Fluorescence Correction Vegetation Index (FCVI): A physically based reflectance index to separate physiological and non-physiological information in far-red sun-induced chlorophyll fluorescence

Highlights

• We propose an index (FCVI) for the effects of physical processes on far-red SIF.

• FCVI is the difference between near-infrared and broadband visible reflectance.

• Normalizing SIF by FCVI and PAR is an estimate of fluorescence emission efficiency.

• FCVI was tested with a field measurement and a numerical experiment.

Abstract

Sun-induced chlorophyll fluorescence (SIF) has been used to track vegetation photosynthetic activity for improving estimation of gross primary productivity (GPP) and detecting plant stress. There are both physical and physiological controls of SIF measured at the surface and retrieved from remote sensing including satellite observations. In order to accurately use SIF for monitoring of plant physiology, the effects of physically-based radiation processes related to leaf and canopy structure, notably photosynthetically active radiation (PAR) absorption and SIF scattering and re-absorption, must be characterized. In this study, we investigate both PAR absorption and SIF scattering processes and find that although it is difficult to quantify their effects individually by using just reflectance, the combined effects of the two processes can be well approximated by a reflectance index. This index, referred to as FCVI (Fluorescence Correction Vegetation Index), is defined as the difference between near-infrared (NIR) and broad-band visible (VIS, 400–700 nm) reflectance acquired under identical sun-canopy-observer geometry of the SIF measurements. The development of the index was based on the physical connection between reflectance and far-red SIF, which was revealed by using the spectral invariant theory. The utility of FCVI to correct far-red SIF for PAR absorption and scattering effects, thus improving the link to photosynthesis, was tested with data from: (i) a field experiment for a growing season; and (ii) a numerical experiment which included a number of scenarios generated by a radiative transfer model. For both the observations and simulations, the FCVI provided a promising estimate of the impact of the physically-based radiation processes on far-red SIF of moderately dense canopies (i.e., FCVI ≥ 0.18). Normalizing the TOC far-red SIF by both the incident PAR (iPAR) and the FCVI provided a good estimate of the far-red fluorescence emission efficiency of the canopies examined. This approach enhances our ability to generalize retrievals for vegetation processes as they change through natural growth phases and seasons. Taken together, far-red SIF and FCVI may enable the assessment of the light partitioning of vegetation canopies, an essential step to facilitate the use of far-red SIF for tracking physiological processes.

Introduction

Remote sensing of sun-induced chlorophyll fluorescence (SIF) is becoming a frequently used technique in plant physiology monitoring as a non-invasive measurement of canopy photosynthetic activities. Top-of-canopy (TOC) SIF is the final outcome of three sequential processes: absorption of sunlight by chlorophyll followed by fluorescence emission by photosystems, and both re-absorption and scattering after this emission. Among these processes, the emission of leaf-level fluorescence is regulated by a number of physiological mechanisms and describes the partitioning of photosynthetically active radiation (PAR) into photosynthesis, heat dissipation and fluorescence and by using pulse-amplitude-modulation (PAM) techniques one can derive the partitioning (Baker, 2008, Maxwell and Johnson, 2000, Middleton et al., 2018). Because of the sensitivity of TOC SIF to physiological processes, SIF observed with remote sensing tools has been used to infer photosynthetic capacity (Zhang et al., 2014), improve gross primary production (GPP) estimation (Campbell et al., 2019, Guanter et al., 2014, Migliavacca et al., 2017), reveal vegetation stress (Ač et al., 2015, Rossini et al., 2015), and estimate transpiration (Lu et al., 2018, Shan et al., 2019).

The potential of SIF for plant physiology monitoring has not been fully explored. A substantial portion of the SIF variations observed at different spatial and temporal scales are due to variations in vegetation biochemical constituents as well as leaf and canopy structure, rather than changes in plant physiology (Migliavacca et al., 2017, Van der Tol et al., 2016). Consequently, extracting physiological information from SIF presents a challenge because the SIF signal suffers from the interferences caused by PAR absorption and SIF scattering into the viewing direction, which are controlled by physical factors, such as soil background, leaf biochemical constituents, canopy structure and sun-observer geometry. Prior investigations have shown that both TOC SIF and GPP are largely explained by vegetation structure (Badgley et al., 2017, Badgley et al., 2018) and the positive correlation between SIF and GPP can be dominated by light absorption process rather than the functional link at photosynthetic level (Miao et al., 2018, Yang et al., 2018).

