On the potential of principal component analysis for the reconstruction of full-spectrum SIF emission and emulated airborne-to-satellite upscaling

Abstract

Solar-induced fluorescence (SIF) emitted by plants as a byproduct of photosynthesis provides critical insights into vegetation health and climate regulation. However, detecting the weak SIF signal from small telluric oxygen absorption features remains challenging. ESA’s upcoming Fluorescence Explorer (FLEX) mission will retrieve full-spectrum SIF data at 300 m spatial resolution. In the meantime, we propose an alternative approach to reconstruct full-spectrum SIF from $O_{2}A$ and $O_{2}B$ bands using Principal Component Analysis (PCA) and the Soil Canopy Observation, Photochemistry, and Energy fluxes (SCOPE) model. Based on 100,000 SCOPE simulations (640–850 nm at 1 nm resolution), the SIF signals in the $O_{2}A$ (760 nm) and $O_{2}B$ (687 nm) bands showed high correlations with adjacent spectral regions and the full spectrum ($R^2>0.89$). From this data, we derived linear regression functions linking SIF at the $O_{2}A$ (760 nm) and $O_{2}B$ (687 nm) bands to the first two principal components (PCs), enabling inverse PCA transformation to reconstruct full-spectrum SIF with $R^2>0.98$ and RMSE $<0.12{\text{mW m}}^{-2}{\text{nm}}^{-1}{\text{sr}}^{-1}$. Applying the functions to HyPlant airborne SIF maps (1.7 m resolution) in northeastern Spain, and subsequent PC transformation, successfully reconstructed full-spectrum SIF including peaks and total emitted flux ($SI{F}_{Tot}$) with propagated uncertainties. To transfer this airborne full-spectrum SIF data to the satellite scale, we then trained an emulator with PRecursore IperSpectrale de la Missione Applicativa (PRISMA) Bottom of Atmosphere (BOA) reflectance spectra as input to produce spaceborne synthetic full-spectral SIF maps at 30 m resolution, and resampled to the nominal 300 m FLEX resolution. Despite PRISMA’s lower spectral resolution for SIF retrieval, the emulator reliably produced full-spectrum SIF with $R^2$ of 0.69 and 0.52 for $SI{F}_{760}$ and $SI{F}_{687}$ bands, enabling FLEX-like SIF products (e.g., peaks, $SI{F}_{Tot}$). This reconstruction and upscaling approach demonstrates its utility for generating FLEX-compatible SIF datasets, supporting FLEX’ mission preparation and cal/val activities.