COMPASS ID  ·  Research Evidence Atlas
MARKET PRESSURES · INDIGENOUS AND COMMUNITY LANDS  /  V1.0 · OCTOBER 2026
Last updated November 2026
Indexed research on commercial sectors and the lands of Indigenous Peoples and local communities

Market Pressures on Indigenous and Community Lands: Research Atlas

Of 7,045 indexed studies of commercial pressure on Indigenous and community lands, 2,812 (39.9 percent) concern mineral extraction.

The atlas reads 7,045 peer-reviewed studies on how mining, oil and gas, agriculture, timber, tourism, carbon markets and other sectors that sell beyond the community affect the lands, territories, rights, livelihoods and health of Indigenous Peoples and local communities, and on how those effects are governed and contested. Study sites fall in 141 countries. 1,472 studies name at least one specific Indigenous people.

Explore the interactive map →

7,045
core studies, selected from 32,948 candidates by an audited relevance screen
106,833
tags on the core studies; 80.0 percent judged correct in a blind check of 600 tags
141
countries with at least one located study site (5,132 of 7,045 studies located)
1,472
core studies naming at least one specific Indigenous people
01

Corpus and method

Every figure on this page counts core studies in the indexed record, unless a caption says otherwise.

Population. A study is core when the relationship between a commercial activity and the lands, territories, resources, rights, livelihoods or health of Indigenous Peoples or local communities is a subject of the study, not background. The activity must serve markets beyond the community. Sectors follow the Land Matrix Initiative's categories of investment intention, with commercial fisheries and aquaculture added. Conservation counts only where control of land or resources passes to an outside actor through a purchase, lease, concession or market scheme. The people concerned follow the term of the Convention on Biological Diversity, Indigenous Peoples and local communities, recorded as each study states it and never inferred.

Selection. Candidates were harvested from OpenAlex (articles, reviews and book chapters with an abstract) by crossing terms for the people concerned with terms for each sector. A small model sorted each candidate into a tier, a stronger model confirmed or changed the tier, and an independent model audited two random draws of 300 candidates: core precision 94.7 and 93.2 percent (95 percent intervals 85.6 to 98.2; 83.8 to 97.3), recall of core from the other tiers 84.8 and 91.4 percent.

TierStudiesShare of candidates
Core (atlas population)7,04521.4%
Borderline (held back, not counted)6,47319.6%
Peripheral10,15630.8%
Out of scope9,26428.1%
Unjudged100.0%

Tags. Each core study was tagged against the shared COMPASS codebook (version 2.0) by a panel of domain-expert prompts. A tag is kept only if its value is in the codebook, its evidence is found in the abstract, and a second model confirms it; a third model then re-checked every kept tag and removed the ones it rejected. A blind check by a fourth model, on 120 random tags from each of five batches, judged 480 of 600 tags correct, 80.0 percent (95 percent interval 76.6 to 83.0). About one tag in five is therefore wrong; counts built from tags carry that error.

Blank or small counts mean no or few indexed studies were found, never that no research exists.
02

Studies by sector

A study can concern several sectors; 1,485 of the 7,045 core studies (21 percent) concern more than one.

Core studies concerning each sector · share of 7,045 core studies
Mineral extraction
2,812 · 40%
Agriculture
1,319 · 19%
Oil and gas
1,042 · 15%
Timber
768 · 11%
Tourism
663 · 9%
Carbon trading
541 · 8%
Fisheries and aquaculture
496 · 7%
Renewable energy
452 · 6%
Conservation deals
442 · 6%
Livestock
128 · 2%
Biofuels
87 · 1%
Industry
82 · 1%
Mineral extraction
Gold mining, Amazon
Coordenação-Geral de Observação da Terra · CC BY-SA 2.0 · Wikimedia Commons

Extractive sectors dominate: mineral extraction and oil and gas together appear in 3,854 sector assignments, against 541 for carbon trading and 442 for conservation deals. Livestock (128), biofuels (87) and industrial zones (82) carry the fewest studies.

03

Studies by people concerned

Each core study records whether it concerns Indigenous Peoples, local communities, or both, as the study itself states it.

Timber extraction
Logging trucks, Sabah
CEphoto, Uwe Aranas · CC BY-SA 3.0 · Wikimedia Commons
Core studies by people concerned · share of 7,045 core studies
Indigenous Peoples
3,277 · 47%
Local communities
3,055 · 43%
Both
713 · 10%
SectorStudiesIndigenous PeoplesLocal communitiesBothShare concerning Indigenous Peoples
Mineral extraction2,8121,4471,10426161%
Agriculture1,31949364717951%
Oil and gas1,04263530410371%
Timber76835231610059%
Tourism6632633653545%
Carbon trading54119823311057%
Fisheries and aquaculture4962142384452%
Renewable energy4522361833360%
Conservation deals4421632295048%
Livestock12883242181%
Biofuels8718541538%
Industry824632461%

Across all sectors, 3,990 core studies (57 percent) concern Indigenous Peoples, alone or with local communities. The share is highest for livestock (81 percent) and lowest for biofuels (38 percent).

04

Named Indigenous peoples

Specific peoples named in the 3,990 core studies that concern Indigenous Peoples; 1,472 of them name at least one.

