Systematic Assessment of Retrieval Methods for Canopy Far-Red Solar-Induced Chlorophyll Fluorescence Using High-Frequency Automated Field Spectroscopy

Remote sensing of solar-induced chlorophyll fluorescence (SIF) offers potential to infer photosynthesis across scales and biomes. Many retrieval methods have been developed to estimate top-of-canopy SIF using ground-based spectroscopy. However, inconsistencies among methods may confound interpretation of SIF dynamics, eco-physiological/environmental drivers, and its relationship with photosynthesis. Using high temporal- and spectral resolution ground-based spectroscopy, we aimed to (1) evaluate performance of SIF retrieval methods under diverse sky conditions using continuous field measurements; (2) assess method sensitivity to fluctuating light, reflectance, and fluorescence emission spectra; and (3) inform users for optimal ground-based SIF retrieval. Analysis included field measurements from bi-hemispherical and hemispherical-conical systems and synthetic upwelling radiance constructed from measured downwelling radiance, simulated reflectance, and simulated fluorescence for benchmarking. Fraunhofer-based differential optical absorption spectroscopy (DOAS) and singular vector decomposition (SVD) retrievals exhibit convergent SIF-PAR relationships and diurnal consistency across different sky conditions, while O(2)A-based spectral fitting method (SFM), SVD, and modified Fraunhofer line discrimination (3FLD) exhibit divergent SIF-PAR relationships across sky conditions. Such behavior holds across system configurations, though hemispherical-conical systems diverge less across sky conditions. O(2)A retrieval accuracy, influenced by atmospheric distortion, improves with a narrower fitting window and when training SVD with temporally local spectra. This may impact SIF-photosynthesis relationships interpreted by previous studies using O(2)A-based retrievals with standard (759-767.76 nm) fitting windows. Fraunhofer-based retrievals resist atmospheric impacts but are noisier and more sensitive to assumed SIF spectral shape than O(2)A-based retrievals. We recommend SVD or SFM using reduced fitting window (759.5-761.5 nm) for robust far-red SIF retrievals across sky conditions.