Methodology: How WoodComp Builds the Dataset

What is in scope

WoodComp maps wood fibre consumers: facilities that take in roundwood or wood chips and turn them into products. Wood fibre here means primary wood raw material of any kind, sawlogs as well as pulpwood, so a sawmill is as much a wood fibre consumer as a pulp mill even though it never separates a fibre.

Facility categories in the published dataset are pulp mills, wood-consuming biorefineries, and (as the dataset grows) sawmills, pellet plants and wood-fired energy plants. Paper-only, tissue and converting sites exist in our records but are outside the published scope, because they consume pulp or recovered paper rather than wood. The published dataset currently holds 2804 facilities across 58 countries.

Facility status

Each facility carries one status:

• Operating: in production. Planned maintenance stops do not change status.
• Idled: production halted indefinitely for market or ownership reasons, but the plant is intact and capacity could return. Sometimes called curtailed or mothballed.
• Closed: permanently shut. The capacity is gone.
• Converted: rebuilt to another use that no longer consumes wood. A mill that still consumes wood after a rebuild, for example a kraft line converted to dissolving pulp, stays operating and carries the conversion as a dated event.
• Project: announced or under construction, not yet producing.
• Unknown: status could not be verified against an acceptable source.

The distinction between idled and closed matters for fibre balance: idled capacity is latent supply-side demand, closed capacity is removed. Status changes are recorded as dated events with sources.

Location and coordinates

Coordinates are never guessed. Every map pin is the product of a two-part rule: a named source establishes which facility it is, and satellite imagery establishes exactly where it is.

• Identity comes from a source. Depending on country, that source is a national or regional register (for example Ontario's MNRF mill licences, the British Columbia Mill List, the US EPA ECHO facility register, Québec's MRNF register, Chile's RETC, ARESCAT in Catalonia, Norway's Brønnøysund register), an OpenStreetMap industrial feature, or the company's own published address.
• Precision comes from imagery. Register coordinates usually point at the registered address rather than the production site, so every candidate position is checked against satellite imagery before it gets site precision.
• Where no register or map object exists, a position may be derived directly from satellite imagery under a fixed rule: the locating pass reports only an image tile and pixel, the coordinate is computed deterministically by a script, and an independent second pass on fresh imagery must confirm the site before anything is written. Pins produced this way carry the source label satellite-vision.
• Every pin is confirmed by two independent review passes. When the two passes fully disagree, a third pass decides.

Each facility carries a location precision flag: site (verified to the plant itself), city, country, or unverified. Only site-verified facilities appear as map pins, and this applies to every published category, pulp mills and sawmills alike. The rest are listed without a pin until verification completes.

Nearest port

Every site-verified facility shows its nearest port. It is a computed figure, not a verified fact about where the facility ships from, and it is labelled accordingly.

• Two registers, combined. The port list is the union of the NGA World Port Index (about 2 950 harbours with a size class, including many mill quays) and the World Bank's Global International Ports (about 840 ports with documented international container trade). Neither contains the other: the first lists the quay at Husum but not the port of Gävle, the second the reverse. Duplicates are removed by UN/LOCODE or when two entries lie within 3 km.
• Major port is the headline. The figure shown is the distance to the nearest *major* port, meaning one in the World Bank set or of World Port Index size medium or large. Without this rule the nearest entry is often a fishing harbour or an island jetty.
• Own quay as a second line. When any harbour lies within 5 km of the facility, it is shown as the facility's own quay. Coastal pulp mills typically load at their own berth, and this line says so without claiming it as sourced fact.
• Straight line. Distances are great-circle, in kilometres, rounded to whole kilometres above 10 km. Road and rail distance would be a different measure and is not implied. Where the nearest major port is more than 300 km away, as for mills in central Russia or inland Canada, the figure says more about geography than logistics, and the row is flagged as such in the data.
• Recomputed, never edited. The whole table is replaced by a script on every run, so a corrected coordinate changes the port distance with it. Facilities without site precision have no figure rather than an approximate one.

Rail siding

Whether a facility can load rail wagons on its own ground is a fact about the plant, so it is treated like one: it is shown as yes only when it has been verified, and the verification is dated.

