The national average hides the defining feature of 2026: a crop split sharply across regions, fields and parts of the same field. This record brings the map, the weekly crop history and the public harvest evidence together—without forcing unlike sources into artificial agreement.
±16,000 ha · 90% interval
200,161 field geometries classified as winter wheat.
11.10–12.90 Mt · 90% interval
Area multiplied by field-level yield, then aggregated.
Before comparing the numbers
A satellite field estimate, a farm survey, a crop-condition score, a variety trial and a national balance sheet are all useful. They are not interchangeable measurements of one thing.
“The aim is not to decide which number wins. It is to show what each instrument observed, where the evidence aligns, where it pulls apart and what the final statistics will test.”
The harvest split before it averaged
Spring rainfall divided England. The field map, produced without weather as an input, shows the same broad north–south pattern. That agreement is a useful external consistency check, not proof that rainfall alone caused the yield pattern.
Bars are scaled from 7.0 to 8.5 t/ha to reveal the regional spread. The full measured record — every region, with its mapped area — follows below the distribution.
Both a severe low-yield tail and a substantial high-yield tail can coexist. Field-level coverage shows where each sits and how much area it represents.
Median field yield for all ten regions on a declared 7.0–8.5 t/ha scale, so the spread is visible, coloured by the published spring-rainfall split (north ~90% of normal; south ~50%, eastern counties 33–35%). Our yields use no weather input, so the agreement is a consistency check, not attribution. Tap the chart to open it full size.
The season can be replayed
The crop reached its national canopy maximum in the third week of May, then matured rapidly. The weekly record lets a reader see when the estimate stabilised—and whether a revision came from mapped area, crop state or calibration.
Each weekly issue behind this timeline is archived with its cutoff in the dated report; the canopy record it derives from is below.
Leaf-area trajectories for 2019–2026, shaded by each year’s final yield. 2026 peaked fourth-highest in late May, then declined steeply — the dates the timeline marks are visible in the curve itself. Darker lines are higher-yielding years; the tallest canopies are not simply the best harvests. Tap the chart to open it full size.
Historical association with Defra all-wheat yield: r = 0.90, n = 7 settled seasons.
Historical association with Defra all-wheat yield: r = 0.89, n = 7 settled seasons.
Late-season leaf area fell to 32% of peak—the third-fastest decline in the eight-season record. It remains an unresolved stress or early-maturity signal, not a one-way adjustment.
Season-integrated canopy against Defra final all-wheat yield, 2019–2025, with 2026 read off each fit. These checks share imagery, masks and processing with the main system, so they corroborate rather than independently validate. Tap the chart to open it full size.
Season-integrated LAI and NDVI use fewer modelling assumptions than the main yield system. They do share Sentinel-2 observations, crop masks and processing, so they are corroborative rather than fully independent validations.
England’s July 2026 rainfall was only 6.5 mm—10% of normal—so the wet-harvest mechanism seen in 2023 was less likely. Post-cutoff loss is still possible; it stays a disclosed risk here, and no adjustment is applied for it — none was pre-specified and validated out of sample.
From observation to supply
Satellite reflectance is observed. Crop classification, leaf area, yield and uncertainty are derived or modelled. Showing those steps makes the result easier to interrogate and harder to overstate.
Our raw winter-wheat series follows the year-to-year movement in Defra’s final all-wheat series across 2019–2025.
The targets differ: Great Britain winter wheat versus United Kingdom all wheat. This supports year-to-year behaviour; it is not a complete accuracy assessment.
The raw GB winter-wheat series (band: 90% interval) against Defra final UK all-wheat. The level gap of about +0.35 t/ha is expected from the differing targets; the co-movement is what validates. In 2023 heavy July rain cut what reached the barn — adjusting that year for rain alone brings it to ≈8.5, keeping the usual premium over the 8.1 final. Tap the chart to open it full size.
A set of instruments—not a scoreboard
The most important redesign is to replace the single ranked number line. The matrix below lets agreement and disagreement remain useful without implying that every source estimates the same crop, geography, date or statistical quantity.
200,161 classified fields; yield from imagery to 1 August, mapping to 17 August; outputs from field to national level.
1.558 Mha · 12.00 Mt · 90% intervals published
Needs an explicit bridge before comparison with UK all-wheat finals.
Survey yields are nationalised using AHDB’s published method.
85% harvested · range 3.9–11.5 · data to 10 August
60 farms and about 38,500 ha across all surveyed crops; no published sampling interval.
Area shares classified from very poor to excellent.
good or excellent on 22 June; down from 75% at end-April
It is evidence of changing crop condition, not tonnes per hectare.
Control means and variety performance relative to multi-year trial benchmarks.
18 treated sites in the 18 August file used by the current evidence page
AHDB explicitly cautions against direct comparison; the reporting site set evolves.
Specific weight, protein and Hagberg provide context on sampled grain quality.
Group 1 specific weight · protein around 13.5% in the report
Specific weight also depends on variety, density, grain size and packing; the panel is not a national yield sample.
A monthly production estimate used in an international commodity framework.
August · revised from 13.50 Mt in July
Any yield comparison also requires a common area assumption.
The benchmark against which the frozen 2026 call can later be evaluated.