One way to examine the effects of PAR absorption and SIF scattering on TOC SIF is to employ radiative transfer models (RTMs), which explicitly simulate light interactions with canopies based on physical laws, and link soil and vegetation properties with TOC SIF. RTMs have provided valuable tools for understanding of the sensitivity of SIF to its controls. For example, Verrelst et al. (2015) carried out a sensitivity analysis with the 1-D RTM SCOPE (Soil Canopy Observation, Photochemistry and Energy fluxes, Van der Tol et al., 2009) and ranked the effects of canopy structure, leaf pigment content and photosynthetic capacity on TOC SIF at different wavelengths. Similarly, Hernández-Clemente et al. (2017) used a 3-D RTM to successfully estimate the structural effects of forest canopy on TOC SIF measurements and demonstrated the significance of these effects. Furthermore, Celesti et al. (2018) used SCOPE to explore physiological information related to SIF and found that fluorescence emission efficiency increased when photosynthesis was inhibited. Despite the powerful capabilities of RTMs, their usage in quantifying the impact of physiological and physical processes may not be practical in many cases, since RTMs require prior information about canopy structure and leaf properties which are often unknown (Porcar Castell et al., 2014).

Reflectance provides valuable information for interpreting SIF measurements. In two earlier studies, we used reflectance to estimate leaf and canopy properties by model inversion (Van der Tol et al., 2016, Yang et al., 2019). After these properties were estimated, their effects on TOC SIF were evaluated through characterizing the light absorption and SIF scattering processes, predicted with RTMs. However, the estimation of vegetation canopy parameters by model inversion requires multi-spectral or preferably hyperspectral reflectance rather than reflectance at two or three bands (e.g., reflectance indices) (Verrelst et al., 2019). Both the model inversion and simulation of SIF radiative transfer are computationally expensive and have limits in large-scale applications. Apart from these limitations, the model inversion is typically ill-posed and there are uncertainties in the estimated vegetation properties, which may be propagated to the estimated effects of the light absorption and SIF scattering processes (Yang et al., 2019).

In contrast to the explicit RTMs, a simple light use efficiency (LUE) model has been formulated to incorporate the effects of the three processes acting on TOC SIF. The formulation takes the form of the LUE concept in GPP estimation (Monteith, 1972), which defines the intensity and spectral properties of a TOC SIF signal observed by a remote sensor as:

$$F_{toc}=\text{iPAR}\times {\text{fAPAR}}_{chl}\times {ϵ}_F\times {\sigma }_F$$

where iPAR denotes the available incident PAR for a canopy, fAPAR_chl is the fraction of absorbed PAR (APAR) by chlorophyll, which is often approximated by fAPAR (i.e., the fraction of PAR absorbed by the entire canopy not just chlorophyll). ϵ_F is canopy SIF emission efficiency and σ_F is the scattering of SIF in the viewing direction, also known as fluorescence escape probability p_esc or escape fraction f_esc (Guanter et al., 2014) (i.e., σ_F = f_esc = p_esc and 0 ≤ σ_F ≤ 1 due to re-absorption). In this expression, the impacts of PAR absorption, SIF emission and SIF scattering are quantified by fAPAR_chl (or fAPAR), ϵ_F and σ_F, respectively.

Based on Eq. (1), a number of practical approaches to correct for the impact of the physical processes on SIF have been explored. Normalizing TOC far-red SIF by canopy APAR (i.e., the product of fAPAR and iPAR) has been commonly used to account for the variation in the light availability and absorption process to allow the exploration of the link between the SIF/APAR ratio and photosynthesis, an approach that has met with varying success. For example, at a local scale, Yang et al. (2015) measured APAR of a temperate deciduous forest by using a set of quantum sensors and found that the far-red SIF/APAR ratio was positively correlated with photosynthetic LUE, while Miao et al. (2018) found that this ratio was negatively correlated with photosynthetic LUE of a soybean field. Wieneke et al. (2018) reported that more than 50% of the observed diurnal and seasonal variation of photosynthetic LUE can be explained by this ratio, but they also found an even stronger relationship between the far-red SIF/APAR ratio and a structural vegetation index. Yoshida et al. (2015) computed APAR from MODIS fAPAR products and found GOME-2 SIF products normalized by APAR clearly declined over the region impacted by the 2010 Russian heat wave, while Wohlfahrt et al. (2018) showed that SIF had limited potential for quantitatively monitoring photosynthesis during heat waves in the absence of large changes in APAR (i.e., variations due to the efficiency ϵ_F). The inconsistent findings between the far-red SIF/APAR ratio and photosynthetic LUE are at least partly caused by the fact that normalizing SIF measurements for APAR results in a signal that is still contaminated by canopy structural effects due to the variations in SIF scattering. Without considering the scattering process, the ‘ pure’ physiological information (i.e., ϵ_F) cannot be completely separated from the structural and illumination effects. Our goal and challenge are to isolate the physiological parameter ϵ_F from fAPAR and σ_F (Eq. (1)) to obtain reliable physiological status information of vegetation.

