Data-driven Prediction of Species-Specific Plant Responses to Spectral-Shifting Films from Leaf Phenotypic and Photosynthetic Traits

TL;DR

This work tackles the problem that spectral-shifting films produce species- and cultivar-specific yield responses in crops. It combines leaf phenotypic and photosynthetic traits with daily light integral, using a variational autoencoder to augment a small experimental dataset and train a feedforward neural network that can predict whether SF will significantly enhance yield for unseen crops, achieving 91.4% test accuracy and 0.97 AUC. SHAP analysis reveals key predictors such as SPAD, leaf thickness (LMA), chlorophyll content, DLI, and LSPO, linking leaf physiology and light environment to SF efficacy. The approach provides a scalable, data-driven tool to guide SF deployment in controlled environments, potentially reducing trial-and-error trials and enabling tailored spectral management across diverse crops.

Abstract

The application of spectral-shifting films in greenhouses to shift green light to red light has shown variable growth responses across crop species. However, the yield enhancement of crops under altered light quality is related to the collective effects of the specific biophysical characteristics of each species. Considering only one attribute of a crop has limitations in understanding the relationship between sunlight quality adjustments and crop growth performance. Therefore, this study aims to comprehensively link multiple plant phenotypic traits and daily light integral considering the physiological responses of crops to their growth outcomes under SF using artificial intelligence. Between 2021 and 2024, various leafy, fruiting, and root crops were grown in greenhouses covered with either PEF or SF, and leaf reflectance, leaf mass per area, chlorophyll content, daily light integral, and light saturation point were measured from the plants cultivated in each condition. 210 data points were collected, but there was insufficient data to train deep learning models, so a variational autoencoder was used for data augmentation. Most crop yields showed an average increase of 22.5% under SF. These data were used to train several models, including logistic regression, decision tree, random forest, XGBoost, and feedforward neural network (FFNN), aiming to binary classify whether there was a significant effect on yield with SF application. The FFNN achieved a high classification accuracy of 91.4% on a test dataset that was not used for training. This study provide insight into the complex interactions between leaf phenotypic and photosynthetic traits, environmental conditions, and solar spectral components by improving the ability to predict solar spectral shift effects using SF.

Figures (11)

• Figure 1: Graphical abstract of this study. The workflow is divided into three main steps. (1) Data acquisition: crop growth, environmental data, and leaf reflectance spectra (400–750 nm) were collected from 24 different species and cultivars grown in a greenhouse covered with a spectral-shifting (SF) film. (2) Data augment & model training: to overcome the small dataset size (n=210), a variational autoencoder (VAE) was trained to generate a large augmented dataset (n=6400). This combined dataset was then used to train a feedforward neural network. (3) Inference: the final model predicts the binary (Yes/No) yield enhancement response to SF with 91.4% accuracy. Furthermore, model interpretation using SHAP identified significant features for the prediction.
• Figure 2: Comparison of relative transmittance between polyethylene film (PEF) and spectral-shifting film (SF).
• Figure 3: Schematic diagram of experimental procedure to develop the binary classification model of this study.
• Figure 4: Visual representation of a variational autoencoder for data augment.
• Figure 5: Relative yield of crops grown at polyethylene film (PEF) or no cover versus spectral-shifting film (SF). Bars without error bars are average. Data are Means$\pm$S.D. The asterisks indicate significant differences (Student's t-test, $^*$P-value $<$ 0.05). zRice was set to open-field as a control.