Improving Full-Spectrum Reconstruction of Solar-Induced Chlorophyll Fluorescence Using Principal Component Analysis Based on an SIF-Free Leaf Reflectance Dataset

Solar-Induced Chlorophyll Fluorescence (SIF), spanning 650–800 nm, is a powerful indicator of plant vitality and productivity. The full SIF spectrum offers comprehensive insights into plant photosynthetic dynamics, surpassing single-band SIF. While current full-spectrum SIF reconstruction algorithms often rely on mathematical functions or basis spectra (principal components or singular vectors) derived from simulated datasets to model reflectance shapes. These methods struggle to accurately capture reflectance, especially in the red and red-edge regions, leading to significant retrieval errors. To address this limitation, we developed an improved full-spectrum SIF reconstruction algorithm using Principal Component Analysis (PCA) based on an SIF-free reflectance dataset (FSIF-iPCA). The SIF-free leaf spectra, obtained using an ASD field spectrometer with a FluoWat leaf clip and long-pass filter and under clear-sky condition, were upscaled to the canopy level using the SCOPE model to create a new SIF-free canopy reflectance dataset. FSIF-iPCA algorithm derives the principal components (PCs) through PCA optimization on this SIF-free reflectance dataset, thereby minimizing reflectance reconstruction artifacts induced by conventional simulated training data. Finally, validation experiments using field leaf datasets and tower-based canopy measurements demonstrated that FSIF-iPCA significantly outperformed previous methods. At the leaf level, FSIF-iPCA achieved an average RRMSE of 11.20%, with RRMSEs of 5.67% and 5.93% and R2values of 0.989 and 0.987 for the O2-A and O2-B bands, respectively. Tower-based measurements confirmed its superior performance, with O2-A band RRMSEs of 3.56% (DM site) and 2.49% (GC site), and O2-B band values of 13.98% and 10.47%. Compared to other methods, FSIF-iPCA consistently delivered higher retrieval accuracy, particularly in the red and red-edge regions, demonstrating the efficacy of PCs derived from SIF-free reflectance dataset for improving SIF spectral reconstruction.