Improving GOME-2C red solar-induced chlorophyll fluorescence retrieval by denoising on radiance spectra

Solar-induced chlorophyll fluorescence (SIF) serves as a direct indicator of vegetation photosynthetic activity, with red SIF (RSIF) originating from photosystem II and providing particularly direct physiological insights. However, RSIF retrieval remains challenging because the red spectral region is characterized by weaker Fraunhofer absorption features and lower signal-to-noise ratios (SNR) than the far-red region. Previous efforts to improve RSIF retrieval have mainly focused on retrieval algorithms and spectral fitting strategies, while the role of radiance-level denoising on retrieval performance has received comparatively little attention. To address this issue, we introduce a low-rank matrix recovery (LRMR)-based denoising framework for enhanced RSIF retrieval from GOME-2C radiance observations. The LRMR method exploits the intrinsic low-rank spatial–spectral structure of hyperspectral imagery to separate signal from noise while preserving spectral features required for fluorescence retrieval. The denoised radiance spectra were subsequently used for RSIF retrieval within the 669–686 nm spectral window. Results indicate that LRMR denoising substantially suppresses radiance noise without noticeably altering spectral shapes, with mean radiance changes below 1%. The retrieved RSIF (RSIFLRMR) exhibits clear spatial patterns, with mean retrieval errors reduced from ~0.3 mW m−2 sr−1 nm−1 (from original radiance, RSIForiginal) to below 0.01 mW m−2 sr−1 nm−1. RSIFLRMR shows strong consistency with independent datasets related to vegetation photosynthetic physiology, achieving monthly correlation coefficients of 0.82 with FLUXCOM-X GPP, 0.58 with GOME-2 FRSIF, and 0.45 with TROPOMI RSIF—markedly surpassing the values of 0.21, 0.13, and 0.12, respectively, obtained from RSIForiginal. This study highlights the critical role of radiance-level denoising in alleviating noise in satellite-based RSIF retrieval, providing a readily implementable pathway to generate high-quality global RSIF products within red spectral windows for the moderate spectral resolution satellites, such as TROPOMI, FLEX, and TECIS-1.