LoginSign Up
← Back to all sparks
B

b3gbi

ANALYTICS
Velocity2.5

R package b3gbi by b-cubed-eu — release notes from GitHub.

github.com/b-cubed-eu/b3gbi ↗

b3gbi pulled confidence intervals out of its indicator workflow and handed them to dubicube.

biodiversitygbifuncertaintybootstrappingspatial gridsr package
◆Current state
b3gbi computes biodiversity indicators from GBIF occurrence cubes for the B-Cubed project, and sits at 0.9.4 in a JOSS review run-up. The 0.9 release decoupled uncertainty from indicator calculation: confidence intervals are no longer produced inline but added afterward with add_ci(), backed by whole-cube bootstrapping from the sibling dubicube package. Everything since has been grid-parsing and compatibility repair around that split.
◆Where it's heading
Two forces are shaping releases. Internally, the uncertainty split produced an indicator-specific rule book — species-level indicators bootstrap the whole cube, raw counts resample within year, evenness gets a logit transform — and that rule book is where the statistical thinking now lives. Externally, GBIF's taxonomic backbone migration to the Catalogue of Life forced string taxon keys through process_cube() and the plotting paths, while recurring EEA and MGRS grid-code fixes mark coordinate parsing as the least settled area.
◆Prediction
The 0.9.4 notes are entirely JOSS review items — contributors, examples, tracked datasets — so the next release is most likely a JOSS-accepted 1.0 rather than new indicator work.

◆Recent moves

  1. 4d ago

    JOSS review fixes: contributors, examples, tracked data

    Entirely JOSS review response: contributors added to DESCRIPTION, the Montenegro bryophytes example dataset tracked in git so examples run out of the box, roxygen examples added to five exported functions, and on.exit()/print() hygiene. No user-facing behavior change.

    View source ↗
  2. 1mo ago

    EEA grid coordinates no longer scaled by resolution

    Fixes EEA grid code parsing where coordinates at some resolutions were multiplied by the resolution value instead of a fixed 1,000, and corrects bounding boxes that added raw resolution units without converting to meters first. This is the third EEA coordinate correction in three releases.

    View source ↗
  3. 1mo ago

    add_ci() no longer crashes on completeness indicators

    Corrects the add_ci() exclusion list so completeness and relative_occupancy warn and return cleanly rather than throwing a fatal error, and documents the uncertainty treatment for every indicator outside add_ci()'s scope. Follow-up cleanup on the decoupling shipped in 0.9.

    View source ↗
  4. 1mo ago

    String taxon keys for GBIF's Catalogue of Life backbone

    Prepares for GBIF's taxonomic backbone migration to the Catalogue of Life, which replaces integer keys with alphanumeric strings: process_cube() stops coercing key columns to numeric, and the plotting and uncertainty paths preserve character taxonKey values. Compatibility work forced by an upstream standard change rather than a b3gbi decision.

    View source ↗
  5. 1mo ago

    Uncertainty split out of the indicator workflow into add_ci()

    ⚡ SPARK

    Confidence intervals leave the core indicator workflow and become a separate add_ci() step applied to any indicator_ts object, with the bootstrap delegated to the sibling dubicube package. The default moves to whole-cube resampling at 1,000 replicates — more robust, slower — with per-indicator defaults encoded in a rule book.

    View source ↗
  6. 1mo ago

    FAIR column mapping doc and Zenodo DOI badge

    Adds FAIR_MAPPING.md describing how output columns map to standard naming conventions, plus a Zenodo DOI badge. Archival and metadata groundwork for the JOSS submission rather than a functional change.

    View source ↗