Abstract. Data from satellite, aircraft, drone, and ground-based measurements have already shown that canopy-scale sun-induced chlorophyll fluorescence (SIF) is tightly related to photosynthesis, which is linked to vegetation carbon assimilation. However, our ability to effectively use those findings are hindered by confounding factors, including canopy structure, fluctuations in solar radiation, and sun–canopy geometry that highly affect the SIF signal. Thus, disentangling these factors has become paramount in order to use SIF for monitoring vegetation functioning at the canopy scale and beyond. Active chlorophyll fluorescence measurements (FyieldLIF), which directly measures the apparent fluorescence yield, have been widely used to detect physiological variation of the vegetation at the leaf scale. Recently, the measurement of FyieldLIF has become feasible at the canopy scale, opening up new opportunities to decouple structural, biophysical, and physiological components of SIF at the canopy scale. In this study, based on top-of-canopy measurements above a mature deciduous forest, reflectance (R), SIF, SIF normalized by incoming photosynthetically active radiation (SIFy), FyieldLIF, and the ratio between SIFy and FyieldLIF (named Φk) were used to investigate the effects of canopy structure and shadows on the diurnal and seasonal dynamics of SIF. Further, random forest (RF) models were also used to not only predict FyieldLIF and Φk, but also provide an interpretation framework by considering additional variables, including the R in the blue, red, green, red-edge, and near-infrared bands; SIF; SIFy; and solar zenith angle (SZA) and solar azimuth angle (SAA). Results revealed that the SIF signal is highly affected by the canopy structure and sun–canopy geometry effects compared to FyieldLIF. This was evidenced by the weak correlations obtained between SIFy and FyieldLIF at the diurnal timescale. Furthermore, the daily mean SIF‾y captured the seasonal dynamics of daily mean F‾yieldLIF and explained 58 % of its variability. The findings also revealed that reflectance in the near-infrared (R-NIR) and the NIRv (the product of R-NIR and normalized difference vegetation index (NDVI)) are good proxies of Φk at the diurnal timescale, while their correlations with Φk decrease at the seasonal timescale. With FyieldLIF and Φk as outputs and the abovementioned variables as predictors, this study also showed that the RF models can explain between 86 % and 90 % of FyieldLIF, as well as 60 % and 70 % of Φk variations under clear-sky conditions. In addition, the predictor importance estimates for FyieldLIF RF models revealed that R at 410, 665, 740, and 830 nm; SIF; SIFy; SZA; and SAA emerged as the most useful and influential factors for predicting FyieldLIF, while R at 410, 665, 705, and 740 nm; SZA; and SAA are crucial for predicting Φk. This study highlighted the complexity of interpreting diurnal and seasonal dynamics of SIF in forest canopies. These dynamics are highly dependent on the complex interactions between the structure of the canopy, the vegetation biochemical properties, the illumination angles (SZA and SAA), and the light conditions (ratio of diffuse to direct solar radiation). However, such measurements are necessary to better separate the variability in SIF attributable to radiation and measurement conditions from the subtler variability attributable to plant physiological processes.