A model listed each people a study names in its title or abstract, with a quote that had to be found in the text; collective labels such as "Indigenous peoples", "First Nations" or "Aboriginal peoples" are not counted. Each name was then matched to Glottolog 5.3, the reference catalogue of the world's languages, where a match fits the study; names left unmatched are kept as the study writes them. A blind check of 100 studies found 93.2 percent of extracted names correct (95 percent interval 88.3 to 96.2), 97.4 percent of named peoples found (93.6 to 99.0), and 95.0 percent of Glottolog matches correct (89.5 to 97.7). Glottolog names languages, so the table shows the people by the name the studies use most, with the language entry beside it.

People (as the studies name it)Glottolog entryStudiesCountries
SámiSaami saam128159Finland, Norway, Sweden
MapucheMapudungun mapu124530Argentina, Chile
Dayaknot matched28Indonesia, Malaysia
InuitInuit inui124628Canada
MāoriMaori maor124626New Zealand
MaasaiMasai masa130017Kenya, Tanzania
NavajoNavajo nava124317United States
Yanomaminot matched15Brazil
Anishinaabenot matched14Canada, United States
AtacameñoKunza kunz124414Argentina, Chile
WayuuWayuu wayu124314Colombia
Kichwanot matched13Ecuador, Peru
MayaMayan maya128713Belize, Guatemala, Mexico
NenetsNenets nene125113Russia
WaoraniWaorani waor124012Ecuador
HaidaHaida haid124811Canada, United States
Standing Rock Siouxnot matched11United States
CreeCree cree127210Canada
Marindnot matched10Indonesia
Mi'kmaqMi'kmaq mikm123510Canada
OgoniOgonoid ogon124010Nigeria
Orang Rimbanot matched10Indonesia
HeiltsukHeiltsuk-Oowekyala heil12469Canada
Métisnot matched8Canada
Oraonnot matched8India
Tłı̨chǫDogrib dogr12528Canada
Denenot matched7Canada
KayapóMebengokre kaya13307Brazil
MundurukuMundurukú mund13307Brazil
ShuarShuar shua12577Ecuador

The 2,047 names cover 873 matched to 347 Glottolog entries; the rest are not matched. Some name a subgroup, a nation or a community (Ngāti Awa, Standing Rock Sioux) rather than a language group; others are peoples Glottolog does list that the matching step missed (Yanomami, Anishinaabe), so an unmatched name is not a statement that no entry exists. The blind check measured the correctness of matches made, not how many fitting entries were missed.

05

Governance, actors and effects studied

The companies, states, markets and rules through which the pressure arrives, and its effects on the people concerned, from the codebook tags of the core studies.

Effects on people studied (social dimension tags) · share of 7,045 core studies
Environmental Justice
683 · 10%
Socio-environmental Conflict
508 · 7%
Community Response and Resistance
462 · 7%
Dispossession and Displacement
426 · 6%
Livelihoods and Wellbeing
351 · 5%
Gender and Social Differentiation
89 · 1%
Health and Environment
50 · 1%
Governance mechanisms studied · share of 7,045 core studies
Indigenous and Customary Rights
2,387 · 34%
Land Acquisition and Concessions
1,252 · 18%
Certification and Voluntary Standards
149 · 2%
Corporate Due Diligence and Supply-Chain Rules
127 · 2%
Trade and Investment Policy
41 · 1%

Customary and Indigenous rights (2,387 studies) and land acquisition and concessions (1,252) are the mechanisms most studied. Market-side rules are rare in the record: certification and voluntary standards appear in 149 core studies, corporate due diligence and supply-chain rules in 127, and trade and investment policy in 41.

Actors studied · share of 7,045 core studies
Indigenous Peoples
3,237 · 46%
Governments and State Agencies
3,069 · 44%
Local Communities
2,684 · 38%
Companies and Industry
2,601 · 37%
Smallholders, Farmers and Resource Users
911 · 13%
NGOs and Civil Society
428 · 6%
Financial Institutions and Investors
95 · 1%
International and Intergovernmental Organisations
86 · 1%
Consumers and Consumer Markets
26 · 0%

Companies (2,601 studies) and states (3,069) are studied far more often than the investors who finance them (95) or the consumer markets that buy from them (26).

06

Effects studied by sector

Share of each sector's core studies carrying each social dimension tag.

SectorStudiesDispossession and DisplacementSocio-environmental ConflictCommunity Response and ResistanceEnvironmental JusticeLivelihoods and WellbeingHealth and EnvironmentGender and Social Differentiation
Mineral extraction2,8127%9%7%11%4%1%1%
Agriculture1,3198%6%4%9%6%1%2%
Oil and gas1,0426%9%7%9%4%1%1%
Timber7687%8%5%8%5%0%1%
Tourism6635%6%7%10%6%0%1%
Carbon trading5415%3%7%11%5%0%2%
Fisheries and aquaculture4965%6%11%7%7%0%1%
Renewable energy4525%9%8%8%5%0%1%
Conservation deals4426%7%9%10%6%0%2%
Livestock1285%5%5%13%6%0%3%
Biofuels8710%5%0%13%9%1%1%
Industry827%5%5%6%6%0%1%

Gender and social differentiation is tagged on 89 core studies (1.3 percent) and health on 50 (0.7 percent), against 426 (6 percent) for dispossession and displacement.

07

Study designs and methods

How the core studies obtain their evidence, from the study type and social-science method tags.

