The Desertification Research Atlas
Of 28,111 indexed desertification studies, 15,016 (53.4 percent) measure degradation and 3,942 (14.0 percent) evaluate an intervention.
We analyzed 28,111 peer-reviewed studies on desertification, land degradation and drought, tagged across dozens of dimensions and mapped to 176 countries. The evidence base is weighted toward measurement over intervention, soil over social factors, and remote sensing over fieldwork. The UNCCD is convening on Restoring Land, Restoring Hope.
Corpus and method
Every claim below is scoped to the indexed scientific record . The corpus is a documented, reproducible set of studies.
The United Nations Convention to Combat Desertification (UNCCD) frames its science around three linked phenomena, desertification, land degradation and drought (DLDD), and around Land Degradation Neutrality, SDG 15.3.1. To read that field completely, we harvested the deduplicated union of a versioned Boolean search over OpenAlex, then passed every record through a two-stage relevance gate: a proposer classifies each abstract, an independent skeptical model confirms or rejects, and a third model audits the boundary. The result is a clean core of 28,111 genuinely on-topic studies (a further 24,565 degradation-adjacent papers are retained but held separate).
Using a term is a weaker signal than engaging the topic. Only 15,446 of the core studies, 55%, use any of the three phrases in their title or abstract at all. The other 12,675 never say “desertification,” “land degradation” or “drought,” yet each is on the subject: soil erosion, salinization, rangeland loss, sand encroachment, dryland restoration. A keyword search on the three terms would miss nearly half of this field; classifying by meaning is what keeps it complete. You can put this to the test in the: search the corpus for desertification, land degradation or drought, filter it by country, and read the studies behind every count.
Each study is then tagged by a domain-expert panel. A fast model proposes tags, a second, stronger model reviews each against the codebook, and only tags that are in-vocabulary, supported by the abstract, and confirmed by the reviewer are kept. Re-judged blind by an independent model on the finished corpus (10 October 2026, 200 sampled tags): precision 69.0% (95 percent interval 62.3 to 75.0), and 70.5 percent on the population-random slice of 149 tags. The measurement standard tightened on 5 October 2026 when the domain rubric was clarified; the previous tags measure 70.0 percent on the same standard, and the September 2026 figure of 72.5 percent came from 40 tags (interval 57 to 84). Recall was measured at 57% on a reference set during the August 2026 build gate, so counts are more likely under- than over-stated. Every study is geocoded to its study site, the place the fieldwork was carried out, which locates the research where it was done.
Research topics in the corpus
Desertification research is predominantly about soil and measurement, documenting the drivers and processes of degradation in detail.
Far more studies measure degradation than test responses
Sorted by what each study does, the most common activity by a wide margin is monitoring degradation. About one study in seven tests whether an intervention works.
Remote sensing is increasingly used: of the monitoring studies, 7,334 use satellite data. Restoration is the most common response studied (8,167 studies, 29% of the corpus), and it is the one area where evaluation is more common than average. Yet even here, only about one restoration study in four actually tests whether the intervention worked; the rest describe, review, or set a baseline.
Topics with few indexed studies
Three areas are notably under-represented, each aligned with a UNCCD priority.
1 · Human dimensions are under-represented
Desertification involves tenure, migration and livelihoods as well as soil and salinity. 25% of studies, 6,937 of 28,111, engage a livelihoods or socioeconomic dimension; the remaining three in four are purely biophysical. Soil is the recorded medium in 47% of studies; economic or social data appear in 3%. Within that human subset, the topics central to the Convention are the least covered.
Tenure, migration, conflict and gender, central to UNCCD’s second strategic objective, each appear in under 2% of the corpus. Gender appears in 130 of 28,111 studies; migration in 290; conflict and human security in 195. The links most often cited in policy, that degradation drives displacement and conflict, are among the least evidenced in the literature.
