Lopes, CarlosNovello, VittorinoCarmignani, Beatrice2022-09-072022-09-072019Carmignani, B.- Comparison of different methodologies to estimate bunch compactness. Lisboa: ISA, 2019, 71 p.http://hdl.handle.net/10400.5/25413Mestrado em Engenharia de Viticultura e Enologia (Double Degree) / Instituto Superior de Agronomia. Universidade de Lisboa / Faculdade de Ciências. Universidade do PortoBunch compactness (BC) is a key target for wine sector because it affects disease susceptibility, berry ripening among other grapes characteristics. The most common method to estimate BC is the O.I.V. descriptor n°204: manual and subjective. Objective and automated methods are based on indices, using different relations between bunch traits, some obtained manually and other automatically through image analysis (as example: BW – weight; BV – volume; ML – maximum length; A – projected area; MVO – morphological volume; V3 – derived volume; BN – berries number). All the variables were significantly and positively correlated between each other: the highest Pearson correlation coefficient was between BW and BV (r = 0.99) followed by BW and A (r = 0.95). Fourteen compactness indices (CI) were tested (9 published and 5 created) on 61 Syrah bunches. These indices were then correlated with the mode of the O.I.V. descriptor n°204, where 11 were positively correlated and three were negatively correlated (CI-3, CI-3a, CSF). The index CI-10a, which relates bunch weight and maximum length, was the most suitable one to define BC (r = 0.78). In the frame of the EU VINBOT project, to improve BW estimation finding the best explanatory variables, a stepwise regression analysis between BW and the variables considered easy to extract by automated image analysis (A1 – projected area, V3 – volume 3, BN – berries number and CI-10a as index) was performed. The variable which explained best BW was A1 (partial R2 = 0.905), followed by CI-10a and V3 with a much smaller contribution (partial R2 <0.06 and partial R2<0.007, respectively). The variable BN was not selected by the model. We concluded that BC can be estimated in an objective and automatic way using image analysis. Furthermore, such estimations can enhance BW prediction by using BC as one of the explanatory variables which can improve automatic yield estimation methodologiesengbunch traitobjective indicesO.I.V. descriptor 204bunch volumebunch weightmaster thesis203083695