• OpenStreetMap proposes. For every site-verified facility, a script looks for an active spur, siding or yard track within 1 km of the plant in OpenStreetMap. A hit makes the facility a candidate, nothing more: OpenStreetMap lists tracks that were lifted years ago, and a siding next door may belong to the neighbour.
• Satellite imagery decides. Each candidate is judged by two independent passes on satellite imagery centred on the track, under a fixed instruction: the track must be seen entering the facility's own yards, not merely passing its fence. A third pass settles disagreement. Only then is the siding published as yes, with the OpenStreetMap way as its source and the verification date shown on the page. A candidate that has not been judged, or where the passes disagree, is shown as not yet verified.
• A published source can settle what imagery cannot. Where the satellite passes stay inconclusive, a source that states the facility loads rail on site, such as the company's own site, an environmental report or trade press, settles the question. Such a siding is shown as documented with its date, and the source is listed on the facility page. The quote is checked against the page by script before anything is written; a group-level statement about a company does not count for one of its mills.
• No siding mapped. Where OpenStreetMap has no such track within 1 km, the page says so and gives the straight-line distance to the nearest active railway instead. That distance is geography, not logistics: a mill 300 m from a passenger line cannot load there. Where the map has no railway within 50 km, the page says that too. Coverage of industrial tracks in OpenStreetMap is good in Europe and North America and thinner in Russia, Brazil and China, so "no siding mapped" there means less than it does in Sweden.
• Follows the pin. The check is part of the location workflow: a facility that gains or changes its site coordinate gets a new port distance and a new rail check before the change is considered complete.

Capacity

Capacity is stored as three parts: a value, a unit, and what the value measures. For pulp mills that is tonnes per year of pulp; for sawmills it is cubic metres per year of sawn timber or of log intake, as the source states it. Site totals and single-line figures are labelled as such. The figure comes from the operator wherever the operator publishes one. Where it does not, and a source we have read reports actual annual volumes for the mill, the highest of the last three reported years is used; planned figures and design figures are never used, and a register that estimates capacity per mill and year applies in its latest edition. Where neither exists, and a regulator has set the permitted annual production in an environmental permit, that permitted figure is used and its source row says so: a permitted maximum is the closest public statement of capacity that exists for such a mill, and the permit is a primary document, but it is a ceiling rather than a measurement and an operator figure replaces it as soon as one is published. Capacity figures are also kept as dated observations, so a facility's history shows when a figure was reported and by whom, and expansions or closures change the record rather than silently overwriting it. Each observation records what the figure rests on: a company figure, an annual report, a permitted maximum from an environmental permit, a register entry, trade press, or another filing. Where the headline figure has such an observation, the facility page says so in a short line under the number ("Permit ceiling", "Company figure for 2024").

Pellets. Capacity is tonnes of pellets out per year (short tons per year for plants reporting to the US Energy Information Administration; the figure is stored as reported and converted to tonnes only for display, at 0.90718474 tonnes per short ton, the same way board feet are converted to cubic metres). The operator's own stated capacity takes precedence. Where the operator publishes only production, the highest of the three latest reported years stands in for capacity and the basis is recorded. Industry tables (Bioenergy International's World of Pellets, Holzkurier's table of plants in the German speaking countries, Bioenergitidningen's Pelletskarta) are used as registers of existence and as capacity sources when no operator figure exists, with the table row as the quoted source. A production figure, a delivery volume or a press rating in tonnes per hour never replaces a stated capacity; production above table capacity is kept as a flagged inconsistency rather than written over the capacity.

Wood consumption factors, pulp

Wood consumption is never stored, only computed, and the factors are published here so the computation can be checked. All factors are cubic metres solid roundwood under bark per air-dry tonne of pulp (ADt, 90 % dry solids). Quoting the same mill over bark, or per oven-dry tonne, gives a materially different number, so the basis travels with the figure everywhere on this site.

The chemi-mechanical range is the mechanical bracket applied for want of a published chemi-mechanical figure, not a measured range of its own. Sulphite pulping is not listed: the sources state its wood input in bone-dry tonnes of wood per air-dry tonne of pulp rather than in cubic metres, so it is quoted in that form on the pulping processes page instead of being converted here.

Where a mill's fibre furnish is unknown but the process is chemical, we use 4.5 as a mixed value until the furnish is established.

