AI Can Now Help Detect Cannabis Crops From Space

Cannabis cultivation has traditionally been monitored from the ground. Regulators inspect licensed facilities, law enforcement searches for illicit grows, and producers rely on workers and sensors to assess crop conditions. Advances in satellite imaging and artificial intelligence are beginning to add another layer: the ability to analyze cannabis cultivation from above.

A new systematic review¹ examining 105 remote sensing studies published between 2020 and 2025 highlights just how far agricultural AI has progressed. Researchers are combining machine learning with satellites, drones, radar, hyperspectral cameras, thermal imaging, and LiDAR to identify crops, predict yields, detect stress, and monitor how vegetation changes over time.

Among those studies was an especially relevant example for the cannabis industry. Researchers in Indonesia demonstrated that machine learning could identify illicit cannabis plantations from multitemporal satellite imagery with approximately 95% accuracy.

The significance goes beyond identifying illegal grows. The same technological advances could eventually support licensed cannabis cultivation, crop health monitoring, compliance, water management, and production forecasting. As AI becomes increasingly integrated into cannabis cultivation, remote sensing could become another source of data feeding those systems.

How Satellites Can Identify Cannabis Crops

A satellite does not need to photograph the distinctive shape of a cannabis leaf to recognize a cannabis field. Instead, remote sensing systems can measure characteristics of vegetation that are invisible or difficult to distinguish with the human eye.

Multispectral satellites record reflected energy across several wavelengths. These measurements provide information about vegetation density, chlorophyll activity, moisture, plant structure, and other biological characteristics.

The Indonesian cannabis study summarized in the review used Sentinel-2 satellite imagery collected at multiple points in time. Rather than relying on a single image, the researchers analyzed how vegetation changed throughout the growing period.

This temporal dimension is important because crops follow recognizable growth cycles. Cannabis progresses through seedling, vegetative, flowering, and harvest stages. Those transitions alter the amount and type of light reflected by the crop canopy.

The researchers extracted statistical characteristics from red, green, blue, near-infrared, and vegetation-index measurements taken across multiple observations. A backpropagation neural network then classified areas as cannabis plantations, forest, or scrub vegetation.

The review reports an accuracy of approximately 94.9% and a kappa score of 0.92 for the cannabis detection system.

In practical terms, the AI was learning something closer to the crop’s seasonal fingerprint than simply recognizing its appearance.

Agricultural AI Is Moving Beyond Single Images

The cannabis example is part of a much broader shift in remote sensing. The review found that satellite optical imagery appeared in approximately 63% of the studies examined, while synthetic aperture radar appeared in about 25% and UAV-based sensing appeared in roughly 18%. Studies could use multiple sensors, so these categories overlap.

Different AI architectures also proved useful for different types of agricultural problems.

• CNNs were widely used for mapping, segmentation, disease detection, and image analysis.
• LSTM networks and Transformers were increasingly applied to agricultural time-series data.
• Random Forest and gradient boosting remained competitive when useful features could be engineered from smaller datasets.
• Multimodal systems combined optical imagery with radar, climate, soil, thermal, or structural information.

This is important because real agricultural environments are messy. Clouds obscure satellites. Crops may resemble surrounding vegetation. Fields vary in size and planting density. Weather changes growth patterns. A system designed for one region may perform differently somewhere else.

Using multiple observations and multiple data sources gives AI more context with which to make a classification.

What The Review Found About Remote Sensing And AI

The diversity of these approaches suggests that agricultural AI is becoming less dependent on any single sensor or algorithm. Instead, developers can select different combinations depending on the problem being solved.

That distinction could become particularly relevant for cannabis because indoor, greenhouse, and outdoor cultivation environments present very different monitoring challenges.

Satellite Detection Does Not Mean Every Cannabis Plant Is Visible

The roughly 95% accuracy reported for cannabis detection should not be interpreted as meaning that satellites can identify any cannabis grow anywhere with 95% reliability.

The system evaluated a particular geographic environment using specific training data, satellite observations, and known cannabis cultivation sites. The broader review repeatedly warns that agricultural AI models can struggle when transferred between regions, seasons, climates, and management practices.

