A Review Finds Vineyard Drone Models Break Down Across Seasons

A new review in the journal Computers and Electronics in Agriculture says unmanned aerial vehicle remote sensing is moving closer to becoming a practical tool for vineyards, but researchers say major technical and operational problems still stand in the way of wider use on farms.

The paper, “From growth monitoring to berry quality prediction: Advances and challenges in UAV remote sensing for precision viticulture,” was written by Jinyu Yang, Mengyuan Wei and colleagues. It examines how drones, onboard sensors and artificial intelligence methods have been used in vineyards over roughly the past decade, and where the field is still falling short when growers need reliable decisions across seasons, sites and grape varieties.

The review comes as wine producers face increasing pressure to monitor fields more closely. According to figures cited in the paper from the International Organization of Vine and Wine, global vineyard area reached 7.1 million hectares and wine production stood at 225.8 million hectoliters in 2024. Grapevines are highly sensitive to climate, water supply, pests, disease and management practices, which makes timely field information important for both yield and fruit quality.

The authors argue that drones have become attractive because they fill a gap between field scouting and satellite imaging. Traditional field checks can be accurate, but they are labor-intensive and cover limited ground. Satellite systems can collect broad regional data, but they often lack the resolution and flexibility needed for vineyard blocks, especially when growers need images at a specific moment or under changing weather conditions. Drones, flying at low altitude, can capture high-resolution data more quickly and from angles that better match vineyard rows and canopy structure.

The review describes how vineyards are now being surveyed with drones carrying RGB, multispectral, hyperspectral, thermal infrared and LiDAR sensors. These systems can produce orthomosaic images, temperature maps, vegetation indices and 3D point clouds. Researchers have used those outputs for nutrient assessment, growth and vigor mapping, disease and pest detection, water stress monitoring, yield estimation and attempts to predict berry quality before harvest.

The paper highlights berry quality prediction as one of the least mature areas in the field. The authors describe berry quality as a high-dimensional outcome influenced by physiology, biochemistry, cultivar, growth stage, microclimate, soil and vineyard management. Because of that complexity, models that perform well in one vineyard or one season often do not hold up when used elsewhere. The review says this weak transferability remains one of the biggest barriers to operational use.

The study also points to disease detection and berry monitoring with deep learning as a current research hotspot. Advances in computing power and open-source software have pushed vineyard analysis away from simple statistical regressions and toward machine learning and deep learning systems that can extract more features from images and sensor data. In some cases, that has improved predictive accuracy. But the authors say accuracy alone is not enough if the models are hard to interpret, expensive to run, or too fragile to work under different field conditions.

To assemble the review, the researchers searched the Web of Science database using terms including UAV, drone, grapevine, vineyard and viticulture. They then assessed studies spanning hardware, preprocessing, canopy segmentation, feature engineering and modeling methods. The paper cites 105 references and organizes prior work around six main application areas: nutritional status monitoring, growth and vigor monitoring, pest and disease detection, water stress monitoring, yield prediction and berry quality assessment.

One of the central findings is that vineyard conditions create special problems for image analysis. Unlike some row crops with more uniform coverage, vineyards often include discontinuous canopies, exposed soil, weeds between rows, uneven terrain and strong shadows. Those factors can contaminate spectral and thermal signals if software cannot reliably separate vine canopy from the background. The review says this issue can affect nearly every later step in analysis, from vegetation indices to machine learning predictions.

The authors also note that vineyard management itself can destabilize models. Practices such as pruning, shoot thinning, leaf removal and canopy training can change vine structure and reflectance quickly. A model trained on images from one management stage may lose accuracy after a different intervention, even in the same block. That makes repeatable flight planning and canopy-aware data collection especially important for multi-temporal monitoring.

Cost and logistics remain another obstacle. High-resolution imagery, thermal data and 3D products require substantial data storage and computing resources. More advanced sensors, including hyperspectral units and LiDAR, can improve the detail and type of information collected, but they also raise system costs and processing demands. The review says that may limit adoption, especially if growers are expected to run complicated workflows or pay for systems that do not yet deliver stable economic returns.

The paper also raises concerns about how studies are evaluated. According to the authors, many published results focus heavily on high performance within a single site, while reporting less about how a model performs across years, regions or grape varieties. That can create an overly optimistic picture of readiness for commercial use. The review calls for more transparent reporting of data acquisition, preprocessing, canopy handling and validation design so that results can be compared across studies.

In assessing sensors and platforms, the authors say multirotor drones play a dominant role in vineyard research because they are flexible, easier to operate at low altitude and suitable for repeated flights over specific blocks. Spectral sensors remain especially important, but the review suggests that the next step is not simply adding more data. Instead, the field may benefit most from better integration of multiple data types collected over time, combining, for example, spectral, thermal and structural information to improve robustness.

The review also says interpretability needs more attention. Data-driven systems can identify useful patterns, but growers and agronomists often need to know why a model reached a result before changing irrigation, spraying or harvest plans. The authors point to radiative transfer models and crop models as one way to make predictions easier to explain, especially when combined with machine learning methods rather than treated as competing approaches.

Another theme running through the paper is standardization. The authors say differences in flight altitude, sensor calibration, lighting conditions, timing, image correction and model testing can lead to conflicting results even when researchers appear to be studying the same problem. They argue that the field needs more consistent acquisition and evaluation protocols if drone sensing is to move from research plots to routine vineyard management.

The review identifies several priorities for future work. Among them are multi-temporal and multi-modal data fusion, transfer learning and domain adaptation to improve cross-site generalization, targeted data augmentation or synthetic sample generation for small datasets, and the development of lightweight models paired with lower-cost sensors. Those steps, the authors suggest, could make drone-based vineyard monitoring more practical for large-scale use rather than a tool limited to experiments or specialized operations.

The paper presents drone sensing not as a replacement for all other forms of monitoring, but as a system with growing value when used to map variation within vineyards quickly and non-destructively. Its message is that the technology has advanced from early image collection toward more detailed crop monitoring, yet key parts of the workflow still need to become more stable, cheaper and easier to transfer from one vineyard setting to another before growers can rely on it for routine decisions on irrigation, disease control, harvest timing and fruit quality management.