Methodology: GeoJSON check and RO Geo APV

The Identifier is step 1 and is built to be fast: it answers from reference files held on its own machine, calls no outside service, so the first answer for each plot comes within seconds.

On the website you send one GeoJSON file (each Feature is one plot). The Analyzer of one website check takes up to 50 plots; a file of up to 500 plots can be checked as a random sample of 10, 30 or 50 of them. Larger volumes and other file formats go through the dedicated API (see "Outline correction and other formats"). The Identifier has no per-check fee with an API key (the key is given after your e-mail address is confirmed, on the website only); the Analyzer is priced by the number of plots (see "What is charged, and when (website)").

What you send

• The plot: a Point, a Polygon, a MultiPolygon, a Feature holding one of them, or a FeatureCollection. Coordinates in longitude and latitude (WGS 84), longitude first.
• A point stands for a circle of the area you declare (0.01 ha to 10,000 ha); without a declared area, 4 ha.
• The commodity (required): cocoa, coffee, oil palm, rubber, soya or wood. Cattle is not checked: a check of cattle is refused and nothing is charged.
• The country you declare (optional): turns on the test for swapped coordinates.
• The harvest period (optional). A plot is its outline, its commodity and its declared harvest period: the same outline with another commodity or period is another plot.

Limits of an outline

• Up to 20,000 points (all rings and parts counted) and up to 10,000 ha per plot. A larger plot gets the Identifier's answer only and is not counted.
• Refused as they are: lines, several points in one plot, mixed collections, coordinates outside -180..180 / -90..90, outlines crossing the 180th meridian, and coordinates in another system than longitude and latitude.
• A broken outline is never repaired silently. If the border crosses itself, parts overlap, a hole lies outside the outline or there is no area, the answer names the fault in words and the plot is stopped. You can fix the outline yourself, or have it repaired by the outline correction of the dedicated API.

What the Identifier checks

Sources must agree. Where two sources describe the same thing (built-up land from two maps, water from satellite against mapped lakes and rivers) and disagree, the answer shows both readings. It does not average them and does not pick one.

Result of the Identifier

• Stop: the outline cannot be read, or half or more of the plot is surely in water ("surely" means no land lies within about 500 m of the sea part; a lake or water seen from satellite always counts as sure). A stopped plot is not worked further and not counted.
• Warning: water on 1% to under half of the plot; sea close to a shore; the plot crosses a border; the centre lies in a town; a point given for more than 4 ha (the regulation asks for an outline above 4 ha); coordinates that look swapped; sources that disagree.
• Not verified: a map the Identifier needs could not be read this time; the answer names it.
• No warning: none of the above. The answer lists what was looked for.

First figures on the website

On the website, after the price of a check is shown and before you pay, each plot the Identifier let through gets two first figures from our own maps, with no fee: the share of the plot that was forest at the end of 2020 on the EU reference forest map of 2020, and the share in protected areas, where our protected-areas data for the place is verified. They are first figures, not the result: the paid land check reads every source of each part. A figure that could not be made is said "not available yet", with the reason, and never shown as 0. One account gets first figures for up to 300 plots in 24 hours; beyond that the plot says so.

The Analyzer is step 2, the land check of each plot the Identifier did not stop. It reads data we hold on our own machines (maps, registers and statistics we keep copies of, the elevation model included), so its answer does not depend on another service being up. A part that cannot answer does not stop the others.

Part A has three sub-parts: A1 the satellite maps, A2 the cadastre of 2020, A3 the other official registers and sources of 2020. Part B has two: B1 land use since 2020 (the maps), B2 the most recent cadastre. A sub-part for which we hold no data for the place says "No data held" and does not count.

Each part lists its sources, with the role each one plays: a source that decides a figure, a source given as evidence beside it (it can show that sources differ, never a result on its own), and context. A source that applies to the plot but could not be read makes the part "not verified" for what it would have answered: never a "no warning" by silence. A source we do not hold for the place is said as a limit, with its name.

Read as published. No source is altered, and Insighter adds no thresholds of its own to a publisher's definitions.

The one table: the plot from 2020 to now

The report shows the plot and its land in ONE table, from the end of 2020 to the most recent data. Its first rows are the plot itself (country and region, area and outline, coordinates, commodity, water, towns, species declared), what the plot can produce (an expeditive estimate) and what was declared with it. Then come the land rows: each is a part of the plot in hectares and in per cent of its area; the rows of each level add up to the level above, and the first level adds up to the whole plot. Official records come last, apart from the levels (see "Records in the table").

