APCalign: an R package workflow and app for aligning and updating flora names to the Australian Plant Census

Introduction

Taxonomic names and lists are essential in modern e-research infrastructure, enabling us to link concepts about what taxa exist to information about where they occur, how they are related to one another, and what traits they have. However, these linkages are easily broken or missed, owing to inconsistencies in the delineation and naming of taxon concepts among different datasets (Franz and Peet 2009; Sandall et al. 2023), where a taxon concept can be described as ‘a circumscribed set of organisms’ (https://github.com/tdwg/tnc/issues/1) (Whitbread 2018). The scientific name applied at a given point in time indicates a snapshot of our understanding of evolutionary relationships within a taxonomic group. However, names change, as understanding of what morphological or genetic features circumscribe a specific taxon concept is refined. Scientific names are also easily misspelt or mistyped. When dealing with large numbers of taxa, such errors are hard to spot. To maintain the unique link among a scientific name, the associated taxon concept, and associated data, diverse users must be able to automatically align and update a list of possibly inaccurate or outdated scientific names to the currently accepted scientific name.

A range of software tools have been developed to resolve taxonomic mismatches (to the extent possible), each referencing one or more taxonomic datasets (Grenié et al. 2023; Schellenberger Costa et al. 2023); however, as yet, none are specific to the current Australian National Species Lists (auNSL) for vascular plants (https://biodiversity.org.au/nsl/). Typically, these tools input a list of species names, returning a list of said names matched to the most likely currently accepted name. Working out the best or most likely match for a species name can be a computationally intensive task; still, modern tools have found several approaches to do this with reasonable speed and measurable accuracy. Different tools also have different features (Table 1) and are presented in a range of formats, from more accessible browser-based tools, such as the Taxonomic Name Resolution Service (TNRS; Boyle et al. 2013) to more specialised R packages, such as the ‘taxize’ R-package (Chamberlain and Szöcs 2013; Chamberlain et al. 2022). Whereas all existing global tools do handle names of Australian taxa, the taxon lists and taxonomy in these global resources can differ from our National Species Lists, because they are managed by different authorities, each with their own missions and guidance councils (Garnett et al. 2020; Schellenberger Costa et al. 2023).

In Australia, the national taxonomic standard for vascular plants is the Australian Plant Census (APC), supported by a comprehensive list of plant names documented in references, the Australian Plant Names Index (APNI). The APC is endorsed by the Council of Heads of Australasian Herbaria (CHAH), with taxonomy being agreed on by consensus. The APC is updated on a regular basis, following a review of both newly described plant taxa and taxonomic revisions, first by a working group comprising representatives from all major Australasian herbaria, followed by ratification by CHAH. The working group reviews taxon concepts extracted from recent taxonomic publications, ensuring Australia’s botanical community considers newly published names and changed taxon circumscriptions robust. The names documented in APNI and accepted by APC may be out-of-step with those used in international checklists, because each checklist may deem different scientific names or taxonomies as ‘current’. Having a national checklist is also essential for Australia because of the inclusion of large numbers of phrase names within the APC/APNI that are excluded from international checklists. Phrase names are assigned to unique taxa that are yet to be formally described, a nomenclatural process that is particular to Australian plant taxonomy (Barker 2005). These names follow a specific convention, namely, the generic name, a rank indicator, a geographic or morphological descriptor, the collector’s name and a number representing an herbarium voucher for the new taxon concept. Moreover, as our official national list, the APC is the natural (sometimes mandated) focus for documenting and linking data for Australian vascular plants.

Here, we provide a new software tool, ‘APCalign’, that can quickly match plant names to accepted taxon concepts in the APC and to names in APNI. ‘APCalign’ can function both as an R package for scientific and technical users, and in a browser interface for lay users. ‘APCalign’ offers a two-step process for creating a list of names matched to the APC/APNI. In the first step, algorithms align each name input to the best match of any name within the APC/APNI, allowing matches to the infraspecies, species, genus, or family level, as is appropriate. In the second step, aligned names that have been matched to a scientific name within the APC are updated to a currently accepted taxon concept within the APC. Our implementation builds on best practices for global taxonomic resources, as well as additional enhancements (Table 1). It includes a well-considered and well-tested sequence of algorithm matching, including both complex string matches and fuzzy-matching, to maximise meaningful links. The taxonomic updating functions handle taxonomic uncertainties. For scientific names linked to taxon concepts accepted by APC, the final table adds distribution and native status. In total, 11 functions are exported by ‘APCalign’ (Table 2), including functions to download taxonomic resources, functions to clean, align and update names to a name recorded in the APC/APNI, and functions to tabulate information about taxon distribution, diversity, and native status. By working directly with Australia’s National Species Lists for vascular plants, ‘APCalign’ fills an important gap in our biodiversity e-resources.

