Simulation framework for solar-induced fluorescence retrieval and application to DESIS and HyPlant

Fluorescence light emitted by chlorophyll in plants is a direct probe of the photosynthetic process and can be used to continuously monitor vegetation status. Retrieving solar-induced fluorescence (SIF) using a machine learning (ML) approach promises to take full advantage of airborne and satellite-based instruments to map expected vegetation function over wide areas on a regular basis. This work takes a first step towards developing a ML-based SIF retrieval method. A general-purpose framework for the simulation of at-sensor radiances is introduced and applied to the case of SIF retrieval in the oxygen absorption band O${}_2$-A with the spaceborne DESIS and airborne HyPlant spectrometers. The sensor characteristics are modelled carefully based on calibration and in-flight data and can be extended to other instruments including the upcoming FLEX mission. A comprehensive dataset of simulated at-sensor radiance spectra is then assembled encompassing the most important atmosphere, geometry, surface and sensor properties. The simulated dataset is employed to train emulators capable of generating at-sensor radiances with sub-percent errors in tens of $\mu s$, opening the way for their routine use in SIF retrieval. The simulated spectra are shown to closely reproduce real data acquired by DESIS and HyPlant and can ultimately be used to develop a robust ML-based SIF retrieval scheme for these and other remote sensing spectrometers. Finally, the SIF retrieval performance of the 3FLD method is quantitatively assessed for different on- and off-band configurations in order to identify the best band combinations. This highlights how our simulation framework enables the optimization of SIF retrieval methods to achieve the best possible performance for a given instrument.