Local and traditional knowledge is also scarce in the record. Smallholder farmers are the knowledge-holders in 2,270 studies and local communities in 1,519, but indigenous peoples in just 170, and traditional ecological knowledge informs restoration in only 16. Pastoral and indigenous stewardship is largely absent from the evidence base.
2 · Rangelands are under-represented
Rangelands and grasslands are roughly half the world’s drylands and among its most degraded systems. Savanna & Grassland is the study ecosystem in 16% of the corpus (against 42% for croplands and other anthropogenic land); grazing and rangeland management is the intervention in 5%, and pastoral livelihoods in 6%. This coincides with the 2026 International Year of Rangelands and Pastoralists.
3 · Decline is documented more than recovery
84% of studies describe land that is degraded or degrading. 4% document land that is improving or recovering. Whether recovery is rare or under-reported, the literature documents loss far more than repair.
Study-site countries compared with degraded land
Research effort and degradation burden are essentially uncorrelated at the country scale.
One country accounts for more geolocated desertification science than the next six combined. China hosts 5,229 core study sites; The African drylands, where degradation affects the most livelihoods, are studied far less relative to their degradation burden.
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.
The map above is fixed. The interactive version plots the same study sites and lets you filter them by this atlas’s own tags, country, year and citations, then read the papers behind any cluster.
Open the interactive map →The skew is biogeographic, too. 29% of studies sit in the Palearctic, that is China, Central Asia, the Mediterranean and North Africa, against 13% in the Afrotropics, where dryland populations are largest and fastest-growing. The human-dimensions literature runs the other way: it is disproportionately African. Ethiopia is its second-largest contributor after China, and Kenya, Burkina Faso and Nigeria rank above their overall research share.
The UNCCD’s own indicator of degradation burden is SDG 15.3.1, the proportion of land degraded. Cross-referenced against it, the two are barely related. Ranking dryland countries by degradation and by research effort produces near-independent lists (Spearman’s ρ ≈ 0). Highly degraded drylands such as Somalia, Eritrea and the Gambia carry heavy burdens and few studies each.
Geographic distribution of study sites
15,766 of 28,111 studies name a locatable study site; where those sites fall, and how far they sit from the nearest large city.
1 · Species ranges against study sites
This is a theme atlas with no species census, so there is no range against which to score the study sites. The two measures below, where the sites are and how far they sit from access points, apply to the 15,766 of 28,111 studies that name a locatable study site.
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. Most cells hold one site; the fieldwork of the whole corpus concentrates in a small number of them.
The densest window, enlarged
The dashed rectangle on the map above marks the window that holds the most study sites (67°E to 121°E, 17°N to 46°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.
3 · 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, 99 sites within 100 km), Bandung (Indonesia, 52 sites within 100 km), Umuahia (Nigeria, 41 sites within 100 km); 5.1% of all sites lie within 100 km of one of the ten busiest.
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; 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; 15,766 of 28,111 studies (56.1%) have one. Density and distance use precise sites only (not a whole country, not a rounded centroid): 7,727 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.
5 · Explore these layers
The coverage cells are a layer on the interactive map, switched on when you open it from here, with the study sites and this atlas's tag filters underneath. Any country, ecoregion, administrative region or drawn area can be queried against the whole database in the place explorer, which lists the papers inside it and the topics that are thin there.
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. 44 of 534 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 0.066 across 534 ecoregions and −0.035 across 1,910 admin1 regions; for drying it is 0.007 across 237 ecoregions and 0.033 across 770 admin1 regions. 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 · Forest loss 2001 to 2024
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 and each first-level administrative region 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.