• The grid. All maps are read on one grid of cells (about 10 to 30 m, by the map). A cell touching the outline counts by the share of it inside the outline, so the table adds up to the plot's area.
• The levels. 1. The land at the end of 2020: forest, agriculture, pasture or grassland, built land, water, other land (shrub, bare land, trees not counted as forest). A tree-crop plantation of 2020 is agriculture. 2. For the forest of 2020 of a wood plot, its type: natural forest, plantation, or natural or plantation (undecided) where the sources cannot tell. 3. In a protected or indigenous area, or outside (part C). 4. Burned after 2020, or not. 5. What happened since. Forest: not changed (trees at the end of 2020 and since), trees cut after 2020 (for a crop, trees cleared), unclear. Land not forest in 2020: not changed, changed to one class (trees two years in a row, or another class in the two newest years), or unclear.

A level that does not split on the plot is not printed as its own row; what it says is added to the row above it.

• Trees cut after 2020 are split, cell by cell, into parts that add up to the row: converted to other land (crop, pasture, built land, a tree crop or water two years in a row after the loss; or no trees again on the yearly vegetation index since a cut of two years or more), natural forest to plantation, harvest (trees back after the loss, or a plantation with trees every year; for a crop, trees growing back), too recent to tell (under two years seen since the cut) and not known yet (open land with no use shown, or not read). The Source cell names the use or the evidence of each part. Only "converted to other land" and "natural forest to plantation" (a proven conversion) are marked orange; "trees not back yet" (cut recently, too early to tell) is marked yellow when it is over 0 ha; a cut alone is not marked.
• Unclear is the land where the sources disagree: the land cover shows a change, or crop or built land in 2020, that the loss maps did not record (a cut they missed, a partial removal under 30 m, or a land cover in error). It is shown as its own row, never hidden in another one.
• Protected or indigenous land and burned after 2020 are always two rows at the same place, right after the whole plot, also when there is none: "None (8 sources asked)", "None (VIIRS, MODIS)", or "Not read on the cells" with the reason.

Reference and range. Each source answers in its own way (definition, cell size, year, method), so sources give different figures for the same land. Each row gives one reference figure, from one named source or rule (the Source column), and a range over the sources held: the lowest and the highest figure, each end one source's own figure on the same land (never a mix of sources), with the two end sources named. "1 source" means no other source gives that row; a narrow range means the sources agree. A source that covers under 99% of the land of a row is named beside the range, never used as its end. Where a "not changed" figure lies below its range, the Source cell says why (the reference needs no loss on any loss map and trees on the land cover).

The reference sources. One reference per thing, fixed by rule for each country:

The yearly vegetation index (NDVI) decides only where the maps disagree. Our copy of the yearly cloudless Landsat composites (30 m, from 2020 to the newest year it holds; later years said as not held) gives each cell its greenness and canopy-moisture index for every year. Reference cells are chosen on the plot by fixed rules: sure natural forest, sure plantation (every type map that gives a value agrees, no loss on any loss map, trees every year) and sure open land. At least 20 cells of 30 m are needed per class; with fewer, the same rule is used on a ring of 1 km, then 3 km, around the plot (said); still too few, that class has no signature. A class's signature is the spread of its cells' values year by year (10th to 90th percentile); the noise is the yearly change seen on the sure forest cells. Each unclear cell is then placed, first that applies: open land already in 2020; trees cut in a year after 2020 (below its forest's spread and clearly below its own value of 2020, and not back the year after); still forest; else it stays unclear. Natural forest that became a plantation with no cut is a change of type and is not tested by the index. The Source column says how much of a row the index decided ("+ NDVI (0.28 ha)"). The index is a range source for the forest of 2020 and its type; it is not a range end of the change rows, because it judges only the cells the maps disagree on.

Typing the forest the type source leaves undecided (rule tf-1). A forest cell of 2020 that the type source does not type is typed natural forest or plantation only where at least two independent sources say the same type and none says the other. Each source counts once: JRC Global Forest Types 2020 and forest management 2020 at 100 m together as one source (related work; the 100 m map from plots of 4 ha), ForTy 2020, Natural Forests of the World 2020 (natural forest only), the same land-use series in 2019 and 2021, and the yearly vegetation index (the cell's values of 2020 inside one type's spread and outside the other's). The type source's own classes never move. Any other cell stays "natural or plantation (undecided)", with why (the sources disagree, one source only, no source). The Source column gives the hectares typed so, and the vegetation index then judges the unclear cells again on the new types.