Two honesty notes. First, we publish ranges rather than point values, because the literature genuinely spans a range and a single decimal implies precision that does not exist; the point values exist only inside the model. Second, no source publishes a CTMP-specific volumetric factor: every statistical source lumps CTMP into "mechanical pulp". Our 2.6 is derived from a published chemi-mechanical yield range of 86 to 94 percent and a basic density of 390 to 420 kg/m³, and is therefore a derived figure, not a measured one.

Wood consumption factors, sawn timber

Sawmill capacity has conversion factors of its own.

Sawnwood recovery. Softwood sawnwood recovery is roughly half the log volume. Reported national factors run from about 1.5 to just over 2.0 cubic metres of roundwood per cubic metre of sawnwood, and much of that spread reflects where in the process the output is measured, since the same chain moves from about 1.6 rough green to about 2.1 once the timber is dried and planed. Log diameter drives recovery on top of that: about 0.41 cubic metres of sawnwood per cubic metre of logs at a 15 centimetre top diameter, against 0.63 at 60 centimetres.

Log intake and sawn output are different metrics, and a capacity stated as log intake is roughly twice the same mill's sawn output. They are never summed. Where a mill reports only log intake, the list view shows an estimated sawn output so that every row in the column measures the same thing. The factor is 2.0 cubic metres of roundwood per cubic metre of sawn timber: the round figure the source itself leads with, and the conservative end of the range above, so the estimate understates rather than flatters. Every figure derived this way is labelled as an estimate and shown together with the reported intake it came from, so the conversion can be checked or redone with another factor. The facility page keeps the figure exactly as the source stated it.

Board feet. North American capacities reported in thousand board feet (MBF, lumber tally) are converted for display using the factor of 589 board feet per cubic metre: the nominal 424 board feet per cubic metre adjusted by the 72 percent actualization factor used in FAO and UNECE statistics. That is, 1 MBF is roughly 1.70 cubic metres of actual sawn timber. Converted and estimated figures are shown rounded to the nearest thousand cubic metres, because the conversion factor carries no finer precision. The cubic-metre figure leads everywhere, and the facility page keeps the source's own figure next to it, unaltered, so the conversion can be checked or redone.

Sawmill chips. The residue side of the same material balance: with recovery at roughly half the log, the remainder is chips, sawdust and shavings, and the source's worked small-log example splits the log 41 percent sawnwood, 43 percent chips, 9 percent sawdust and 7 percent shavings. For the country fibre balance we estimate chips at 30 to 40 percent of sawlog volume, the span from large-log to small-log industries, applied to national sawlog removals. The chip estimate is always shown as its own supply component next to the pulpwood removals, never merged with them, per the double counting rule below.

These factors carry the country fibre balance shown on country pages: estimated industry demand against reported removals, derived at render time, printed as ranges, with the coverage of our capacity data stated in the same breath.

Double counting: why consumption columns are never summed

Sawmills sit upstream of pulp and board mills: a large share of the chips a pulp mill consumes began as sawlogs at a sawmill. Consider a sawmill that takes in 100 cubic metres of logs and sends 30 cubic metres of chips to a nearby pulp mill. The sawmill consumed 100, the pulp mill consumed 30, and adding them gives 130, yet only 100 cubic metres of wood ever left the forest.

The dataset therefore applies a fixed rule:

• Wood consumption is reported separated by assortment: roundwood, chips and other byproducts, and recovered wood in distinct columns.
• Those columns are never summed into a single figure on the public site: not per facility, not per country, not per region.
• A roundwood equivalent is computed internally for balance modelling, counting each stream once at the point where the log first entered industry, but it is not published as a total.
• Where a chip flow between two named facilities is known and sourced, it is stored as a relation between the two facilities rather than as an attribute of either one, so it can be netted out instead of counted twice.

The chain is not always two steps. A pellet plant buying sawdust and an energy plant buying bark sit downstream of the same original log, so the rule is stated in terms of where the log first entered industry rather than in terms of sawmills and pulp mills specifically.

