Ceremis × Hyperplan — From reactive collection planning to silo-level confidence, weeks before harvest
Ceremis is a cooperative group in northern France that brings together five regional grain collection organizations (OS). Together, they manage a dense network of grain silos across some of France’s most productive cereal-growing departments — from the Aisne and the Oise to the Nord, Pas-de-Calais, Somme, and Seine-et-Marne.
Like most French cooperatives, Ceremis faces the same structural challenge every season: submitting collection forecasts — the prévisionnel de collecte — before the data they need actually exists.
Every spring, each OS within Ceremis must commit to a collection forecast by crop and by silo. These forecasts drive storage allocation, staffing, transportation planning, and procurement negotiations. They are high-stakes numbers that ripple across the entire logistics chain.
The problem is timing. Official statistics from Agreste (the French Ministry of Agriculture) arrive late and at a coarse departmental level. Field visits by technical advisors provide local color but cover a fraction of the territory. And the cooperative’s own internal surveys — built on spreadsheets and phone calls — are slow to aggregate and hard to compare across regions.
The result: forecasts built on last year’s averages, gut feel, and late-season corrections. When volumes concentrate at three silos instead of six, or peak intake arrives two weeks early, the cooperative absorbs the cost in emergency trucking, storage congestion, and margin erosion.
But the cost isn’t only operational. Without early, reliable volume signals, the cooperative also lacks visibility into how much storage capacity it will actually need — and that affects how it manages the grain it’s still holding from the previous harvest.
Ceremis began working with Hyperplan to bring satellite-derived crop intelligence directly into its collection planning workflow. The collaboration is structured around monthly touchpoints between Hyperplan and the OS teams, with data flowing continuously in between.
Hyperplan maps every declared parcel in the Ceremis territory and assigns each to the nearest silo within a defined radius (typically 10–15 km, set by each OS). This produces a table of estimated surfaces by crop, by silo, updated throughout the season as satellite classification improves.
Starting mid-April, Hyperplan overlays yield models based on vegetation indices, cumulative thermal time, and meteorological variables. These estimates converge progressively — stable on wheat by early June, on maize by early August — giving OS teams a view on expected volumes per silo before official statistics are published.
A dedicated module shows how Hyperplan’s estimates have evolved week by week across the season. Teams can see whether surfaces in their territory are trending up or down relative to the previous campaign, and when the estimates have stabilized.
Before in-season yield data is available, OS teams can build custom tables showing historical yields by silo across multiple years, creating an immediate baseline for their preliminary forecasts.
The workflow Hyperplan recommends — and that Ceremis has adopted — follows a clear logic:
Pull estimated volumes by silo from Hyperplan (surfaces × yields, once available).
Calculate historical market share per silo by comparing past Hyperplan volumes against actual collection data.
Apply that market share ratio to in-season Hyperplan volumes for the current campaign.
This produces a silo-level collection forecast that is grounded in satellite-observed field reality — not extrapolated from last year’s averages.
Hugues Desmet, who leads collection planning at one of the Ceremis OS, has used this method for multiple campaigns. He tracks hundreds of ground-truth data points from his own territory and uses Hyperplan outputs as the primary input into his forecast. He also feeds Hyperplan’s silo-level estimates into a downstream flow optimization tool, creating a two-stage pipeline from satellite signals to operational logistics decisions.
The most concrete impact came during a season where Hyperplan’s in-season yield estimates pointed to a lower-than-expected harvest. That signal arrived weeks before official confirmations.
Because the cooperative could see early that incoming volumes would be below average, it made a deliberate decision: slow down the sale of grain still held from the previous campaign. The silos wouldn’t need to be fully emptied to make room for a record intake — so there was no urgency to sell into a soft market.
That single decision — enabled by earlier, more reliable volume visibility — translated into an estimated impact of approximately 1€ per ton. On the scale of a regional cooperative’s collection, that is a material margin improvement driven not by a new process, but by better-timed information.
