China Crop-Weather Risk Monitor — SoftSignal Research

What this is — and isn't. A phenology-gated monitor of threshold-crossing weather events in China's two grain regions that matter to the world balance — the Northeast (NE), the corn and soybean belt, monitored at its frost-marginal far north in Heilongjiang, and the North China Plain (NCP), the winter-wheat belt across Hebei, Henan, Shandong, and northern Jiangsu and Anhui — paired with a measure of how much room the rest of the world has if China needs to buy. We do not extrapolate Chinese production, model internal stocks or reserve policy, or publish an import forecast — those are opaque at best. This states what happened in the window and what it would cost the world if China comes — two separate statements, neither a tonnage claim.

NE Frost Risk — Equal-Weighted Monitored Points

Share of Monitored Points Under Elevated Frost Risk · HEADLINE

Equal-weighted % of the 5 monitored points at/above each class, by day (no verified per-point production-share data exists for these zone-derived grid points).

Frost-Risk Calendar — current season · None → Severe

Cell = daily risk class from the estimated leaf minimum. Rows: Zone VI (marginal) above Zone III (contrast).

Monitored Points · ZONE-DERIVED

3 Zone VI (marginal/coldest, computed from accumulated temperature) + 2 Zone III (contrast).

NCP Dry-Hot-Wind Risk — Wheat-Area-Weighted Stations

Share of Wheat Belt Under Dry-Hot-Wind Risk · HEADLINE

Wheat-area-weighted (FAS shares) % of the 5 monitored stations tripping Light+ / Severe, by day. Station-count based, never averaged.

Dry-Hot-Wind Calendar — current season · None / Light / Severe

GB/T-consistent: Tmax, RH, and wind speed at 14:00 Beijing Time, all thresholds AND'd.

Monitored Stations · 5 PROVINCES

Representative city coordinates (starting proxy pending confirmed CMA station locations).

NCP Harvest-Window Rain — Quality Risk

A quality signal, not a yield signal — the 2023 Henan flagship case: prolonged rain during wheat harvest sprouted 4.8-10 Mt, downgraded to feed grade, wheat imports rose months later. Two-sided: bearish domestic corn demand (feed wheat displaces it), bullish wheat import needs. Each province has its own harvest window (harvest sweeps north over ~3 weeks) — see the legend for exact dates; a province only shows data during its own window.

Wheat-Area-Weighted Share at Elevated+ Harvest-Rain Risk · HEADLINE

% of the 5 monitored provinces' wheat area (FAS-weighted) at/above Watch and Elevated, by day, within each province's own window.

Area-Fraction by Province — current season · % OF PROVINCE WHEAT AREA

Fraction of each province's area with a 5-day cumulative rain sum >15mm, thresholded at the cell then aggregated (never averaged first). Each line spans only its own harvest window — the staggered starts are the harvest sweeping north.

Persistence by Season · DAYS AT ELEVATED+

Days inside each province's own window at Elevated+ (≥30% of area). Duration, not peak — peak area-fraction saturates at 100% in most years, so it separates nothing; persistence is what the sprouting literature keys on.

Monitored Provinces · REAL FAO/GAUL BOUNDARIES

Anhui/Jiangsu clipped north of the Huai River (~33°N, the actual wheat/rice climatic boundary) — not the whole province.

Ex-China / Major-Exporter STU Amplifier

Tells you what a Chinese import call would cost the rest of the world — not whether China will come (policy, opaque, un-modelable). Chinese buying itself lowers ex-China stocks, so using this to forecast Chinese demand would regress an effect on its own cause.

Corn — World Less China STU

Wheat — World Less China STU

Engine: NE frost — per-point coldest-hour-of-night extraction from ERA5-Land (via native Earth Engine), an estimated leaf/canopy minimum (radiative offset for clear/calm/dry nights, floored at the dewpoint), classified None/Watch/Moderate/High/Severe. NCP dry-hot-wind — GB/T-consistent classification (Tmax, relative humidity derived from dewpoint via Magnus/Tetens, wind speed) at the fixed 14:00 Beijing Time reading hour, station-count aggregation. NCP harvest-window rain — per-province rolling 5-day cumulative precipitation, thresholded at the cell (15mm) then aggregated to area-fraction (never averaged first), each province gated to its own harvest-risk window. All three classifiers are ported/proxy-calibrated against historical event backtests (`backtest_china_events.py`) — not yet independently validated for a full season. Generated 2026-08-02.

Non-goals: no Chinese production estimate, no internal-stocks or reserve-balance model, no import forecast, no generic drought/NDVI monitoring (saturated, wrong instrument for an irrigated NCP), no NE soybean leg (~20 Mt domestic vs 105-112 Mt imports = noise against the import book).

Data: ECMWF ERA5-Land via native Google Earth Engine (not BigQuery, not Open-Meteo) — see `kb/china-crop-weather-methodology.md`. STU amplifier from USDA WASDE (`data/wasde/*.csv`), reusing `generate_wasde_grains.py`'s stocks-to-use helpers unchanged.