For the wine industry, yield is agronomically defined as the quantity of harvest, either expressed in grape mass or wine volume units, that has been collected per surface unit area and per crop cycle. Since the introduction of wine regulations at the beginning of the 20th century, grape and wine production has been seen as a trade-off between harvest quantity, i.e. yield, and quality (Ravaz and Sicard, 1911, Champagnol, 1984, Guilpart et al., 2014). However, this trade-off is not bijective i.e. a given harvest quality does not imply a unique yield but can exist across a range of possible yields (Tardáguila et al., 2008, Intrigliolo and Castel, 2009, McClymont et al., 2012, Martínez et al., 2016). Therefore, grape yield can be optimised for a given harvest quality by applying appropriate technical operations throughout the production chain. To this end, decisions on operations, both in the vineyard and in the winery, are based on an expected final yield and expected growing conditions from the start of the season. Numerous approaches to report the expected final yield have been proposed in the scientific literature (Clingeleffer et al., 2001, Diago et al., 2012, Cunha et al., 2016, Nogueira Júnior et al., 2018, Sirsat et al., 2019, Zhu et al., 2020). In this paper, they are referred to as yield assessment methods when considered as a whole. Most of these studies are conducted in a context of research experiments aimed at statistically linking total yield to a yield component (Diago et al., 2012, Lopes et al., 2016, Cunha et al., 2016), another plant-related variable (Cunha et al., 2010, González-Flor et al., 2014, Sun et al., 2017) or environmental variables (Nogueira Júnior et al., 2018, Sirsat et al., 2019, Zhu et al., 2020).
However, there are only a few studies that consider the adaptation of yield assessment approaches to the operational context. As a result, yield assessment methods are often adopted by the wine sector on the basis of scientific work although they have not necessarily been defined to be effective or even valid in such operational conditions. The operational context includes additional needs and constraints to be met to ensure the smooth running of the production chain of any commercial wine-growing structure. The operational context also generates data that differs from data collected for a research experiment. Operational data, e.g. weather data or field observations, is collected throughout the season for immediate decision-making purposes but it is usually not intended to support any statistical analysis. Therefore, there is an issue of whether or not the scientific studies addressing yield assessment have properly accounted for operational needs, constraints and capabilities in terms of data acquisition in order to enable wine sector professionals to rigorously apply the methods proposed in the scientific literature in a production context.
Such a question is also of real interest from a scientific research perspective since adapting to production conditions requires reporting on a wide variety of situations for grape yield development, which is likely to generate knowledge on this subject. In that respect, operational datasets often offer larger amounts of data, particularly in terms of time series, which can be used to improve yield modelling by supporting novel statistical approaches. However, the development of operational methods presents scientific challenges related to the quality, heterogeneity and low number of site-specific data to be taken into account. It is also dependent on the definition of indicators to be used, which in turn depends on the working habits of each vineyard/winery. Finally, yield assessment methods constitute both a technical and social issue for the wine industry and it is the role of scientific research to address both. In particular, the development of more relevant yield assessment methods should encourage their adoption by the wine industry, promoting a virtuous approach to developing agri-services that collect data for their own improvement and to support further research studies.
The few studies that have adapted yield assessment to operational conditions have often focussed on improving only one step of the yield assessment process. For example, some studies have attempted to improve measurement issues by working on the automation of yield components counting (Aquino et al., 2015a, Aquino et al., 2015b, Aquino et al., 2015a, Aquino et al., 2015b, Liu et al., 2020) or total yield weighing (Tarara et al., 2014). Others studies have sought to optimize sampling strategies (Araya-Alman et al., 2019, Oger et al., 2021). Although relevant, these studies remain limited in the way they respond to operational issues since they aim to improve only one constitutive step of the yield assessment process. As a consequence, the yield assessment methods that are currently available to the industry have significant limitations, such as a high degree of imprecision and an inability to characterize the uncertainty associated with the yield assessment. Moreover, the extent of these limitations is difficult to quantify in regard to what can be reasonably expected under operational conditions as no single yield assessment approach has dealt with the problem in its entirety.
