Machine learning in determination of water saturation deficit in wheat leaves on basis of Chl a fluorescence parameters

Abstract

Water saturation deficit (WSD) is a parameter commonly used for detection of plant tolerance to temporary water shortages. However, this parameter does not meet criteria set for screening. On the other hand, measurement of chlorophyll (Chl) a fluorescence is a fast and high-throughput method. This work presents the application of learning systems to set up a model between WSD and Chl a fluorescence parameters allowed for development of a new screening test. Multilayer perceptron (MLP) was trained to predict WSD values on the basis of Chl a fluorescence. The best MLP consisted of three inputs: maximal quantum yield of PSII photochemistry, approximated number of active PSII reaction centres per absorption, and measure of forward electron transport, three hidden nodes and one output (WSD). The MLP precision was 82% with a correlation coefficient of 0.98. Continuous improvement of MLP structure and model adaptation to new data takes place. (1)