Prediction of Leaf Net Photosynthetic Rate in Greenhouse-Grown Zucchini Using a GA-Optimized Adaptive Neuro-Fuzzy Inference System

Abstract

Accurate prediction of the net photosynthetic rate (Pn) is important for understanding crop responses to environmental conditions in protected cultivation. The Pn responds nonlinearly to the temperature, photosynthetic photon flux density (PPFD), and CO2 concentration, making accurate prediction under combined environmental conditions challenging. However, these effects make it difficult for mechanistic models to precisely predict the Pn. In this study, 1800 repeated Pn records of zucchini leaves were obtained during flowering and fruiting stages and used to construct and evaluate a GA-optimized adaptive neuro-fuzzy inference system (GA-ANFIS), ANFIS, random forest (RF), and radial basis function (RBF), as well as backpropagation (BP) models. The maximum Pn values were 43.22 and 45.21 μmol·m−2·s−1 during flowering and fruiting, respectively, at 32 and 28 °C, a PPFD of 1800 μmol·m−2·s−1, and a CO2 concentration of 1200 μmol·mol−1. On the test set, the GA-ANFIS achieved R2 values of 0.9811 and 0.9901 and RMSE values of 1.4808 and 1.1889 μmol·m−2·s−1, respectively. It also achieved the highest R2 and lowest RMSE on the additional temperature interpolation set. The test set regression slopes were 0.9876 and 1.0001, with intercepts of 0.1060 and −0.0465. Overall, the GA-ANFIS showed the best predictive performance among the five models under the evaluated conditions.