Any proposal for an operational yield assessment method cannot be based solely on the optimisation of any single step of the yield assessment process. Instead, it must collectively assess issues associated with measurement and sampling approaches for both the estimation of explanatory variables (covariables) and of the yield response to be explained, as well as modelling issues for the development of a yield assessment model. New methods will also require an analytical approach that considers the entire yield development process in relation to the operational needs and constraints resulting from the production context. However, there is no holistic synthesis of the existing literature on yield assessment methods to help to identify the key research and industry questions that remain to be addressed in order to achieve a robust, accurate method of yield assessment in commercial vineyards. To address this deficit, this paper first provides an overview of the yield development process under operational conditions and a summary of the subsequent operational needs and constraints related to yield assessment to identify the challenges to be accordingly addressed. Secondly, a knowledge framework of yield assessment methods is proposed. It is framed in terms of measurement, sampling and modelling approaches for a yield assessment purpose. For each of these three topics, issues and literature propositions are presented. Finally, the yield assessment methods proposed in the literature are reviewed with regard to their characteristics in terms of measurement, sampling and modelling and to the challenges identified in the first part. In conclusion, concrete proposals for a new grape yield assessment method are discussed. These considerations are primarily aimed at the production of wine grapes, which constitutes the vast majority of published literature, but could easily be transposed to the production of table grapes, juice grapes or potentially to other perennial crops. As the terminology and nuances around yield assessment are often complex and diverse, a series of definitions for the terms used in the paper are also provided.Definitions
GRAPE (ACTUAL) YIELD: quantity of harvest that is effectively reached, expressed in mass (kg or t) or volume units (L or hL) per plant or surface unit area (ha or a)
INPUT INDICATOR: a variable that influences yield development without being reciprocally influenced by it
MEASUREMENT: observation in the vineyard or in the winery that may be performed with or without the help of instrumentation
MEASURAND: real value of a particular quantity to be estimated
MODELLING: establishing a statistical relationship between explanatory variables and a response variable to be explained (here, grapevine yield)
OPERATIONAL CONDITIONS/DATA: referring to the conditions/data of a commercial vineyard or winery, as opposed to research ones. This often implies that not all conditions are equal because different choices can be made from one year to the next or from one block to the next for agronomic, logistical and human reasons.
OUTPUT INDICATOR: a variable that stands as an outcome of yield development, mainly referring to yield components
SAMPLING: choice of measurement sites (spatial sampling) and dates (temporal sampling)
SITE-SPECIFIC: including effects of the environment (soil, climate, topography, etc.), cultural practices as well as operational constraints, needs and strategies in particular the qualitative orientation of the production
SYSTEMIC INDICATOR: a variable that influences yield development and that is reciprocally influenced by it
VINEYARD: refers to both grapevine blocks and the company that cultivates it, as understood in vineyard estate
WINERY: refers to both the cellar in which the operations of wine-making take place and the company that actually produces wine and commercializes it. N.B.: sometimes the terms vineyard and winery refer to the same enterprise
YIELD ASSESSMENT: any kind of yield information, including yield estimation, prediction and forecast without distinction
YIELD ASSESSMENT UNCERTAINTY: a distribution of values attributable to the real yield performance once the measurement, sampling and modelling steps have been performed
YIELD COMPONENTS: grapevine reproductive anatomical structures that are successively settled during the vineyard part of a yield development cycle
YIELD DEVELOPMENT CYCLE: overall process of grape or wine production, which includes different stages depending on the enterprise and target markets. N.B. this is not limited to the vineyard and may be extended into the winery
YIELD ESTIMATE: yield assessment made in the same unit, time and space than the measurement and sampling processes it results from
YIELD FORECAST: yield assessment made in different units, time or space than the measurement and sampling processes it results from. It is most commonly associated with a yield performance that is expected to be reached in the same space but at a future date and consequently expressed in different units. A yield forecast corresponds to a statistical distribution of expected yield values.
YIELD PREDICTION: yield assessment made in different units, time or space than the measurement and sampling processes it results from. It is most commonly associated with a yield performance that is expected to be reached in the same space but at a future date and consequently expressed in different units. A yield prediction corresponds to a single expected yield value (statistical expectation) .