2  Releases

This book is published once per release. The page you are reading documents v7, and every figure in it comes from that release’s own published tables — so a v3 page reports v3’s numbers, not the newest ones. Use the version switcher in the sidebar to move between builds.

This page is the exception: it reads all releases, because the comparison is exactly what a single-version page cannot show.

Notev7 — Validity decoupled from Program Areas

Released 2026-06-12. This is the promoted release.

2.1 Every release at a glance

Table 2.1: Published releases. Species counts are read from each release’s own taxon table and are not comparable across releases, because the definition of a valid species changed at v3 and again at v7 (see below).

Version	Released	Status	Title	Datasets	Valid species	In Program Areas	Grid
v8	2026-07-28	prerelease	Marine Atlas	8	17,112	9,632	global05
v7	2026-06-12	released	Validity decoupled from Program Areas	8	16,153	9,230	usa05
v6	2026-04-09	retired	IUCN range outside US EEZ	8	9,424	9,226	usa05
v5	2026-03-24	retired	MMPA spatial floor	8	9,795	9,597	usa05
v4b	2026-03-19	retired	Turtle multiplicative merge	8	9,795	9,597	usa05
v4	2026-03-01	retired	SWOT sea turtles	8	9,795	9,597	usa05
v3	2026-02-01	retired	Merged models + extinction risk	7	9,795	9,621	usa05
v2	2026-01-01	retired	Program Areas	6	17,353	9,792	usa05
v1	2023-09-01	retired	Planning Areas	5	17,379	—	usa05

WarningWhy the species count is not a trend line

The count moves for three different reasons, only one of which is “more data”:

• v3 tightened the taxonomy — resolving every source name to a canonical identity, dropping non-marine and extinct taxa — which cut the count roughly in half.
• v6 removed species whose expert range lies entirely outside US waters, whose apparent US presence was an artifact of environmental-envelope prediction.
• v7 stopped requiring Program-Area overlap for validity. Nothing was added; the definition changed, and the count went up accordingly.

Read the column as “what this release considered valid”, never as a measurement of biodiversity that changed between releases.

Table 2.2: What each release published, read from its manifest.

Version	Model id	Metrics	Spatial units
v8	mdl_key	17	Ecoregions (12), Program Areas (20), Subregions (5)
v7	mdl_seq	17	Ecoregions (12), Planning Areas (36), Program Areas (20), Subregions (5)
v6	mdl_seq	17	Ecoregions (12), Planning Areas (36), Program Areas (20), Subregions (4)
v5	mdl_seq	17	Ecoregions (12), Planning Areas (36), Program Areas (20), Subregions (4)
v4b	mdl_seq	17	Ecoregions (12), Planning Areas (36), Program Areas (20), Subregions (4)
v4	mdl_seq	17	Ecoregions (12), Planning Areas (36), Program Areas (20), Subregions (4)
v3	mdl_seq	17	Ecoregions (12), Planning Areas (36), Program Areas (20), Subregions (4)
v2	mdl_seq	19	Ecoregions (12), Planning Areas (36), Program Areas (20), Subregions (4)
v1	mdl_seq	19	Ecoregions (12), Planning Areas (36), Subregions (4)

2.2 History

Each release is summarized under the same four headings, so two versions can be read line for line. The same summary heads that version’s own landing page.

• Datasets — inputs added or changed, and the role each plays (contributing a value, masking a range, or carrying extinction risk spatially).
• Methods — the scientific rules: how models merge, how extinction risk is scored, which statutory floors apply.
• Scope — what counts and where: the definition of a valid species, the spatial units, the study area, which taxa are eligible.
• Technology — formats, identifiers, serving and applications.

WarningScope changes move the count without any data changing

The distinction between Methods and Scope is the one to watch. v3 tightened the taxonomy, v6 excluded ranges outside the EEZ, and v7 decoupled validity from Program Areas — each moved the headline species count substantially, and none of them added or removed a single input. Reading those jumps as growth or loss is the single easiest mistake to make with this table.

NoteWhat changed in v8, relative to v7

Marine Atlas

Datasets — The same eight contributing inputs. NOAA SEFSC and NCCOS density models are ingested and catalogued but deliberately excluded from the merge, because their units (individuals per km²) have no agreed mapping onto the 0–100 suitability scale — they are flagged is_scored = FALSE so no published count includes them.