The package is coded in R (ver. 4.3.1, https://www.R-project.org/), an open-source language widely used in the ecological sciences (R Core Team 2024). Within R, we leveraged open source packages, including diverse packages from tidyverse (dplyr, purr, readr, rlang, stringr) (Wickham et al. 2019), arrow (Richardson et al. 2024), curl (Ooms et al. 2024), httr (Wickham 2023), jsonlite (Ooms 2014), stringdist (van der Loo 2014), stringi (Gagolewski 2022), and testthat (Wickham 2011). All analyses within the paper use APCalign v1.0.0, with APC/APNI data downloaded 29 April 2024 (ver. 0.0.5.9000).

Loading taxonomic resources

The taxonomic resources underpinning ‘APCalign’ are the datasets within Australia’s National Species List (auNSL) pertaining to terrestrial vascular plants, the Australian Plant Census (APC) and the Australian Plant Names Index (APNI). The APC is a compilation of endorsed taxon concepts, including both currently accepted taxon concepts and taxon concepts with an alternative taxonomic status, such as synonyms or orthographic variants. Nearly all taxon concepts with a taxonomic status other than ‘accepted’ can be matched to an accepted taxon concept through an identifier (‘acceptedNameUsageID’). For ‘APCalign’, taxon concepts (and their associated scientific names) within the APC are divided into accepted taxon concepts and taxon concepts with an alternative taxonomic status, such as taxonomic synonym, basionym, orthographic variant, and misapplied statuses. Taxon concepts within the APC that have a taxonomic status other than ‘accepted’ are collectively referred to hereafter as ‘APC-synonym’. The APNI is a dataset of all scientific names documented as having been applied to a native or naturalised plant growing in Australia, including both the ~110,000 names included in the APC and ~24,500 names not linked to an APC taxon concept. For ‘APCalign’, only the subset of APNI names not also in APC is considered.

The function ‘load_taxonomic_resources()’ loads copies of the APC and APNI, either a default version (the latest archived version), the current data downloaded directly from the APC, or a specified version number. On loading, the function creates several data frames by filtering and selecting data from the loaded tables, and dividing the resources on the basis of the taxon concepts’ taxonomic status (for APC) and the taxon rank of the taxon concepts (for APC) or scientific names (for APNI). A separate function, (‘standardise_taxon_rank()’, then standardises taxon ranks, replacing Latin rank with their English equivalents.

Taxonomic alignment and updating

Distribution and establishment means

The APC also specifies state-level distribution for the accepted taxa, including information about their native or introduced status; this piece of data is termed ‘establishment means’ in DarwinCore (Wieczorek et al. 2012). Information about state-level establishment means is a result of a complex consultation process involving the state herbaria. A static version of this is available (Martín-Forés et al. 2023), along with R scripts. A very similar product is available here, but in a more user-friendly R package with automatic file handling that also dynamically updates as APC moves forward.

One complex feature of the Australian flora is internal out-of-range introductions, with taxa from one part of the continent becoming naturalised in another (Fig. 1). Some of these taxa become very successful outside of their ‘native’ range. A specific type of introduction that may become more important under future climate scenarios is where taxa from mainland Australia have been very recently introduced to Tasmania and other islands. The full complexity of taxon distribution and native status documented in the APC data is available via two functions, namely ‘create_species_state_origin_matrix()’ and ‘native_anywhere_in_australia()’. The function ‘create_species_state_origin_matrix()’ can be especially informative for specific taxa (Fig. 1). For the reduced species lookup table, treating all of Australia as an entity, there is a function ‘native_anywhere_in_australia()’. A related function, ‘state_diversity_counts()’ calculates state-level diversity for native taxa, introduced taxa, and taxa with more complicated origins (Table 2).

