Taxonomic Mapping to WoRMS

• Taxonomic mapping to WoRMS is the process of aligning diverse biological annotations to a globally recognized taxonomic schema, ensuring clarity and consistency.
• The methodology employs systematic preprocessing, exact and fuzzy matching using the WoRMS REST API, and manual curation to resolve label ambiguities.
• Expert quality control and rigorous filtering reduce mapping errors significantly, making the process reliable for large-scale marine biodiversity studies.

Taxonomic mapping to the World Register of Marine Species (WoRMS) refers to the process of aligning biological annotation labels—often arising from disparate sources and with varying nomenclatural conventions—to the nomenclatural and taxonomic schema curated by WoRMS. This is a critical workflow for ensuring interoperability, reproducibility, and extensibility in large-scale biodiversity informatics, particularly for taxa with complex or shifting taxonomy. ReefNet (Battach et al., 2025) provides a canonical example of large-scale, genus-level taxonomic mapping using ~925,000 coral annotations across 77 sites, standardizing all entries to valid Scleractinia genera (plus Fungiidae) with explicit linkage to WoRMS AphiaIDs, thus enabling rigorous global benchmarking for automated monitoring and conservation (Battach et al., 19 Oct 2025).

1. Input Corpus and Mapping Objectives

ReefNet's taxonomic alignment pipeline ingests sparse point annotations at varying taxonomic resolutions aggregated from 76 curated CoralNet sources and one site in the Red Sea. The annotations comprise approximately 920,000 distinct points reflecting raw label diversity—including scientific names, vernacular labels, abbreviations, and misspellings. The central goal is to collapse this heterogeneity to a parsimonious set of genus-level hard-coral labels (plus Fungiidae family) and map each to its unambiguous, globally recognized WoRMS taxon using AphiaIDs, producing a benchmark-ready, ML-compatible schema with 44 unique genus or family labels.

2. Preprocessing and Label Normalization

The pipeline systematically extracts the set $L_0$ of unique strings across all data sources, capturing significant syntactic and semantic variation (e.g., “Staghorn coral”, “Acropora cervicornis”, “P. porites”). Preprocessing entails:

• Lower-casing, punctuation removal, and whitespace normalization.
• Whitespace-based tokenization to facilitate string matching.
• Filtering for genus-level candidates by discarding labels naming only families (with the exception of “Fungiidae”), higher-orders, or standalone species epithets.
• Reduction of species- or morphotype-level strings to the appropriate genus (e.g., “Acropora cervicornis” → “Acropora”), and mapping of frequently used common names (e.g., “Staghorn coral”) to their corresponding genus via a manually curated lookup.

This initial phase imposes uniformity and creates a candidate set suitable for programmatic taxonomic alignment.

3. Automated WoRMS Alignment and Disambiguation

Automated mapping leverages the WoRMS REST API and proceeds in two phases:

• Exact-match Pass: For each candidate genus $g$, a direct query is issued to the API; valid genus records require status="accepted" and rank="Genus". Upon match, the AphiaID, accepted name, and higher-taxa (family, order, class) are recorded.
• Fuzzy-match Pass: For unmatched or ambiguous labels, a token-overlap similarity is computed as $$S(g,h) = \frac{|\mathrm{tokens}(g)\cap \mathrm{tokens}(h)|}{\max(|\mathrm{tokens}(g)|,|\mathrm{tokens}(h)|)}$$ against all WoRMS genus names. A candidate genus $h$ is accepted if $S(g,h) \geq \tau$ with $\tau=0.5$; otherwise, manual curation is triggered.

Ambiguities arising from homonyms (e.g., the same genus name used in non-coral taxa) are resolved by retaining only entries whose “order” field is “Scleractinia”. Synonyms and spelling variants are normalized by following WoRMS “synonym_of” fields and leveraging Levenshtein distance for noisy matches. Common name ambiguities are addressed via a curated lookup table.

4. Mapping Table Schema and Filtering

The finalized mapping table encodes, for each retained label:

input_label	cleaned_label	aphia_id	accepted_name	status	rank	family	order
Staghorn coral	Acropora	207749	Acropora	accepted	Genus	Acroporidae	Scleractinia

Each field is directly derived from the API or from manual curation steps. After initial mapping, additional biologically informed filters are applied:

• Only hard-coral genera (“Scleractinia” order) and Fungiidae family retained.
• Minimum total frequency (≥ 100 annotations).
• Presence in at least 3 independent sources, each with ≥ 10 annotations.
• Visual consistency and distinguishability as assessed by experts.
• WoRMS “status=accepted”.

This process reduces the candidate list from approximately 50 taxa to the final 44 genera (including Fungiidae).

5. Expert Quality Control and Verification

A stratified random sample of 8,962 image patches (10 per genus per source) is subjected to expert review by 10 coral taxonomists via a custom web-based tool. Labels are scored as “Correct”, “Incorrect”, “Low-quality image”, or “Hard-to-decide”. The initial micro-averaged agreement (fraction labeled “Correct”) is 73%. Quality control proceeds by:

• Excluding any source with agreement < 50% (raising average to 78% across 70 sources).
• Excluding any genus with agreement < 50% (raising average to 78% across 39 genera).
• Further filtering source×genus pairs to retain only those with ≥ 70% agreement, yielding a high-confidence subset: 479,027 annotations across 40 genera at 92% agreement.

6. Challenges and Resolution Strategies

Principal challenges include:

• Missing or recently described genera not represented in WoRMS, requiring manual curation or provisional omission.
• Homonymy across taxonomic kingdoms—disambiguated by checking the “order” field in WoRMS.
• Taxonomic instability and synonym drift (splits/merges)—addressed by recursively resolving the “synonym_of” chain to an “accepted” name.
• Vernacular/common-name ambiguity and spelling errors—resolved via a combination of manual mapping tables and algorithmic string similarity methods.

Quality assurance metrics indicate a mapping error rate decrease from 27% before QC to 8% after all curation filters are applied, demonstrating the necessity and impact of expert review and systematic filtering.

7. Reproducibility and Data Schema

All mapping scripts, the input label list, WoRMS API queries, and the final mapping table (with explicit AphiaIDs and taxonomic fields) will be released on GitHub. AphiaIDs serve as persistent, globally recognized identifiers tethered to the evolving WoRMS taxonomy. This infrastructure enables ongoing updates as genus concepts change or new taxa are described and ensures reproducibility for downstream machine learning and ecological research (Battach et al., 19 Oct 2025).