How the sources agree on this plot. Under the table, a short table gives, for each level (forest at the end of 2020; the type of 2020; not changed or cut), the share of the land where two sources give the same class: the vegetation index against each map, and the maps among themselves (lowest to highest, with the lowest pair). A map that agrees with the type source on less than 70% of the forest both type (compared on 0.5 ha or more) does not vote in the typing of that plot, and the report says so.

Records in the table. The cadastre of 2020, the other official registers of 2020, the most recent cadastre and a declared property or farm land are rows of their own, outside the levels ("none held for this place" where none is held). A cadastre or register of 2020 that covers 99% of the plot or more is also a range end of the forest row; one that covers a part is named beside the range. The forest row says who decided it, for example "decided by the cadastre 2020 on 80% of the plot, the satellite maps on the rest". A declared property or farm land is said as declared, not verified.

Several plots: one report per submission. For two plots or more, the report gives one table for the whole submission: each land row is the sum of the plots' rows, the per cent are of the whole area, and a range end is one source's figure added over the plots of the row. A source that does not give the row on every plot is named "(covers n of m plots)" and left out of the range. The plot rows are merged (one value said once, else each value with its plots; areas added). Every result, warning and signal line names the plots it comes from ("Plot 2", "Plots 1, 4", "All 5 plots"). The figures of each plot are kept per plot, the cover lists the plots with their results, the plot facts are one row per plot, and the maps of each plot come at the end.

A. Land use 2020

Part A uses the regulation's definition of forest (Article 2(4)): land of more than 0.5 hectares with trees higher than 5 metres and a canopy cover of more than 10%, or trees able to reach those thresholds in situ, excluding land that is mainly in agricultural or urban use. Orchards and tree crops in agricultural use are not forest; plantations of trees for wood are forest.

Who decides each cell, in this order: (1) a parcel of the cadastre of 2020 with a recorded land use (A2); else (2) an official register or zoning of 2020 - official forest records, declared farm fields of the paying agencies (A3); else (3) the satellite maps (A1). An official source decides only with a state dated on or before 31 December 2020; an undated record decides nothing and is shown as evidence. Where forest and another class overlap, forest wins (said). Where the records say "not forest" and the maps say forest, the report shows it and part B still follows those cells. A declared property never decides.

The satellite maps (A1). The reference forest map of 2020 (JRC Global Forest Cover 2020, 10 m, made for this regulation) and the forest temporarily without trees at the cut-off (trees before - tree cover of 10% or more in 2000, or forest in an earlier year of a land cover by year -, a recorded loss between 2010 and 2020, and land neither crop nor built in 2020) make the forest of 2020. The other maps of all forest held for the place (national forest maps, the forest type maps, the land-cover maps of 2020: ESA WorldCover, MapBiomas in Brazil, the yearly Impact Observatory map where held) give the range. Maps of natural, primary or tropical moist forest only never vote on "forest or not": they are context. For a crop, a forest cell of 2020 that a dated map of 2020 or earlier, with a published accuracy, shows as a tree crop (oil palm, rubber, the tree-crop classes of MapBiomas) is a tree-crop plantation, not forest; later or undated tree-crop maps are shown beside them and take no cell out.

The result of A (counted):

• Wood: forest at the end of 2020 on at least half of the land of the plot is no warning; under half is a warning ("check the outline and the land"), unless part B shows trees grown since on the land not forest in 2020 that bring it to half.
• An agricultural commodity: forest at the end of 2020 on at least 10% of the plot or 0.5 ha, whichever is larger, is a warning: the plot is likely not where the commodity is grown, or the outline holds forest. When it reaches that only through land the records call not forest and the maps call forest, A is not verified (sources on both sides). Less forest than that: no warning.
• A2 and A3 (crops): a warning where the records of 2020 say forest land.
• Not verified when nothing could be read, when a register that applies could not be read, or when the satellite maps could not be read on the cells they decide. A warning always stays a warning.

B. Land changes after 2020

Question: what happened after 31 December 2020 to the land of 2020, and what is it on the most recent data?

