Skip to content
Frontiers
Search

Original Research

Frontiers in Remote Sensing ·

Volume 7 - 2026 | https://doi.org/10.3389/frsen.2026.1819365

Hyperspectral indices and global fluorescence: how PACE vegetation indices correlate with terrestrial photosynthesis

  • Caleb S. Harmon1*corresponding author
  • K. Fred Huemmrich2
  • Randolph H. Wynne1
  • Valerie A. Thomas1
  • Rui Cheng3
  1. 1Forest Resources and Environmental Conservation, Virginia Polytechnic Institute and State University, Blacksburg, VA, United States
  2. 2GSTAR‐II, University of Maryland Baltimore County, Baltimore, MD, United States
  3. 3Kravis Department of Integrated Sciences, Claremont McKenna College, Claremont, CA, United States
Article metrics0 viewsView details
PDFCiteMetrics

Abstract

Solar-induced chlorophyll fluorescence (SIF) is a direct remotely sensed indicator of photosynthetic activity and plant physiological status. However, its application is limited by coarse spatial resolution and data gaps. The NASA PACE mission offers hyperspectral observations that may help address these limitations through its Ocean Color Instrument and associated Land Vegetation Index (LANDVI) products. This study evaluates the strength of the relationship between PACE-derived vegetation indices and TROPOMI SIF across diverse global biomes. Using 8-day global composites from 2024, we demonstrate strong spatial and temporal correlations between PACE vegetation indices and TROPOMI SIF across fifteen biome-stratified regions, including forests, grasslands, agricultural areas, and xeric landscapes. Even in the absence of photosynthetically active radiation (PAR) and albedo corrections, simple univariate linear models show that the Enhanced Vegetation Index (EVI) and Chlorophyll Index Red Edge (CIRE) explain a large fraction of SIF variance (R 2 = 0.80 and 0.77, respectively). The stability of these relationships across seasons highlights the key role of canopy structure and chlorophyll content in driving global SIF variability. Seasonal analyses further reveal that while EVI and CIRE perform robustly in most forests and agricultural systems, other moisture- and pigment-sensitive indices can provide better performance during dry seasons in water-limited ecosystems. Spatial residual analyses indicate minimal global bias, though systematic deviations occur in certain regions (e.g., boreal forests and parts of the tropics), consistent with known effects of canopy architecture and fluorescence escape probability. These strong correlations suggest that PACE vegetation indices capture key biophysical drivers of SIF and point to the potential for improved SIF prediction and downscaling in future studies, particularly when combined with additional variables such as PAR corrections, albedo, or escape probability factors. Given the availability of EVI from moderate-resolution sensors (e.g., Sentinel-2 and Landsat), such relationships also offer promising avenues for bridging coarse-resolution SIF to finer management-relevant scales.

Keywords

1 Introduction

Terrestrial ecosystems are critical components of the global carbon cycle, with photosynthetic activity serving as a fundamental measure of ecosystem productivity and resilience. SIF is a faint electromagnetic radiation emitted by chlorophyll molecules during photosynthesis, typically within the 650–850 nm spectral range, as a byproduct of absorbed light energy not used for photochemistry or dissipated as heat (Porcar-Castell et al., 2014). Because SIF is directly linked to photosynthetic electron transport and light use efficiency (LUE), it provides a mechanistic proxy for gross primary productivity (GPP), the total carbon fixed by vegetation (Li et al., 2022). In contrast, traditional vegetation indices (VIs) such as the normalized difference vegetation index (NDVI) and EVI infer photosynthetic activity indirectly through canopy greenness. EVI, which incorporates near-infrared, red, and blue reflectance bands, reduces soil and atmospheric effects to better capture canopy structure and greenness (Huete et al., 2002). However, because canopy greenness often remains relatively stable, especially in evergreen systems, VIs may not fully represent the physiological seasonality of photosynthesis. Empirical analyses have shown that EVI and related indices exhibit moderate correlations with GPP (R 2 ≈ 0.5–0.7), though the strength and form of these relationships vary across vegetation types and environmental conditions (Sims et al., 2008; Liu et al., 2021).

SIF, by contrast, captures the dynamic efficiency of the photosynthetic apparatus and has been shown to track GPP seasonality across a range of temporal scales. In winter-dormant conifer forests, SIF correlates strongly with GPP (R 2 = 0.62–0.92) due to changes in photoprotective pigments and photosystem II efficiency rather than variations in absorbed PAR (Magney et al., 2019). Similarly, Wu et al. (2022) found that in soybean fields, SIF peaks 14–17 days earlier than GPP, a difference explained by canopy structural development and chlorophyll content. These findings emphasize that while SIF directly reflects the efficiency of light use in photosynthesis, its seasonal dynamics may precede carbon fixation due to biophysical canopy processes.

At larger scales, Tao et al. (2024) provided a high-resolution, multi-decadal SIF dataset demonstrating strong coherence between SIF and GPP across Chinese ecosystems, further validating the potential of satellite-based SIF as a reliable photosynthetic indicator. Satellite-based SIF retrievals have also demonstrated high correspondence with tower-measured GPP across biomes, with global correlations typically ranging from R 2 ≈ 0.6–0.9 (Frankenberg et al., 2018; Li et al., 2022).

