Kuva Space takes on illicit crop monitoring with hyperspectral satellites and AI

Kuva Space says it has finished a pilot that used hyperspectral satellites and AI to improve monitoring of illicit crops at regional scale. The test focused on opium poppy in Afghanistan and paired Kuva Space’s Hyperfield imagery with Sentinel-2, the EU’s multispectral Earth-observation mission. The company says the combined approach reduced false positives compared with Sentinel-2 on its own.

The scale is the point. In Helmand Province, the system identified 248,889 agricultural fields and marked 8,835 of them as likely poppy fields. Those are model predictions, not 8,835 individually verified plots, and Kuva Space is careful about that distinction. The data is meant to help authorities narrow the search, not pretend every flagged field has already been proven guilty.

The company captured 363 hyperspectral images between January and June 2026, covering 648,450 km². Because precise field maps for Afghanistan were not available, Kuva Space first built a separate model to trace field boundaries, then fine-tuned it with 10-metre Sentinel-2 data to account for small fields and desert terrain. Olli Elliranta, a data scientist at the company, said the boundary model outperformed publicly available boundaries and gave geolocation accuracy averaging 21.75 metres.

Once the parcels were drawn, Kuva Space’s in-house foundation model looked for the spectral fingerprints of opium poppy at the parcel level. That matters because pixel-by-pixel maps are harder to use operationally. Hyperspectral sensors can see across hundreds of narrow bands, letting the model pick up differences in chlorophyll, water content and pigmentation that ordinary imagery can miss, especially when poppy is growing alongside similar-looking legal crops.

The company says the combined model reached 75 per cent accuracy, versus 71 per cent with Sentinel-2 alone. Elliranta said the real win was fewer false positives, since screening nearly 249,000 fields means even small improvements can remove a lot of wasted follow-up. Kuva Space is now pushing toward 90 per cent accuracy with more ground-truth data, higher-quality Hyperfield-2 data and other inputs such as SAR, weather data and DEMs. It is already extending the work to Myanmar, and says the same approach could eventually be used for coca or cannabis too.