2023: UK wheat wet harvest, 6 of 6 high-confidence regions correct

Crop year 2022/23 · published 2026-05-02

In one paragraph
The 2022/23 UK wheat year was a textbook compound stress at flowering and ripening case. The walk-forward ensemble called 6 of 6 high-confidence regions correctly as below-average, with the strongest signal in the eastern wheat belt. UK average yield came in at 6.7 t/ha, well below trend.

By the numbers

-0.33
Predicted anomaly (t/ha), below-average
-0.77
Actual anomaly (t/ha), below-average
6 of 6
Tradeable regions correct

Walk-forward, the model was trained only on years before this one. Every figure is a plotted point on the track-record scatter.

The macro story

Spring 2023 was relatively dry across the UK. Crops looked promising through May. Then July and August delivered unusually wet conditions, cool, overcast, with persistent rainfall that prevented timely harvest across most of England.

The wet weather hit at two critical stages simultaneously: late flowering (when kernel set was already underway) and ripening (when grain fill was supposed to complete). Disease pressure, particularly fusarium and septoria, built rapidly under prolonged leaf wetness. Combine availability lagged demand because every farm was waiting for the same dry windows.

What the model said

The walk-forward ensemble (trained on 1999–2021, with 2022 data missing from DEFRA so excluded from training) flagged compound stress across flowering and ripening stages as the wet July developed. Six regions hit high-confidence below-average calls; the eastern wheat belt showed the strongest signal because the rainfall was concentrated there.

The model's threshold-rules component, the deterministic third of the ensemble, fired specifically on the flowering-stage rainfall anomaly, which is one of the strongest historical predictors. All three components agreed, so the consensus filter let the call through as high-confidence.

What actually happened

Why it matters for the methodology

2023 demonstrated the model's ability to fire on a "second-half" stress event, one that develops after the May checkpoint. The May call was already mildly bearish based on the dry spring; the July deterioration intensified it through to the August harvest. The system's daily refresh catches these shifts in close to real time, with the compound stress score updating each morning as new ERA5 weather data lands.

The 6/6 high-confidence rate was the highest hit-rate result in the backtest other than 2019's 8/9. Together they show the model's accuracy is concentrated in the years where it actually emits high-confidence calls, the low-confidence and "no tradeable" years are appropriately excluded by the consensus filter rather than flooding the published record with weak signals.

The economic benefit, by user type

What the call was worth, by user type, the same signal, three different decisions:

Grain merchant. A months-early, regionally-resolved below-average call is time to secure supply and set basis before the market reprices the shortfall. On a 100,000 t book, a ~0.8 t/ha regional shortfall is tens of thousands of tonnes to re-source, moving early on that is the difference between paying the pre- and post-repricing price.
Crop insurer. A below-average signal a season ahead lets you reserve earlier and re-weight exposure away from the regions heading into compound stress, before the loss crystallises at harvest. One quarter of reserving lead time on a bad year is materially valuable.
Commodity desk. A leading (not coincident) directional signal with a published hit rate is a sized edge: hold or add to a long feed-wheat view through the deterioration, or use it as an independent confirmation layer on your own fundamentals.

Illustrative, the figures show the magnitude and direction of the decision, not a guaranteed return. The model is directional and probabilistic (62.3% walk-forward). See who this is for for per-segment detail.

Who this is relevant to

CropIntel's signal, and the underlying system (available to acquire), is relevant across the UK and European arable value chain:

Representative firms by segment, illustrating breadth, not claimed clients or relationships. If your firm is on this map and the track record is interesting, start a conversation.

Related: All case studies · Methodology · Track Record · 2019 disaster year · 2014 bumper year