Introduction
Measuring evapotranspiration (ETc) is critical for managing vineyard irrigation efficiently. Traditional approaches use the two-step crop coefficient method, where ETc is modeled as the product of crop coefficient (Kc) and reference evapotranspiration (ETo). While models such as the FAO Penman-Monteith equation provide reliable estimates, vineyard-specific measurements are necessary for precise management. Existing methods like grid paper and Paso Panel sensors are labor-intensive, bulky, or insufficiently …
Methodology
Videos of shaded vineyard ground were captured with a smartphone camera during sunny conditions. A computer vision pipeline was developed, incorporating structure-from-motion (SfM), 3D surface reconstruction, and U-Net segmentation to classify shaded and unshaded areas. Paso Panel measurements were collected concurrently for comparison and calibration. Polynomial fitting was applied to better model the relationship between Paso Panel current and shaded area, improving on earlier linear assumptions.
Results
The CV-based method demonstrated a correlation of $R^2 = 0.68$ with Paso Panel readings. The U-Net segmentation achieved a validation intersection-over-union of 0.91, and the approach successfully generated voxelized reconstructions of shaded ground under vines. This method proved faster and simpler than Paso Panel use, with data collection taking only about one minute per row using a handheld smartphone.
Discussion
Compared to aerial or satellite-based approaches, the proposed ground-based CV method provides higher spatial and temporal resolution at lower cost, requiring only a smartphone. It enables mapping of crop coefficient variation at the panel or vine level, facilitating integration into variable-rate irrigation systems. Although current processing times remain high, pipeline optimization and GPU acceleration are expected to make the method practical for wider deployment. The method is particularly useful for …
Conclusion
This study demonstrates the feasibility of using smartphone-based computer vision to estimate crop coefficient in vineyards. The approach is affordable, easy to use, and offers improved spatial resolution over existing field methods. With further validation and optimization, this method has potential for practical adoption by growers and integration into vineyard decision support systems.