Yield estimations in a vineyard based on high-resolution spatial imagery acquired by a UAV

Abstract

Yield estimation in vineyards is an essential issue for performing cropping practices to reach the desired production quality or quantity. Sporadic manual measurements have been traditionally made. In the last decade, vegetation indices (VIs) and geometric parameters obtained from unmanned aerial vehicle (UAV) imagery have been related to different vine biophysical features. This research aimed to evaluate the potential of VIs and green canopy cover (GCC; as a measure of plant vigour) obtained from conventional (or red, green, blue; RGB) and multispectral sensors that monitor spatial intraplot variability for yield predictions. The yield components traditionally sampled in an early growth cycle stage (pea berry size) were combined with UAV imagery-based products. The proposed methodology was applied to a vineyard in southeastern Spain during the 2019 and 2020 growing seasons. Rain-fed and irrigated treatments were implemented. Flights were performed throughout the growth cycle using RGB and multispectral cameras mounted on a UAV. Orthoimages were generated. Computer vision techniques were used to segment these orthoimages to obtain vegetation-only masks. Simple and multiple linear regression techniques were evaluated by using VIs alone, VIs combined with GCC and yield components as predictors. The RMSE values ranged from 0.21 kg vine −1 to 0.39 kg vine −1 when yield components, RGB or multispectral VIs were employed. Therefore, with all the advantages that their use entails, RGB and multispectral sensors are a good option for estimating the final yield of vineyards despite calibration being required for each season and grapevine plot. • An approach was proposed to estimate yield in vineyards with UAVs. • Multispectral and RGB VIs and grapevine vigour were employed. • The method used yield components sampled in early growth cycle stages. • Estimations were made with small errors combining UAV data with yield components.