Hyperspectral remote sensing of grapevine drought stress

In karst landscapes stony soils have little water holding capacity; the rational use of water for irrigation therefore plays an important management role. Because the water holding capacity is not homogenous, precision agriculture approaches would enable better management decisions. This research was carried out in an experimental vineyard grown in an artificially transformed karst terrain in Dalmatia, Croatia. The experimental design included four water treatments in three replicates: (1) fully irrigated, based on 100% crop evapotranspiration (ETc) application (N100); (2 and (3) deficit irrigation, based on 75% and 50% ETc applications (N75 and N50, respectively); and (4) non-irrigated (N0). Hyperspectral images of grapevines were taken in the summer of 2016 using two spectral-radiance (W sr−1 m−2) calibrated cameras, covering wavelengths from 409 to 988 nm and 950 to 2509 nm. The four treatments were grouped into a new set consisting of: (1) drought (N0); and (2) irrigated (the remaining three treatments: N100, N75, and N50). The images were analyzed using Partial Least Squares-Discriminant Analysis (PLS-DA), and treatments were classified using PLS-Single Vector Machines (PLS-SVM). PLS-SVM demonstrated the capability to determine levels of grapevine drought or irrigated treatments with an accuracy of more than 97%. PLS-DA identified relevant wavelengths, which were linked to O–H, C–H, and N–H stretches in water, carbohydrates and proteins. The study presents the applicability of hyperspectral imaging for drought stress assessment in grapevines, even though temporal variability needs to be taken into account for early detection.