Predict the Harvest from Satellite to Set Purchase Contracts
An olive oil cooperative buys olives from 4,000 farmers and sells the oil with forward contracts signed in spring. If it overestimates the harvest, it cannot fulfil the contracts; if it underestimates, it sells late and cheap. Today the forecast is made with technician visits to a sample of farms. Free satellite images are available every 5 days for all plots.
Where the money comes from
What to do now
- 1Validate the model with the last 2 seasons.
- 2Commit to contracts only the low percentile of the forecast.
- 3Update the forecast every month and sell in tranches.
For your committee
One page with the decision, the impact, the risks, the steps and the KPI to track.
The print dialog opens: choose "Save as PDF".
Watch out: more volume is not always more profit
Committing the entire forecast volume to contracts to lock in the price: if the harvest turns out smaller, you have to buy oil externally at a higher price. Commit in tranches according to the uncertainty.
KPI for the committee
Production forecast error and average selling price.
Detail for your data team
Proposed technique: Neural Networks (Deep Learning / CNN)
Why: The signal is in multispectral images that change over time: canopy density, vigour, water stress. Convolutional neural networks (combined with temporal layers) extract those patterns from millions of pixels and relate them to the historical production of each plot.
Cost of getting it wrong: Overestimating forces buying oil externally to meet contracts (expensive); underestimating forces selling the surplus late and at a worse price.
Data needed: Multispectral satellite images of 30,000 plots every 5 days (vegetation health, moisture).; Actual production of each plot over the last 8 seasons.; Weather: rain, heatwaves, frosts at flowering.; Plot data: age of the grove, irrigated or rain-fed, variety.