Study type · share of 7,045 core studies
Qualitative or Documentary
2,813 · 40%
Literature Review
1,071 · 15%
Mixed or Other
361 · 5%
Field Study
101 · 1%
Modeling or Simulation
88 · 1%
Observational
71 · 1%
Experimental
8 · 0%
Social-science method · share of 7,045 core studies
Policy, Legal and Document Analysis
2,147 · 30%
Ethnographic Fieldwork and Participant Observation
403 · 6%
Participatory and Community-based Methods
371 · 5%
Indigenous-led or Co-produced Research
57 · 1%

Qualitative and documentary studies (2,813) and literature reviews (1,071) make up most of the record; field studies number 101 and modelling studies 88. Policy, legal and document analysis is the most common method (2,147 studies), ethnographic fieldwork appears in 403, participatory and community-based methods in 371, and research led or co-produced by Indigenous Peoples in 57 (0.8 percent of core studies).

08

Study-site countries

5,132 of 7,045 core studies (73 percent) have at least one located study site, at 3,480 distinct places in 141 countries. A study can have sites in several countries.

Each dot is one located study site: a place the study examines, such as a field site, a community, a mine or concession, or a country or province whose laws and policies it analyses. Larger, darker dots mean more studies; grey land has no indexed study.
5,132 of 7,045 core studies (73 percent) name a locatable study site. Studies with no locatable site (global reviews, theory, studies of a whole region) are not drawn.
Each dot = a located study site ·  smaller / fainter = fewer studies  to  larger / darker = a dense cluster ·  grey land = no indexed study

For where research sits against coverage cells and access distances, see section 9; for the same sites set against environmental change and split by decade, section 10.

Core studies with a study site in each country · top 20 of 141 countries · share of 5,132 located studies
Indonesia
1,196 · 23%
Canada
506 · 10%
Australia
279 · 5%
Brazil
277 · 5%
India
214 · 4%
United States
180 · 4%
Peru
140 · 3%
Chile
133 · 3%
Nigeria
129 · 3%
Ecuador
126 · 2%
Mexico
119 · 2%
Russia
117 · 2%
Colombia
108 · 2%
Philippines
87 · 2%
Malaysia
82 · 2%
Kenya
75 · 1%
South Africa
72 · 1%
Ghana
71 · 1%
Argentina
69 · 1%
Papua New Guinea
68 · 1%
Plantation agriculture
Oil palm, Borneo
Ben Sutherland from Catford, London, United Kingdom · CC BY 2.0 · Wikimedia Commons

Five countries (Indonesia, Canada, Australia, Brazil, India) hold 2,472 of the 5,571 study-country pairs (44 percent). Indonesia alone has 1,196 studies; its most-studied sector is agriculture.

CountryStudies with a site thereMost-studied sector
Indonesia1,196Agriculture
Canada506Mineral extraction
Australia279Mineral extraction
Brazil277Mineral extraction
India214Mineral extraction
United States180Oil and gas
Peru140Mineral extraction
Chile133Mineral extraction
Nigeria129Oil and gas
Ecuador126Oil and gas
Mexico119Renewable energy
Russia117Mineral extraction
Colombia108Mineral extraction
Philippines87Mineral extraction
Malaysia82Agriculture
Kenya75Conservation deals
South Africa72Mineral extraction
Ghana71Mineral extraction
Argentina69Mineral extraction
Papua New Guinea68Mineral extraction
09

Geographic distribution of study sites

5,132 of 7,045 studies name a locatable study site; where those sites fall, and how far they sit from the nearest large city.

1 · 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. Most cells hold one site; the fieldwork of the whole corpus concentrates in a small number of them.

Study sites per resolution-2 cell (summed over child cells), log ramp from 1 to 571382057 (most)
2,533 precise study sites fall in 1,432 hexagonal cells (H3 resolution 3, about 12,400 km2 each); the ten fullest cells hold 5.6% of the sites and 921 cells hold a single site. At this scale a resolution-3 cell is a few pixels wide, so the main map aggregates them to H3 resolution 2 (about 86,800 km2 each): 778 cells, each holding the summed sites of its child cells. The inset below keeps resolution 3.
Equirectangular, the same basemap as the map above; every cell is drawn on its true H3 boundary, so the cells tile without overlap; 1 cell crossing the antimeridian is split at it and drawn at both edges of the map.

The densest window, enlarged

The dashed rectangle on the map above marks the window that holds the most study sites (94°E to 148°E, 14°S to 14°N). Here it is at a larger scale and at the grid's own resolution 3 (about 12,400 km2 per cell), with its own log ramp.

Study sites per resolution-3 cell, log ramp from 1 to 231251023 (most)
Inset denominator: 268 resolution-3 cells inside the window with 817 of the 2,533 precise study sites (32.3%). Most of those sites are in Indonesia, Malaysia and Australia. The window is chosen by the sum of sites, not by area or country.

2 · How far the sites are from a city

Distance from each study site to the nearest populated place of 100,000 people or more is a rough measure of how reachable the fieldwork was. The busiest access points, counted greedily so that no site is counted twice, are Salatiga (Indonesia, 40 sites within 100 km), Denpasar (Indonesia, 31 sites within 100 km), Mojokerto (Indonesia, 29 sites within 100 km); 9.0% of all sites lie within 100 km of one of the ten busiest.

84.2 km
median straight-line distance from a study site to the nearest city of 100,000 or more (2,533 sites)
32.4%
of study sites lie within 50 km of such a city
78.9%
of study sites lie within 200 km of such a city
Denominator: 2,533 precise study sites (distinct places with coordinates, not paper-weighted). Quartiles 36.3 and 172.6 km; the farthest site is 3,562 km from any such city.

