4. Discussion
4.1. Validity of the method: the risk of auto-correlation was low The three years (2010–2012) of experiment 1 were characterized by contrasting water and nitrogen availability due to both weather variability and treatment application. This allowed us to explore (i) a wide range of stresses (water and nitrogen) and (ii)various successions of levels of stresses within and between years, which are suitable characteristics to study statistical relationships between yield formation and resource availability as presented in this paper. However, one of the concerns when examining linear regression analyses between a yield component at harvest in season 2 and an indicator of the crop water or nitrogen status at any time in season1 is auto-correlation. Ranking of treatments with respect to values of such indicators may be the same from year to year independently of weather variability. Plotting any yield component against those indicators measured at any time would yield significant relation-ships despite any real underlying physiological process. It was not the case in the present study for two reasons. First, bud fertility and berry number per bunch at harvest in season 2 were significantly correlated with LNC and predawn at flowering in season 1 whereas there was almost no correlation with LNC and predawn at flowering in season 2 for either cultivar (Fig. 2).