Sites and co-located facilities. Many pellet plants, and some pulp mills, stand on a sawmill's own site and press or cook that sawmill's residues. WoodComp records each production unit as its own facility row, because each has its own capacity in its own unit, its own status and its own products, but the rows share a site key (`site_id` in the dataset): the lowest facility reference on the plot. A map, a country page or a dataset filter can therefore show either facilities or places. Pellet capacity in tonnes is never added to sawn timber capacity in cubic metres or to pulp capacity in tonnes, on the same site or anywhere else, and the input materials of a pellet plant are recorded with their origin (on-site residues, purchased or mixed) rather than as a wood consumption figure that could be summed with the sawmill's own. A pellet plant that sits beside a sawmill on another plot, or buys its sawdust from a neighbour, is a free-standing facility with its own site key: adjacency is not co-location.

A missing aggregate is a gap. A wrong aggregate is a false statement, and it propagates into every analysis built on it.

Sources, confidence and certification

Every material fact links to at least one source, graded by type: company disclosures and environmental statements first, then regulator filings, then credible trade press. From source count, source quality and data completeness we compute a confidence score per facility. Certification fields (FSC, PEFC) are strictly evidence-based: true and false both require a source, and where the evidence names only the company but not the site, the field stays unknown rather than assumed.

Verification

High-impact changes (status changes, capacity corrections, deletions, merges) go through independent second review: a separate research pass attempts to refute the claim against primary sources before it is written. Findings that cannot be confirmed stay flagged as open questions instead of entering the dataset as facts. The review queue is part of the dataset, not a hidden backlog.

Production. Where a facility publishes its actual production, the page shows the latest reported year next to capacity, tagged with the year and the kind of source ("2024, environmental report"). Production and capacity are different quantities and are never mixed or derived from one another: no utilisation rates, no estimates. Only the most recent year is shown; earlier years are collected but not published. European figures come mostly from EMAS statements and per-site environmental reports, which report actual annual production per facility.

Update cadence. A nightly sweep re-checks about twenty facilities against sources, field by field, which works through the whole register in about a quarter; it has run since 24 August 2026. Proposed changes from the sweep are reviewed by a person before anything is written. Corrections from readers and operators land continuously as they are verified. Each source listed on a facility page shows when it was last checked, and the timeline shows dated capacity observations as they are recorded.

Measuring our own error rate

Most databases can tell you about the errors they caught. They cannot tell you about the ones they missed. We try to measure exactly that: at regular points we pull a random sample from data that has already passed every check, and ask an independent reviewer to prove it wrong.

This works. One such sample found wrong locations in 2 of 16 approved coordinates, so every coordinate in that batch was re-checked and satellite verification was added to the pipeline; the same measurement on the satellite-verified results found no errors. Where sampling has shown a whole class of evidence to be unreliable, that class was dropped from the dataset rather than patched row by row.

The first capacity sample, in August 2026, measured the hardest field we carry. We drew 60 published capacity figures at random from records that had already passed every check, and re-checked each against a primary source. Of the 56 that could be verified, 18 were wrong: 32 percent, with a 95 percent confidence interval of 21 to 45 percent. Two thirds were wrong values, usually a figure that predates an expansion or a rebuild. The rest were wrong in a way that is harder to spot, where the number itself was accurate but measured something else, most often a paper or board capacity recorded as pulp. Pulp mills scored worse than sawmills, 45 percent against 26 percent, because a pulp and paper site publishes several different capacity figures and they are easy to mix up. Capacity is the field to read with most caution, and the one we check hardest.

That sample drove two full reviews. Every published pulp mill capacity was re-verified against primary sources in August 2026, and 61 records were corrected or cleared. Every published sawmill capacity followed in September 2026: 348 figures were re-verified in four regional passes (the Nordics, the German-speaking countries, North America and the rest of the world), on top of the figures the nightly sweep had already checked. Of the 305 that could be verified, 67 were wrong, 22 percent; 20 of those measured the wrong quantity, most often a log intake recorded as sawn timber or the reverse, and 12 turned out to be group totals, tonnage figures or finished-product volumes with no per-mill capacity behind them, so the field was cleared. The rest of the world scored worst at 29 percent and had the most figures that could not be verified at all, because many sources give a volume of "wood" without saying whether it is logs in or timber out. After each pass a second, independent reviewer tried to prove a random sample of the approved figures wrong, and every proposed correction was challenged the same way before it was written: 3 of 59 approved figures fell, 5 percent with a 95 percent confidence interval of 2 to 14 percent, and 6 of 64 corrections fell, 9 percent with an interval of 4 to 19 percent. Those are the error rates of our review, as distinct from the error rate of the data before it. In total 60 sawmill records were corrected or cleared.

