Introducing the spectral characteristics index: A novel method for clustering solar radiation fluctuations from a plant-ecophysiological perspective

Solar radiation (SR) is a critical environmental factor influencing plant ecophysiology and ecosystem dynamics, not merely as an energy source but through its spectral characteristics, including critical wavelength ratios (CWRs) that trigger photomorphogenic responses in plants, the diffuse fraction (DF), that affect light distribution within canopies, and the variability of SR. This study presents the Spectral Characteristics Index (SCI), a novel method that integrates spectral quality and energy flux to classify daily SR conditions. Data were collected using a rotating shadow-band spectroradiometer. The study applied agglomerative hierarchical clustering (AHC) based on cumulative Euclidean distance matrices and identified five SR clusters ranging from clear (SCI-01) to overcast (SCI-05) conditions, with spectral shifts from red to blue. Significant differences in DF, global solar irradiance (GSI), and CWRs were observed across clusters ( p < 0.0001, F > 27). Given the challenges in obtaining comprehensive spectral data in certain regions, machine learning models replicated SCI clustering using easily accessible environmental variables (DF, GSI, variability, airmass, and vapor pressure). The support vector machine (SVM) model achieved 88.03 % validation accuracy and 94.29 % test accuracy, providing a practical alternative where spectral measurements are not available. While long-term data collection across various climatic zones could improve the validity and adaptability of SCIs to different geographical locations, the current model demonstrates high accuracy and efficiency. This innovative approach enhances the understanding of SR dynamics and advances ecological research on plant responses and ecosystem functions. • A method for clustering daily solar radiation properties using both spectral and energy data. • Analysis of all-weather observations from a new rotating shadowband spectroradiometer. • Five distinct categories were identified, ranging from clear to cloudy. • Even days with similar energy flux conditions were classified with different spectra. • New machine learning model able to reproduce clustering from simple observational data.