting to note was how the effort to include additional data from vegetation plot databases proved to be particularly valuable for locations such as the Lord Howe Island Group and New Caledonia (Fig. 3) with some species being solely represented by data from sPlotOpen (Fig. 4). We would therefore recommend that plot data be regularly considered as part of plant occurrence data compilation, as this additional effort can fill gaps in knowledge that would result from relying on GBIF federated data alone.

Data quality is another issue, especially as much of the species occurrence data is originally derived from the iNaturalist citizen science platform. These issues were also addressed in a recent analysis by López-Guillén et al. (2024). We discovered and corrected numerous and widespread species mis-identifications that had achieved ‘research grade’ status, and incorrect reporting of observations as wild rather than their true cultivated status distorts natural distribution patterns. Such errors are problematic in themselves, but mis-identifications have the potential to bias the underlying computer vision species identification algorithms of iNaturalist (iNaturalist 2023), which result in compounding errors and the geographically localised collections of mis-identifications we observed. While we corrected many errors as part of our review of the data, these changes do require a majority of the iNaturalist community to agree with our re-identifications before they can become ‘research grade’ again, and as such we would encourage those interested in developing our knowledge of Metrosideros distributions to become similarly engaged in reviewing iNaturalist data. A strength of the iNaturalist platform is that not only are the species observations uploaded by numerous contributors, but the curatorial components (identifications, annotations etc.) are similarly crowdsourced. To support that effort, we identified some general trends in mis-identification.

Metrosideros kermadecensis has smaller and more rounded leaves than the closely related M. excelsa, but relative leaf size can be difficult to interpret from uploaded images; a more reliable character is the sporadic flowering season of M. kermadecensis throughout most of the year compared to a tight peak flowering period (November–January in the Southern Hemisphere) typical for M. excelsa. Reliable identification of M. kermadecensis in cultivation is complicated by the occurrence of hybrids between M. kermadecensis and M. excelsa, and this combination has only recently been added to the iNaturalist platform.

With more than 40 name cultivars (Dawson and Heenan 2010), M. excelsa is the most cultivated species of the genus worldwide. Its preference for warm-temperate coastal climates exposes mis-identifications in areas not suited to its cultivation, including common mis-identification for Callistemon/Melaleuca in drier inland areas of the USA (e.g. Arizona and Texas). Although in the same family (Myrtaceae) and sharing colourful red flower filaments, the flowers and leaves of Callistemon/Melaleuca (and Corymbia ficifolia, another mis-identification frequently encountered) are easily separable from Metrosideros, most likely illustrating the occurrence of local confirmation bias among observers.

Members of different genera and families, namely several Calliandra (Fabaceae), Combretum (Combretaceae), and Stifftia (Asteraceae) species were also mis-identified as Metrosideros due to a superficial similarity of flowers. Lonicera ligustrina (box-leaved honeysuckle; Caprifoliaceae) was frequently mis-identified in the UK, France, Germany, and elsewhere as one of the New Zealand endemic climbing species of Metrosideros (Metrosideros colensoi, Metrosideros diffusa, Metrosideros fulgens, or Metrosideros perforata) due to similar growth form of stems and leaves. A range of other unrelated plants (e.g. Peperomia rotundifolia, Zanthoxylum beecheyanum, several orchids and ferns) were also occasionally mis-identified as these climbing Metrosideros species. Species from these unrelated genera are well outside of the natural and cultivated range of New Zealand endemic Metrosideros. With varying degrees of certainty, numerous records were found marked as wild when they were actually observations of cultivated plants.

A similar comment around improving data quality can also be made regarding the use of OpenStreetMap data because any errors that are detected by a user of this data can be corrected within OpenStreetMap by anyone, meaning that the research community can again build a data resource that everyone can benefit from. While we have already noted that crowdsourced data can have quality issues (Basiri et al. 2019), the maturity and quality of OpenStreetMap is evident from its use by many mainstream companies (Mooney and Minghini 2017).

We only removed occurrences greater than 10 km from land or that had a locational uncertainty greater than 10 km, so users of our data should consider other potential data issues and remedies (see: Zizka et al. 2020; Sillero and Barbosa 2021). While errors in species observation data can be corrected, another data quality issue is that of temporal and geographical sampling bias. Much of the data is concentrated into more recent time and into certain areas of the Pacific due to differences in survey effort and data mobilisation resources. If ignored, such bias can lead to misleading distributional analyses; therefore, care must be taken to use approaches that are insensitive to the inevitable presence of biased data.

We have also only considered distribution information for naturally occurring Metrosideros species. However, many Metrosideros species, in particular M. excelsa, have undergone selection to produce almost 100 cultivars (Dawson and Heenan 2010). This clearly indicates that the distributions of natural Metrosideros species presented here most likely underpredict true distributions of Metrosideros species as many cultivars are likely to be found in human-dominated environments such as gardens and parks. These localised concentrations of Metrosideros individuals could have implications for issues such as plant pathology surveillance and disease control and should not necessarily be ignored.

Despite these possible future developments, our goal is that by making this Metrosideros distributional information openly available, and by updating the information as new and better occurrence data becomes available, we can meet an immediate and critical need for data that will support the development of pan-Pacific surveillance and management plans for Ceratocystis (and potentially) other potential Metrosideros pathogens.

While more data can be collected or mobilised and errors in existing data can be addressed, urgency related to the possible arrival of the two Ceratocystis pathogens causing ROD requires at least an initial mapping of Metrosideros species across the Pacific region, as spreading across these island chains is one of the possible invasion pathways. Despite the recognised limitations in our occurrence data, our datasets represent the most comprehensive data to support such efforts, and both the GIS and occurrence data described here are made freely available under permissive open licences.