Hyperspectral imaging data combined with climate data to predict stomatal conductance and transpiration of grapevine plants
Résumé
Digital agriculture driven by new intelligent sensors is one of the main ways to improve farm
management. Accessing physiological variables such as transpiration (E) and stomatal
conductance (gs) in real time with optical instruments is challenging. These are the
privileged variables to detect water stress. In this study, the objective is to evaluate visible-
near-infrared spectral imaging data combined with climate data to predict transpiration (E)
and stomatal conductance (gs) of grapevine (Vitis vinifiera L.) plants by using Sequentially-
Orthogonalized Partial-Least-Square Regression (SO-PLS). A water stress gradient was obtained using pots of three grape varieties (Syrah, Merlot,
Riesling) tested under two water conditions where precise monitoring of physiological
variables was performed. Hyperspectral images were acquired and a weather station
provided radiation (Rg), relative humidity (RH), temperature (Ta) and wind speed (Ws). For
gs, best model is obtained by using only spectral data (R²= 0.656, bias=8.76, RMSE=64.7
mmol.m².s-1). For E, the best model is obtained by using both blocks (R²= 0.699, bias=0.055,
RMSE=0.614 mmol.m².s-1). While E prediction model has a lower performance using only
spectral data (R²= 0.625, bias=-0.02, RMSE=0.67 mmol.m².s-1). These encouraging results
offer prospects for the use of spectral imaging to detect water stress of grapevine plants
Origine : Fichiers produits par l'(les) auteur(s)