Crops: how these numbers are made

The crops channel answers one question, in 165 countries and 3,273 sub-national crop regions reporting this dekad, of 3,430 that ASAP defines: how does this growing season compare with the same point in that same place's own 26 years?

It does not rank countries against each other, and it does not say what caused anything. Both of those are deliberate, and both are explained below.

This page is written for someone deciding whether to cite us. It states what the numbers are, what they are not, and which of them we decline to make a claim about even when the arithmetic would allow one.

The source

Every figure comes from the Joint Research Centre's ASAP system (Anomaly hot Spots of Agricultural Production), the European Commission's operational agricultural monitoring service. We fetch its published indicators and re-express them; we do not run a crop model and we do not produce our own remote sensing.

ASAP publishes on a dekad: 36 times a year, on the 1st, 11th and 21st of each month, each covering the following ten days except the 21st which runs to month end.

We restrict every pull to ASAP's "crop during growing cycle" class. A region that is out of season is not measured rather than measured as zero, which is why the number of reporting regions changes through the year.

What is measured

Six instruments, of which five are reporting as of the current dekad. Each carries its own unit, its own direction, and its own observation window.

The order above is the order they appear on every page, shortest observation window first, because a shorter window can move sooner. That order is emitted as data rather than chosen by a renderer, so it cannot differ between pages.

Vegetation, cumulative adds the season up, so one bad dekad barely moves it. It is structurally the last of the six to move and reads calmest exactly when a fast deterioration is under way, which is both the reason it is shown and the reason it is excluded from the ordering key.

It is the instrument most often read as a proxy for outcome, and that is a use rather than a property. Every one of these instruments measures greenness, or water, or temperature. Any friendlier name for one of them, "the harvest measure", "a drought index", "crop damage", is a claim about what the measurement means rather than a description of what it is. This page tries not to make those, and where our own pages have, we have been correcting them.

The comparison

Every claim on this channel has the same shape: one unit against its own history, at the same dekad of the year.

• The baseline is 2001 to 2025, 25 prior observations, at the same dekad. Never a rolling window.
• The current year is not in the baseline, so a z-score has no (n-1)/√n ceiling. A rank of "N of 26" counts the 25 baseline years plus the current one.
• Ranks use competition ranking, so ties are stated rather than broken arbitrarily.

A region is compared against itself in August against its own Augusts. It is never compared against a neighbouring region, and never against its own spring.

Country figures are weighted by cropland area

A country value is the area-weighted mean of its regions, weighted by ASAP's own km2_crop crop mask from boundary layer gaul1_asap_v06. This changed in methodology version 2.0 and it matters: under the previous unweighted mean, England carried a quarter of the United Kingdom figure while holding 85.6% of its cropland.

Two consequences are disclosed per country rather than left implicit:

• A region with no cropland in ASAP's mask gets no vote. The count of such regions is published per country.
• Weights are applied to every year, not only the current one, so the rank is a like-for-like comparison.

Where one region dominates, the page says so by name and share, and the sentence is computed from the same weights it describes rather than written by hand.

The one number that is ours

Everything above is a re-expression of a published indicator. The composite severity figure is not. It is the mean position of a country's instruments within their own histories, and no source publishes it. It is labelled in the payload as our construction rather than an observation, and it carries three qualifiers that travel with it everywhere:

1. It is not a cross-place ordering. The value places a country against itself. The rank also saturates: 37 of the 165 published countries, about one in four, sit at rank 1, so sorting on it produces a large tied block. For how much of a country is abnormal, use the proportion of its regions at a record, which is comparable between places. For how fast, use the rate.

2. Its spread differs by place. A higher composite elsewhere can be a less unusual year than a lower one here, because the year-to-year spread of the composite is itself a property of the place. The rank is the comparable figure, not the value.