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".
| Ecoregion | Country | Forest lost | Papers with a site | Per 10,000 km2 |
|---|---|---|---|---|
| Madagascar humid forestsTropical & Subtropical Moist Broadleaf Forests | Madagascar | 67.7%88% forested in 2000 | 0 | 0.00 |
| Northern Khorat Plateau moist deciduous forestsTropical & Subtropical Moist Broadleaf Forests | Thailand, Laos | 61.3%25% forested in 2000 | 0 | 0.00 |
| Khangai Mountains conifer forestsTemperate Conifer Forests | Mongolia | 55.5%9% forested in 2000 | 0 | 0.00 |
| Santa Lucia Montane Chaparral & WoodlandsMediterranean Forests, Woodlands & Scrub | United States | 53.9%52% forested in 2000 | 0 | 0.00 |
| Petén-Veracruz moist forestsTropical & Subtropical Moist Broadleaf Forests | México, Guatemala, Belize | 51.9%81% forested in 2000 | 0 | 0.00 |
| Knysna-Amatole montane forestsTropical & Subtropical Moist Broadleaf Forests | South Africa | 51.7%82% forested in 2000 | 0 | 0.00 |
| Nyanga-Chimanimani Montane forest-grasslandMontane Grasslands & Shrublands | Zimbabwe, Mozambique | 49.3%35% forested in 2000 | 0 | 0.00 |
| Xingu-Tocantins-Araguaia moist forestsTropical & Subtropical Moist Broadleaf Forests | Brazil | 49.1%90% forested in 2000 | 1 | 0.04 |
| Madagascar mangrovesMangroves | Madagascar | 47.5%29% forested in 2000 | 0 | 0.00 |
| Klamath-Siskiyou forestsTemperate Conifer Forests | United States | 46.8%87% forested in 2000 | 0 | 0.00 |
| Region | Country | Forest lost | Papers with a site | Per 10,000 km2 |
|---|---|---|---|---|
| Bình Dương | Vietnam | 96.3%37% forested in 2000 | 0 | 0.00 |
| Bong | Liberia | 95.5%100% forested in 2000 | 0 | 0.00 |
| Aveiro | Portugal | 93.7%62% forested in 2000 | 0 | 0.00 |
| Porto | Portugal | 90.9%33% forested in 2000 | 0 | 0.00 |
| Montserrado | Liberia | 90.4%88% forested in 2000 | 0 | 0.00 |
| Bomi | Liberia | 89.3%99% forested in 2000 | 0 | 0.00 |
| Melaka | Malaysia | 88.7%83% forested in 2000 | 0 | 0.00 |
| Margibi | Liberia | 87.3%95% forested in 2000 | 0 | 0.00 |
| Otdar Mean Chey | Cambodia | 84.7%45% forested in 2000 | 0 | 0.00 |
| Braga | Portugal | 82.2%48% forested in 2000 | 0 | 0.00 |
Forest loss 2001 to 2024, cell by cell
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 and each first-level administrative region 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.
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.
| Ecoregion | Country | Aridity trend | Papers with a site | Per 10,000 km2 |
|---|---|---|---|---|
| Southern Andean steppeMontane Grasslands & Shrublands | Argentina, Chile | −13.7%dry sub-humid | 0 | 0.00 |
| Sudd flooded grasslandsFlooded Grasslands & Savannas | South Sudan, Ethiopia | −10.4%semi-arid | 3 | 0.16 |
| Dry ChacoTropical & Subtropical Grasslands, Savannas & Shrublands | Argentina, Paraguay, Bolivia | −10.3%semi-arid | 20 | 0.25 |
| Khangai Mountains alpine meadowMontane Grasslands & Shrublands | Mongolia | −9.8%dry sub-humid | 1 | 0.27 |
| Gobi Lakes Valley desert steppeDeserts & Xeric Shrublands | Mongolia | −9.6%arid | 2 | 0.14 |
| Western Australian Mulga shrublandsDeserts & Xeric Shrublands | Australia | −9.5%arid | 2 | 0.04 |
| Knysna-Amatole montane forestsTropical & Subtropical Moist Broadleaf Forests | South Africa | −9.2%dry sub-humid | 0 | 0.00 |
| High MonteMontane Grasslands & Shrublands | Argentina | −9.2%arid | 0 | 0.00 |
| Southern Andean YungasTropical & Subtropical Moist Broadleaf Forests | Argentina, Bolivia | −8.5%dry sub-humid | 3 | 0.40 |
