Monitoring the photosynthetic performance of grape leaves using a hyperspectral-based machine learning model

Photosynthesis is a direct expression of the crop growth status and an important indicator predicting yield and quality. Rapid and accurate monitoring of the dynamics of photosynthetic is key to field management. In this study, we obtained photosynthetic pigments and water status parameters at the leaf scale during different growth periods of grape. The potential maximum quantum yield (Fv/Fm) of photosystem II (PSII) under dark adaptation and the light response curve (LRC) of the PSII electron transfer rate (ETR) under light adaptation were measured using a pulse amplitude modulated (PAM) chlorophyll fluorometer, while leaf spectral information was recorded using a hyperspectral imager. The maximum ETR (ETRmax) and initial quantum efficiency (k(alpha)) were calculated using the LRC model. A Bayesian neural network (BNN) model (implemented in Tensorfolw2.8) was developed to predict Fv/Fm, ETRmax and k(alpha) by quantifying the spectral response indices of photosynthetic pigments and water status parameters. A comparison was made with the partial least squares (PLS) and photochemical reflectance index (PRI). The results show that BNN, PLS model and PRI have better predictive performance for Fv/Fm than ETRmax and k(alpha). Compared with the PLS and PRI, the BNN model was able to significantly improve the prediction accuracy, where the validation results for Fv/Fm were R-2 of 0.78, ETRmax of 0.57 and k(alpha) of 0.53. In addition, the importance of the BNN model input features varied with Fv/Fm, ETRmax and k(alpha), with the vegetation index describing the photosynthetic pigments having the highest importance. The PRI had the worst predictive performance probably because the deep-oxidation state of the xanthophyll cycle pigments is strongly influenced by temporal changes. The model developed in this paper for monitoring photosynthetic performance parameters in grape leaves can simplify the complex photosynthetic reaction process, expand the application of PAM technology and provide a method for rapid and accurate monitoring of photosynthetic performance.

Automated summary

Photosynthesis is a critical indicator for predicting crop yield and quality, and accurately monitoring its dynamics is of great importance in field management. This study developed a Bayesian neural network model to predict photosynthetic performance parameters in grape leaves by quantifying spectral response indices of photosynthetic pigments and water status. The results showed that the developed model had better predictive performance compared to other models, and it could simplify the complex photosynthetic reaction process and provide a rapid and accurate method for monitoring photosynthetic performance.