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.

3 · 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; grid = H3 resolution 3 (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; 5,132 of 7,045 studies (72.8%) have one. Density and distance use precise sites only (not a whole country, not a rounded centroid): 2,533 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.

Provenance. Generated 2026-10-28 from the published database. Figures in this section are recomputed by phase1/spatial_shortfall.js and checked by the figures audit.

10

Research against change

Three questions of the same study sites: where the environment is changing fastest with the least located research, whether research effort tracks change at all (it does not: the correlation is near zero), and when the located research was published. 31 of 290 ecoregions sit in the high-change, low-research quadrant for forest loss.

In this corpus the rate of environmental change and the density of located research are essentially unrelated: Spearman's rank correlation between the two, unit by unit, is near zero. For forest loss it is −0.022 across 290 ecoregions and 0.048 across 1,042 states and provinces; for drying it is −0.100 across 237 ecoregions and −0.113 across 770 states and provinces. These are descriptive figures: they do not show that change causes or deters research, and the units differ in area, accessibility. What they do show is where fast change and little located research coincide, which is the list each block below ranks.

1 · Forest loss 2001 to 2024

The forest frontier
Slash-and-burn clearing
shankar s. from Dubai, united arab emirates · CC BY 2.0 · Wikimedia Commons

Change here is the share of the area that was forest in 2000 (tree cover of 30 percent or more) with a loss event between 2001 and 2024, from Hansen Global Forest Change v1.12 at 30 m. Each ecoregion (an area of broadly similar natural vegetation and climate, from the RESOLVE 2017 ecoregion map) and each state or province (the first level of division inside a country: a state, province, region or its equivalent, from GADM 4.1) is placed in one of four quadrants: the top third of the universe by forest loss 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.

31 ecoregions
197 states and provinces
high change, low research, of 290 ecoregions and 1,042 states and provinces
66 ecoregions
151 states and provinces
high change, more research, of 290 ecoregions and 1,042 states and provinces
66 ecoregions
444 states and provinces
less change, low research, of 290 ecoregions and 1,042 states and provinces
127 ecoregions
250 states and provinces
less change, more research, of 290 ecoregions and 1,042 states and provinces
Thresholds. Ecoregions: high change = 13.53% of the 2000 forest lost or more (67th percentile of 290 units); low research = 0.11 papers per 10,000 km2 or fewer (33rd percentile). States and provinces: high change = 15.14% of the 2000 forest lost or more (67th percentile of 1,042 units); low research = no located paper at all (a third or more of the units have none, so the bottom tercile is zero) (33rd percentile). Spearman rho between forest loss and research density across the universe: ecoregions: rho −0.022 (n 290); states and provinces: rho 0.048 (n 1,042). Descriptive only.

Loss is any stand-replacing event, including fire and the rotational harvest of plantations, and regrowth is not subtracted, so a high value can be a working plantation landscape as easily as a cleared forest; the 30 percent canopy threshold makes open woodland "not forest".

Ten ecoregions with the most forest loss and bottom-tercile located research, of 31 in that quadrant. Denominator: 290 ecoregions in the universe.
EcoregionCountryForest lostPapers with a sitePer 10,000 km2
Northern Khorat Plateau moist deciduous forestsTropical & Subtropical Moist Broadleaf ForestsThailand, Laos42.9%26% forested in 200000.00
Peninsular Malaysian peat swamp forestsTropical & Subtropical Moist Broadleaf ForestsMalaysia42.9%91% forested in 200000.00
Knysna-Amatole montane forestsTropical & Subtropical Moist Broadleaf ForestsSouth Africa32.8%70% forested in 200000.00
Mato Grosso tropical dry forestsTropical & Subtropical Moist Broadleaf ForestsBrazil28.8%76% forested in 200040.10
Chiquitano dry forestsTropical & Subtropical Dry Broadleaf ForestsBolivia, Brazil28.4%78% forested in 200020.09
Xingu-Tocantins-Araguaia moist forestsTropical & Subtropical Moist Broadleaf ForestsBrazil27.8%85% forested in 200020.08
Madagascar dry deciduous forestsTropical & Subtropical Dry Broadleaf ForestsMadagascar27.6%19% forested in 200010.07
South China-Vietnam subtropical evergreen forestsTropical & Subtropical Moist Broadleaf ForestsChina, Vietnam26.8%39% forested in 200020.09
Guinean montane forestsTropical & Subtropical Moist Broadleaf ForestsGuinea, Côte d'Ivoire, Sierra Leone26.7%45% forested in 200000.00
Madagascar mangrovesMangrovesMadagascar26.1%29% forested in 200000.00
Ten states and provinces with the most forest loss and bottom-tercile located research, of 197 in that quadrant. Denominator: 1,042 states and provinces in the universe.
State or provinceCountryForest lostPapers with a sitePer 10,000 km2
Bình DươngVietnam78.9%36% forested in 200000.00
MelakaMalaysia61.0%69% forested in 200000.00
KrâchéhCambodia56.2%58% forested in 200000.00
Kâmpóng ThumCambodia56.2%54% forested in 200000.00
Kâmpóng ChamCambodia55.3%16% forested in 200000.00
Amnat CharoenThailand53.2%12% forested in 200000.00
AtlantiqueBenin52.0%5% forested in 200000.00
Negeri SembilanMalaysia50.8%84% forested in 200000.00
KindiaGuinea50.7%45% forested in 200000.00
Tbong KhmumCambodia50.2%29% forested in 200000.00