| Khangai Mountains conifer forestsTemperate Conifer Forests | Mongolia | −8.5%semi-arid | 0 | 0.00 |
| Region | Country | Aridity trend | Papers with a site | Per 10,000 km2 |
|---|---|---|---|---|
| Central Equatoria | South Sudan | −12.8%dry sub-humid | 0 | 0.00 |
| San Juan | Argentina | −12.7%semi-arid | 1 | 0.11 |
| Govi-Altay | Mongolia | −12.1%arid | 1 | 0.07 |
| Gafsa | Tunisia | −12.0%arid | 0 | 0.00 |
| La Rioja | Argentina | −11.4%arid | 0 | 0.00 |
| Hovd | Mongolia | −11.1%semi-arid | 1 | 0.13 |
| Libertador General Bernardo O'Hi | Chile | −10.9%dry sub-humid | 0 | 0.00 |
| Warap | South Sudan | −10.9%dry sub-humid | 0 | 0.00 |
| Hilmand | Afghanistan | −10.8%arid | 0 | 0.00 |
| Madaba | Jordan | −10.6%arid | 0 | 0.00 |
Aridity trend 1990 to 2024, cell by cell
3 · When the research happened
The same study sites split by decade of publication. Of the 12,786 located papers, 72.8% were published in the 2010s or 2020s and 51.5% since 2016. The whole corpus of 28,111 papers, located or not, is 72.3% from those two decades, so located research is about as recent as the corpus. Of the 593 ecoregions with a located paper, 50 have none published since 2010; of the 1,468 admin1 regions, 118. 14 located papers have no publication year.
The decade with the most located papers, cell by cell
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 1,000 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 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. 12,786 of 28,111 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. 13 precise places (open sea, Southern Ocean, atoll centroids) fall in no ecoregion and 11 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.
5 · Explore these layers
These layers are on the interactive map as forest loss 2001 to 2024 and aridity trend 1990 to 2024, cell by cell with the value and the located papers on hover, and a decade selector that limits the study-site dots to one decade of publication.
Expected research Beta
How much located research would each place be expected to carry, given its size, dryland character, access, environmental change, biome and national research effort, and where the record falls short of that expectation. Cross-validated pseudo-R2 0.49 for ecoregions and 0.44 for administrative regions. Beta: a first model, stated with its interval and its limits.
Beta This section fits a model of research attention: the number of papers of this corpus with a precise study site inside each ecoregion and each first-level administrative region, predicted from the unit's area, dryland character (the TerraClimate mean aridity index and whether the unit is dryland), 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 0.49 for ecoregions and 0.44 for administrative regions (admin1). Each unit carries a 90 percent prediction interval from the negative-binomial count spread about its expectation. The rankings order units by the tail probability of the observed count, the same quantity the interval label thresholds, and carry the label alongside.
1 · Ecoregions: expected against observed Beta
Fit. 760 ecoregions in the universe, 577 with a located paper; negative binomial, dispersion theta 0.85, mean by Poisson quasi-likelihood, ridge lambda 300. 5-fold spatial cross-validation blocked by realm: pseudo-R2 0.49 (0.19 without the national-effort covariate), Spearman rho 0.76 out of fold. Labels against the 90 percent interval: 1 below, 727 within, 32 above.
