Please login first
In-field hyperspectral proximal sensing for estimating grapevine water status to support smart precision viticulture
1, 2 , 2, 3 , 4 , 2, 3 , 4 , 3 , 1 , 3 , * 2, 3
1  Federal Rural University of Amazon, 2501 Presidente Tancredo Neves Av., 66077-830, Belém-PA, Brazil
2  Department of Geosciences, Environment and Spatial Planning, Faculty of Sciences of the University of Porto, Rua do Campo Alegre, S/N, 4169-007, Porto-Portugal
3  INESC TEC - Institute for Systems and Computer Engineering, Technology and Science, Campus da Faculdade de Engenharia da Universidade do Porto, Rua Dr. Roberto Frias, S/N, 4200-465, Porto-Portugal
4  Associação para o Desenvolvimento da Viticultura Duriense, Edifício Centro de Excelência da Vinha e do Vinho Parque de Ciência e Tecnologia de Vila Real, Régia Douro Park, Portugal
Academic Editor: Leifeng Guo


Predawn leaf water potential (Ψpd), commonly accessed by using a Scholander type pressure chamber, is the main parameter to determine plant water status and it has been largely used to support irrigation management. However, this methodology is laborious, time-consuming and invasive, limiting its usage in mapping the water status of plants in large plantation areas. In this study, a low-cost hyperspectral proximal sensor to estimate Ψpd in grapevine (Vitis vinifera L.) was examined. For this, both Ψpd and spectral reflectance (340-850 nm) were accessed in grapevines in a commercial vineyard located at Douro Wine Region, northeast Portugal. A machine learning algorithm was tested and validated to assess grapevine water status. The experiment was performed in a randomized design with 12 grapevines (Touriga Nacional) per irrigation treatment, which were: non-irrigated, irrigation to replace 30% evapotranspired (Etc) water volume, and 60% Etc. The Ψpd and spectral data were determined weekly over six consecutive weeks, totaling 216 observations. The dataset was analyzed using Principal Component Analysis (PCA), and the machine learning regression algorithm applied was Random Forest (RF). Results from the validation dataset (n = 65 observations) for the RF tested exhibited a root mean square error (RMSE) of 0.25 MPa, mean absolute error (MAPE) of 29.55% and an R² value of 0.94. These results demonstrate that the hyperspectral sensor and RF algorithm can be accurately used to predict Ψpd in vineyards regardless of plant water status. This methodology emerges as a tool to assist vineyards producers in making decisions on irrigation management.

Keywords: Leaf water potential; machine learning; plant water status; point-of-measurement