3. It is a reading of conditions to date. It carries no statement about the rest of the season.

That saturation figure is computed at build time from the published set. It read "roughly one country in seven" at 123 countries, "one in nine" at 165 on the old boundary layer, and "one in four" on the new one, and none of those was typed.

How many records to expect anyway

This is the number most easily misread, so it is published rather than buried. With 3,273 units each compared against 26 years, over a hundred will sit at a record in any given dekad by chance alone. An even spread of records would put about 126 units at their worst at any moment.

So the count only means something against its own history:

• This dekad: 128 units at their worst on record.
• Recent twelve years at this dekad: mean 86.3, ranging from 45 to 160.
• The full series runs back to 2001, whose 320 dwarfs anything since.

A count near the even-spread figure is therefore an ordinary reading, and we say so on the page rather than presenting it as an alarm.

What we decline to claim

Our own rate instrument currently shows 39 countries deteriorating fast, against a prior-year mean of 5.0 and a prior maximum of 18, with no year in the 25-year record at or above it.

We do not call that a record, and the payload states the refusal in the same field as the number. The count is not controlled for the level each fall started from, nor for drift in the series. A steep fall from a high starting level is partly regression toward the mean, and a raw count that tops the record is not a record until that is accounted for.

Where a single country's fall does survive that control, the page says so. Where it does not, the page says that too, in the same sentence as the rank, so the two cannot be separated by a layout.

What this channel never says

It never says why. The driver is unidentified for most places, including every region of the United Kingdom. Where a page notes that "vegetation here usually tracks water availability", that is a statement about a historical relationship in that region, not a claim about this year.

This matters because the instruments frequently disagree. Of the 440 regions currently at a record low on current vegetation, 66, about one in seven, have water satisfaction or rainfall sitting in its best third at the same moment. Angola is the clearest case: five regions at a record on the harvest measure while the water instruments read ordinary to favourable. A page reports where each instrument sits and leaves the causal question open.

Sequence is treated the same way. A window can manufacture the order it appears to reveal, so the per-country history covers the whole calendar year to the published dekad rather than a chosen number of dekads, and each cell carries the fullest coverage that place reached so a reader can tell genuinely low extent from thin coverage.

Freshness

The age of the newest observation cycles rather than drifting: it falls to about 9 days when a dekad lands and reaches about 19 before the next. That is a property of a dekadal source, not staleness.

Our own bound is 30 days from the dekad label, being 21 days of reader relevance past the window close plus the 9 days from label to close. Separately, an automated check compares what ASAP has published against what we hold, and fails when the source has moved and we have not. That check exists because the collection job failed silently for five days in August 2026 and every relevance-based check stayed green throughout.

A page is never built from a fetch. The build refuses outright if the six instruments do not all sit on the same dekad, so a page cannot be assembled from mixed vintages.

The boundary layer changed on 25 August 2026

Every figure on this channel published before 17 September 2026 was computed on ASAP's boundary layer v05. Everything since is on v06, and the two are not comparable at region level.

ASAP published gaul1_asap_v06 on 25 August and switched its export to it in early September. Under v06, every country id changed, and 87 of our 165 countries carry different regions: renamed, re-keyed, and in many cases subdivided to the current administrative map. France is now 13 regions rather than 22. Estonia, which was a single unit, is 15 counties; Switzerland is 25 cantons. The channel went from 2,126 sub-national units reporting to 3,273, of 3,430 defined, without a single country being added.

What stays comparable. Each region is still compared only against its own 26-year series, which ASAP supplies in full for every v06 region. A rank of "lowest of 26" on a v06 region is exactly as sound as it was on a v05 region. Country-level figures that are weighted means of regions shift with the region set but remain each country against itself.

What does not. "N of M regions at a record" changed denominator for half the channel. The dekad history restarts on v06. A card or a Note built on v05 region counts cannot be continued on v06 counts, and this page will not pretend otherwise.