• Loss of trees by year on the forest of 2020: Global Forest Change (University of Maryland, 30 m, loss years to 2025) and, in the humid tropics, the Tropical Moist Forest map (JRC). Official records of forest cover loss (for example INPE PRODES in Brazil) and felling permit areas, where held, are shown beside them.
• What the land became: the newest land-cover years held (MapBiomas to 2025 in Brazil, the yearly Impact Observatory map to 2024 where held, else ESA WorldCover 2021), the crop maps held, and the yearly vegetation index for the cells the maps disagree on.
• Signals: confirmed disturbance alerts (GLAD Landsat, GLAD Sentinel-2, RADD, DIST-ALERT where they cover the place; low-confidence alerts are shown and not counted), burned area (VIIRS and MODIS on the cells of the table) and the yearly vegetation index, each with its first year. They are signals, not proof of a clearing, and are not added to the table's sum.
• The most recent cadastre (B2): the newest parcel register held, where one is held.

Each figure carries its own data date and source: the sources end in different years, so the report says "most recent data", never one shared year. A loss after the newest map year is seen only by the alerts and the burned area.

Words. A loss seen on the maps is said as forest cover loss; the report says that land was converted to another use only where the new use is shown. A loss can be a harvest, a storm or a fire, so the report says what the land became.

The result of B (counted):

• B1, warning: any proven conversion after 2020, whatever its size: forest of 2020 converted to other land (the land cover of the years after shows another use, crop or built land, on a part of it) or natural forest turned into a plantation. Under 0.5 ha the warning carries a note that such a small area is near the accuracy of the maps. Also a warning: forest of 2020 lost after 2020 whose recovery is not shown (the land cover of the years after cannot say it yet, or was not read).
• B1, no warning: no forest cover loss after 2020 on the forest of 2020; or a loss after which the land cover shows trees again on every cell lost (the loss is said as a signal).
• B1, not verified: the forest cover loss after 2020 could not be read for the plot.
• B2 (crops): a warning where the current parcel or zone has a use that excludes the commodity; not verified where a register held could not be read.
• Fire and alerts never give a warning by themselves. What the cut land became (converted, natural forest to plantation, harvest, too recent, not known yet) is given with its plots and hectares.

C. Protected and indigenous land

The plot is compared with the national and regional registers of protected areas held for its countries, an international register, and the registers of indigenous and community land (national registers and world collections of community-mapped land), each in its newest published version. Designated areas count; proposed areas, areas given only as points and areas for information only are named apart. A cell is inside when its centre is. The table gives the part of the plot inside such areas (all sources joined, never counted twice) and the registers by kind and number. Where the law of the country recognises indigenous peoples' land, the report names the act; where no complete register is held for the place, it says that the absence is not confirmed.

Part C also gives the other conservation areas with no legal protection (row 9.5) and the species recorded near the plot: threatened species on the IUCN Red List (9.3) and CITES-listed species (9.4), from GBIF records since 2000, each with its records, nearest distance and newest year. No record nearby is not proof that a species is absent.

Information: part C is shown in the table and in the report and does not change the result. Land in a protected area or on indigenous land is marked in the warning colour in the table, with its share and the sources: a point to look at, not a refusal.

Plot facts (page 1)

Page 1 of the report gives the facts of the plot beside the summary. They are information only and never change the result.

• Terrain: altitude and slope (mean, maximum, share over 30 degrees) and the main orientation of the slopes, from our copy of the Copernicus elevation model (GLO-30, 30 m, read with a margin of 300 m around the plot).
• Climate: the averages of 1981-2010 (CHELSA, about 1 km) and the climate since 2020 (TerraClimate, about 4 km), with the difference between the two.
• Water, access and settlements: rivers, streams and water bodies, roads and power lines, settlements, services and cultural sites on the plot, within 1 km and within 5 km, and the nearest, from our copies of OpenStreetMap. A place not mapped is not seen, so "nothing mapped" never means "nothing there".
• Soil: pH and clay at 0-30 cm (SoilGrids, 250 m), soil carbon where a map is held (MapBiomas Solo in Brazil, the European soil carbon map in Europe, else SoilGrids), and an estimate of soil erosion in tonnes per hectare and year (RUSLE, from the rain, the soil, the slope and the land cover).
• Cuts by year and the harvest period: where a harvest period was declared, the trees cut inside it and outside it, and the cuts by year (rows 1.x.2 of the table).
• Carbon: one line, rounded to 10 t C: the carbon in trees at the end of 2020 and now, and the change, from biomass maps and IPCC default values on the cells of the table. A rough estimate; the full carbon estimate is a dedicated API.

