In-season prediction of yield variability in an apple orchard

Prediction of the spatial variability of apple yield as early as possible in the season is very important for farm managers. Many researchers used colour images of the apple trees and algorithms in order to develop prediction models of apple yield. The objective of this research was to study the spatial variability in an apple orchard and to develop methods for predicting yield variability within the growing period and as early as possible to permit management decisions. A commercial point-and-shoot and a multi-spectral camera were utilised to obtain images of the apple trees during the flowering period under daylight in a cv. Fuji apple orchard in Greece. The images were taken in April 2010 and 2011. Fruit yield was recorded at harvest each year. Supervised classification was used to isolate and calculate the flowers' pixel density of the whole image. For both years of the study, the results showed that, with both cameras, the estimated distribution of the flowers was correlated with the final yield distribution; however, for the second year, the correlation was slightly lower, probably due to adverse climatic conditions during and after the pollination period, which resulted in low yield. Multi-spectral images gave the best results in both years (r = 0.859 in 2010 and r = 0.827 in 2011).