2026 final due later · our frozen call for it: 7.18, 90%: 7.01–7.36 (fitted route 7.75, 90%: 7.17–8.33) — derivation in 05
Its strength is the final national statistical estimate, not within-season spatial monitoring.
A single-axis summary of the matrix. It is a convenience, not a scoreboard: the sources answer different questions, and the intervals mean different things. Our single entry is the Defra call (ordinary-year 90%, backtested above) — the raw 7.70 winter-wheat headline is not on this axis, and the fitted-route alternative (7.75) lives in the audit record, not on the chart. Tap the chart to open it full size.
Our GB winter-wheat production (band: 90% interval) against USDA’s UK all-wheat balance-sheet figure, revised from 13.50 Mt in July to 12.00 Mt in August. The scopes differ — UK all wheat adds spring wheat and Northern Ireland — so convergence, not equality, is the observation. Tap the chart to open it full size.
Condition deteriorated through spring; harvested farm results span a very wide range; trial yields sit below their recent benchmark; the map concentrates losses in drier southern and eastern areas; and sampled milling grain often retained strong quality.
Our record estimates 7.70 t/ha for Great Britain winter wheat. AHDB’s early harvest estimate is 6.7 t/ha for wheat reported so far. The gap is material; its meaning depends on scope, sample, timing and moisture convention.
One measurement, three public yardsticks
The same measurement meets three unlike yardsticks: a farm survey read mid-harvest (6.7), the official all-wheat final still to come (our call: 7.18), and a global balance sheet (12.00 Mt). Each comparison is information — provided its basis travels with it.
Field-population model · yield from imagery to 1 August · 90% interval 7.13–8.28.
Participating farm returns · 85% harvest progress · reported range 3.9–11.5 t/ha.
AHDB adjusts survey yields above 14.5% moisture down but does not uplift yields below 14.5%. Defra has used the same asymmetric convention since 2022. Our historical target spans that convention change unless it has been independently normalised to a fixed moisture or dry-matter basis.
A common-basis reconstruction is needed before attributing a precise share of the 1.0 t/ha difference to moisture.
The interval is empirical, not assumed: replayed over the settled seasons (below), the same route lands within 0.13 t/ha of Defra in every ordinary year (MAE 0.06). It is an ordinary-year interval — the two abnormal years missed by −0.40 and +0.51, the tail risk it does not cover. The purely fitted route — raw plus a fitted rain term, regressed on Defra finals — reads 7.75 (90%: 7.17–8.33); its dry-year term rests on 2022 alone, a year the model under-read. Both routes are frozen in the audit record (06).
Level with the weak years — not a recovery. 7.18 (90%: 7.01–7.36) lands beside 2025 (7.2) and 2024 (7.3): a third consecutive weak Defra year, ahead of only 2020 at the central value and well below the seven-year average of 7.84 — yet still above the farm survey’s 6.70. On Defra’s all-wheat area (≈1.66 Mha) it implies ≈11.92 Mt of production (11.64–12.22), bracketing the 12.00 USDA moved to in August.
The caveat we publish ourselves: the record attenuates extreme years (out-of-fold slope 0.42). If 2026 truly is survey-weak, Defra can land below 7.0 — our reading cannot rule that out. Either way this call is frozen, and will be scored.
The route behind the call, applied year by year. Ordinary years land within 0.13 t/ha of the final; the two abnormal years miss in the two known ways — 2022 (dry, strong; under-read) and 2023 (post-cutoff rain losses) — which is why the 2026 interval is labelled ordinary-year. Tap the chart to open it full size.
Later northern harvest returns could change the farm-survey estimate; the final wheat area could move the production comparison; and official regional results will test whether the north–south ranking appears in a different measurement system.
They answer different questions and are based on only seven or eight historical seasons. The trial network also changes as results arrive. They remain useful technical context, but converting them into co-equal national yield numbers creates more apparent precision than the evidence supports.
USDA publishes production, not yield, so a yield comparison exists only under a stated area assumption. The ledger converts the 12.00 Mt August figure using our mapped 1.558 Mha and a 93–96% winter share of UK wheat area, giving an implied 7.06–7.43 t/ha; the production chart compares the tonnage directly, where the scopes differ only by spring wheat and Northern Ireland.
A forecast should leave a record
A reader should be able to retrieve the data cutoff, model version, scope, transformation and uncertainty method after the harvest outcome is known. The audit card below shows the minimum public record.
Make the capability understandable
Rather than listing product attributes, follow one real, anonymised field from the frozen issue: its boundary, its weekly estimate, its within-field variation and the caution each layer deserves.
10 m yield · 17 Aug issue · same 4–10 t/ha scale as the live map
A real parcel from the frozen 17 August issue — 25.7 ha in Lincolnshire, East Midlands, coloured exactly as the live map. The estimate is a model output, not a combine record; the shaded band is the model’s pixel-level 5–95% range averaged over the field.
What this record adds
The value becomes visible in the evidence itself: where the national total came from; how it changed through the season; why public sources do or do not line up; what remains uncertain; and whether the frozen call survives the final result. Two numbers leave this page: 7.70 t/ha for the winter wheat we measure, and 7.18 (90%: 7.01–7.36) for the all-wheat figure we expect Defra to publish.