Commodity and place (from the production the country records, the climate, the altitude and slope, the soil and the land cover: fits, unusual for this place, or unlikely) is given in the report's data file.

Online checks (dedicated API only)

Some sources can only be read online at the moment of the check (Whisp, recent satellite images, radar, national services we keep no copy of). They are read in a separate call of the dedicated API, not in the website check. They are evidence: shown beside the parts, they can show that sources differ, never a "no warning" on their own.

The report and its conclusion

When the Analyzer has run to its end, the report is made by fixed rules (no AI writes it):

• Page 1, the summary: the overall result, the results of parts A, B and C (for several plots, an index of the plots with the result of each part) and the plot facts.
• Page 2, the one table: the land of the plot from 2020 to the most recent data, rows numbered 1 to 11 - 1 to 8 the land at the end of 2020 by class and what became of it, 9 protected, indigenous and conservation areas (9.1 protected by law, 9.2 indigenous or community land, 9.3 threatened species recorded nearby, 9.4 CITES-listed species recorded nearby, 9.5 other conservation areas with no legal protection), 10 burned after 2020, 11 the cadastre (11.1 of 2020, 11.2 most recent). Orange marks a proven conversion (any proven conversion after 2020 is a warning of part B); yellow marks trees cut recently that are not back yet (too early to tell).
• Page 3, the sources used: every source pages 1 and 2 cite, with what it was used for, its publisher and its published accuracy and limits.
• Then the maps of each plot: the land at the end of 2020 and the most recent.

Its conclusion comes from the Identifier and parts A and B only:

Part C and the plot facts are information and never make the conclusion a warning; what they found is always listed. The report gives context and warnings, never a rating: you decide what they mean for your due diligence.

These are all the sources the website check reads (data of 2026-11-02): 91 world and regional sources, and 1,302 national and regional layers of 98 countries. Each report lists only the sources it used for that plot. Maps are estimates made from satellite images; registers show what was recorded, on the date they were read.

Identifier: place, borders, water and settlements

Part A, land use 2020: forest maps

Part A, land use 2020: kind of forest (natural, primary, plantation)

Part A, land use 2020: tree crops, crops and grassland

Parts A and B: vegetation index and images

Part B, land changes after 2020: forest cover loss and land cover by year

Part B, land changes after 2020: disturbance alerts

Burned land after 2020 (row 10)

Part C: protected, indigenous and conservation areas (rows 9.1, 9.2, 9.5), international registers

Part C: species recorded near the plot (rows 9.3, 9.4)

Plot facts (page 1): terrain, climate, soil, soil carbon and erosion

Plot facts (page 1): water, access, settlements and works

Plot facts: commodity and place (report data file)

Plot facts (page 1): the carbon line (one rounded estimate)

The maps of the report

National and regional sources

Official layers of national and regional agencies, read where the plot lies. A country not named has none of that kind held.

Read only by the dedicated API

Whisp (FAO / Forest Data Partnership), recent Sentinel-2 images and Sentinel-1 radar read online, the Global Wildfire Information System and national services asked online are read in the online checks of the dedicated API, not in the website check.

What is charged, and when (website)

The website check takes the plots only (GeoJSON).

• No fee: sending the file, the Identifier (no per-check fee with an API key), the first figures and the check of the file at upload. A file refused there is never charged.
• The price of the Analyzer is shown after the upload, by the number of plots counted: up to 10 plots 10 EUR, up to 30 plots 20 EUR, up to 50 plots 25 EUR, excluding VAT. A plot the Identifier stops, the same plot twice in one file, and a plot your account checked recently are not counted. More than 50 plots: choose a random sample of 10, 30 or 50, or use the dedicated API. The prices are those the website shows at the time of the quote; a check keeps the price it was quoted.
• You pay when you confirm, before the check runs.
• What fails on our side is given back at the end of the check: a plot that could not be checked, when it changes the price. Paid from credits: the credits come back to your account (shown as a line given back) and stay valid at least 30 days after they come back. Paid by card for that check: the card payment is reverted for that amount.