finebooks/bhl-metadata

BHL Metadata (Parquet)

A faithful Parquet conversion of the Biodiversity Heritage Library (BHL) monthly TSV catalog exports. Each BHL table is published as its own dataset config; the relational model is preserved exactly — no pre-joins, no column renames, no de-duplication — so you can join what you need in DuckDB, pandas, or polars.

Column names here are the original BHL PascalCase names (TitleID, PageID, BarCode, SequenceOrder, …), so the official BHL Export Schema PDF documents every column in this dataset directly. The underlying open-data project (S3 layout, file descriptions, special cases) is documented in the gbhl/bhl-open-data repository.

This dataset is self-contained: the original BHL .txt.gz exports it was built from are included under source/ for provenance, alongside the derived Parquet in data/.

This export: snapshot dated 2026-06-06 (from S3 LastModified). 294,490,955 rows across 13 tables (2.0GB of Parquet, 47 languages).

References & documentation

Resource	Link
BHL on the AWS Open Data Registry	https://registry.opendata.aws/bhl-open-data/
BHL Export Schema (column-by-column reference for every table)	https://www.biodiversitylibrary.org/data/TSV/BHLExportSchema.pdf
BHL Open Data project README (S3 layout, URL patterns, copyright notes)	https://github.com/gbhl/bhl-open-data/blob/main/README.md
BHL Open Data NOTES.md (OCR file-organization special cases)	https://github.com/gbhl/bhl-open-data/blob/main/NOTES.md
BHL Copyright & Reuse (metadata is CC0; per-item content rights)	https://about.biodiversitylibrary.org/help/copyright-and-reuse/
Biodiversity Heritage Library home	https://www.biodiversitylibrary.org/

Configs

Config	Description	Primary key(s)
title	Bibliographic metadata for journals and monographs	TitleID
title_identifier	Standard identifiers (ISBN, ISSN, OCLC, Wikidata, etc.) for titles	TitleID, IdentifierName
subject	Subject headings for titles	TitleID, Subject
creator	Authors and other creators linked to titles	TitleID, CreatorID
creator_identifier	Standard identifiers for creators	CreatorID, IdentifierName
doi	DOIs assigned to BHL entities	EntityType, EntityID, DOI
item	Digitized volumes/books with scanning metadata and rights	ItemID
page	Individual scanned pages with sequence and type information	PageID
page_name	Scientific names identified on pages by TaxonFinder	PageID, NameConfirmed
part	Articles, chapters, and treatments within items	PartID
part_creator	Authors linked to parts	PartID, CreatorID
part_identifier	Standard identifiers for parts	PartID, IdentifierName
part_page	Mapping between parts and their pages	PartID, PageID

Dataset statistics

Collection

Metric	Count
Titles	192,039
Digitized volumes (items)	324,356
Pages	68,657,888
Parts (articles/chapters)	404,793
Scientific-name occurrences	216,874,371
Distinct creators	79,892
Distinct scientific names	4,142,825

Languages (by title)

Language	Titles	Share
English	169,236	89.8%
French	6,249	3.3%
German	6,213	3.3%
Latin	2,323	1.2%
Chinese	929	0.5%
Spanish	861	0.5%
Italian	787	0.4%
Dutch	433	0.2%
Portuguese	293	0.2%
Swedish	240	0.1%
Japanese	239	0.1%
Russian	180	0.1%

Publication period (1175–2025)

Century	Titles	Share
Pre-1700	890	0.5%
1700s	3,321	1.8%
1800s	39,646	20.9%
1900s	142,205	75.0%
2000s	3,561	1.9%

Files

Row counts, shard counts, and on-disk Parquet sizes:

Config	Rows	Shards	Size
title	192,039	1	12.9MB
title_identifier	557,706	1	6.8MB
subject	1,112,073	1	6.9MB
creator	380,562	1	3.2MB
creator_identifier	150,176	1	1.3MB
doi	303,701	1	3.0MB
item	324,356	1	14.1MB
page	68,657,888	1	313.0MB
page_name	216,874,371	4	1.6GB
part	404,793	1	17.9MB
part_creator	554,656	1	9.3MB
part_identifier	300,302	1	1.1MB
part_page	4,678,332	1	9.6MB
Total	294,490,955	16	2.0GB

Join relationships

• Title → Item (via TitleID)
• Item → Page (via ItemID)
• Title → Creator (via TitleID)
• Title → Subject (via TitleID)
• Title → TitleIdentifier (via TitleID)
• Creator → CreatorIdentifier (via CreatorID)
• Item/Page → DOI (via EntityType + EntityID)
• Part → PartPage (via PartID)
• PartPage → Page (via PageID)
• Part → PartCreator (via PartID)
• Part → PartIdentifier (via PartID)
• Page → PageName (via PageID)