Forest loss 2001 to 2024, cell by cell

Share of the 2000 forest lost 2001 to 20240 to 5%5 to 10%10 to 20%20 to 35%35 to 50%50% or morenot applicable
Dots: papers with a site in the cell (summed over child cells), log ramp from 1 to 611382061 (most)
Denominator: 1,432 resolution-3 cells with at least one precise study site from this corpus (2,689 of 7,045 papers); 1,131 have an applicable forest loss value and 301 do not. They are drawn on their true H3 boundaries and aggregated to 778 cells at H3 resolution 2 (about 86,800 km2 each): a drawn cell takes the valid-area-weighted mean of its child cells' values and the sum of their papers; 614 of the 778 drawn cells carry a value and 164 carry none (dashed outline). 1 cell crossing the antimeridian is split at it. Cells with no located research are not on this map; the unit tables above include them.

2 · Aridity trend 1990 to 2024

Change here is the linear trend of the annual aridity index (precipitation over potential evapotranspiration, P/PET) from 1990 to 2024, as percent change per decade of the 1990 to 1999 mean, from TerraClimate at about 4.6 km; a negative number means drying. Each ecoregion (an area of broadly similar natural vegetation and climate, from the RESOLVE 2017 ecoregion map) and each state or province (the first level of division inside a country: a state, province, region or its equivalent, from GADM 4.1) is placed in one of four quadrants: the top third of the universe by drying 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.

45 ecoregions
217 states and provinces
high change, low research, of 237 ecoregions and 770 states and provinces
34 ecoregions
41 states and provinces
high change, more research, of 237 ecoregions and 770 states and provinces
85 ecoregions
395 states and provinces
less change, low research, of 237 ecoregions and 770 states and provinces
73 ecoregions
117 states and provinces
less change, more research, of 237 ecoregions and 770 states and provinces
Thresholds. Ecoregions: high change = drying of 4.28% of the 1990s index per decade or more (67th percentile of 237 units); low research = no located paper at all (a third or more of the units have none, so the bottom tercile is zero) (33rd percentile). States and provinces: high change = drying of 4.77% of the 1990s index per decade or more (67th percentile of 770 units); low research = no located paper at all (a third or more of the units have none, so the bottom tercile is zero) (33rd percentile). Spearman rho between drying and research density across the universe: ecoregions: rho −0.100 (n 237); states and provinces: rho −0.113 (n 770). Descriptive only.

TerraClimate is an interpolated climate surface, so a 35-year trend at 4 km carries station-density and interpolation error, and a relative trend is unstable where the baseline is near zero; the ranking is restricted to drylands (mean P/PET under 0.65) and excludes hyper-arid units (mean under 0.05) for that reason.

Ten ecoregions with the most drying and bottom-tercile located research, of 45 in that quadrant. Denominator: 237 ecoregions in the universe.
EcoregionCountryAridity trendPapers with a sitePer 10,000 km2
Khangai Mountains alpine meadowMontane Grasslands & ShrublandsMongolia−9.8%dry sub-humid00.00
Gobi Lakes Valley desert steppeDeserts & Xeric ShrublandsMongolia−9.6%arid00.00
Caspian lowland desertDeserts & Xeric ShrublandsKazakhstan, Russia, Turkmenistan, Iran−9.5%arid00.00
Knysna-Amatole montane forestsTropical & Subtropical Moist Broadleaf ForestsSouth Africa−9.2%dry sub-humid00.00
Namaqualand-Richtersveld steppeDeserts & Xeric ShrublandsSouth Africa, Namibia−8.6%arid00.00
Succulent Karoo xeric shrublandsDeserts & Xeric ShrublandsSouth Africa−8.6%arid00.00
Khangai Mountains conifer forestsTemperate Conifer ForestsMongolia−8.5%semi-arid00.00
Edwards Plateau savannaTemperate Grasslands, Savannas & ShrublandsUnited States−8.2%semi-arid00.00
Canary Islands dry woodlands and forestsMediterranean Forests, Woodlands & ScrubSpain−8.1%arid00.00
North Arabian highland shrublandsDeserts & Xeric ShrublandsSaudi Arabia−8.1%arid00.00
Ten states and provinces with the most drying and bottom-tercile located research, of 217 in that quadrant. Denominator: 770 states and provinces in the universe.
State or provinceCountryAridity trendPapers with a sitePer 10,000 km2
LakesSouth Sudan−13.0%dry sub-humid00.00
MendozaArgentina−12.8%semi-arid00.00
BiskraAlgeria−12.8%arid00.00
Central EquatoriaSouth Sudan−12.8%dry sub-humid00.00
CórdobaArgentina−12.2%dry sub-humid00.00
Govi-AltayMongolia−12.1%arid00.00
San LuisArgentina−12.1%semi-arid00.00
MisratahLibya−12.1%arid00.00
La RiojaArgentina−11.4%arid00.00
KalmykRussia−11.3%semi-arid00.00

Aridity trend 1990 to 2024, cell by cell

Trend of the aridity index P/PET, percent of the 1990s mean per decadedrier by 10% or more per decade5 to 10% drier1 to 5% drierwithin 1% (no trend)1 to 5% wetter5 to 10% wetterwetter by 10% or morenot applicable
Dots: papers with a site in the cell (summed over child cells), log ramp from 1 to 611382061 (most)
Denominator: 1,432 resolution-3 cells with at least one precise study site from this corpus (2,689 of 7,045 papers); 1,165 have an applicable drying value and 267 do not. They are drawn on their true H3 boundaries and aggregated to 778 cells at H3 resolution 2 (about 86,800 km2 each): a drawn cell takes the valid-area-weighted mean of its child cells' values and the sum of their papers; 660 of the 778 drawn cells carry a value and 118 carry none (dashed outline). 1 cell crossing the antimeridian is split at it. Cells with no located research are not on this map; the unit tables above include them.

