Automatic drought stress detection in grapevines without using conventional threshold values

Because the water status of grapevines strongly affects the quality of the grapes and resulting wine, automated and early drought stress detection is important. Plant measurements are very promising for detecting drought stress, but strongly depend on microclimatic changes. Therefore, conventional stress detection methods require threshold values which define when plants start sensing drought stress. There is however no unique method to define these values. In this study, we propose two techniques that overcome this limitation. Two statistical methods were used to automatically distinguish between drought and microclimate effects, based on a short preceding full-irrigated period to extract plant behaviour under normal conditions: Unfold Principal Component Analysis (UPCA) and Functional Unfold Principal Component Analysis (FUPCA). Both techniques aimed at detecting when measured sap flow rate or stem diameter variations in grapevine deviated from their normal behaviour due to drought stress. The models based on sap flow rate had some difficulties to detect stress on days with low atmospheric demands, while those based on stem diameter variations did not show this limitation, but ceased detecting stress when the stem diameter levelled off after a period of severe shrinkage. Nevertheless, stress was successfully detected with both approaches days before visible symptoms appeared. UPCA and FUPCA based on plant indicators are therefore very promising for early stress detection.