Automatic alerts when a crop deteriorates or improves fast on satellite, when conditions stray far from normal, when our nowcast moves a lot, and when our model points to a large USDA revision. Every threshold comes from the history of the series itself, and every rule is backtested below.
RSS feed · checked every day after the 12:00 UTC run
Vegetation health 54 (anomaly +9.8 vs the normal for the week), +6.4 points of anomaly in 4 weeks. A 4-week change this large ranks at the 96th percentile of the same weeks in earlier seasons. Backtest since 2011: this signal fired in 6 seasons (2011-2025) and the yield ended above trend in 6.
Nothing matches.
Rules
Percentiles, not hand-picked levels: a crop is flagged when its reading is rare for that time of the season, compared with earlier seasons only (point in time, at least 10 of them). A rule that has fired stays quiet for 4 weeks for the same origin and direction. Satellite rules skip the first 6 weeks after sowing and, for winter crops, the dormancy weeks from December to early March.
| Alert | Source | Statistic | Fires when | Applies to |
|---|---|---|---|---|
| Crop deterioration / improvement | NOAA satellite VHI, weekly | 4-week change of the VHI anomaly, ranked against the same weeks (±2) of earlier seasons | bottom 10% with the crop in the bottom 30% for the week (deterioration), or top 10% with the crop in the top 30% (improvement); strong at 5% | every origin in season except China (irrigated; no skill in the backtest) |
| Conditions far from normal | NOAA satellite VHI, weekly | percentile of this week's VHI among the same weeks (±2) of earlier seasons | ≤ 10% or ≥ 90% for 2 weeks in a row; strong at 5% / 95% | every origin in season except China (EU and France included) |
| Large nowcast move | Live nowcast ledger | % change of our yield p50 since the previous publication for the season | above the 80% percentile of absolute changes between consecutive forecast days in the origin's backtest (at least 1.5 %); strong above the 95% | every origin with a live nowcast |
| USDA revision expected | Live nowcast ledger, USDA PSD and WASDE | our yield p50 against USDA's yield in force (%) | above the 80% percentile of USDA's absolute monthly production revisions for the country (and at least 2 %); major above the 90% percentile with USDA outside our 80% interval | origins with a WASDE line |
Thresholds
Move needed to trigger the two model-based alerts. The USDA threshold is the 80th / 90th percentile of USDA's own month-to-month production revisions for the country since 2002 (minimum 2%).
| Origin | Nowcast move (notable / strong) | USDA revision expected (notable / major) |
|---|---|---|
| United States | 3.4% / 4.8% | no WASDE line |
| France | 3.8% / 6.7% | no WASDE line |
| Russia | 4.2% / 11.1% | 2.3% / 4.8% (48% of reports move it) |
| Ukraine | 4.5% / 6.9% | 3.8% / 6.9% (39% of reports move it) |
| Argentina | 5.4% / 10.6% | 3.8% / 7.3% (40% of reports move it) |
| Australia | 7.0% / 16.1% | 4.5% / 7.7% (40% of reports move it) |
| Canada | 6.1% / 15.2% | 3.3% / 5.6% (38% of reports move it) |
| European Union | 2.5% / 6.3% | 2.0% / 2.0% (73% of reports move it) |
| China | 1.5% / 1.5% | 2.0% / 2.0% (17% of reports move it) |
| India | 1.9% / 6.8% | 2.0% / 2.0% (20% of reports move it) |
| Pakistan | 1.5% / 1.5% | no WASDE line |
| Turkey | 2.0% / 3.0% | no WASDE line |
| Kazakhstan | 9.1% / 22.4% | 4.2% / 8.0% (31% of reports move it) |
Backtest
Every satellite alert the rules would have fired in 2011–2025, with the same point-in-time thresholds, against the season's final yield versus trend. Read the hit rate against the base rate: the signals are worth most for rain-fed crops (Argentina, Australia, India, Canada, Ukraine) and are close to a coin toss for the EU and France, where every alert says so in its own text. China is left out: its signals pointed the wrong way in most seasons. The level condition of the deterioration / improvement rule was chosen on this same period.
| Signal | Alerts per origin-season | Season ended on the flagged side of trend | Base rate | Mean final yield vs trend |
|---|---|---|---|---|
| Crop deteriorating | 0.41 | 56% (62 seasons) | 45% | -3.4% |
| Crop improving | 0.48 | 69% (71 seasons) | 55% | +5.3% |
| Conditions well below normal | 0.35 | 55% (55 seasons) | 45% | -3.5% |
| Conditions well above normal | 0.35 | 67% (52 seasons) | 55% | +5.6% |
All satellite rules together, one vote per origin, season, rule and direction.
| Origin | Signals | Season ended on the flagged side |
|---|---|---|
| Argentina | 10 | 90% |
| Australia | 17 | 88% |
| India | 25 | 80% |
| Canada | 14 | 79% |
| Ukraine | 19 | 68% |
| Turkey | 28 | 64% |
| Russia | 15 | 60% |
| United States | 25 | 56% |
| Kazakhstan | 13 | 54% |
| Pakistan | 20 | 50% |
| European Union | 27 | 44% |
| France | 27 | 41% |
Scored
Each "USDA revision expected" alert is scored at the first WASDE after it: did USDA move the country's production in the direction we flagged?
No "USDA revision expected" alert yet.