Methods — The expert range constrains AquaMaps rather than merely competing with it, and whether a taxon has a range is decided globally — so a species whose range lies wholly outside US waters contributes no US cells. Governing extinction risk is applied at merge time, to the range cells. A BirdLife↔︎WoRMS crosswalk de-duplicates seabirds that earlier releases split across two taxa. Species categories come from taxonomy rather than keyword matching, removing the catch-all other and surfacing primary_producer; birds are additionally culled to marine/coastal families with at least 5% of their global range over ocean.

Reported units — Scores are reported on the 20 Program Areas, and ecoregion and subregion rollups are restored alongside them — so the release covers every spatial unit it defines, rather than Program Areas alone. (measured: 20 programarea, 12 ecoregion, 5 subregion scored on the composite metric.)

Scope — Ingest becomes global and whole-range — no land mask, no Program-Area gate — while scoring stays over US waters. A species map now shows the species rather than the analysis extent. Non-turtle reptiles and amphibians remain excluded from scoring.

Technology — Releases are published as partitioned Parquet on S3 with a STAC catalog and a searchable STAC API, readable directly by DuckDB including in a browser. Score surfaces become pre-rendered COGs, retiring the tile factory that ran SQL per request. A stable mdl_key replaces the auto-increment mdl_seq, which renumbered on every rebuild and broke saved links. One application renders every release from a ?ver= parameter, so the per-version forks are retired.

Measured against v7: valid species 16,153 → 17,112; grid changed (usa05 → global05), so cell_id denotes a different place; public model id mdl_seq → mdl_key; spatial units changed.

NoteWhat changed in v7, relative to v6

Validity decoupled from Program Areas

Methods — Scores are recomputed over the full set, with negligible effect: 18 of 20 Program Areas are unchanged.

Reported units — Scores are reported on the 20 Program Areas. A single subregion also carries a composite score, which is a by-product of the study-area rollup rather than a reporting unit. (measured: 20 programarea, 1 subregion scored on the composite metric.)

Scope — Validity is decoupled from Program Areas. Earlier releases silently required a species to overlap a current Program Area to count as valid, which hid biodiversity outside the program cycle’s footprint. Validity now means a taxonomically valid species with a modeled distribution anywhere in the study area, and Program-Area membership becomes a spatial query answered afterwards. The count rises sharply without a single species being added.

Measured against v6: valid species 9,424 → 16,153; spatial units changed.

NoteWhat changed in v6, relative to v5

IUCN range outside US EEZ

Datasets — Three missing marine-mammal IUCN ranges ingested from a more complete download.

Reported units — Scores are reported on 20 programarea — unchanged from the previous release.

Scope — Species whose IUCN expert range lies entirely outside the US EEZ are excluded. Their apparent presence in US waters was an edge-of-range artifact of environmental-envelope prediction, with near-zero suitability, so the effect on scores is negligible even though the species count drops.

Measured against v5: valid species 9,795 → 9,424.

NoteWhat changed in v5, relative to v4b

MMPA spatial floor

Methods — The MMPA floor is corrected for FWS-managed marine mammals — walrus, polar bear, sea otter, manatee and dugong — which had been missed because the protection flag was derived from the NMFS Species Directory alone.

Reported units — Scores are reported on 20 programarea — unchanged from the previous release.

Measured against v4b: valid species 9,795 → 9,795.

NoteWhat changed in v4b, relative to v4

Turtle multiplicative merge

Methods — The multiplicative turtle merge: for sea turtles the DPS extinction-risk surface is multiplied into the suitability surface during merging, with critical habitat overriding by maximum. Risk therefore varies spatially within a species. These taxa pass er_score = 100 through at scoring so the risk already embedded in their values is not counted twice. From here the toolkit has two merge strategies rather than one.

Reported units — Scores are reported on 20 programarea — unchanged from the previous release.

Measured against v4: valid species 9,795 → 9,795.

NoteWhat changed in v4, relative to v3

SWOT sea turtles

Datasets — SWOT sea turtle ranges and NOAA Distinct Population Segment polygons added, replacing the IUCN range as the global mask for all six sea turtle species. Their role is unusual: the DPS polygons carry ESA listing status spatially (Endangered vs Threatened by region), rather than as one scalar per species — the first input to vary extinction risk within a species’ range.

Reported units — Scores are reported on 20 programarea — unchanged from the previous release.

Scope — Turtle distributions become substantially more restrictive, since the IUCN and AquaMaps ranges both extended well beyond where the species actually occur.

