Every month, in the Brazilian coffee belt, Fundação Procafé publishes a plant-health bulletin telling growers how much leaf rust its monitored plots are carrying. Each bulletin is a snapshot — and a snapshot can't answer the question a trader or an agronomist actually has: is this reading high, or is it just what this month always looks like? So we read the whole archive — 529 monthly bulletins, 2010–2026 — and parsed it into one tidy time series of 50,673 station-level observations. Here is what it told us.

The record nobody had read end-to-end The Record

Procafé's bulletins are a grower-alert service: a signal that pressure is present in an area, not a census of the crop. We downloaded 529 of them spanning four regions — Sul de Minas, Alto Paranaíba/Triângulo Mineiro, Alta Mogiana, and Matas de Minas — and parsed the disease, climate, and phenology tables into a single time series. As far as we can tell, this record doesn't exist anywhere else in this form. It let us do the one thing a single bulletin can't: put today's reading in the context of every month like it for sixteen years.

June 2026 was genuinely low — not just low for June The Baseline

The headline for Sul de Minas in June 2026 was 19.2% of leaves infected. Against a single month that reads "moderate." Against sixteen prior Junes it reads very differently: the June norm is ~44%, and 19.2% ranks third-lowest in seventeen years (z = −1.2; 88% of prior Junes ran higher). Only the drought Junes of 2021 and 2024 were lower.

The seasonal cycle the archive reveals is the reason it matters: rust builds from January, peaks in May–June, then collapses to near zero through the dry season and harvest before rebuilding. A June number means nothing without June's own climatology behind it.

Our weather model checks out — on temperature The Check

Before trusting any comparison of modeled rust risk against observed rust, we had to know our own weather was right. Procafé publishes each station's coordinates and its monthly temperature and rainfall — so we compared our ERA5 reanalysis for the grid cell containing the Varginha experimental station against Procafé's own gauge, 196 months deep.

• Temperature: excellent. Monthly correlation r = 0.98, bias −0.6 °C. Our gridded temperature tracks the station almost perfectly.
• Rainfall: good but noisy. Correlation r = 0.91, but individual wet months scatter by ±30–45 mm, worst on convective downpours a coarse reanalysis smooths away.

This matters because the 2026 rust split tracked temperature, not rainfall — and temperature is exactly the variable our model gets right. It also flags a real limit: a rain-driven wetness input can under-call regions where leaf wetness comes from dew and fog rather than rain (the humid eastern hills like Matas de Minas). We'd rather name that than paper over it.

Warmer regions carry more rust — but only on the average The Read

Line the three regions up by temperature and they sit in a fixed ladder: Sul de Minas (coolest, ~16.7 °C in June) below Triângulo (17.9 °C) below Alta Mogiana (warmest, 18.9 °C). Their long-run mean rust follows the same order — 43% → 51% → 54%. Warmer region, more rust.

But that's an average, not a rule. In any single June the ranking reshuffles as weather varies region by region; the pooled temperature-to-rust correlation across region-years is only +0.21, and Sul de Minas was the lowest-rust region in just 6 of 10 shared Junes. The honest lesson: comparing rust levels across regions is the weakest, most-confounded test — management, altitude, and a copper protectant spray every station receives all differ. The trustworthy read is a region against its own history.

Leafing is next year's crop, showing now The Lead

The bulletins also report enfolhamento — leaf retention. It's a quieter number than rust and a more forward-looking one: the canopy a tree carries into flowering previews the 2027 crop, long before any production estimate. Across the archive, rust and leafing move opposite each other — rust strips the canopy.

2026 shows it starkly. Alta Mogiana, under 100% rust, fell to 16% leaf retention — the lowest in its record; Triângulo dropped to 30%; cool, barely-touched Sul de Minas held 56%. That gap is productive capacity the Cerrado will spend the dry season trying to recover — a supply signal visible months ahead, and invisible to anyone reading only balance sheets.

What we changed — and what we're still chasing The Payoff

The point of the exercise was to make our own leaf-rust signal better, and it did. Our leaf-rust report now carries an observed-incidence ground-truth panel for the Sul de Minas belt — the realized rust the conduciveness calendar aims to lead, shown as an anomaly against its own climatology and lagged the ~3–6 weeks it takes rust to appear. We lean on the temperature axis with confidence, treat rainfall as the coarser input, and the engine is now ground-truth-checked against 16 years of field observations — a claim we couldn't make a month ago.