Fig. 1.

Distribution of the number of vascular plant species in several major Australian families, split by family, state/territory, and establishment means. Note that individual species may appear in different states with different statuses, and the striking variation in geographic diversity patterns across families.

Performance and comparison with existing tools

The ‘APCalign’ algorithms were explicitly developed to standardise and align diverse lists of taxonomic names submitted to the AusTraits plant-trait database (Falster et al. 2021). These taxon lists were derived from ecological field studies, taxonomic revisions, online floras, experimental datasets, and government biodiversity reports. Each category of dataset offered unique challenges that were considered in algorithm development. In particular, aligning the many phrase names in common use in Australian taxonomic literature required distinct tools not yet available in any other existing taxonomic resolution tools worldwide. Although the APC prescribes an exact syntax for phrase names, such as the absence of spaces between the collector’s initials (Barker 2005), phrase names in different resources rarely align perfectly with the APC and fuzzy matches cannot reliably match names with missing collectors or voucher numbers. The solution within ‘APCalign’ was, late in the algorithm sequence, to perform matches on just the first two or three ‘real’ words.

To test the performance of ‘APCalign’ across diverse datasets, 47,698 unique taxon names contained within the 362 datasets in AusTraits v4.2.0 (Falster et al. 2021) were submitted to ‘APCalign’. The distinct names in the AusTraits dataset include minor syntax variants within the dataset (e.g. ‘subsp.’ versus ‘ssp.’), more significant syntax variants (absence of taxon designations such as ‘subsp.’ and ‘var.’; incomplete phrase names), spelling mistakes (in the original datasets submitted to AusTraits), taxonomic synonyms, and incomplete names (e.g. Acacia sp. [long leaves]). In total, 95.7% of the AusTraits names had perfect matches to names in the APC (whether accepted or synonyms, with or without authorship) (Fig. 2); the effort to create alignment algorithms focused on the final 4.3% of names. Most of the additional matches were names where a conservative fuzzy match (changes of fewer than three characters and 20% of characters) aligned the name to the APC (1.8%) or names that were documented in the APNI (0.4%). However, the algorithms for other alignments captured important classes of name. In particular, many ecological datasets include names that indicate a taxon that the observer could not resolve to species, with names that follow some variant of ‘genus sp.’ or ‘genus species A/species B’ or ‘genus sp. (notes)’.

Fig. 2.

Frequency of different name-handling classes found during alignment of 47,698 names from 362 raw inputs to the AusTraits compilation (Falster et al. 2021). Each cell is sized to represent the proportion of taxon names within AusTraits that were matched by a specific algorithm, with the broad algorithm categories being indicated in the legend and all algorithms included in Table 3. Although the vast majority of names were ‘perfect’ matches, almost all APCalign algorithms were triggered by some of the names submitted.

A sequence of different algorithms was used to identify all genus- and family-level names and confirm that the name was listed within the APC or APNI. In all, 0.7% of all taxon names could be resolved only to genus- or family-level names. The matches that looked at only the first two or three words, ignoring all infraspecific taxon designations and punctuation, also were shown to be essential to accurately align many taxon names documented in AusTraits datasets. These matches aligned an additional 0.8% of taxa to an accepted name, mostly aligning (1) two or three words from the submitted name string where the name string included notes and (2) phrase names. Just 240 names (0.1%) could not be aligned by any of the matching algorithms; manual inspection of these names indicated that alignments to an APC/APNI name were possible for just 32 of these, mostly to the family level. Importantly, of the 99.9% of names that were aligned, there were <0.1% that the AusTraits team deemed to be incorrect alignments during comprehensive manual checks of names matched by anything other than the exact matches to species names and conservative fuzzy-matching algorithm, and spot checks (~25%) of names matched by the conservative fuzzy-matching algorithm; these were almost entirely ‘imprecise fuzzy matches’, which are known to often introduce errors and are, by default, turned off within the ‘APCalign’ functions. When turned on, 0.1% of taxon names were aligned by the imprecise fuzzy matches, but about half of these were subsequently reverted (manually), indicating that they are a useful tool for exploring possible name alignments, but not for mass processing when each aligned name will not be reviewed. When names of non-native, non-naturalised taxa are submitted to APCalign, if the genus is not present in the APC/APNI the name is usually unmatched (54 of the unmatched taxa from AusTraits). If the genus, but not species, is in APC/APNI, the name will be aligned only to the genus level.

