The Antarctic Research Atlas

Antarctica is the only continent run entirely by treaty. The Antarctic Treaty System, the Madrid Protocol on environmental protection, and CCAMLR govern everything from where a research station can be built to how much krill can be fished. To read the indexed science of the continent, we harvested Antarctic and Southern Ocean literature spanning ice, climate, and the biology of the species that live there, then tagged every study against a codebook of 46 dimensions shared across domains plus domain-specific extensions: a domain-expert model proposes tags, a second, stronger model adversarially judges each against the codebook, and only in-vocabulary, evidence-grounded, judge-confirmed tags persist.

Re-judged blind by an independent model on the finished corpus (5 October 2026, 200 sampled tags): precision 79.0% (95 percent interval 72.8 to 84.1). Recall was measured at 55% on a 41-paper reference set during the August 2026 build gate, so counts are more likely under- than over-stated; they are not exact, and the record is compiled to be measured, never assumed.

Every one of this corpus's study-site coordinates was re-checked automatically in August 2026 against the geographic context of the abstract and, for species papers, against known range; where the location was ambiguous it was re-resolved from the full abstract. A 40-site manual audit of the result found 83% correct. The same pass also caught and corrected sites that had drifted to the wrong hemisphere entirely.

The corpus is a climate-science corpus first

By knowledge domain, the picture sharpens further: more than half the corpus sits inside climate and environmental science, with biology and taxonomy trailing well behind.

By study type, field study is the largest single class, 1,602 studies (29% of the corpus), ahead of modeling and simulation, 993 (18%), lab analysis, 578 (10%), and observational work, 382 (7%). A continent this difficult to reach still carries a substantial amount of fieldwork, not only modeling from a distance.

1 · Governance is essentially absent from the indexed record

Antarctica's entire legal and environmental-management framework depends on science to function. That framework is the Antarctic Treaty, the Madrid Protocol and CCAMLR's fisheries limits. Yet just 93 of 5,550 indexed papers (1.7%) name an Antarctic governance instrument at all (the Antarctic Treaty System, CCAMLR or its Conservation Measure 91-05, the Madrid Protocol, or an ASPA/ASMA designation), many only in passing (a permit citation, a station-operations note). Fewer still, 62 papers (1.1%), have policy or human dimensions as their research focus, and only 40 papers do both. The continent most dependent on an international agreement to function is one of the least directly studied as a governance question in the indexed literature.

2 · Fieldwork clusters on a handful of places

The 3,967 studies with a locatable study site carry 6,978 study-site placements (one study at one place), and the map groups those into 466 distinct locations. Antarctica is enormous and largely inaccessible; research concentrates heavily around the stations, coastlines, and features that logistics actually reach (McMurdo Sound, the Antarctic Peninsula, the Ross Sea), while vast stretches of the interior are represented, if at all, by satellite monitoring and modeling rather than site-level study.

3 · Species attention follows the same pattern: a handful get most of it

The corpus tracks 3,851 distinct species and taxa, a wide biological net. But research attention inside that net is just as skewed as the geography: 42 species, 1.1% of those tracked, account for 1,413 of 6,111 total species-paper links, nearly a quarter of all species-level research attention. The rest is a long tail: 94% of tracked species appear in only one or two papers. The most-studied are a predictable mix of the charismatic (Weddell seals, Adélie and Emperor penguins) and the ecologically foundational (phytoplankton, diatoms, Antarctic krill). These are the animals visible from a ship, and the organisms holding up the food web beneath them.

Nearly every trusted study site in this corpus sits below 56° south, off the edge of an ordinary world map. This view is centered on the pole instead: each concentric ring is a line of latitude, each spoke a line of longitude, and each dot a geolocated study site scaled to how many studies it carries.

For where research sits against species ranges, coverage cells and access distances, see section 5; for the same sites set against environmental change and split by decade, section 6.

1 · Species ranges against study sites

This atlas does not score range coverage. The known-range test rests on GBIF occurrence countries, and for Antarctic taxa that list is weak: AQ is a single country code for the whole continent, many of the 3,851 taxa are cosmopolitan or marine, and only 1,191 of them (30.9%) carry any range countries at all. The site density and station distances below are the useful measures here.

2 · Where the study sites are

Each cell is one hexagon of the H3 grid, drawn on its true boundary so that the cells tile without overlap; darker cells hold more distinct study sites. The pattern follows the stations and the coast.

The densest window, enlarged

The dashed rectangle on the map above marks the window that holds the most study sites (a square window in the polar projection). Here it is at a larger scale and at the grid's own resolution 4 (about 1,800 km² per cell), with its own log ramp.

3 · How far the sites are from a research station

On this continent the meaningful access point is a research station, not a city: the nearest city of 100,000 or more is a median 3,234 km from a study site and only 3.9% of sites lie within 200 km of one. Measured against the 83 COMNAP stations instead, the fieldwork sits close to the bases.

These are straight-line distances, not travel times; roads, terrain, sea ice and protected-area access are not modelled. Study-site coordinates are geoparsed from abstracts, and region-level places sit at their centroid.

4 · Method note

Sources. access points = Natural Earth ne_10m_populated_places_simple, GitHub master 5.2.0-pre (nvkelso/natural-earth-vector, downloaded 2026-09-29) (populated places with 100,000 people or more), public domain; research stations = COMNAP Antarctic Facilities, v3.5.0, facilities of type Station, used for educational and non-commercial purposes only; grid = H3 resolution 4 (h3-js 4.5.0), drawn on the true cell boundaries from h3-js and aggregated one level (h3 cellToParent) on the main maps. A study site is a place a paper's abstract names as the location of the work, resolved to coordinates by a gazetteer and kept only when it came from one of the six trusted geocode sources; 3,967 of 5,550 studies (71.5%) have one. Density and distance use precise sites only (not a whole country, not a rounded centroid): 462 distinct places.

