AnswersPublished14 sources
How Kuva Space Used Hyperspectral Satellites to Screen for Opium Poppy Fields in Afghanistan
Kuva Space’s Helmand pilot screened 248,889 agricultural fields and flagged 8,835 as likely opium poppy fields. The workflow first delineated small field parcels from Sentinel 2 imagery, then used seasonal observations and separate hyperspectral and multispectral AI models to identify fields on which both models agr...
Published byEdited with GPT-5.6 TerraImages generated with GPT Image 2
Research answer
Key takeaways
- Kuva Space’s Helmand pilot screened 248,889 agricultural fields and flagged 8,835 as likely opium poppy fields.
- The workflow first delineated small field parcels from Sentinel 2 imagery, then used seasonal observations and separate hyperspectral and multispectral AI models to identify fields on which both models agreed.
- The main operational benefit is triage: satellite analytics can focus scarce high resolution imagery, analyst review and field investigation on a smaller set of higher probability locations.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How did Finnish spacetech company Kuva Space use its Hyperfield-1 hyperspectral satellites, Copernicus Sentinel-2 imagery, multi-temporal da. Article summary: Kuva Space used hyperspectral and conventional optical imagery as a parcel-level screening system—not as conclusive proof of illicit cultivation. In Helmand, the combined model improved overall classification accuracy fr. Topic tags: general, government, general web. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake n
Kuva Space’s Afghanistan pilot shows how satellite data can be used to prioritize suspected illicit crop fields at regional scale. The company combined its Hyperfield-1 hyperspectral imagery with Copernicus Sentinel-2 observations, seasonal crop-timing data and AI models. The output was a list of likely poppy parcels for follow-up—not evidence sufficient on its own to establish illicit cultivation. 18
What Kuva Space analyzed in Helmand
From January through June 2026, Kuva Space collected 363 Hyperfield-1 hyperspectral images across Afghanistan’s Helmand region, representing 648,450 km² of total image coverage. It combined that imagery with Copernicus Sentinel-2 data and AI-based classification. 18
Hyperfield sensors measure reflected light across many narrow visible and near-infrared bands. Kuva says Hyperfield-1 operates in the 475–880 nm range, enabling detailed characterization of vegetation and other surface materials from their spectral properties. 1 That additional spectral detail is intended to help distinguish crop signals that may look similar in conventional optical imagery.
Step 1: Build a map of individual fields
A central challenge was that precise agricultural parcel boundaries were not available for the study area. Kuva therefore used a separate parcel-delineation model, fine-tuned with Sentinel-2 imagery for Helmand’s small, heterogeneous, desert-mixed fields. 18
This matters because the system assessed fields rather than individual pixels. Parcel-level screening is more useful operationally: a reviewer needs a defined location to order sharper imagery or direct other verification work.
Step 2: Use the crop’s seasonal pattern alongside spectral data
The pilot drew on observations made over time rather than relying on a single satellite pass. That multi-temporal approach captures crop phenology—the changes in a crop’s appearance across its growing cycle—and can help separate poppy from visually similar legal crops when their timing differs. 18
Kuva’s reported workflow used very-high-resolution Pléiades imagery and seasonal Sentinel-2 observations to create a poppy reference set. With limited field access, those image-derived labels are an important qualification when interpreting the reported performance. 18
Step 3: Require agreement between two AI models
Kuva trained a hyperspectral classifier using its Hyperfield-1 foundation model, which it says had been pre-trained on more than 1.2 million Hyperfield image patches. It evaluated Hyperfield and Sentinel-2 models separately, then flagged a parcel as likely poppy only when both models agreed. 18
That agreement rule is designed to be conservative. Rather than maximizing the number of alerts, it seeks to reduce false positives—fields incorrectly identified as poppy—which can consume scarce analyst time and costly high-resolution imagery.
Results: 8,835 fields flagged from 248,889 mapped parcels
The system delineated 248,889 agricultural fields and flagged 8,835 as likely poppy fields, equal to roughly 3.6% of the identified fields. 18
Kuva reported 75% overall accuracy for the combined approach, compared with 71% for Sentinel-2 alone, alongside fewer false positives. 18 The four-percentage-point gain is meaningful chiefly as a triage improvement: it can produce a more manageable set of places to inspect.
In a geolocation test using three Hyperfield scenes overlapping Pléiades reference imagery, Kuva reported average CE90 accuracy of 21.75 metres, or 0.87 Hyperfield pixels. The company said that precision was enough to associate detections with the correct small, fragmented fields. 18
Why this kind of monitoring can be useful
Satellite screening is particularly useful where cultivation is geographically dispersed, remote or difficult to access. Instead of acquiring very-high-resolution imagery or conducting field checks across an entire province, investigators can use likely-poppy alerts to prioritize the next step. 18
The same basic data pipeline—field boundaries, repeated observations and crop-specific classification—also has relevance for agricultural monitoring. But a model trained and assessed for poppy in Helmand should not be presumed to work reliably for a different crop or country without local training data and validation.
What the pilot does not prove
A likely-poppy classification is not proof of illicit activity. Kuva frames the technique as a screening tool that narrows the search area; confirmation should rely on higher-resolution imagery, corroborating intelligence and, where feasible, field checks. 18
The reported 75% accuracy is a company-reported pilot result, not an independently published field-validated assessment. Its reference labels depended on high-resolution imagery and seasonal interpretation in a setting with limited ground access. 18
Several factors can complicate classification:
- small, mixed or poorly delineated fields;
- legal crops with similar appearance at particular growth stages;
- clouds, shadow and changing vegetation conditions; and
- incomplete or imperfect reference labels.
18
Deliberate concealment, such as fragmentation, intercropping or altered planting timing, could add to that ambiguity. Seasonal analysis and hyperspectral signatures offer more information than a one-date image, but they do not eliminate the need for verification.
What could improve the next generation of models
The evidence available for this pilot supports Hyperfield-1, Sentinel-2 fusion, parcel delineation and multi-temporal analysis. Kuva separately states that Hyperfield-2 is planned for launch in 2026, with expanded spectral coverage and improved spatial resolution. 13
Further accuracy gains would still need to be demonstrated. More local ground-truth data is especially important: Kuva notes that custom projects may use ground truth to improve insight accuracy. 29 Other inputs often proposed for crop monitoring—such as radar imagery, weather and terrain data—may be useful additions, but the supplied evidence does not establish them as validated parts of this Helmand workflow.
The key takeaway is practical rather than absolute: hyperspectral satellites and AI can make wide-area crop monitoring more targeted. They can identify fields worth investigating first, while the final determination remains a job for corroborating evidence and human review. 18
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "How Kuva Space Used Hyperspectral Satellites to Screen for Opium Poppy Fields in Afghanistan"?
Kuva Space’s Helmand pilot screened 248,889 agricultural fields and flagged 8,835 as likely opium poppy fields.
What are the key points to validate first?
Kuva Space’s Helmand pilot screened 248,889 agricultural fields and flagged 8,835 as likely opium poppy fields. The workflow first delineated small field parcels from Sentinel 2 imagery, then used seasonal observations and separate hyperspectral and multispectral AI models to identify fields on which both models agreed.
What should I do next in practice?
The main operational benefit is triage: satellite analytics can focus scarce high resolution imagery, analyst review and field investigation on a smaller set of higher probability locations.
Sources
science.nasa.govCSDA Vendor - Kuva Space - NASA Science