Yield forecasts with reasons
A forecast per field, with a range and the factors behind it.
- DATA
- Sentinel-2, weather, soil, StatCan
- OUTPUTS
- Forecast, range, reasons per field
- TYPICAL TIME
- Mid-season onward

What you can do with it
- 01Plan storage, sales and logistics with a range, not a single number
- 02See which factors raise or lower each field’s forecast
- 03Check the forecast against what you know about the field
- 04Compare the forecast with the regional baseline
METHOD
How it is made
Every number names its data and its limits.
- DATA USED
- Season crop-health record per field, daily weather (Open-Meteo), soil properties (SoilGrids), regional yields (Statistics Canada).
- METHOD OR MODEL
- Gradient-boosted model trained on past Ontario seasons. Each forecast comes with its factor contributions (SHAP) in t/ha and a 90% range.
- ACCURACY AS MEASURED
- R² 0.89 on seasons the model never saw, measured on the AgriLens demo farm. Your fields: measured on your data during discovery.
- WHAT THIS DOES NOT TELL YOU
- The forecast reflects conditions seen so far; a late storm or early frost is not in it. Small or unusual fields have wider ranges.
- LANGUAGES
- English and French

WHAT YOU RECEIVE
Maps and data in your formats
- Forecast table (CSV)
- Factor chart per field (PDF)
- GeoPackage with forecasts per field
- Dashboard (AgriLens)
WHAT WE NEED FROM YOU
- Field outlines and crop per field
- Past yields per field, if available (improves the model)
- Your harvest timing
WHAT WE PREPARE
- Season record per field
- Model training and testing on held-out seasons
- Factor explanations and ranges
- Report and dashboard
Steps and time
- Discovery2–4 weeksModel tested on your past seasons, accuracy reported
- Build3–4 weeksForecast pipeline and explanations for your fields
- RunEach seasonForecasts updated through the season
DISCUSS THIS SOLUTION
Tell us about your area
We reply with a first view on whether this fits, and what discovery would cover.