3 · When the research happened

The same study sites split by decade of publication. Of the 2,689 located papers, 91.1% were published in the 2010s or 2020s and 75.6% since 2016. The whole corpus of 7,045 papers, located or not, is 90.3% from those two decades, so located research is about as recent as the corpus. Of the 454 ecoregions with a located paper, 17 have none published since 2010; of the 706 states and provinces, 29. 9 located papers have no publication year.

Located papers by decade of publication. Denominator = 2,689 papers with a precise study site; the tick marks the whole corpus's share of the same decade (7,045 papers).
before 1990
11 (0.4%) · corpus 0.4%
1990s
46 (1.7%) · corpus 1.6%
2000s
172 (6.4%) · corpus 7.5%
2010s
906 (33.7%) · corpus 34.7%
2020s
1,545 (57.5%) · corpus 55.6%

The decade with the most located papers, cell by cell

Peak decade of publicationbefore 1990 (3)1990s (7)2000s (34)2010s (243)2020s (490)
Denominator: 1,430 resolution-3 cells with at least one located paper of known year, drawn on their true H3 boundaries and aggregated to 777 cells at H3 resolution 2 (about 86,800 km2 each; decade counts summed over child cells); a cell is shaded by the decade in which most of its located papers were published (2,689 located papers in all; a tie goes to the later decade). Darker is more recent. Decades are publication years, not fieldwork years.

4 · Method note

Change layers. Forest loss 2001 to 2024 = UMD/hansen/global_forest_change_2024_v1_12 (v1.12 (2024 update)), 2001-2024, sampled at 30 m (native 30 m), CC BY 4.0; Aridity trend 1990 to 2024 = IDAHO_EPSCOR/TERRACLIMATE (TerraClimate monthly, EE catalogue copy (1958-01 to 2024-12 at run time)), 1990-2024, sampled at 4,638 m (native 4,638 m), TerraClimate: CC0 1.0 (public domain). MODIS: NASA, no restrictions; dryland classes after UNEP 1992. Values are computed in Google Earth Engine 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); boundary pixels are weighted fractionally. Forest loss sums every 30 m pixel by its area for cells and units up to 25,000 km2; a larger unit is estimated from 1,000 windows of about 1 km2 placed at random inside it, each measured at 30 m. The other layers are area means of a systematic sample at their native resolution. Layers are remote-sensing and reanalysis products, not field measurements.

Revision. Forest loss was recomputed on 28 October 2026 at the native 30 m resolution. The earlier version averaged tree cover and loss to a coarser grid before applying the 30 percent forest threshold, which overstated loss in mixed and fragmented landscapes; the order of units changed little (rank correlation 0.94 to 0.98 between versions), but individual values fell, most in plantation and mosaic regions.

Research. 2,689 of 7,045 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. 20 precise places (open sea, Southern Ocean, atoll centroids) fall in no ecoregion and 11 in no state or province; 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, 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: a study whose question is not tied to a place, such as a global, laboratory, trade-flow, policy or review paper, counts in the corpus total above but has no site to place in any unit, so absence of a located study is not absence of research. The geoparsing caveat of the section above applies.

Provenance. Generated 2026-11-03 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.

11

Expected research Beta

How much located research would each place be expected to carry, given its size, access, environmental change, biome and national research effort, and where the record falls short of that expectation. Cross-validated pseudo-R2 0.66 for ecoregions and 0.49 for states and provinces. Experimental: a first statistical model, still under test and stated with its interval and its limits. The sections above do not depend on it.

Beta This section fits a model of research attention: the number of papers of this corpus with a precise study site inside each ecoregion (an area of broadly similar natural vegetation and climate, from the RESOLVE 2017 ecoregion map) and each state or province (the first level of division inside a country: a state, province, region or its equivalent, from GADM 4.1), predicted from the unit's area, its distance to the nearest city of 100,000 people, forest loss and the aridity trend, its biome, realm and latitude band, and how much located research the rest of its country carries. The model predicts where research tends to go. It does not predict ecological importance, threat or need, and a unit with less research than expected is a candidate gap: a place where the record is thinner than in comparable places. It is never proof of neglect. Fit quality is measured out of sample by 5-fold spatial cross-validation (whole realms or whole countries held out): pseudo-R2 0.66 for ecoregions and 0.49 for states and provinces. Each unit carries a 90 percent prediction interval from the negative-binomial count spread about its expectation. Located research is strongly clustered, so that interval is wide, and in this run no unit falls below it. The rankings therefore order units by the tail probability, the chance of a count at or under the observed one given the expectation, and carry the interval label alongside.