The sawmill review also fixed a definition. Where a company publishes its own capacity figure for a mill, that figure is used. Where it does not, the capacity is the highest actual annual volume reported in the last three years, from a source we have read, and never a planned figure or a design figure. Where a register publishes an estimated capacity per mill and year, as the British Columbia mill list does, the latest edition applies.

The pellet plants, added in October 2026, were measured the same way as they were built, region by region. Every row rests on a register or an industry table and was then checked against the operator's own figure by one agent and challenged by another: Nordic and Baltic plants 0 errors in a sample of 15 approved figures, German-speaking and Central European plants 0 in 15 (2 uncertain), the rest of Europe 1 in 14, the World of Pellets supplement 1 in 10 approved and 2 in 8 corrections, North America 0 in 13, Asia, Oceania, Latin America and South Africa 2 in 13 (both on what the figure measured, a production volume and a torrefied agricultural pellet, neither a wrong number). China, Russia and Belarus, added last, were not measured: the Chinese rows rest on one industry table and no company source could be read, and the Russian and Belarusian rows rest on a certificate register with no source after 2022, so they are published with a flag rather than with an error rate. The locations were verified in satellite imagery with an independent control pass on every site-level hit, which downgraded about a third of the first-pass hits to town level; those downgrades are recorded, not hidden.

These measurements are snapshots of specific work, not a running accuracy score. We publish them because a data source that only reports the errors it caught is telling you half the story; to our knowledge, no closed data vendor in this industry measures or publishes its own error rate at all.

What WoodComp is not

WoodComp records what exists: facilities, owners, capacities, status, sources. It does not model what production costs, and it does not carry machine-level data: no paper machine registers, no cost curves, no benchmarking of individual facilities' economics. Those are different products, sold by closed vendors on different terms. The boundary is deliberate, and it is why everything here can be open.

Confidence score

Every facility page shows a data confidence score from 0 to 100, computed at render time from the same sources listed on the page. It grades the paper trail behind the record, not the facility: a well-run mill with thin public documentation scores lower than a mediocre one that publishes everything.

Scores of 75 and above read as High, 50 to 74 as Medium, below 50 as Low. A record that carries a capacity figure nobody has verified is capped at 74: it cannot read as High on documentation alone, because the capacity sprint showed that a well-documented record can still carry the wrong number. Where a check exists, the facility page says so in the short provenance line under the figure ("Permit ceiling · verified" with the date); the source itself is listed under Sources. In rare cases an editor sets the score manually, for example after a merge; the same scale applies.

Permanent identifiers

Every facility in the dataset carries a WoodComp ID: the number shown on its page and used as the join key throughout the database. The ID is permanent. It is never reused for another facility, and it survives renames, ownership changes and page moves. It is written with five digits, so facility 5 appears as 00005, and leading zeros are cosmetic: 5 and 00005 are the same identifier and both resolve.

When two records turn out to describe the same physical facility, the records are merged and the retired ID is not deleted: it becomes a redirect to the surviving record, and the old page address does the same. A reference to a merged ID therefore keeps resolving to the right facility instead of going dead.

The stable way to link to a facility is https://woodcomp.world/id/ followed by the ID, which redirects to the facility's current page. When citing a specific facility, cite the ID alongside the name: names change, the ID does not.

Licence and citation

The WoodComp dataset and this knowledge base are published under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0). You may copy, redistribute, adapt and build on the material, including commercially, provided you give attribution and license your adaptations under the same terms.

The bulk CSV download is closed. Every figure remains free to read and reuse on the pages themselves under the licence above; for bulk access, contact us.

How to cite: WoodComp (2026). *WoodComp facility database*, version 2026-11. Accessed the date you retrieve it. https://woodcomp.world

Third-party sources cited on these pages are not covered by our licence. They are cited with attribution and remain subject to their own terms.

Corrections

If you find an error, we want to know. Every correction with a source is reviewed and credited in the facility's change history. Contact