ABSTRACT
In order to enhance the understanding of vine yield development and facilitate the design of innovative agricultural practices in viticulture (i.e., new estimation methods), it is essential to have accurate and detailed data on vine yield components, including unproductive vines, number of bunches, and bunch weight. However, obtaining accurate and high spatial resolution yield data at the vine scale is costly and difficult to have for the main yield components (number of bunches, weight of bunch, missing plants, etc.). As a result, existing vine yield data are frequently estimated or measured at the field level. Unfortunately, the accuracy of these vine yield data is insufficient to study the intricate relationships between different yield components and their spatial distribution within vineyards. In this context, this article proposes a complete vine yield dataset that was specifically collected to develop and to test new sampling protocols in precision viticulture. This dataset comprises a comprehensive mapping of vine yield at the plant scale over two vine fields located in the southern region of France. Both vine fields were planted with the Vitis vinifera: cv. Syrah. The first field (Field 1) occupies 0.8 ha and data were collected in 2022, while the second field (Field 2) has an area of 0.5 ha and data were collected in 2008. Throughout the growing season, information regarding unproductive vines, inflorescence number, and bunch weight was collected for both vine fields. For both fields, at the flowering stage, the location of each productive and unproductive vines (dead and missing vines) was georeferenced, and the number of inflorescences was manually counted for all productive vines. For Field 1, at harvest, all bunches of the field were manually weighed with an accuracy of ±1 gram and georeferenced precisely (one point per vine). For each vine, total yield (grams per vine) was then computed as as the sum of the weight of its bunches. For Field 2, at harvest, the total yield per vine was estimated based on the weighing of representative bunches obtained from several regularly spaced set of 5 vines. In addition to the yield data, two ancillary data, including soil apparent resistivity measurements and common vegetative index derived from remote sensed imagery, are provided for both vine fields. Overall, the dataset consists of 3644 vines, with 2151 being productive, along with a total count of 33354 inflorescences and 19635 manually weighed bunches at harvest. This dataset is of interest as it contains information on grape yield organization at the within-field level. This dataset could be used to assess the impact of unproductive vines on neighbouring vines yield, as well as the correlations between available ancillary data and all yield components.
ABSTRACT
Detailed and precise knowledge of production parameters (yield, quality, health status, etc.) in agriculture is the basis for analyzing the effect of any agricultural practice. Fine mapping of production parameters makes it possible to identify the origin of observed variability, whether associated with environmental factors or with agricultural practices. In viticulture, in real commercial context, these data are rare because monitoring systems embedded on harvesting machines for grape yield and quality are not yet available. As a result, they are costly and/or cumbersome to acquire manually. As an alternative, a research project has been proposed to test low-cost methods using GNSS tracking devices for yield and harvest quality mapping in viticulture. The data set was acquired as part of this research. The methodology was applied on a commercial vineyard of 30 ha during the whole 2022 harvest season. The method has identified harvest sectors (HS) associated to measured production parameters (grape mass and harvest quality parameters: sugar content, total acidity, pH, yeast assimilable nitrogen, organic nitrogen) and calculated production parameters (potential alcohol of grapes, yield, yield per plant, percentage of unproductive plants) over the entire vineyard. The grape mass was measured at the vineyard cellar or at the wine-growing cooperative by calibrated scales. The harvest quality parameters were measured from samples on grape must at a commercial laboratory specialized in oenological analysis (Institut Coopératif du Vin, Montpellier, France) with standardized protocols. The percentage of unproductive plants of a harvest sector was calculated from the manually geolocation of each unproductive plants (dead plants + missing plants) over the entire vineyard, the plantation density of blocks, and the geolocalization of the harvest sector. The mean area of these harvest sectors is 0.3 ha. The data set is supplemented by climatic data from a weather station deployed in the center of the vineyard. It provided three climatic parameters (relative humidity, rainfall, air temperature) every 15 min, for the 2020, 2021 and 2022 years. It was also supplemented by a complete description of the vineyard blocks (grape variety, plantation year, area, inter-row distance and vine distance). The proposed data set constitutes a unique and interesting resource for research in agronomy, vine ecophysiology and remote sensing. It can be used for any research in vine ecophysiology aimed at identifying potential relationships between yield and harvest quality parameters for different grape varieties. The data set only covers one year, which is a limitation for studying inter-annual variability of the parameters measured. Another limitation of the method concerns the footprint (0.3 ha on average) of the parameters measured.
ABSTRACT
Applying a calibration model onto hyperspectral (HS) images is of great interest because it produces images of chemical or physical properties. HS imaging is widely used in this way in food processing industries for monitoring product quality and process control. In this context, one of the main difficulties in the application of regression models to HS images is to evaluate the error of the obtained predictions, since in a proximal imaging set up, the size of the pixels is usually much smaller than the area required to obtain a wet chemical reference. Moreover, the selection of regression model parameters, such as the number of latent variables (LV) in a partial least squares (PLS) model, can modify the appearance of the prediction maps. The objective of this work is to propose an approach based on geostatistical indices to use spatial information of prediction maps for supporting the evaluation of regression models applied to HS images. This work stablishes a theoretical connection between linear regression model performance estimates and the spatial decomposition of variance in prediction maps, when the ground truth to estimate is spatially structured. This approach was tested in a simulated dataset and two real case studies. Geostatistical indices of the prediction maps were compared to model performance metrics for PLS models with increasing number of LV. The theoretical framework was proven by the results on the simulated dataset. In particular, the evolution of the nugget effect, C0, corresponded with the evolution of the random error of the model. Conversely, the error term of the model related with the slope of the model corresponded with the evolution of the structured variance observed in the prediction maps. On the real case studies, geostatistical indexes, extracted from the prediction maps, allowed to quantitatively evaluate the spatial structure of the estimations and complement the Root Mean Standard Error of Cross Validation (RMSECV) for the choice of optimal number of LV to consider in the model. The main advantage of this approach is that it does not require ground truth values. It could be used as a source of information for supporting the choice of optimum calibration parameters, such as the number of latent variables, or the choice of pre-treatments, complementing the traditional visual inspection of prediction maps with quantitative and objective metrics.
