The new workflow, developed by Elena Berhocoïrigoin and colleagues at CNRS, Université de Lorraine, Sorbonne Université, and partner institutions, addresses a deceptively simple bottleneck. Data-aggregation initiatives have made hundreds of millions of biodiversity records accessible online, but accessibility is not the same as usability. Before occurrence records can be fed into ecological models, they must be reconciled across three dimensions at once: space, time, and taxonomy. Records must be checked for coordinates that fall on land, for missing or implausible depths, for dates that are vague or absent, and for species names that are misspelled, outdated, or refer to organisms that do not live in the sea at all.

Marinedb is designed as a modular toolkit rather than a rigid pipeline. Users can freely combine its components into workflows tailored to their own study objectives and the quirks of their data. That flexibility matters because marine biodiversity data arrive from an extraordinary variety of sources: research vessels, museum collections, diver surveys, acoustic monitoring, and citizen-science programs, each with its own conventions and quality standards. The package is openly available on GitHub, and the curated datasets it produced have been deposited with persistent identifiers on Zenodo, allowing other researchers to reproduce or extend the work without starting from scratch.

To demonstrate the tool’s power, the team applied it to three of the largest marine biodiversity resources in existence: the Global Biodiversity Information Facility (GBIF), the Ocean Biogeographic Information System (OBIS), and the Jellyfish Database Initiative (JeDI). Together these archives span hundreds of millions of records located at sea and more than one hundred thousand species. The researchers first used marinedb to assess data quality across all three sources non-destructively, meaning the original records were left intact while copies were scrutinized. This diagnostic step alone produced one of the most comprehensive quality audits of marine occurrence data ever attempted.

The audit revealed substantial, and strikingly source-dependent, problems. Some records were located on land, an obvious red flag for marine species. Many lacked depth information entirely, even though depth is a fundamental environmental variable for ocean-dwelling organisms. Temporal information was heterogeneous, incomplete, or of low precision, making it difficult to align sightings with environmental conditions. Other records lacked species-level identification, referred to non-marine taxa, or carried weak or unspecified evidential support for the observation. The team also found absence records mixed in with presences, and a strong taxonomic imbalance, with some groups of organisms heavily documented while others remain nearly invisible in the global record.

These issues are not mere technical annoyances; they can distort the conclusions of large-scale ecological analyses. Species distribution models, which predict where species occur by relating occurrence records to environmental predictors such as temperature, salinity, and depth, are only as good as the data behind them. A single mislabeled coordinate can place a coral reef species on a mountain range, and a poorly dated record can blur the climate signal that researchers are trying to detect. By quantifying these problems systematically, marinedb gives the research community both a clearer picture of the data’s weaknesses and a practical means of correcting them.

After the diagnostic phase, the team applied a tailored dataset-production workflow to generate a ready-to-use global marine occurrence dataset designed specifically for large-scale modeling of present-day species distributions. The result comprises nearly 159 million records representing almost 140,000 species, each cleaned and standardized so that modelers can link them directly to environmental predictors without repeating the painstaking curation process themselves. The dataset is split into source-specific curated versions for GBIF, OBIS, and JeDI, as well as an integrated global product, giving users the option of working with a single resource or the combined whole.

The timing of this release is significant. Machine learning, and deep learning in particular, is transforming how ecologists study biodiversity, but neural networks are notoriously data-hungry. Multi-species and deep-learning approaches require extensive and diverse occurrence records to perform well, and until now the technical burden of assembling such datasets has kept many research groups out of the game. By lowering that barrier, marinedb could accelerate a wave of large-scale studies of marine species distributions, from forecasting how fish stocks will shift as oceans warm to identifying priority areas for marine protected areas.

The open-source ethos of the project is central to its design. Because the code is freely available and the curation steps are explicit and reproducible, other researchers can inspect exactly how the data were treated, adapt the workflow to their own needs, or challenge particular decisions. This transparency stands in contrast to older practices in which data cleaning happened behind closed doors, buried in undocumented spreadsheets and personal scripts. The modular architecture also means the framework can accommodate future data sources and new quality checks as the biodiversity informatics landscape evolves, preserving flexibility for use cases the original authors did not anticipate.

For a field that has long aspired to a complete picture of life in the ocean, the arrival of a curated, model-ready dataset of this size marks a genuine milestone. Nearly 159 million cleaned records spanning almost 140,000 species will not answer every question about marine biodiversity, and the audit makes clear that gaps and biases remain embedded in the underlying archives. But it transforms the starting point for anyone modeling where ocean species live today and how those distributions may change tomorrow. As climate change reshapes marine ecosystems at unprecedented speed, tools that turn scattered, messy observations into reliable, reproducible science have never been more urgently needed.

Subject of Research: Open-source curation and integration of global marine biodiversity occurrence data for species distribution modeling

Article Title: marinedb: An open-source, flexible and scalable Python workflow for curating, standardizing, and integrating marine biodiversity data

Article References: Berhocoïrigoin, E., Le Noac'h, P., Laviale, M., Ayata, S.-D., Fix, J., & Pradalier, C. (2026). marinedb: An open-source, flexible and scalable Python workflow for curating, standardizing, and integrating marine biodiversity data. https://doi.org/10.5194/essd-2026-771

Image Credits: AI Generated

DOI: 10.5194/essd-2026-771

Keywords: marine biodiversity, marinedb, Python, GBIF, OBIS, JeDI, species distribution modeling, data curation, open-source software, occurrence records, data quality, ocean ecology