Relationship between yield components and multispectral images at the within-field level: A first step to improve sampling for yield estimation

Abstract

Most of the vineyard yields estimation methods promote the use of random sampling assuming yield variability is random and normally distributed. However, this assumption is in contradiction with recent results obtained with yield monitoring systems (Taylor et al., 2005). These results showed that the yield variability is not random but strongly organized spatially at the within-field level. Also, many authors highlighted a relationship between yield and vegetative indices, like NDVI. Therefore, improving yield sampling based on multispectral image information may significantly improve the estimation of mean field yield. However, in order to design relevant sampling methods, a better knowledge on how each yield components is related to NDVI is required. To this purpose, NDVI information derived from two multispectral images, and yield components (number of bunches, number of berries per bunch and berry weight at veraison and harvest) were measured in 330 sampling sites defined over 9 fields with 2 different varieties. By means of a principal component analysis (PCA), a significant relationship was detected between the weight of berries at veraison and harvest, and the NDVI. A significant improvement of the two-step method (Wolpert et al., 1992) can be proposed, according to spatial variability observed by remote sensing.