Extracting fluorescence efficiency from TROPOMI satellite observations: is it better to work on individual observations before gridding into data cubes?

Solar-induced chlorophyll fluorescence (SIF) is a plant signal that can currently be retrieved from satellites at regional to global scale. Since SIF originates from the pool of excitation energy absorbed by chlorophyll molecules that also provides the energy for the photosynthetic CO2 assimilation, it has potential for diagnosing vegetation stress, particularly before the stress becomes apparent by optical decreases in greenness. However, the interpretation of satellite-observed SIF (SIF_obs) remains challenging because it integrates multiple confounding factors beyond plant physiology, including variations in illumination conditions, canopy structure, and observation geometry. For applications aiming to detect early stress signals, it is essential to disentangle the physiological component, i.e., fluorescence efficiency (ΦF). SIF_obs are strongly influenced by illumination conditions which change with actual differences in overpass times that can occur from one day to the next. Also, the canopy structure determines the fraction (f_esc) of fluorescence that escapes the canopy to the sensor. Consequently, SIF_obs are affected by the spatial heterogeneity of vegetation elements within the satellite footprint. A common practice to account for these effects is to apply corrections after multiple instantaneous SIF_obs have been aggregated onto a regular grid of a geographic coordinate system, which may underestimate the uncertainty from spatio-temporal mismatches. We propose that these procedures should be applied prior to spatial gridding to ensure they are done over the correct spatio-temporal supports. We hypothesize that doing so will ensure consistency within the same support of all contributing variables and reduce uncertainties arising from spatial and temporal mismatches.

Here we derive ΦF from TROPOMI observations by normalizing SIF_obs with radiation and canopy features prior to  gridding. We normalize SIF_obs by photosynthetically active radiation (PAR) and near-infrared reflectance of vegetation (NIRv), where NIRv serves as a proxy for canopy structure and vegetation greenness status. We explore ΦF using multi-source NIRv and PAR datasets in combination with TROPOMI SIF from three independent retrieval products. PAR is approximated using downward shortwave radiation products with multiple spatio-temporal resolutions (e.g., MSG, ERA5, TROPOMI estimation of radiance). NIRv, derived from other sources (e.g., MODIS, Sentinel-3, and Sentinel-2), is aggregated to the TROPOMI footprint and compared against the native TROPOMI  top-of-atmosphere reflectance product. To evaluate the performance of ΦF derived at the individual footprint level, we compare it against flux tower observations from the Austro-SIF dataset. Austro-SIF is a fluorescence-specific dataset that integrates both active and passive measurement approaches from multiple European sites collected over different time periods between 2018 and 2022. It combines meteorological data with photosynthetic measurements of vegetation at both leaf and canopy scales, capturing comprehensive ecosystem responses to environmental variation. Using this dataset, we further assess the cross-scale consistency and uncertainty of ΦF across ecosystems spanning diverse biomes.