Uncertainty analysis of SVD-based spaceborne far–red sun-induced chlorophyll fluorescence retrieval using TanSat satellite data

Abstract

The singular value decomposition (SVD) method, a sun-induced chlorophyll fluorescence (SIF) retrieval approach, has been used widely in far-red SIF spaceborne retrievals on a global scale. However, due to its semi-empirical nature, setting different parameter values may affect its retrieval accuracy, and ultimately have a large impact on its application. Hence, in this study, we evaluated the impact of parameter selection on the far-red SIF retrieval of this approach using TanSat satellite data. We first retrieved the far-red SIF within a narrow spectral window of 757.4–759.2 nm using the first four singular vectors (SVs) that were derived from the snow and soil spectra in a globally distributed training set. The retrievals are highly consistent with TanSat mission SIF (R2 = 0.72), indicating the reliability of our retrieval results. Then, the uncertainty was executed based on the above SIF retrievals and evaluated from five metrics: the number of SVs, length of the fitting window, type of training set elements, proportion of training set elements, and spatial distribution of the training set. Results showed that an unwise selection of the number of SVs would result in large retrieval errors (R2 = 0.03). Meanwhile, a fitting window that is too short or does not include a strong Fraunhofer line would cause severe errors and a large number of negative values (R2 = 0.06). Further, failure to include training samples below the equatorial zone in the Southern Hemisphere, and high latitude samples in the Northern Hemisphere, leads to poorer outcomes (R2 = 0.57). In contrast, the other global sampling metrics had little effect on the retrieval results (generally, R2 > 0.97). Accordingly, the relevant suggestions for using this method in the future are listed in this study for reference, with the potential to improve the accuracy of SIF retrievals from ultra-high spectral resolution instruments in the far-red spectrum.