We also compared the alignment algorithms of ‘APCalign’ with those of three existing global tools, namely, (taxonomic name resolution service (‘TNRS’); Boyle et al. 2013), ‘taxize’ (Chamberlain and Szöcs 2013; Chamberlain et al. 2022), and ‘kewr’ (Walker 2021). ‘taxize’ describes itself as a taxonomic toolbelt, simultaneously searching 20 separate global taxonomic resources to find matches, whereas ‘kewr’ matches names to six different taxonomic resources. There is a significant overlap between these resources. ‘TNRS’ uses the World Flora Online (Borsch et al. 2020) and the World Checklist of Vascular Plants (Govaerts et al. 2021). These R-based tools all allow lists of names to be read in, offering the rapid alignment of entire datasets. Across the Tree of Life, many such packages exist, referencing either global or regional taxonomic datasets (Boyle et al. 2013; Grenié et al. 2023).

Of the tools compared, ‘APCalign’, ‘TNRS’, and ‘taxize’ include fuzzy-matching algorithms, required to correct the minor typographic errors that proliferate in scientific names (Table 1). Meanwhile, ‘kewr’ does not perform syntax standardisation or fuzzy matches and, therefore, returns ‘NA’ when an exact match is not detected. Another feature of ‘APCalign’ that is shared by ‘TNRS’ and ‘taxize’ is an output column that identifies the resolution of the output name, indicating, if, for instance, a name can be matched to a genus or family, but not a species (Table 1). The outputs from each tool vary according to the encoded matching algorithms. Although ‘TNRS’ and ‘taxize’ also incorporated fuzzy matching, a distinctive feature of the ‘APCalign’ fuzzy-matching function is that the first letter of each word cannot change. This occasionally misses a valid alignment, but without this feature, there are many additional false alignments made. ‘APCalign’ also allows the fuzzy-matching level to be adjusted, should a user wish to institute more or less conservative fuzzy matching.

Another feature of ‘APCalign’, not shared by the other tools, was detecting a series of symbols/punctuation that indicated that the name could be aligned only to a genus, not a species. These include ‘x’ (a hybrid), ‘aff.’ (given taxon shows an affinity to the named taxon) ‘cf.’ (given taxon should be compared to the named taxon), ‘--’ (taxon is an intergrade between two taxa), and ‘/’ (two taxon names submitted, because the author is unsure which is correct). The comparison tools all prioritise looking at just the first words in the string, which leads to the submitted name being aligned to the first of the names listed, whereas ‘APCalign’ automatically flags these names as being appropriate to align only to a genus. The initially submitted string is then inserted in square brackets following the genus, so the information is not lost (e.g. ‘genus sp. [genus species × genus species]’). In APCalign, the two-word and three-word matches are made only after names with such syntax patterns are flagged as being suitable to align only to genus resolution, and hence are not passed through these matching algorithms.

One ‘APCalign’ feature that is not available in any of the other outputs is the addition of a custom identifier for names resolved to the genus or family level. This does not improve alignment or increase our taxonomic interpretation of a specific name, but does offer a means to indicate that a taxon name that is resolvable only to a particular genus or family represents an unidentified species within a given dataset or location. For AusTraits, this is essential, because many datasets have submitted the name ‘Acacia sp.’ and a mechanism is required to indicate that these do not all refer to the same taxon concept, but instead to a multitude of different Acacia species that the dataset contributors could not identify to species.

An important feature of ‘APCalign’ shared by ‘TNRS’ and ‘taxize’ is the ability to update outdated taxon concepts to the names associated with their currently accepted taxon concept. Because lists of taxonomic names quickly become outdated, it is time-consuming to manually determine which names must be updated, then change them one by one. Because taxonomic concepts are a moving target, different taxon lists will consider different names to be ‘current’. As such, although ‘TNRS’ and ‘APCalign’ agree on the currently correct name for the majority of taxa, there are instances where ‘TNRS’ returned a recently updated name, whereas ‘APCalign’ reported an older name that is still considered current by the APC. This asynchrony will always exist across taxonomic lists, because each is updated by a different panel who, for some taxa, will make divergent judgements on the application of each name and review name changes at different times. All detected names that were updated by ‘TNRS’, but not ‘APCalign’, were manually checked and all were names that had been changed in the past 5 years. However, APCalign is the only resource that offers suggestions of alternative matches for scientific names that are ambiguous as a result of a previous taxonomic split, because the APC is one of the few taxonomic lists (regional or global in scope) that documents splits.

