Evaluation of photosynthesis estimation from machine learning-based solar-induced chlorophyll fluorescence downscaling from canopy to leaf level

Solar-induced chlorophyll fluorescence (SIF) is strongly correlated with gross primary productivity (GPP). Satellite-observed canopy SIF (SIFobs) captures only a part of the total leaf-emitted SIF (SIFtotal); therefore, SIFobs may hinder the interpretation of the physiological mechanism for GPP estimation. Furthermore, there are still significant discrepancies in the estimated SIFobs escape ratio (fesc) from the canopy to the leaf level with current methods. Here, we selected several vegetation canopy variables and downscaled SIFobs based on the Soil Canopy Observation, Photochemistry and Energy fluxes (SCOPE) model from the canopy to the leaf level using machine learning (ML) algorithms and then applied our method to the TROPOspheric Monitoring Instrument (TROPOMI) near-infrared (NIR) SIFobs. The results showed that simulating the fesc with SIFobs, TROPOMI NIR reflectance, and the fraction of photosynthetically active radiation (FPAR) avoided the effects of different sun-sensor geometry conditions introduced by different sensors and was more suitable for satellite-observed SIFobs downscaling. Our downscaled SIFtotal also correlated well with the flux site GPP in areas with sparse vegetation types. SIFtotal better reflected the photosynthetic differences among vegetation types and showed an enhanced relationship with absorbed photosynthetically active radiation (APAR) compared with SIFobs. We provide an efficient canopy-toleaf SIFobs downscaling method improved SIFtotaland GPP estimation, and our results also demonstrated the potential for using SIFobs as vegetation information in sparse coverage areas.