Interpreting wine aroma: from aroma volatiles to the aromatic perception

A large varietal collection including over 1700 varieties was maintained in Conegliano, ITA, since the 1950s. Phenological data on a subset of 400 grape varieties including wine grapes, table grapes, and raisins were acquired at bud break, flowering, veraison, and ripening since 1964. Despite the efforts in maintaining and acquiring data over such an extensive collection, the data set has varying degrees of missing cases depending on the variety and the year. This is ubiquitous in phenology datasets with significant size and length. In this work, we evaluated four state-of-the-art methods to estimate missing values in this phenological series: k-Nearest Neighbour (kNN), Multivariate Imputation by Chained Equations (mice), MissForest, and Bidirectional Recurrent Imputation for Time Series (BRITS). For each phenological stage, we evaluated the performance of the methods in two ways. 1) On the full dataset, we randomly hold-out 10% of the true values for use as a test set and repeated the process 1000 times (Monte Carlo cross-validation). 2) On a reduced and almost complete subset of varieties, we varied the percentage of missing values from 10% to 70% by random deletion. In all cases, we evaluated the performance on the original values using normalized root mean squared error. For the full dataset we also obtained performance statistics by variety and by year. MissForest provided average errors of 17% (3 days) at budbreak, 14% (4 days) at flowering, 14.5% (7 days) at veraison, and 17% (3 days) at maturity. We completed the imputations of the Conegliano dataset, one of the world’s most extensive and varied phenological time series and a steppingstone for future climate change studies in grapes. The dataset is now ready for further analysis, and a rigorous evaluation of imputation errors is included.

The project VINIoT pursues the creation of a new technological vineyard monitoring service, which will allow companies in the wine sector in the SUDOE space to monitor plantations in real time and remotely at various levels of precision. The system is based on spectral images and an IoT architecture that allows assessing parameters of interest viticulture and the collection of data at a precise scale (level of grape, plant, plot or vineyard) will be designed. In France, three subjects were specifically developed: evaluation of maturity, of water stress, and detection of flavescence dorée. For the evaluation of maturity, it has been decided first to work at the berry scale in the laboratory, then at the bunch scale and finally in the vineyard. The acquisition of the spectral hyperstal image as well as the reference analyzes to measure the maturity, were carried out in the laboratory after harvesting the berries in a maturity monitoring context. This work focuses on a case study to predict sugar content of three different grape varieties: Syrah, Fer Servadou and Mauzac. A robust method called Roboost-PLSR, developed in the framework of this work (Courand et al., 2022), to improve prediction model performance was applied on spectra after the acquirement of hyperspectral images. Regarding the evaluation of water stress, to work with a significant variability in terms of water status, it has been worked first with potted plants under 2 different water regimes. The facilities have allowed the supervision of irrigation and micro-climatic conditions. The regression models on agronomic variables (stomatal conductance, water potential, …) are studied. To detect flavescence dorée, the experimental plan has consisted of work at leaf scale in the laboratory first, and then in the field. To detect the disease from hyper-spectral imaging, a combination of multivariate curve resolution-alternating least squares (MCR-ALS) and factorial discriminant analysis (FDA) was proposed. This strategy proved the potential towards the discrimination of healthy and infected leaves by flavescence dorée based on the use of hyperspectral images (Mas Garcia et al., 2021).

Agriculture, natural resource management and the production and sale of products such as wine are increasingly data-driven activities. Thus, the use of remote and proximal crop and soil sensors to aid management decisions is becoming commonplace and ‘Agtech’ is proliferating commercially; mapping, underpinned by geographical information systems and complex methods of spatial analysis, is widely used. Likewise, the chemical and sensory analysis of wines draws on multivariate statistics; the efficient winery intake of grapes, subsequent production of wines and their delivery to markets relies on logistics; whilst the sales and marketing of wines is increasingly driven by artificial intelligence linked to the recorded purchasing behaviour of consumers. In brief, there is data everywhere!

Opinions will vary on whether these developments are a good thing. Those concerned with the ‘mystique’ of wine, or the historical aspects of terroir and its preservation, may find them confronting. In contrast, they offer an opportunity to those interested in the biophysical elements of terroir, and efforts aimed at better understanding how these impact on vineyard performance and the sensory attributes of resultant wines. At the previous Terroir Congress, we demonstrated the potential of analytical methods used at the within-vineyard scale in the development of Precision Viticulture, in contributing to a quantitative understanding of regional terroir. For this conference, we take this approach forward with examples from contrasting locations in both the northern and southern hemispheres. We show how, by focussing on the vineyards within winegrowing regions, as opposed to all of the land within those regions, we might move towards a more robust terroir zoning than one derived from a mixture of history, thematic mapping, heuristics and the whims of marketers. Aside from providing improved understanding by underpinning terroir with data, such methods should also promote improved management of the entire wine value chain.

The study of climate is relevant as an element conditioning the typicity of a product, its quality and sustainability over the years. The grapevine development and growth and the final grape and wine composition are closely related to temperature, while climate components vary at mesoscale according to topography and/or proximity to large bodies of water. The objective of this work is to assess the mesoclimate of the Atlantic region of Uruguay and to determine the effect of topography and the ocean on temperature and consequently on Tannat grapevine behavior.

Grapevine yield is a key indicator to assess the impacts of climate change and the relevance of adaptation strategies in a vineyard landscape. At this scale, a yield model should use a number of parameters and input data in relation to the information available and be able to reproduce vineyard management decisions (e.g. soil and canopy management, irrigation). In this study, we used data from six experimental sites in Southern France (cv. Syrah) to calibrate a model of grapevine yield limited by water constraint (GraY). Each yield component (bud fertility, number of berries per bunch, berry weight) was calculated as a function of the soil water availability simulated by the WaLIS water balance model at critical phenological phases. The model was then evaluated in 10 grapegrowers’ plots, covering a diversity of biophysical and technical contexts (soil type, canopy size, irrigation, cover crop). We identified three critical periods for yield formation: after flowering on the previous year for the number of bunches and berries, around pre-veraison and post-veraison of the same year for mean berry weight. Yields were simulated with a model efficiency (EF) of 0.62 (NRMSE = 0.28). Bud fertility and number of berries per bunch were more accurately simulated (EF = 0.90 and 0.77, NRMSE = 0.06 and 0.10, respectively) than berry weight (EF = -0.31, NRMSE = 0.17). Model efficiency on the on-farm plots reached 0.71 (NRMSE = 0.37) simulating yields from 1 to 8 kg/plant. The GraY model is an original model estimating grapevine yield evolution on the basis of water availability under future climatic conditions.  It allows to evaluate the effects of various adaptation levers such as planting density, cover crop management, fruit/leaf ratio, shading and irrigation, in various production contexts.