| Ecoregion | Country | Observed | Expected, 90% interval | Tail probability | Label |
|---|---|---|---|---|---|
| North Tibetan Plateau-Kunlun Mountains alpine desertMontane Grasslands & Shrublands | China | 2 | 177.2590%: 6 to 563 | 0.028 | below |
| Sierra Madre Occidental pine-oak forestsTropical & Subtropical Coniferous Forests | Mexico | 0 | 18.1190%: 0 to 58 | 0.071 | within |
| West Saharan montane xeric woodlandsDeserts & Xeric Shrublands | Algeria, Niger, Mali | 0 | 17.7990%: 0 to 57 | 0.072 | within |
| Southern Andean steppeMontane Grasslands & Shrublands | Argentina, Chile | 0 | 15.5290%: 0 to 50 | 0.080 | within |
| High MonteMontane Grasslands & Shrublands | Argentina | 0 | 14.9490%: 0 to 48 | 0.083 | within |
| Karakoram-West Tibetan Plateau alpine steppeMontane Grasslands & Shrublands | Pakistan, India, China | 3 | 56.2590%: 1 to 179 | 0.092 | within |
| West Sahara desertDeserts & Xeric Shrublands | Algeria, Libya, Mauritania | 5 | 64.8390%: 2 to 206 | 0.115 | within |
| Paropamisus xeric woodlandsDeserts & Xeric Shrublands | Afghanistan | 0 | 9.7990%: 0 to 32 | 0.116 | within |
| Maputaland coastal forests and woodlandsTropical & Subtropical Moist Broadleaf Forests | Mozambique, South Africa, Eswatini | 0 | 8.9490%: 0 to 29 | 0.125 | within |
| Eastern Anatolian deciduous forestsTemperate Broadleaf & Mixed Forests | Türkiye | 3 | 37.7990%: 1 to 121 | 0.127 | within |
| Ecoregion | Country | Observed | Expected, 90% interval | Tail probability | Label |
|---|---|---|---|---|---|
| Caatinga Enclaves moist forestsTropical & Subtropical Moist Broadleaf Forests | Brazil | 48 | 0.6790%: 0 to 3 | under 0.001 | above |
| Iceland boreal birch forests and alpine tundraBoreal Forests/Taiga | Iceland | 13 | 0.1790%: 0 to 1 | under 0.001 | above |
| Fiji tropical moist forestsTropical & Subtropical Moist Broadleaf Forests | Fiji | 5 | 0.0390%: 0 to 0 | under 0.001 | above |
| Atlantic coastal pine barrensTemperate Conifer Forests | United States | 31 | 1.2490%: 0 to 5 | under 0.001 | above |
| Mascarene forestsTropical & Subtropical Moist Broadleaf Forests | Réunion, Mauritius | 3 | 0.0390%: 0 to 0 | under 0.001 | above |
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 · Administrative regions (admin1): expected against observed Beta
Fit. 2,700 administrative regions (admin1) in the universe, 1,386 with a located paper; negative binomial, dispersion theta 0.67, mean by Poisson quasi-likelihood, ridge lambda 300. 5-fold spatial cross-validation blocked by country: pseudo-R2 0.44 (0.20 without the national-effort covariate), Spearman rho 0.66 out of fold. Labels against the 90 percent interval: 0 below, 2,587 within, 113 above.
| Region | Country | Observed | Expected, 90% interval | Tail probability | Label |
|---|---|---|---|---|---|
| Xinjiang Uygur | China | 0 | 15.4090%: 0 to 54 | 0.119 | within |
| Gafsa | Tunisia | 0 | 14.8890%: 0 to 52 | 0.122 | within |
| West Bank | Palestine | 2 | 46.5090%: 0 to 162 | 0.128 | within |
| Cascades | Burkina Faso | 0 | 13.7190%: 0 to 48 | 0.128 | within |
| Hims | Syria | 0 | 10.4590%: 0 to 37 | 0.153 | within |
| Balkan | Turkmenistan | 0 | 9.7290%: 0 to 34 | 0.160 | within |
| Illizi | Algeria | 0 | 9.6490%: 0 to 34 | 0.161 | within |
| As-Sulaymaniyah | Iraq | 2 | 25.8190%: 0 to 90 | 0.186 | within |
| La Rioja | Argentina | 0 | 7.3890%: 0 to 26 | 0.189 | within |
| Adrar | Algeria | 1 | 15.6490%: 0 to 55 | 0.194 | within |
| Region | Country | Observed | Expected, 90% interval | Tail probability | Label |
|---|---|---|---|---|---|
| Hong Kong | China | 78 | 1.2390%: 0 to 5 | under 0.001 | above |
| Al Gharbiyah | Egypt | 48 | 0.9090%: 0 to 4 | under 0.001 | above |
| Western | Fiji | 5 | 0.0490%: 0 to 0 | under 0.001 | above |
| Lima Province | Peru | 3 | 0.0190%: 0 to 0 | under 0.001 | above |
| Lac | Chad | 8 | 0.2190%: 0 to 1 | under 0.001 | above |
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 0.56 and 1.10, top decile 1.14; administrative regions (admin1): deciles 3 to 8 between 0.90 and 1.10, top decile 1.33. 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. 12,786 of 28,111 papers in this corpus have a precise trusted study site (45.5 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: 14,610 unit-papers over 760 ecoregions and 13,964 over 2,700 administrative regions (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_mean_aridity_index, dryland, 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. Richness is a country-level proxy: every unit of a country is credited with the national list.