Measured against v3: datasets added: rng_turtle_swot_dps; valid species 9,795 → 9,795.

NoteWhat changed in v3, relative to v2

Merged models + extinction risk

Datasets — NMFS core areas added — designated regions of concentrated use, a targeted high-confidence habitat indicator rather than a broad-coverage model.

Methods — Extinction risk is reworked into a single governing er_score per taxon, taken as the most protective across its datasets with a US national listing overriding the international IUCN category, plus statutory floors for the MMPA (marine mammals) and MBTA (migratory birds). Merging now runs for every taxon, masked to the IUCN range where one exists.

Reported units — Scores are reported on the 20 Program Areas only. The Planning Area geometry is still carried in the database, and a couple of non-composite metrics with it, but no composite score is computed for it — from here on, Planning Areas are history rather than a reporting unit. (measured: 20 programarea scored on the composite metric.)

Scope — Taxonomy is tightened: every source name resolves to a canonical WoRMS or BirdLife identity, and non-marine and extinct taxa are dropped. Sea turtles become their own category, and non-turtle reptiles stop being scored. This roughly halves the valid-species count — a change in definition, not a loss of data.

Measured against v2: datasets added: ca_nmfs; valid species 17,353 → 9,795.

NoteWhat changed in v2, relative to v1

Program Areas

Datasets — IUCN Red List range maps added, and used as the global mask — constraining AquaMaps predictions to the expert-assessed range rather than contributing a value of their own.

Methods — A model-merging pipeline is introduced, producing one merged surface per taxon where a taxon has more than one source.

Reported units — The transitional release: scores are reported on both the 36 Planning Areas and the new 20 Program Areas, so the two can be compared directly on the same underlying data. This is the only release that scores both. (measured: 36 planarea, 20 programarea scored on the composite metric.)

Scope — The BOEM Program Areas of the 2026 program cycle are introduced. They do not yet replace the Planning Areas — this release scores both, and the changeover completes at v3.

Technology — Individual model layers are published, and distributions are stored as cloud-optimized GeoTIFFs.

Measured against v1: datasets added: rng_iucn; valid species 17,379 → 17,353; spatial units changed.

Notev1 — the basis everything since builds on

Planning Areas

Datasets — AquaMaps global environmental-envelope models supply the distribution of nearly every species, with BirdLife expert ranges for birds. NMFS/FWS critical habitat and FWS ranges are present only as a small pilot. Continuous suitability is used as published; expert range polygons take a 50% presence value by convention, since an extent-of-occurrence boundary asserts presence somewhere within it rather than throughout.

Methods — A cell’s score is the sum, over the species present, of presence × extinction-risk weight, computed separately per species category. Weights are the IUCN equal-steps scale used by the Red List Index and the STAR metric, rescaled so Least Concern keeps a non-zero weight — so overall biological richness stays part of the signal rather than the score being driven entirely by the few threatened species. Cell scores are then rescaled within each BOEM Ecoregion and aggregated by area-weighted mean.

Reported units — Scores are reported on the 36 OCS Planning Areas and the 12 BOEM Ecoregions. v1 predates the Program Areas entirely — its database contains none, which is why a Program Area view of v1 is empty rather than merely uninteresting. (measured: 36 planarea, 12 ecoregion scored on the composite metric.)

Scope — Marine relevance is decided spatially — any model whose distribution reaches the study area is included — with no independent test of whether the species is marine. Models are merged only for the handful of taxa that have more than one source; every other species is scored directly from its single source model.

Technology — A 0.05° geographic (WGS84) grid, chosen because every input arrives on a latitude/longitude grid: analysing in geographic coordinates resamples each input once rather than twice. Area distortion is corrected arithmetically rather than by projection.

2.3 Reading an older release

Every release remains published and readable:

• Documentation — this book, at /{version}/; the sidebar switcher moves between builds and marks any version whose docs are not built.
• Applications — /scores/?ver=v3 and /species/?ver=v3, substituting any published version. The per-version application forks are retired; their old URLs 301 to the equivalent ?ver=, so links printed in earlier reports still resolve to the release they always showed.
• Data — marine-atlas/{version}/tables/*.parquet on S3, readable directly by DuckDB (including in a browser) without downloading a database.

Promotion is deliberate and separate from publishing. Writing data does not make a release current: the pointer that applications follow by default is only rewritten under an explicit flag, and may only name a released — never a pre-release — version. It currently names v7.