Researchers refer to this as a generalization problem.

A model trained on cannabis plantations surrounded by Indonesian forest may encounter very different conditions in California, British Columbia, Morocco, or an agricultural area where numerous visually similar crops are grown nearby.

Cloud cover also presents a practical limitation for optical satellites. Radar can help address that issue because synthetic aperture radar does not depend on visible light and can collect information through clouds. The review found growing interest in combining optical and radar observations specifically because the two sensors provide complementary information.

That creates a likely direction for future cannabis detection systems: not one perfect photograph, but multiple measurements accumulated across time.

What Remote Sensing Could Mean For Legal Cannabis

Surveillance of illicit cultivation is the most obvious cannabis application discussed in the review, but the underlying technology is not inherently limited to enforcement.

Remote sensing AI can also estimate water stress, nutrient conditions, plant health, biomass, and yield in conventional agriculture. The review includes systems designed to detect disease, identify irrigation requirements, monitor chlorophyll and nitrogen levels, and predict production before harvest.

Similar concepts could eventually complement existing cultivation technologies within legal outdoor or greenhouse cannabis operations.

From Surveillance To Crop Management

For licensed producers, remote sensing could potentially support monitoring across larger cultivation areas where manually inspecting every plant or section of a field is difficult.

Changes in canopy temperature, vegetation indices, or plant development could identify areas requiring closer inspection. Instead of replacing growers, AI could help prioritize where growers should look.

This mirrors developments already occurring inside cannabis cultivation facilities, where technologies such as precision irrigation systems increasingly turn crop measurements into management decisions.

Remote sensing adds another potential layer to that data stack.

AI Cannabis Surveillance Raises Regulatory Questions

The technology becomes more complicated when remote sensing moves from agricultural management into enforcement.

The review specifically identifies governance concerns surrounding surveillance applications such as illicit-crop detection. Its authors note that responsible-use frameworks and independent validation remain underdeveloped within the studies they examined.

This distinction matters because classification errors have different consequences depending on how a system is used.

If an agricultural model incorrectly identifies a stressed area of a field, a grower can inspect the location and correct the mistake. If an automated surveillance system incorrectly classifies legal vegetation as illicit cannabis cultivation, the consequences could involve investigations or enforcement activity.

Accuracy therefore cannot be the only metric.

Operational systems would also need reliable uncertainty estimates, external validation, transparent procedures, and some mechanism for confirming AI-generated detections before decisions are made.

The review makes a broader point that applies particularly well to cannabis regulation: moving from a research demonstration to a trusted operational system requires considerably more than producing a high accuracy score.

The Future May Be Continuous Cannabis Monitoring

The next generation of remote sensing AI could become substantially more capable.

The review highlights emerging technologies including Transformers, self-supervised learning, multimodal sensor fusion, 3D neural networks, digital twins, and Earth-observation foundation models. These approaches could reduce the amount of manually labeled data required to build specialized agricultural systems.

That could eventually make crop monitoring much more scalable.

Instead of developing an entirely new model for every location, future systems may begin with a pretrained understanding of vegetation, terrain, seasonal patterns, and land use, then adapt that knowledge to particular crops or regions.

For cannabis, that creates two parallel possibilities.

Regulators and law enforcement could gain better tools for identifying unlicensed outdoor cultivation. At the same time, licensed growers could gain access to increasingly sophisticated monitoring systems capable of identifying stress, estimating production, or tracking crop development remotely.

The important technological change is therefore not simply that AI can detect cannabis from space. It is that satellites, sensors, and AI are increasingly capable of treating agricultural land as a continuously measurable system.

Cannabis is now becoming part of that transformation.

References:

¹ Gracia Rojas, D. A., Gómez Díaz, H. J., & Gil Herrera, R. de J. (2026). Remote sensing-based machine learning and deep learning for agriculture and forestry: Yield estimation, land-cover mapping, and plant stress: A systematic review (2020–2025). Remote Sensing Applications: Society and Environment, 102279. https://doi.org/10.1016/j.rsase.2026.102279