Model and validation. Negative binomial (NB2) GLM with a log link and a ridge penalty on the standardised slopes (ecoregions lambda 300, administrative regions (admin1) lambda 300); 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 (admin1), 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 2026-10-14 from the published database (SELECT only) and the change layers' cache. Refresh with node phase1/expected_research.js desertification, redraw with phase1/gen_atlas_expected_section.js; every figure in this section is checked by the figures audit.
5 · Explore this layer
The same units are on the interactive map as the beta layer "Expected vs observed", ecoregions or administrative regions, with observed, expected, interval and label on hover, and a link from each unit to the place explorer. The place explorer shows the same figures for a selected unit under "Beta: expected research".
Research topics by half-decade, 1990 to 2025
The field has grown roughly eighteen-fold since the early 1990s, shifting toward remote sensing and soil science and away from rangeland and human-dimension research.
Remote sensing is rising
Share of studies using satellite & GIS methods, by half-decade, tripling from 11% to 34%.
Rangeland science is falling
Share of studies on rangelands & pastoral systems, by half-decade, drifting down from 13% to 8%.
Restoration research kept pace with the field’s overall growth (a steady ~33% share); human-dimensions research declined, from about a quarter of studies to a fifth. The trend is toward biophysical measurement and away from the social and pastoral questions the Convention foregrounds.
Six cross-tabulations
The tags reveal finer patterns than the headline numbers, from how studies define “degraded” to how rarely they reference the relevant conventions.
Summary of the record
Nearly half of all studies engage an intervention, and responses are matched to problems consistently. The small fraction that documented recovery points toward what works, but those signals are rarely tested.
Responses are matched to problems
Responses map onto processes. Each degradation process has a characteristic response: soil erosion draws soil-and-water conservation (2,210 studies), sand encroachment draws dune stabilization (622), and vegetation loss draws restoration and revegetation (2,934 and 1,704 studies). Salinization is an exception, 14% of studies, but with no dedicated intervention; it defaults to generic “restoration.”
What the field associates with recovery
1,105 studies (4%) document land that is improving. Cross-referenced with intervention, active revegetation and restoration are the responses most often present where land recovers, each in about 8% of their studies, about twice the rate for soil-and-water conservation or agroforestry (3 to 4%). This is an association, and it has not been examined systematically.
More often named than evaluated
Across the physical interventions, about a quarter of studies evaluate an outcome. Soil-and-water conservation is at 29%, restoration 27%, revegetation 25%. The least-evaluated response is the one the Convention most directly controls: governance and policy interventions are evaluated in about 4% of the studies that invoke them. Policy instruments, tenure reform, subsidies and land-use rules, are described and modeled far more often than measured against outcomes.
Studies that evaluate interventions
Each of these four steps targets a specific gap the tags identify, and each is answerable from the existing corpus.
The COMPASS corpus is complete and queryable; this atlas is one view of it. A drought-resilience atlas, a Sahel deep-dive, a salinization gap map, or a single Party’s national evidence profile are each a query away, on the same reproducible foundation.