taxify

offline taxonomic name resolution

Offline taxonomic name matching against local Darwin Core backbones, with matching done in C.

Hand it a column of messy species names. taxify cleans them, matches them against a backbone you already have on disk, resolves synonyms to accepted names, and returns one standardized data.frame. Every step runs locally against a versioned snapshot, so there are no API calls, no rate limits, and the same input gives the same output on any machine. The matching engine is written in C through the vectra columnar engine.

library(taxify)

# first call installs the default backbone set (COL + GBIF + ITIS, ~4 GB)
taxify(c(
  "Quercus robur",
  "Pinus abies",        # synonym, resolved to Picea abies
  "Quercus robus",      # typo, fuzzy-corrected to Q. robur
  "Taraxacum officinale"
))

Local, not over the wire

The usual route for name resolution, taxize, calls out to around twenty web services (NCBI, ITIS, GBIF, EOL, IUCN, WoRMS, Tropicos, …). That covers everything, but it ties each run to network latency, service uptime, and rate limits, and the answer can change between runs as upstream services update. taxify ships the backbones as pre-built local snapshots and matches against them in C, so a list of thousands resolves in seconds and a result is reproducible from the recorded backbone version.

The closest local analogue is taxadb, which also stores backbone snapshots on disk; the migration vignette walks through the differences in matching strategy, output schema, and enrichment.

Fifteen backbones, one call

taxify ships fifteen backbones as compressed .vtr files, pre-built by the companion taxifydb package, downloaded once, and matched locally. Pass several and they form a fallback chain: a name unmatched by the first backbone cascades to the next.

# WFO first (plants), then GBIF for whatever WFO doesn't cover
taxify(
  c("Quercus robur", "Panthera leo", "Amanita muscaria"),
  backbone = c("wfo", "gbif")
)

list_backbones() returns this table live, with the installed and version status of each. taxify_databases() adds the enrichment layers alongside it for the full picture.

Names are cleaned before matching

Input names are normalized first, so the fuzzy pass only runs on names that genuinely differ from the backbone rather than on names that just carry extra authorship or qualifiers:

"Quercus robur L."            ->  "Quercus robur"        # authorship stripped
"Pinus cf. sylvestris"        ->  "Pinus sylvestris"     # qualifier removed
"Nothofagus x alpina"         ->  "Nothofagus × alpina"  # hybrid sign normalized (x -> ×)
"Betula pendula (Roth) Doll"  ->  "Betula pendula"       # parenthesized author stripped

Fuzzy matching is configurable (Damerau-Levenshtein, Levenshtein, or Jaro-Winkler, with a distance threshold), and runs genus-blocked so a typo only competes against names in the same genus.

Both packages can read the same WFO snapshot, so the comparison below is between the two matching implementations rather than between two versions of WFO. taxify queries the columnar .vtr taxifydb builds from the Zenodo Darwin Core archive; WorldFlora reads the classification table inside that same archive into RAM.

The corpus is 1,000 accepted binomials drawn from the backbone with a fixed seed. The fuzzy corpus is those names with one substituted character in each epithet, so every name has to be resolved by distance – a deliberate worst case, not a mixed list where most names still match exactly.

scripts/benchmark-worldflora.R produces these numbers and scripts/benchmark-worldflora-results.json records the run, including package versions and the backbone snapshot. Both packages were measured back to back on one machine (Windows 11, R 4.6.0, taxify 0.3.21, WorldFlora 1.14.5) that was carrying other work at the time, so treat the ratios as the reliable figures and the absolute times as an upper bound.

Where the gap comes from: taxify runs the fuzzy pass genus-blocked, so a typo competes only against names in the same genus, and the comparison itself runs in C. Scaling five times the names costs 1.4x the time, because the fixed cost of opening the backbone dominates.

taxize is not in the table. It resolves names through the Global Names online resolver rather than a local backbone, so its time is network-bound and varies with service load; there is no way to measure it against a fixed snapshot the way these two can be measured against each other.

What you get back

taxify() returns one row per input name with a fixed schema: the matched and accepted names with their IDs and authorship, rank, family, genus, epithet, synonym / hybrid / ambiguity flags, any taxonomic qualifier, the match type (exact, exact_ci, fuzzy, abbrev, hybrid_formula, out_of_scope, or none), the fuzzy distance, a coarse kingdom / taxon-group label, the backbone, and the backbone version used. summary() prints a compact digest of how the batch resolved.

result <- taxify(c("Quercus robur", "Pinus abies", "Quercus robus", "Taraxacum officinale"))
summary(result)
#> -- taxify results ----------------------------------------------------
#>   backbone: WFO  |  4 names submitted
#>
#>   matched         4  (exact: 2, case-insensitive: 0, fuzzy: 2, abbrev: 0)
#>   --------------------------------------------------------------
#>   taxon groups: vascular plant: 4