1 · Ecoregions: expected against observed Beta

Fit. 777 ecoregions in the universe, 451 with a located paper; negative binomial, dispersion theta 0.90, mean by Poisson quasi-likelihood, ridge lambda 100. 5-fold spatial cross-validation blocked by realm: pseudo-R2 0.66 (0.35 without the national-effort covariate), Spearman rho 0.65 out of fold. Labels against the 90 percent interval: 0 below, 744 within, 33 above.

Tail probability of the observed count under the fitted expectationless than expected, tail probability under 0.05 (below the 90 percent interval) (0)less, 0.05 to 0.15 (6)less, 0.15 to 0.30 (59)near expectation (0.30 or more either way) (510)more, 0.15 to 0.30 (67)more, 0.05 to 0.15 (47)more than expected, under 0.05 (above the interval) (31)not in the model universe
Denominator: 777 ecoregions in the model universe, 451 with a located paper of this corpus; 720 are drawn on their outlines (RESOLVE Ecoregions 2017 and GADM 4.1, 5,000 m simplification, rounded to the canvas); 57 are too small to draw at this scale (one canvas unit is about 40 km at the equator). Shading is the tail probability of the observed count under NB(expected, theta): the two darkest classes are the interval labels below and above; a unit with no located paper and an expectation under 0.5 sits in the neutral class by construction (the zero rule). No unit falls below its interval, so the darkest "less than expected" class is empty; 31 units are above theirs. Unshaded land is outside the universe (under 1,000 km2, or polar). A shaded unit is a candidate gap or a candidate artefact, not a finding.
Ten ecoregions with the least located research relative to the model's expectation, ordered by the tail probability P(count at or under the observed). Denominator: 777 ecoregions in the model universe, 451 with a located paper.
EcoregionCountryObservedExpected, 90% intervalTail probabilityLabel
Borneo montane rain forestsTropical & Subtropical Moist Broadleaf ForestsIndonesia, Malaysia127.3090%: 1 to 860.085within
Somali Acacia-Commiphora bushlands and thicketsTropical & Subtropical Grasslands, Savannas & ShrublandsEthiopia, Somalia, Kenya012.8390%: 0 to 410.086within
Vogelkop montane rain forestsTropical & Subtropical Moist Broadleaf ForestsIndonesia011.4990%: 0 to 370.095within
Luang Prabang montane rain forestsTropical & Subtropical Moist Broadleaf ForestsLaos, Thailand, Vietnam010.4390%: 0 to 330.103within
Northeast Congolian lowland forestsTropical & Subtropical Moist Broadleaf ForestsCongo - Kinshasa, Central African Republic08.0590%: 0 to 260.127within
Central Deccan Plateau dry deciduous forestsTropical & Subtropical Dry Broadleaf ForestsIndia114.5790%: 0 to 460.143within
Sri Lanka dry-zone dry evergreen forestsTropical & Subtropical Dry Broadleaf ForestsSri Lanka05.9190%: 0 to 190.162within
Southern Hudson Bay taigaBoreal Forests/TaigaCanada05.8890%: 0 to 190.163within
Seram rain forestsTropical & Subtropical Moist Broadleaf ForestsIndonesia111.2890%: 0 to 360.176within
Ogilvie-MacKenzie alpine tundraTundraCanada, United States05.1590%: 0 to 170.180within
Five ecoregions with the most located research relative to the model's expectation, ordered by the tail probability P(count at or over the observed). Denominator: 777 ecoregions in the model universe, 451 with a located paper.
EcoregionCountryObservedExpected, 90% intervalTail probabilityLabel
Fiji tropical moist forestsTropical & Subtropical Moist Broadleaf ForestsFiji50.1790%: 0 to 1under 0.001above
Southern Pacific dry forestsTropical & Subtropical Dry Broadleaf ForestsMexico302.6290%: 0 to 9under 0.001above
Caatinga Enclaves moist forestsTropical & Subtropical Moist Broadleaf ForestsBrazil60.3290%: 0 to 2under 0.001above
Arnhem Land tropical savannaTropical & Subtropical Grasslands, Savannas & ShrublandsAustralia313.2290%: 0 to 11under 0.001above
Queen Charlotte Islands conifer forestsTemperate Conifer ForestsCanada80.6390%: 0 to 30.001above

The "more than expected" list doubles as a geocoding audit: a small unit with tens of papers against an expectation under one is as likely a place-name collision, a herbarium or a field station as a genuine concentration of fieldwork.

2 · States and provinces: expected against observed Beta

Fit. 2,715 states and provinces in the universe, 686 with a located paper; negative binomial, dispersion theta 0.64, mean by Poisson quasi-likelihood, ridge lambda 30. 5-fold spatial cross-validation blocked by country: pseudo-R2 0.49 (0.32 without the national-effort covariate), Spearman rho 0.53 out of fold. Labels against the 90 percent interval: 0 below, 2,637 within, 78 above.

