Platform module

Forecasting & prediction

Harvest estimates and growth-stage guidance, with their method and confidence shown.

Overview

What this module does

Harvest estimates combine the recipe baseline with trained-model output when enough suitable data is available. Otherwise the service returns a labelled fallback. Growth-stage estimates currently use rules, recorded observations and recipe timing. Prediction records can be compared with actual outcomes as those outcomes are recorded.

Prediction logs identify the model or estimation method used
Fallbacks are labelled, never disguised as model output
Accuracy is something to establish in a trial, not a figure we can show you
Estimates use available site history; local accuracy needs outcome data

Capabilities

In detail

Harvest-date prediction

Estimate days to harvest from the recipe baseline and available growing history. The model contribution depends on data quality and whether a suitable model is loaded.

Growth-stage prediction

Estimate a BBCH stage using crop age and available observations, then compare it with recipe progression. This is currently a rule-based estimate, not a trained growth-stage model.

Feature pipeline

Hourly and daily aggregates build the features the models use, including cumulative growing-degree-days, photoperiod and humidity stress.

Confidence and completeness

Both services report confidence. Harvest estimation also uses sensor coverage and data completeness to choose between model output and fallback. Confidence is an internal estimate, not a demonstrated accuracy percentage.

Cold-start behaviour

Where a model is unavailable or a batch is too new, the service returns a conservative fallback and flags it as a fallback rather than presenting a guess as a prediction.

Accuracy reconciliation

Every prediction is logged with the method that produced it, and harvests are recorded against the plan, so the two can be compared. The automatic scoring that turns those pairs into an accuracy figure is not in service yet, so treat accuracy as something to establish in a trial.

Next step

Check a forecast against your own outcomes

A 30-minute walkthrough against your zones, crops and sensors. No slides unless you ask for them.

Walkthrough led by someone who has built the system
We map your zones, crops and sensors before the call
Straight answers on what fits and what does not

Book a demo Talk to sales