This dataset comes from two experiments performed on a diversity panel of 279 grapevine (Vitis Vinifera L.) cultivars, designed to be representative of the genetic diversity of the worldwide diversity of cultivated grapevine (Nicolas et al., 2016, doi : 10.1186/s12870-016-0754-z). The panel was first studied in July 2021 on 10 3-years-old potted plants per cultivar, which were grown outdoors within an experimental vineyard in Montpellier, South France, under non-limiting water availability. On the next year (May 2022, 4-years-old plants), 6 plants of each cultivar from this outdoor experiment were placed in the PhenoArch phenotyping platform (https://eng-lepse.montpellier.hub.inrae.fr/platforms-m3p/montpellier-plant-phenotyping-platforms-m3p) and conducted under 3 water treatments: Well-Watered (WW), moderate Water Deficit (WD1) and severe Water Deficit (WD2). Two plants per cultivar and water treatment were studied, hence a total of 1476 plants in the greenhouse. In both experiments (outdoors 2021 and greenhouse 2022), near infrared spectroscopy, porometry and fluorescence data were collected using high-throughput devices on all plants. Simultaneously, measurements of leaf ecophysiological traits, using conventional and mostly low-throughput methods, were acquired on a subset of plants:

Poro-fluorimetry data (high-throughput) were collected with a Li600 device. Spectra (high-throughput) were taken in situ using two complementary spectrometers i/ MicroNIR with a spectral range from 950 nm to 1650 nm and ii/ NeoSpectra with a spectral range from 1350 nm to 2500 nm. Leaves disks were sampled from the measured leaf, weighted and oven-dried at 60°C. A third spectrometer was then used to get spectra from dried leaves: iii/ ASD with a spectral range from 350 nm to 2500 nm. Conventional, mostly low-throughput methods, were used to access leaf mass per area water content, water quantity, net CO2 assimilation, water potential and both water use efficiency intrinsic and instantaneous. Variable number of plants were measured depending on the trait considered.

This dataset was used to evaluate the performance and the robustness of predictive models built using high-throughput data to predict ecophysiological traits. Calibration models using Partial Least Square Regression (PLSR) were constructed in order to relate high-throughput data (spectral or poro-fluorimetry data) to the ground-truth measures of the traits of interest. The robustness of PLSR models was evaluated using different combinations of calibration and validation sets. The robustness of the predictive model of leaf mass per area using spectral data was tested across experiments. The greenhouse experiment was useful to evaluate the predictive ability of PLSR models across the three water treatments using spectral or poro-fluorescence data. Once the predictive model was considered good enough in predicting the trait of interest, the corresponding PLSR model was used to predict this trait of interest on the 1476 plants of the greenhouse experiment. The ultimate goal of this study was to investigate the broad-sense heritability of such predicted traits.

The associated publication is: Coindre E, Boulord R, Chir L, Freitas V, Ryckewaert M, Laisné T, Bouckenooghe V, Lis M, Cabrera-Bosquet L, Doligez A, Simonneau T, Pallas B, Coupel-Ledru A, Segura V Robustness of high-throughput prediction of leaf ecophysiological traits using near infra-red spectroscopy (NIRS) and poro-fluorimetry. (submitted) The scripts for prediction analyses are publicly available in GitLab at https://forgemia.inra.fr/eva.coindre/robustness_ht_prediction_ecophysiological_traits.