Before Hyperplan, Ceremis teams built their prévisionnel de collecte on Agreste department-level statistics (which arrive late and don’t break down below department level) and internal surveys (which are patchy and slow). With Hyperplan, they have parcel-level, silo-level views that update as the season progresses.
Hyperplan’s classification models, enhanced with radar-based sowing detection that works through cloud cover, now deliver reliable total winter cereal acreage estimates by late March on the Ceremis territory. Crop-by-crop distinction (wheat vs. barley, for instance) continues to refine through June, but the total envelope — the number that matters most for early logistics planning — is available months before any official source.
With first yield estimates available from mid-April, Ceremis teams have a view on expected volumes per silo roughly four weeks before peak collection planning deadlines. The thermal-time-based yield model improves accuracy in atypical seasons — years where crop development runs early or late relative to the calendar.
Ceremis contributes ground-truth data — parcel-level crop declarations and yield records from farm management tools — that Hyperplan uses to recalibrate models on their specific territory. This creates a virtuous cycle: better local accuracy leads to higher trust, which leads to more data sharing, which drives further accuracy gains.
During one session, a member of one of the OS teams tested Hyperplan’s yield estimates against his own family farm records and confirmed the wheat estimate was within one quintal per hectare on one parcel. On an adjacent parcel, the estimate was less accurate — which led to a productive conversation about how local ground-truth contributions can close those gaps.
During working sessions, the Ceremis OS teams have been direct about what they value — and where they push for more.
Hugues Desmet uses Hyperplan as the foundation of his collection forecast. He builds his entire prévisionnel around Hyperplan surfaces, yield hypotheses, and historical market shares by silo. When a feature he relied on was temporarily removed from the platform, he flagged it immediately — it was central to his workflow. He described the tool in a review session as something he already uses to feed downstream logistics optimization, making Hyperplan the first link in a chain that runs from field signals to truck routing.
The coordination team uses the monthly touchpoints to challenge Hyperplan on convergence speed and crop-by-crop accuracy. Their question is always the same: when will the distinction between individual cereals be reliable enough to feed into Ceremis-wide resource declarations?
The newer OS users are going through their first full season with Hyperplan and are building their silo configurations, catchment areas, and forecast methods from scratch, supported by dedicated onboarding sessions.
The way Ceremis uses Hyperplan illustrates a broader pattern emerging among French cooperatives. Hyperplan provides the what and where — how much grain, at which silos, how the season is evolving. Downstream optimization platforms then handle the how — which trucks, which routes, which storage allocations.
This combination was also deployed at another major French cooperative, where the two layers together delivered measurable operational gains during the collection campaign. Within Ceremis, interest in this integrated approach is growing — several OS are either already connected to optimization tools or evaluating them.
The 1€/ton impact at Ceremis didn’t come from a new logistics algorithm. It came from knowing earlier what the season was going to look like — and having the confidence to act on that knowledge before the market forced the decision.
|
Organization |
Ceremis (cooperative group, 5 regional OS) |
|
Territory |
Northern France — Aisne, Oise, Nord, Pas-de-Calais, Somme, Seine-et-Marne |
|
Primary use case |
Silo-level collection forecasting (prévisionnel de collecte) |
|
Measured impact |
~1€/ton through better-timed grain sales enabled by early volume signals |
|
Crops monitored |
Wheat, barley, rapeseed, maize, triticale |
|
Hyperplan outputs |
Acreage by silo, yield estimates, acreage evolution tracking, historical yield tables |
|
Integration |
Hyperplan volume forecasts feed into flow optimization tools |
|
Engagement model |
Monthly touchpoints, in-season alerts, annual performance reviews |
|
Ground-truth loop |
Parcel-level crop corrections and yield records fed back into models |
Hyperplan helps agribusiness teams make faster, better decisions by turning in-season, field-level crop intelligence into actionable commercial and supply insights at scale. Ceremis is one of 59 organizations globally that rely on Hyperplan to plan with foresight, not hindsight.