The goal of ‘APCalign’ is to align submitted names to the currently accepted name within Australia’s two vascular plant taxonomic resources, APC and APNI. ‘TNRS’ offers a similar aim, but at a global scale, and is therefore aligned to a global taxonomic resource. In contrast, ‘taxize’ and ‘kewr’, although seeking to standardise names, have the objective of matching an input taxon name to associated taxon names across many resources. For instance, whereas the ‘APCalign’ and ‘TNRS’ workflows aim to replace all outdated taxonomy with a currently accepted name, the outputs of ‘taxize’ and ‘kewr’ are developed to display ‘here are all the names, in all these resources, that are linked to the string that was input’, both leaving more name-selection decisions up to the user and providing a snapshot of the complexity of selecting a ‘best’ taxonomic name across all global resources. APCalign will perform poorly when presented with scientific names of species occurring only outside Australia or horticultural species, because these are not included in APCalign’s taxonomic reference lists, the APC and APNI.

Options for different users

Although ‘APCalign’ was created with researchers who are well-versed in R programming in mind, we also wanted to cater to non-R users. For non-R users who want to align and update plant taxonomic names, we recommend using the ‘APCalign-app’ web interface (https://unsw.shinyapps.io/APCalign-app/). ‘APCalign-app’ provides an easy entry point to the core functionality of the R package by replacing the multi-step, code workflow with a simple interface. The interface is preset with intuitive default settings that can be bypassed. This allows users to feasibly supply their taxonomic names, either by typing into an input box or uploading a .csv file and obtain aligned and updated names that can be downloaded. We hope that this mobile-enabled web interface can improve the consistency of plant taxonomic names for all users in the botanical community.

Conclusions

Australian researchers are fortunate to have the resources provided as part of the Australian National Species List for vascular plants, regularly updated to incorporate taxonomic revisions and the names of newly described taxa. This paper presents a new interface to the dynamic APC, which is the consensus taxonomy for native and naturalised Australian plants. It is essential that research projects, biodiversity assessments, and, indeed, nurseries in Australia, use the scientific names for vascular plants accepted by the APC, to facilitate communication at the national level.

Country or continental efforts will always be complementary to global efforts. Although there exist global taxon-matching tools with similar functionalities, none references the APC. Our comparison with the global tool ‘TNRS’ highlighted the importance of a local tool, as only the APC includes phrase names and aligns names to those specified within the Australian National Species Lists for vascular plants. Unlike global taxonomic alignment tools, ‘APCalign’ also considers taxonomic splits, allowing the user to decide what scientific name they consider appropriate for taxon concepts that have been subdivided during taxonomic revisions. The Brazil Flora 2020 effort has a similar country-specific tool with a shiny interface (http://www.plantminer.com/), specific to the Brazilian flora. An added benefit of working with a continent- or country-specific list is that fuzzy matching and synonym correction are less prone to false positive errors with smaller lists than with longer lists.

Although ‘APCalign’ was released as a complete R package only in late 2023, the underlying code has been tested on multiple datasets during its development, including the 47,000+ taxon names submitted to AusTraits, 12,600+ taxon names submitted to the AIslands project on the floras of Australia’s islands (J. Schrader, pers. comm.), and 15,700+ taxon names within the New South Wales BioNet Atlas (Department of Planning and Environment; https://www.environment.nsw.gov.au/topics/animals-and-plants/biodiversity/nsw-bionet). These projects and others have each merged disparate datasets collected across more than a century, successfully outputting harmonised compilations of updated taxon names. Taxonomy will always be a moving target, but we have built the ‘APCalign’ package to be stable as the APC continues to evolve and improve. We hope that others can use the tool to synchronise their research with the APC, leveraging the actively developed, high-quality taxonomic resource that the APC provides.