Usage

from datasets import load_dataset

pages = load_dataset("finebooks/bhl-metadata", "page")
items = load_dataset("finebooks/bhl-metadata", "item")
titles = load_dataset("finebooks/bhl-metadata", "title")

Or query the Parquet directly from the Hub with DuckDB:

-- Page -> Item -> Title
SELECT p.PageID, i.BarCode, t.FullTitle
FROM 'hf://datasets/finebooks/bhl-metadata/data/page/*.parquet' p
JOIN 'hf://datasets/finebooks/bhl-metadata/data/item/*.parquet'  i USING (ItemID)
JOIN 'hf://datasets/finebooks/bhl-metadata/data/title/*.parquet' t USING (TitleID)
WHERE t.LanguageCode = 'ENG'   -- BHL stores language codes uppercase (ENG, FRE, GER, LAT, ...)
LIMIT 10;

-- All pages bearing a given scientific name (via TaxonFinder output)
SELECT pn.PageID, pn.NameConfirmed
FROM 'hf://datasets/finebooks/bhl-metadata/data/page_name/*.parquet' pn
WHERE pn.NameConfirmed = 'Coleoptera';

-- All creators for a title
SELECT c.CreatorID, c.Name
FROM 'hf://datasets/finebooks/bhl-metadata/data/creator/*.parquet' c
WHERE c.TitleID = 12345;

Constructing media URLs

BHL stores page images and OCR text on the same open S3 bucket. URLs are derived from columns in the item and page tables:

-- Page image (JPEG-2000):
--   https://bhl-open-data.s3.amazonaws.com/images/{BarCode}/{BarCode}_{SequenceOrder:04d}.jp2
SELECT printf('https://bhl-open-data.s3.amazonaws.com/images/%s/%s_%04d.jp2',
              i.BarCode, i.BarCode, p.SequenceOrder) AS image_url
FROM 'hf://datasets/finebooks/bhl-metadata/data/page/*.parquet' p
JOIN 'hf://datasets/finebooks/bhl-metadata/data/item/*.parquet' i USING (ItemID)
LIMIT 5;

-- OCR text:
--   https://bhl-open-data.s3.amazonaws.com/ocr/item-{ItemID:06d}/item-{ItemID:06d}-{PageID:08d}-{SequenceOrder:04d}.txt
SELECT printf('https://bhl-open-data.s3.amazonaws.com/ocr/item-%06d/item-%06d-%08d-%04d.txt',
              p.ItemID, p.ItemID, p.PageID, p.SequenceOrder) AS ocr_url
FROM 'hf://datasets/finebooks/bhl-metadata/data/page/*.parquet' p
LIMIT 5;

Data quality & conventions

• As-is. BHL strives to produce clean data in these exports but cannot guarantee it. Use the surrogate primary keys (TitleID, ItemID, PageID, PartID, CreatorID) for joins.
• Duplicate rows. The exports contain known duplicate rows. They are preserved verbatim — de-duplicate in your own queries if needed.
• NULL vs empty. BHL uses \N for NULL in the TSVs; these become true NULLs. Empty strings are preserved as empty strings.
• Types. Integer key columns are stored as int64 (dirty values that fail the cast become NULL). Dates (e.g. CreationDate) are kept as strings because BHL's date formats are inconsistent. All other columns are nullable strings.
• Encoding. UTF-8, preserved as-is — HTML entities and special characters are not decoded.
• Sparse columns. Some columns are populated only when applicable. Notably, page_name.NameBankID (the uBio NameBank identifier) is present only for names that were reconciled to NameBank — roughly 28% of rows; the rest are NULL. This reflects the source data, not the conversion.
• Update cadence. BHL refreshes the TSV exports monthly; images and OCR are updated weekly. This dataset is a snapshot of one monthly export.

License

This dataset (BHL metadata) is dedicated to the public domain under CC0 1.0 Universal. Per the BHL Copyright & Reuse policy, BHL makes its metadata available under CC0, allowing reuse, modification, repurposing, and redistribution for any purpose — commercial or non-commercial — with no need to ask permission. It is provided "as is", without warranty as to accuracy, reliability, or fitness for any particular application. (The BHL open-data export files themselves are released under MIT-0; BHL's stated metadata-reuse dedication is CC0.)

Note on the underlying works. The CC0 dedication covers the metadata in this dataset. The digitized content it describes — page images and OCR text, which are not included here — carries varying copyright status. That status is recorded per item in the item config's CopyrightStatus, RightsStatement, LicenseType, and RightsHolder columns; consult those before reusing any linked image or OCR file.

Citation

Biodiversity Heritage Library. BHL Open Data exports (metadata, CC0 1.0). https://www.biodiversitylibrary.org/ Data model / schema: https://www.biodiversitylibrary.org/data/TSV/BHLExportSchema.pdf · Project: https://github.com/gbhl/bhl-open-data