Caveat. Study-site coordinates come from automated geoparsing; a 40-site manual check found 31 plausible; a 100 to 200 pin human check is planned. Range coverage is a country-level test, so a study in the right country but far from the species' actual distribution still counts as in range.

Provenance. Generated 2026-10-14 from the published database. Figures in this section are recomputed by phase1/spatial_shortfall.js and checked by the figures audit.

The change layer that applies here covers very few units, so the correlation below is reported for completeness rather than as a finding. For sea-ice change it is −0.168 across 8 ecoregions. These are descriptive figures: they do not show that change causes or deters research, and the units differ in area, accessibility and richness. What they do show is where fast change and little located research coincide, which is the list each block below ranks.

1 · Sea-ice trend 1990 to 2025

Change here is the linear trend from 1990 to 2025 of the annual mean sea-ice concentration within 100 km of the cell or unit, in percentage points per decade, from the NOAA OISST v2.1 ice field at 0.25 degrees (ice-free ocean counted as zero). Each ecoregion and each first-level administrative region is placed in one of four quadrants: the top third of the universe by sea-ice change is high change, the bottom third by located research density (papers of this corpus with a precise study site inside the unit, per 10,000 km²) is low research. The high-change, low-research quadrant is the candidate gap list.

No dedicated sea-ice concentration collection exists in the Earth Engine catalogue, so the OISST ice analysis stands in for one; and a straight line through 36 annual means hides the shape of the record, in which Antarctic sea ice grew to 2014 and collapsed after 2016, so the trend understates both. Ranking uses the absolute trend: loss and gain are both change.

Sea-ice trend 1990 to 2025, cell by cell

2 · When the research happened

The same study sites split by decade of publication. Of the 3,892 located papers, 56.9% were published in the 2010s or 2020s and 40.4% since 2016. The whole corpus of 5,550 papers, located or not, is 57.6% from those two decades, so located research is about as recent as the corpus. Of the 27 ecoregions with a located paper, 6 have none published since 2010; of the 14 admin1 regions, 5. 3 located papers have no publication year.

The decade with the most located papers, cell by cell

3 · Method note

Change layers. Sea-ice trend 1990 to 2025 = NOAA/CDR/OISST/V2_1 (OISST v2.1 daily, EE catalogue copy (1981-09 to 2026-09 at run time)), 1990-2025, sampled at 27,830 m (native 27,830 m), NOAA, public domain. Values are area means of a systematic sample over each H3 cell of the grid (the maps above draw them one resolution coarser, as valid-area-weighted means of the child cells) and each unit outline (RESOLVE Ecoregions 2017 and GADM 4.1, simplified for display), computed in Google Earth Engine; boundary pixels are weighted fractionally. Layers are remote-sensing and reanalysis products, not field measurements.

Research. 3,892 of 5,550 papers in this corpus have a precise trusted study site (the same selection as the density map above); a paper counts once per cell and once per unit it has a site in, and a unit's density is papers per 10,000 km² of its area. Only units of 1,000 km² or more are ranked, so that a city-state's density is not set against a province's. 44 precise places (open sea, Southern Ocean, atoll centroids) fall in no ecoregion and 45 in no admin1 region; they are in the cell maps but not in the unit tables.

Reading it. A near-zero correlation says that, across these units, more change does not go with more or with less located research. It does not say that change deters research or that research prevents change, and units differ in area, accessibility, richness and national research capacity. A unit in the high-change, low-research quadrant is a candidate gap, not a proven one: papers whose text names no locatable site are invisible here, and the geoparsing caveat of the section above applies.

Provenance. Generated 2026-10-14 from the published database and the Earth Engine cache. Figures in this section are recomputed by phase1/research_vs_change.js, drawn by phase1/gen_atlas_change_section.js and checked by the figures audit.

Beta On the other atlases this section fits a model of research attention per ecoregion and administrative region and reports where the located record falls short of the model's expectation. For this atlas the model is not fitted: of 5,550 papers, 3,892 have a precise study site, and those sites fall in only 19 ecoregions and 10 administrative regions, under the floor of 40 units with a paper that a cross-validated model needs (one of them, the Antarctic "Rock and Ice" ecoregion, holds most of the record). The observed counts are listed below and no expectation is claimed; a polar model would need its own covariates (station distance, ice-free area, sea-ice access) rather than the terrestrial set used elsewhere.

1 · Ecoregions: observed counts only

Only 19 ecoregion units in the universe carry a located paper of this corpus; the floor for fitting a model is 40. Observed counts are listed; no expectation is claimed.

2 · Administrative regions (admin1): observed counts only

Only 10 admin1 units in the universe carry a located paper of this corpus; the floor for fitting a model is 40. Observed counts are listed; no expectation is claimed.

3 · Method note Beta

Rule. A unit type is fitted only when at least 40 of its units carry a located paper of the corpus; otherwise the JSON lists observed counts with the reason. The universe is every unit of 1,000 km² or more with a centroid and at least one census taxon whose GBIF range countries include one of its countries. Located paper = a paper with a precise trusted study site (the same selection as the sections above) inside the unit or snapped to it as the nearest within 25 km.

Provenance. Generated 2026-10-14 from the published database (SELECT only) by phase1/expected_research.js antarctic; the observed counts here are checked by the figures audit. The interactive map's beta layer is not offered for this atlas for the same reason.