Navigate the backbone, not just match it

taxify() resolves a name to its accepted name. The same local backbone file answers the related lookups too — synonyms, children, ancestors, descendants — with nothing else to download:

synonyms("Picea abies")                     # every synonym of an accepted name
children("Quercus")                         # accepted species in a genus
downstream("Fagaceae", downto = "genus")    # all genera under a family
upstream("Quercus robur", to = "family")    # the family a species sits in

Decompose a name without matching it, or go from a backbone ID back to a name:

parse_name("Quercus robur (L.) H.Karst.")   # genus / epithet / author / rank, no lookup
id2name("2878688", backbone = "gbif")        # GBIF usage key -> name + classification

Cross the vernacular boundary in either direction, and shape a lineage into a tree:

comm2sci("pedunculate oak")                 # common name -> scientific
sci2comm("Quercus robur")                   # scientific -> common names
class2tree(species)                         # taxonomy as a Newick / ape phylo tree
lowest_common(species)                      # deepest shared rank (the MRCA)

For migrating an old checklist and recording exactly what you matched against:

reconcile(old_species_list)                 # how each name maps onto the current backbone
lock <- taxify_lock(result)                 # freeze the resolved backbone + enrichment versions
cite(result)                                # formatted citations for every source used

Trait and status enrichment

Nearly ninety enrichment layers join published trait and status data to your results through the backbone-resolved accepted name, so synonyms in either dataset land on the same key:

# plants
taxify(plant_names) |>
  add_iucn() |>                  # IUCN Red List
  add_griis("AT") |>             # GRIIS
  add_zanne() |>                 # Zanne et al. woodiness
  add_eive()                     # EIVE indicator values

# fish
taxify(fish_names, backbone = "col") |>
  add_fishbase() |>              # FishBase morphology & ecology
  add_fishmorph()                # FISHMORPH functional traits

# one trait, gathered across every source that has it
taxify(plant_names) |>
  add_trait("seed_mass")         # Diaz + GIFT, harmonized to mg

Sources span all kingdoms: IUCN, GRIIS, GBIF common names, WCVP, EIVE, Diaz et al., LEDA, GIFT, FungalTraits, FUNGuild, AlgaeTraits, EltonTraits, AVONET, PanTHERIA, AmphiBIO, FISHMORPH, FishBase, AnAge, GloNAF, LepTraits, AnimalTraits, and regional plant-trait sets for France (Baseflor), Britain (Ecoflora), and Germany (FloraWeb), and more. The enrichments vignette lists the full set with references and licenses, and list_enrichments() returns it in R; list_traits() browses the cross-source trait vocabulary that add_trait() draws on.

To join your own table, add_data() auto-detects the species column, matches it through the same backbone(s) used in the original call, and left-joins. It accepts data.frames, CSV, CSV.GZ, XLSX, SQLite, and .vtr.

result |> add_data("TRY_traits.csv")
result |> add_data("TRY_traits.csv", cols = c("LeafArea", "SLA", "PlantHeight"))

Check a list before you trust it

inspect() returns only the names that look wrong, each labelled with what stands out and the name to use instead. It catches typos, retired synonyms, made-up genera, near-duplicate spellings, and the lone animal in a list of plants, and ranks them by whether they need a decision, a second look, or just optional cleanup.

inspect(field_names)                  # offline register and list checks
inspect(field_names, backbones = TRUE) # also typos, synonyms, ambiguity

For regional field lists, region steers a fuzzy correction toward species that actually occur where you work, so a misspelling resolves to the plant that grows there even when a one-letter neighbour from another continent sits just as close in spelling. Pass a region name, a TDWG code, or coordinates.

taxify(field_names, region = "Belgium")
taxify(field_names, coords = c(4.35, 50.85))

Installation

install.packages("taxify")

Or the development version from GitHub:

install.packages("pak")
pak::pak("gcol33/taxify")          # vectra is installed automatically

Documentation

• Getting started
• Choosing and combining backbones
• Fuzzy matching
• Constraining matches to a region
• Enrichments
• Custom data
• Inspecting a name list
• Hybrids and aggregates
• Migrating from taxize, WorldFlora, and related tools
• Large-scale workflows

Support

“Software is like sex: it’s better when it’s free.” — Linus Torvalds

I’m a PhD student who builds R packages in my free time because I believe good tools should be free and open. I started these projects for my own work and figured others might find them useful too.

If this package saved you some time, buying me a coffee is a nice way to say thanks. It helps with my coffee addiction.

License

MIT (see the LICENSE file)

Citation

@software{taxify,
  author = {Colling, Gilles},
  title  = {taxify: Offline Taxonomic Name Matching Against Darwin Core Backbones},
  year   = {2026},
  url    = {https://github.com/gcol33/taxify}
}

Cite the backbones and enrichment layers you actually used with cite(result), which pulls each source’s own reference from the manifest.