Attempts to correct SIF for the effects of scattering (σ_F) to achieve a better physiological indicator have been made in recent years. He et al. (2017) performed angular normalization of TOC SIF and partially corrected the variation of σ_F caused by different viewing angles. They found that such normalization provided a better proxy of GPP as compared to the use of SIF uncorrected for angular scattering. Recently, we linked σ_F of far-red SIF with TOC near-infrared (NIR) reflectance by comparing the radiative transfer of incident radiation and emitted SIF (Yang and Van der Tol, 2018). We found that for dense vegetation canopies, σ_F was proportional to NIR reflectance and to the reciprocal of canopy interceptance (i₀) (i.e., σ_F = R_nir/i₀), which is the portion of the incident photons that interact with the canopy (see Smolander and Stenberg, 2005, and Section 2.1) . Liu et al. (2018) utilized this link and estimated σ_F from TOC reflectance by using a machine learning approach. By normalizing the TOC SIF by this estimated value for σ_F, they obtained an estimate for canopy total emitted far-red SIF. Although there is a strong connection between TOC reflectance and far-red SIF, the estimation of σ_F from reflectance only is still to be resolved because the estimation requires canopy interceptance knowledge apart from TOC reflectance (Yang and Van der Tol, 2018).

Zeng et al. (2019) extended the relationship between σ_F and NIR reflectance to sparse vegetation canopies and proposed to use the NDVI (normalized difference vegetation index, Rouse et al., 1973) to approximate the contribution of ‘ pure’ vegetation signals. They derived σ_F as the ratio of the product of NIR reflectance and NDVI to i₀ (i.e., σ_F = R_nir ×NDVI/i₀). Furthermore, Zeng et al. (2019) proposed to use fAPAR to approximate i₀ for practical usage and obtained σ_F = R_nir ×NDVI/fAPAR. Using this expression of scattering of far-red SIF, we can estimate the product of fAPAR and σ_F as R_nir ×NDVI straightforwardly. A limitation of this approach is that a number of steps in the derivation for σ_F are not fully consistent with radiative transfer theory, and therefore are not universally valid. For example, 1) the use of NDVI as a measure for ‘ pure’ vegetation signals could be debated as NDVI is still dependent on the soil spectrum to some degree, 2) fluorescence excited by scattered radiation is not considered, and 3) using fAPAR for the spectral invariant i₀ relies on the assumption that both soil and leaves are perfectly absorbant (i.e., leaf albedo and soil reflectance are both zero). Furthermore, the accurate estimation of fAPAR is difficult and thus additional errors may be introduced in the estimation of σ_F when implementing that approach in practice.

Direct fAPAR or APAR measurements are not commonly available, especially for global applications. Most remote sensing fAPAR products are derived from the canopy reflectance signal, but the use of canopy reflected signals to estimate canopy absorbed signals (i.e., fAPAR or APAR) is an ill-posed problem as the relationship between fAPAR and reflectance is regulated by canopy transmittance and soil absorptance. For example, the use of reflectance indices (e.g., EVI and NDVI) to estimate fAPAR (preferably fAPAR_chl) is possible (Viña and Gitelson, 2005), even globally (Huete et al., 2002, Myneni et al., 2002), but the empirical coefficients for the reflectance index approach are limited to specific cases (e.g., a regional area and/or at a specific period).

In this study, we evaluate the possibility to link fAPAR to TOC reflectance analytically, since far-red σ_F has already been related to TOC NIR reflectance in our previous study (Yang and Van der Tol, 2018). We demonstrate that although it is difficult to estimate fAPAR and far-red σ_F individually by using just TOC reflectance, their product can be well approximated by a reflectance index. The new reflectance index (i.e., Fluorescence Correction Vegetation Index, FCVI) is given as the difference of NIR and broadband visible (VIS) reflectance acquired under identical sun-canopy-observer geometry of the SIF measurements. In what follows, we develop the theoretical basis for FCVI, which is based mainly on the spectral invariant radiative transfer theory (Huang et al., 2007, Knyazikhin et al., 2011, Lewis and Disney, 2007, Stenberg et al., 2016). Further, we evaluate the utilities of the FCVI by using field measurements of a corn canopy and simulations from the RTM SCOPE. Our objective is to improve estimation performance for the physiological information captured with far-red SIF, by reducing the influence of physical processes that contaminate the observed SIF signal.