M. Epee, Paul T.2023-05-292023-05-292022https://hdl.handle.net/10182/16159In most New Zealand vineyards planted in Sauvignon blanc, cane pruning is an essential management practice with direct consequences on yield, ripeness, and balance. In Marlborough, the largest wine grape producing region of New Zealand, Sauvignon blanc is generally pruned to three or four canes contrary to the traditional two-cane pruning system. Not only is four-cane pruning more labour demanding and more challenging for the selection of fruiting canes, but also it results in the production of more above-ground biomass as the vine is loaded with more nodes compared with two-cane pruning. Cane pruning requires skilled labour, which has become a rare commodity to find and retain. Artificial Intelligence (AI) technologies can support least skilled human pruners and guide automated pruning systems, thus alleviating this labour shortage. However, training an AI system in cane pruning requires learning from the decisions of expert human pruners. Besides being a specialised task, cane pruning also ensures that grapevines are kept in balance, and node number retention is one of the initial and cost-effective ways of achieving it. At the pre-flowering growth stage, the vine budburst percentage is used to assess capacity and infer balance, with values around 100% being regarded as optimal. However, this metric does not accurately account for other underlying physiological conditions such as number, distribution and location of double shoots, non-count shoots and blind nodes, which are also key indicators of potential balance. From veraison to harvest, ratios such as LA/FM – leaf area to fruit mass, ELA/FM – exposed leaf area to fruit mass, FM/PM – fruit mass to pruning mass, FM/CM – fruit mass to cane mass as well as fruit composition through ripening provide measures of the vine performance and balance. The interpretation of these metrics requires their comparison with optimal ranges, which in most cases, are often non-existent for many grapevine cultivars and vineyard conditions. When such metrics exist, their wide range makes their use impractical for specific growing conditions warranting research to find optimal values that are specific to cultivars, locations, and growing conditions. Therefore, to train an AI system in cane pruning and to enhance the monitoring and assessment of vine balance, retained dormant shoot attributes were characterised and the effect of retained node numbers on the vegetative and reproductive growth of Sauvignon blanc were analysed. To characterise winter cane pruning decisions for the selection of fruiting canes and renewal spurs, eight dormant shoot attributes – diameter, length, vertical distance from the bottom fruiting wire (VD), horizontal distance from the vine trunk (HD), node count, internode length, origin, and position relative to bottom fruiting wire (PFW) – were measured before and after pruning. This first experiment was conducted in two vineyard sites located in the Awatere Valley, Marlborough (New Zealand) over two growing seasons (2019/20-2020/21). In one site (Site 1), vine canopies were modified according to a 5 [total node numbers on canes: 10, 20, 30, 40, 50] x 3 [total node numbers on spurs: 1, 2, 3] factorial design and in the other site (Site 2) there was no canopy modification. The second experiment was conducted to investigate the physiological response of Sauvignon blanc to retained node numbers at budburst, pre-flowering, veraison, harvest, and winter dormancy in three different sites over two growing seasons (2019/20-2020/21). For this second experiment, the three sites used were coded Site 1, Site 2 and Site 3. Site 1 was the same as in the first experiment (site with vine canopies modified), Site 2 was located within the same vineyard row as Site 1, and Site 3 was in a vineyard block in Waipara, Canterbury. At Site 1, vine node numbers were set according to the same factorial design as in the first experiment (i.e. 5 [total node numbers on canes: 10, 20, 30, 40, 50] x 3 [total node numbers on spurs: 1, 2, 3]). At Site 2 and Site 3 vine node numbers were set according to a 5 [total node numbers on canes: 10, 20, 30, 40, 50] x 2 [total node numbers on spurs: 1, 2] factorial design. Cane pruning stripped the vine of 82% dormant shoots, the remaining 18% formed the next season’s canes and spurs. Retained and non-retained dormant shoots significantly differed in all their attributes except for node count. On average, retained dormant shoots at Site 1 were 9.2±0.07 mm diameter, 104.7±0.93 cm long, 11.4±0.65 cm HD and 83.2±0.54 cm VD. Retained dormant shoots that formed spurs had the shortest HD (9.8±0.93 cm) and VD (77.8±0.7 cm) and originated by order of preference from the vine head, basal nodes or first nodes on old spurs and old canes. Still at Site 1, retained dormant shoots that formed fruiting canes were 9.1±0.08 mm diameter, 106.4±0.89 cm long, 85.7±0.69 cm VD, 12.2±0.85 cm HD and originated by order of preference from old canes, old spurs and vine head. Increasing the vine node load in 2019 resulted the following winter (2020) in more dormant shoots for spurs being selected from old canes (49%), less from the vine head (25% reduction) and old spurs (14% reduction). Moreover, the