Search

Efficient use of scarce water resources for optimum crop yield and quality is a major concern in dry lands. Traditionally, in-situ methods such as measurements of leaf water potential (L) or soil moisture have been used for estimating plant water stress in vineyards. However, these methods are time, labor, and cost-intensive, and limited to point measurements of water stress that do not offer much spatial resolution. In this study, hyperspectral images acquired from a ground-based utility vehicle were analyzed to evaluate the applicability of hyperspectral data in classifying plant water stress measured as L into three classes: no to mild water stress (L > -0.8 MPa), moderate water stress (-0.8 MPa = L = -1.2 MPa) and severe water stress (L < -1.2 MPa). A field trial was carried out in an experimental vineyard in arid southeastern Washington, USA. Grapevines were subjected to either full irrigation (FI) or regulated deficit irrigation (RDI). Several vegetative indices (VIs) derived from hyperspectral reflectance information were assessed to estimate the level of water stress. Linear relationships between VIs and L, and relative variable importance based on Gini impurity in different test models were evaluated to select optimal VIs for vine water stress classification. Green Normalized Difference Vegetation Index (GNDVI), Photochemical Reflectance Index (PRI), and Anthocyanin Index (ANT) showed consistently good trends in the data obtained over two growing seasons and were selected as the optimal VIs. In addition, air relative humidity, evapotranspiration, minimum soil temperature, and solar radiation demonstrated high significance in model development and were included in the final optimized random forest classifier (RFC) and artificial neural network (ANN) models. The RFC model had an accuracy of 73% while the ANN model had an accuracy of 70% in classifying plant water condition into three classes. Both models demonstrated good classification abilities and can be applied to analyze vine water stress and developing decision support tools for precision irrigation in vineyards. [ABSTRACT FROM AUTHOR]

Efficient use of scarce water resources for optimum crop yield and quality is a major concern in dry lands. Traditionally, in-situ methods such as measurements of leaf water potential (ψL) or soil moisture have been used for estimating plant water stress in vineyards. However, these methods are time, labor, and cost-intensive, and limited to point measurements of water stress that do not offer much spatial resolution. In this study, hyperspectral images acquired from a ground-based utility vehicle were analyzed to evaluate the applicability of hyperspectral data in classifying plant water stress measured as ψL into three classes: no to mild water stress (ψL > -0.8 MPa), moderate water stress (-0.8 MPa ≤ ψL ≤ -1.2 MPa) and severe water stress (ψL < -1.2 MPa). A field trial was carried out in an experimental vineyard in arid southeastern Washington, USA. Grapevines were subjected to either full irrigation (FI) or regulated deficit irrigation (RDI). Several vegetative indices (VIs) derived from hyperspectral reflectance information were assessed to estimate the level of water stress. Linear relationships between VIs and ψL, and relative variable importance based on Gini impurity in different test models were evaluated to select optimal VIs for vine water stress classification. Green Normalized Difference Vegetation Index (GNDVI), Photochemical Reflectance Index (PRI), and Anthocyanin Index (ANT) showed consistently good trends in the data obtained over two growing seasons and were selected as the optimal VIs. In addition, air relative humidity, evapotranspiration, minimum soil temperature, and solar radiation demonstrated high significance in model development and were included in the final optimized random forest classifier (RFC) and artificial neural network (ANN) models. The RFC model had an accuracy of 73% while the ANN model had an accuracy of 70% in classifying plant water condition into three classes. Both models demonstrated good classification abilities and can be applied to analyze vine water stress and developing decision support tools for precision irrigation in vineyards. [ABSTRACT FROM AUTHOR]