How it was caught. The first pull on the new ids fetched other countries' data under the old names, 148 of 148 mislabelled, and the sign-off gate held it from rendering. The pull now refuses when ASAP's own country name for an id differs from the file it would replace. The episode is recorded as D-301.

Who is not here

Three countries that were in ASAP's v05 list are not in v06: Greenland, the French Southern and Antarctic Territories, and the Falkland Islands (Malvinas). Under v05 ASAP reported no cropland inside a growing cycle for any of them; under v06 they are simply not listed. That is the source's determination, not an exclusion of ours. Every other country in ASAP's crop export is published here.

Forty-two countries were excluded until 30 August 2026 because ASAP reported them as a single national unit. That threshold was wrong, and under v06 it is also largely moot: only 6 countries remain single-unit, because the new layer subdivides most of the 42. Every claim here is one unit against its own 26 years, and that comparison is exactly as sound for a country reported whole as for one split into eighteen. The six that remain carry no map and no count of regions at a record, and say so.

Known limits

A reading is not restricted to standing crop, and for most of the channel there is no calendar that could restrict it. ASAP publishes a crop calendar giving planting and harvest dekads per unit per crop, and it covers 74 countries. 91 of the 165 published here are not in it, including every European country we track: France, Germany, Spain, Italy, Poland, Hungary, Austria, the United Kingdom, Estonia and Romania among them. The calendar's coverage follows ASAP's food-security mandate, so it is thorough exactly where the channel's European set is absent.

For those 91, nothing in the data tells us when a season closes. France and the United Kingdom report every crop region at all 36 dekads of the year; Poland drops out only in deep winter. So a mid-August reading in northwest Europe covers whatever ASAP's crop mask holds, which at that date includes both harvested cereal ground and standing summer crops.

What this does and does not invalidate. The comparison stays like-for-like: mid-August 2026 is compared against 25 prior mid-Augusts, which contained the same mix. "The lowest of 26 observations for this point in the season" therefore stands as written. What it does not support is reading a late-summer European vegetation record as a statement about the cereal harvest, which by then is in. Where the two readings diverge, the honest phrasing is about what is standing in the fields now, not about a crop already in the barn.

The calendar was not republished for v06. It is keyed on v05 region ids, so it no longer joins to the regions on this channel. Which countries have a calendar still resolves; where inside a country a season sits does not, until ASAP reissues it.

For the 74 countries that do have a calendar, check where the dekad falls in it. Angola's maize, millet and sorghum harvest ends between dekads 15 and 19 and its planting resumes around dekad 24, so a reading at dekad 23 sits between cycles, and its cumulative measure describes the season that has just finished rather than one under way.

Soil moisture reads one dekad behind the others. It stalled at source from 1 July to the v05-to-v06 change and is publishing again under v06, one dekad behind by design. The build uses only same-dekad readings, so every country currently reads five instruments of six and the composite says which were used. This is channel-wide and is not a country-specific failure.

Coverage varies within a season. A region drops out when its growing cycle closes, so a country's denominator changes through the year. A fraction computed on a collapsed denominator is a statement about coverage wearing the clothes of a statement about extent, so the fullest coverage each place reaches is published alongside.

A missing value is never a zero. An instrument with no reading is marked absent with the reason attached at the datum: no current value, undefined at this dekad, too few comparable years, or not published for that country.

Version history

3.0. ASAP boundary layer v06, from 17 September 2026. Every country id and most region ids and names changed at source; 40 single-unit countries became multi-unit; 2,126 units became 3,273. Region-level baselines re-keyed to v06 regions, each against its own 26-year series. The dekad history restarts. Area weights from gaul1_asap_v06. Not comparable with 2.0 at region level.

2.0. Country figures area-weighted by ASAP's km2_crop. Previously an unweighted mean over regions. Affects every country value and therefore the composite, the magnitude, the rate and the ordering key. Ranks are computed on weighted values for every year so the comparison stays like for like.

1.0. Unweighted mean over a country's regions.