Tail probability of the observed count under the fitted expectationless than expected, tail probability under 0.05 (below the 90 percent interval) (0)less, 0.05 to 0.15 (0)less, 0.15 to 0.30 (17)near expectation (0.30 or more either way) (2,014)more, 0.15 to 0.30 (158)more, 0.05 to 0.15 (91)more than expected, under 0.05 (above the interval) (68)not in the model universe
Denominator: 2,715 states and provinces in the model universe, 686 with a located paper of this corpus; 2,348 are drawn on their outlines (RESOLVE Ecoregions 2017 and GADM 4.1, 5,000 m simplification, rounded to the canvas); 367 are too small to draw at this scale (one canvas unit is about 40 km at the equator). Shading is the tail probability of the observed count under NB(expected, theta): the two darkest classes are the interval labels below and above; a unit with no located paper and an expectation under 0.5 sits in the neutral class by construction (the zero rule). No unit falls below its interval, so the darkest "less than expected" class is empty; 68 units are above theirs. Unshaded land is outside the universe (under 1,000 km2, or polar). A shaded unit is a candidate gap or a candidate artefact, not a finding.
Ten states and provinces with the least located research relative to the model's expectation, ordered by the tail probability P(count at or under the observed). Denominator: 2,715 states and provinces in the model universe, 686 with a located paper.
State or provinceCountryObservedExpected, 90% intervalTail probabilityLabel
Kalimantan UtaraIndonesia220.5190%: 0 to 730.226within
Uttar PradeshIndia113.1290%: 0 to 470.226within
CoahuilaMéxico05.4390%: 0 to 200.237within
OregonUnited States05.3790%: 0 to 200.239within
East SepikPapua New Guinea05.1490%: 0 to 190.245within
BiharIndia05.0490%: 0 to 190.247within
SomaliEthiopia04.5990%: 0 to 170.261within
IdahoUnited States04.3490%: 0 to 160.269within
CaquetáColombia04.3290%: 0 to 160.270within
TocantinsBrazil04.3090%: 0 to 160.271within
Five states and provinces with the most located research relative to the model's expectation, ordered by the tail probability P(count at or over the observed). Denominator: 2,715 states and provinces in the model universe, 686 with a located paper.
State or provinceCountryObservedExpected, 90% intervalTail probabilityLabel
AlbaRomania110.0290%: 0 to 0under 0.001above
IburengerazubaRwanda50.0390%: 0 to 0under 0.001above
TaiwanTaiwan70.1290%: 0 to 1under 0.001above
BrokopondoSuriname40.0490%: 0 to 0under 0.001above
UlaanbaatarMongolia30.0490%: 0 to 0under 0.001above

The "more than expected" list doubles as a geocoding audit: a small unit with tens of papers against an expectation under one is as likely a place-name collision, a herbarium or a field station as a genuine concentration of fieldwork.

3 · Calibration

Out of fold, by decile of expected count (observed / expected): ecoregions: deciles 3 to 8 between 0.89 and 1.79, top decile 1.28; states and provinces: deciles 3 to 8 between 0.55 and 0.98, top decile 1.20. The middle deciles are the best calibrated; the top decile is under-predicted in every model, because the busiest places (long-term field stations, provinces with a herbarium or a university) exceed what any smooth covariate model gives. That is one more reason the residuals are candidate gaps and candidate artefacts rather than findings.

4 · Method note Beta

Response and exposure. 2,689 of 7,045 papers in this corpus have a precise trusted study site (38.2 percent); the response per unit is the number of those papers with a site inside it (point in polygon, or the nearest unit within 25 km, as the place explorer counts), so a paper with sites in two units counts once in each: 3,115 unit-papers over 777 ecoregions and 2,935 over 2,715 states and provinces (of 847 and 3,686 units in all). The exposure is log(area), with log(area) also entering as a slope so the data set the area elasticity.

Covariates. log_area_km2, log_access_km, forest_loss_pct, forest_loss_missing, aridity_trend_pct_per_decade, aridity_trend_missing, log_national_loo_density, national_loo_undefined, biome, realm, lat_band; continuous covariates standardised, categorical levels with fewer than 10 units pooled. Protected-area share is not available (no protected-area table in the database) and is stated as the natural next covariate.

Model and validation. Negative binomial (NB2) GLM with a log link and a ridge penalty on the standardised slopes (ecoregions lambda 100, states and provinces lambda 30); the mean estimator and lambda are chosen on out-of-fold Poisson deviance over the same spatial folds; the dispersion is estimated by maximum likelihood. 5-fold spatial cross-validation blocked by realm (ecoregions) or country (states and provinces), never random rows; a held-out realm's own dummy is unseen in training. The 90 percent interval is the 5th and 95th percentile of NB(expected, theta) and carries no parameter uncertainty, so it is if anything too narrow. Zero rule: a unit with no located paper and an expectation under 0.5 is never labelled below and never ranked.

Limits. Located papers only: papers whose text names no locatable site are invisible here, and the geoparsing error rate of the sections above applies. The national-effort covariate carries most of the between-country signal; the fit without it is printed above. Terrestrial units only. This is a Beta layer: the numbers are honest as stated, and they will move when the geocoding, the corpus or the covariates change.

Provenance. Generated 2026-10-29 from the published database (SELECT only) and the change layers' cache. Refresh with node phase1/expected_research.js market-pressures, redraw with phase1/gen_atlas_expected_section.js; every figure in this section is checked by the figures audit.

12

Studies by half-decade of publication

5,493 of the 7,045 core studies (78 percent) were published in 2015 or later.

Core studies by half-decade of publication, 1990 to 2026 · share of 7,045 core studies
1990 to 1994
36 · 1%
1995 to 1999
74 · 1%
2000 to 2004
190 · 3%
2005 to 2009
339 · 5%
2010 to 2014
868 · 12%
2015 to 2019
1,577 · 22%
2020 to 2024
2,195 · 31%
2025 to 2026
1,721 · 24%

The last half-decade is incomplete: it holds studies published from 2025 to the October 2026 harvest.