number of spurs retained on old canes in winter 2020 increased with node load. Retained dormant shoots on 50-node vines compared with 10-node vines increased in VD (from 83.8 to 87.8 cm; p<0.05) and HD (9.3 to 12.8 cm; p<0.05) but decreased in diameter (9.6 to 9.3 mm; p<0.05). PFW was the only attribute that remained unchanged despite change in node load (canopy modification). Characterising cane pruning by quantifying dormant shoot attributes revealed that: (1) retained, non-retained dormant shoots, canes and spurs exhibited distinct attributes, and that (2) dormant shoot selection in commercial vineyards was a consistent knowledge-based process that followed strict rules. These rules could be used to train inexperienced human pruners, AI pruning systems and fully autonomous pruning robots. Retaining low node numbers (10 nodes) resulted in a very high vine budburst (>100%), a cane budburst close to 100%, the growth of numerous non-count shoots on the vine head (29.5±3.0 shoots, p<0.001) likely caused by an imbalance between potentially more carbohydrate reserves (source) for fewer active count buds (sink). On 50-node vines the response was characterised by vine budburst percentage close to 100%, low cane percent budburst (<<100%), the appearance of several blind nodes (7.6±0.3 nodes, p<0.001) mainly located at the cane’s proximal section to the head. Blind nodes were likely caused by the manipulation of the source-sink resulting in inadequate carbohydrate reserves to stimulate growth at budburst. However, additional experimentation is required to quantify the carbohydrate deficit. Cane percent budburst provided a more accurate assessment of the vine capacity than vine budburst. The number of double shoots was not associated with the vine node load as they appeared on both low-node and high-node vines. Three-node spurs developed more blind nodes than one-node and two-node spurs (p<0.001). Budburst start date was unaffected by node load, commencing at the cane’s most distal nodes to the vine head, and resulting in shoot growth non-uniformity along canes. Cane percent budburst seemed to reflect better the vine capacity than vine budburst at pre-flowering. Results of this research also justified the practice of retaining spurs of one or two nodes at cane pruning. During ripening, 10-node vines accumulated higher concentrations of soluble solids as a result of a high source/sink ratio (ELA/FM) caused by lower yields whereas 50-node vines stored less sugar due to a low source/sink ratio driven by higher yields. The average berry mass, TA and pH were unaffected by node numbers in the first year. A low source/sink ratio induced by high node numbers not only reduced the vine capacity to fully ripen the current crop but also jeopardized the next season’s reproductive potential (1.6 bunches per shoot on high-node vines compared with 1.8 on low-node vines). Carbohydrate storage was not directly measured but inferred through average cane mass and fruiting cane mass which were significantly higher on 10-node vines (respectively 41.1±4.2 g and 126.0±11 g for Site 1 at the end of the first season) compared with 50-node vines (respectively 30.5±3.4 g and 90.2±7.2 g). The source/sink ratio (ELA/FM and LA/FM) for optimum berry ripening (TSS accumulation) at these sites located in Marlborough and North Canterbury was found to be 0.75 m2 kg-1 for ELA/FM and 2.0 m2 kg-1 for LA/FM. Although Ravaz indices (FM/PM and FM/CM) for optimum TSS could not be determined, the ELA/FM of 0.75 m2 kg-1 corresponded to an FM/CM of 3.0– 6.0 kg kg-1 and FM/PM of 2.5 – 4.5 kg kg-1. Moreover, values below or above FM/CM of 3.0 – 5.0 kg kg-1 and FM/PM of 2.0 – 4.0 kg kg-1 were an indication of imbalance: either excess vigour (high carbohydrate reserves), high capacity and non-limiting source size for values below these ranges, or reduced vigour, limited capacity and source restriction for values above these ranges. Pruning decisions made by an AI system may have far reaching consequences which may extend over several seasons. Monitoring the outcome of these decisions is paramount for the success of an automated pruning system. Characterising the pruning decisions of an expert human pruner by quantifying retained dormant shoot attributes provided knowledge and data useful to initiate the training of AI systems. The outcome of pruning decisions on the grapevine vegetative and reproductive growth could be monitored and assessed at budburst with the composite metric for vine capacity (vine budburst, cane budburst, blind node count), at harvest with the source/sink balance metrics (LA/FM, ELA/FM) and at dormancy with Ravaz indices (FM/CM, FM/PM). This assessment of AI pruning decisions will enhance quality pruning and ensure the long-term success of wine grape production in New Zealand vineyards.enblind nodebudburstcanecorrelative inhibitioncrop loaddouble shootgrapevineGuyotpHpruningSauvignon blancRavaz indexspurtitratable aciditytotal soluble solidsvine balancegrape compositiongrape ripeningapical dominancevine capacitycarbohydrate reservessource/sinkThesisANZSRC::300805 Oenology and viticultureANZSRC::300403 AgronomyANZSRC::310806 Plant physiologyhttps://creativecommons.org/licenses/by-nc-nd/4.0/Attribution-NonCommercial-NoDerivatives 4.0 International