Functional unfold principal component analysis for automatic plant-based stress detection in grapevine.

Abstract

Detection of drought stress is of great importance in grapevines because the plant's water status strongly affects the quality of the grapes and hence, resulting wine. Measurements of stem diameter variations show promise for detecting drought stress, but they depend strongly on microclimatic changes. Tools for advanced data analysis might be helpful to distinguish drought from microclimate effects. To this end, we explored the possibilities of two data mining techniques: Unfold principal component analysis (UPCA) - an already established tool in several biotechnological domains - and functional unfold principal component analysis (FUPCA) - a newer technique combining functional data analysis with UPCA. With FUPCA, the original, multivariate time series of variables are first approximated by fitting the least-squares optimal linear combination of orthonomal basis functions. The resulting coefficients of these linear combinations are then subjected to UPCA. Both techniques were used to detect when the measured stem diameter variations in grapevine deviated from their normal conditions due to drought stress. Stress was detected with both UPCA and FUPCA days before visible symptoms appeared. However, FUPCA is less complex in the statistical sense and more robust than original UPCA modelling. Moreover, FUPCA can handle days with missing data, which is not possible with UPCA.