Impact of sample size on yield estimation in commercial vineyards

The accurate estimation of yield is a fundamental for suitable viticulture, playing a pivotal role in the planning of logistics, the allocation of resources and the formulation of commercial strategies. The capacity to make yield projections enables producers to anticipate market demands and optimize operations, enhancing both efficiency and sustainability (Komm Clingeleffer et al. 2001; Carrillo et al. 2015). Historically, the yield estimation has been dependent on manual techniques, such as the counting of bunches on a restricted sample of vines and the extrapolation of results to the entirety of the vineyard. Despite their widespread use, these techniques have notable limitations. They are costly in terms of labour and time, and they are prone to sampling errors due to the lack of representativeness, particularly in large or spatially heterogeneous vineyards (Dami, 2011).To address these shortcomings, advanced techniques have been developed that integrate historical data and refined sampling designs. For example, Araya-Alman et al. (2019) proposed a methodology based on historical yield patterns to identify key sampling areas, thereby reducing estimation errors. Similarly, Oger et al. (2021) investigated the potential for optimising the number of sampled vines to minimise errors, emphasising the value of systematic approaches for more accurately capturing spatial variability. The adoption of emerging technologies has further transformed yield estimation practices. Torres-Sánchez et al. (2021) demonstrated how unmanned aerial vehicles (UAVs) equipped with high-resolution cameras and photogrammetric point cloud analysis can provide detailed spatial data on cluster distribution across extensive vineyard areas. In a further development of the techniques described above, Meyers et al. (2011) devised a dynamic spatial optimisation model which enhances the representativeness of the samples taken while reducing the costs of the operation. Moreover, Nuske et al. (2011) implemented computer vision algorithms for automated cluster detection and counting, thereby eliminating the necessity for destructive sampling. The integration of machine learning has expanded the scope of these technologies. Palacios et al. (2023) applied computer vision and machine learning to enable early yield predictions across different grapevine varieties, facilitating adaptive management strategies early in the growth cycle. Similarly, Íñiguez et al. (2024) developed deep learning models capable of detecting grape clusters even under complex occlusion conditions, improving data accuracy in challenging environments.In this context, all these new methodologies and technologies can help to optimise the sampling strategy. To achieve a satisfactory yield estimation from punctual measurements, the number of measurements must reflect the expected yield variance at the desired scale. However, field measurements represent a significant effort in terms of labour and time, logistics and cost associated with equipment and the technology used. Therefore, a proper definition of sampling size is a key aspect of the success of the yield estimation approach used.