SIFFI: Bayesian solar-induced fluorescence retrieval algorithm for remote sensing of vegetation

Abstract

Remote sensing of solar-induced vegetation chlorophyll fluorescence (SIF) has a rich history of more than 50 years of research covering active and passive techniques from leaf, canopy, and satellite scale. Current satellite-derived SIF products primarily focus on the far-red spectral range, with variations in techniques dependent on sensor capabilities. However, these retrieval methods often rely on parametric spectral models and are constrained to narrow absorption regions. In this paper, we introduce a novel Bayesian retrieval technique, referred to as SIFFI (Siffi Is For Fluorescence Inference), designed for the flexible and robust estimation of spectrally resolved fluorescence spectra. SIFFI utilizes spectral representations for both fluorescence and surface reflectance, enabling its application to distinct spectral ranges, e.g., red, far-red, and full spectral range. Also, its applicability extends to top-of-canopy (TOC) and top-of-atmosphere (TOA) measurements, with the latter being possible when auxiliary information about the atmospheric state is available. For the assessment of SIFFI, we performed an extensive proof-of-concept simulation exercise involving diverse scenarios that integrated measured leaf-level fluorescence and reflectance signals, propagated them to the TOC and TOA levels, and perturbed the resultant signal with instrument Gaussian noise to simulate realistic conditions. Additionally, we extend our assessment exercise to TOC measurements acquired by a fluorescence box (FloX) instrument during two diurnal cycles on sunlit and cloudy conditions. In all the TOC cases, simulations- and measured-based scenarios, we compared our SIF estimates with the results from two well-established methods: the improved Fraunhofer line discrimination method (iFLD) and the Spectral Fitting (SpecFit) method covering the full fluorescence spectra. Notably, our results highlight the versatility and accuracy of SIFFI in estimating spectrally resolved fluorescence, achieving Mean Absolute Error (MAE) values of 0.07 (0.09) $[\mathrm{mW}/\left({\mathrm{m}}^2\mathrm{sr}\mathrm{nm}\right)]$$[\mathrm{mW}/\left({\mathrm{m}}^2\mathrm{sr}\mathrm{nm}\right)]$ in the TOC (TOA) simulation scenarios, improving the SpecFit method estimates, and being aligned with the iFLD method results at the oxygen bands. SIFFI represents a significant advancement in SIF retrieval, providing a robust approach that exploits the full spectral information from the red to the near-infrared regions.