How Freyr Uses Artificial Intelligence

1. What the AI Does

Freyr uses exactly one AI component: the variable-rate zoning that groups a field's NDVI pixels into management zones (k-means clustering), so you can plan where to apply fertiliser or other inputs more precisely. You review and can edit every zone before it is used. Everything else you see — crop-health values, status labels, trends, agronomic guidance — is ordinary arithmetic on measured satellite data, not AI. The AI does not make decisions for you.

2. Where the Data Comes From

Freyr relies on publicly available satellite data from the European Copernicus programme. Sentinel-2 satellites capture optical images of your fields every few days, and Sentinel-1 radar data fills gaps when clouds block the view. We also use your field boundaries, historical crop records you have entered, and weather observations from national meteorological services (FMI, SMHI, MET Norway).

3. How It Works

The satellite images are processed to calculate a vegetation index (NDVI) that measures how healthy and active plant growth is. Our system compares this across your fields and over time to spot trends. When clouds prevent a clear view, short gaps are estimated by interpolating between the clear passes either side, and wider gaps from Sentinel-1 radar. Both are fixed calculations from measured data — not AI, and always labelled as estimates. The variable-rate zones are then produced by grouping the NDVI values into clusters; that clustering step is the AI part.

4. Known Limitations

No system is perfect. Cloud cover in the Nordic region (40–60% in summer) means some observations are estimated rather than directly measured; those are labelled. Regional differences in soil type, local weather, and crop variety can affect accuracy. Zone boundaries are a statistical grouping of the imagery, not a survey of your soil. Early-season and late-season readings can be less reliable when plant cover is sparse.

5. What It Does Not Do

Freyr's AI does not provide certified agronomic advice. It does not replace the expertise of your local agricultural advisor or the requirements of your national authority. It does not autonomously make decisions about your farm — every action remains in your hands. Its outputs are not certified compliance reports under EU Common Agricultural Policy regulations.

6. How Often Models Are Updated

Freyr does not train or retrain a model on your data. The zoning algorithm is unsupervised: it runs fresh on each field's current imagery every time you generate a plan, so there is no learned model that drifts between seasons. Every version of every model we run is recorded in our AI model registry, and when one changes, this page is revised to reflect it.

7. Your Data and Privacy

All data processing happens on EU-based servers (Hetzner, Helsinki). Your field data is never shared with third parties without your explicit consent. Satellite imagery is publicly available and does not contain personal information. For full details, see our Privacy Policy.

8. Questions?

If you have questions about how the AI works or want to know more, we are happy to help.

Reach us at ai@arcteon.space