The estimation of SIF from space relies on detecting the weak fluorescence signal embedded within the reflected solar spectrum. Early satellite missions such as GOSAT and GOME-2 exploited Fraunhofer lines, narrow spectral regions of reduced solar irradiance, to isolate SIF signals (Joiner et al., 2011). Later missions, including the TROPOspheric Monitoring Instrument (TROPOMI) aboard Sentinel-5 Precursor, substantially improved retrieval accuracy and spatial resolution, providing near-daily global SIF coverage at ∼3.5 × 7 km resolution at nadir (Köhler et al., 2018). These advances have enabled high-frequency mapping of photosynthetic activity and support applications in carbon-cycle modeling, drought monitoring, and crop productivity assessment.

However, while TROPOMI provides high-quality SIF observations, it does not distribute derived vegetation indices, and generating such indices from reflectance data requires additional processing. The PACE mission, launched in 2024, fills this gap with its publicly available LANDVI product, which includes ten precomputed vegetation indices: NDVI, EVI, CIRE, Normalized Difference Water Index (NDWI), Normalized Difference Infrared Index (NDII), Normalized Difference Snow Index (NDSI), Chlorophyll-Carotenoid Index (CCI), Carotenoid Content Index (Car), Photochemical Reflectance Index (PRI), and Modified Anthocyanin Reflectance Index (mARI) (Werdell et al., 2019; Caplan and Huemmrich, 2025).

The ten vegetation indices provided from the PACE satellite target distinct structural, biochemical, and physiological properties of vegetation. CIRE uses red-edge reflectance to estimate total canopy chlorophyll content and is particularly sensitive to variations in photosynthetic capacity. NDWI and NDII both exploit near- and shortwave-infrared reflectance to monitor vegetation water content, with NDWI emphasizing liquid water absorption near 860 nm and NDII targeting deeper canopy moisture. NDSI distinguishes snow from vegetation and soil using green and shortwave-infrared bands, aiding phenological studies in high-latitude and mountainous regions. CCI and Car track the balance between chlorophyll and carotenoid pigments, providing insight into photosynthetic downregulation and stress responses. PRI, calculated from reflectance at 531 nm and a reference band (usually 570 nm), serves as a well-established proxy for xanthophyll-cycle activity and light-use efficiency at both leaf and canopy scales. Lastly, mARI detects anthocyanin accumulation in leaves, which often signals senescence, nutrient stress, or protective responses to high light or cold. Together, these hyperspectral indices extend traditional metrics by capturing pigment pool dynamics, water status, and rapid physiological adjustments that directly influence photosynthetic efficiency and carbon uptake (Caplan and Huemmrich, 2025).

Vegetation indices derived from hyperspectral reflectance, such as those in PACE’s LANDVI dataset, offer complementary insight into GPP by capturing both structural and physiological properties of vegetation. The red edge-based CIRE, as stated, is sensitive to canopy chlorophyll concentration and has been shown to align closely with photosynthetic capacity and SIF in croplands and forests (Wu et al., 2024). Similarly, machine-learning approaches integrating EVI, CIRE, and PRI have demonstrated the ability to reconstruct SIF or GPP signals with relatively high accuracy (R 2 up to ∼0.8) in certain ecosystems (Zhang et al., 2023; Wu et al., 2024). Several of these indices, particularly CCI and PRI, provide physiologically relevant measures of photosynthetic efficiency by tracking pigment dynamics associated with the xanthophyll cycle and chlorophyll-carotenoid balance (Gamon et al., 2016). The inclusion of such hyperspectral indices in PACE’s LANDVI dataset enables monitoring of both canopy structure and photoprotective responses in an all-in-one product, offering more convenient insights into vegetation function.

PACE’s Ocean Color Instrument collects hyperspectral measurements from 340–890 nm, providing global 8-day averages at a consistent 0.1° (∼11 km) resolution. This temporal aggregation minimizes cloud-related data gaps, enabling consistent global coverage suitable for monitoring vegetation phenology (Werdell et al., 2019). The enhanced spectral fidelity of PACE improves red-edge and xanthophyll-based indices such as CIRE, and PRI, while EVI benefits from more precise atmospheric correction. Although PACE does not directly measure SIF, its hyperspectral LANDVI dataset provides the necessary inputs to model SIF and bridge spatial or temporal gaps in existing SIF products, offering a new pathway for GPP estimation and vegetation monitoring.

Despite the availability of high-quality satellite SIF products such as TROPOMI’s ∼3.5 × 7 km retrievals, modeling SIF from hyperspectral vegetation indices remains essential for several practical and scientific reasons. First, harmonizing SIF and reflectance-based datasets across missions requires spatial resampling, which can introduce artifacts and increase processing complexity; deriving SIF directly from the vegetation indices yields a native 0.1° dataset that aligns exactly with PACE’s global 8-day averages without additional spatial transformations. Second, PACE also distributes higher-resolution Ocean Color Instrument reflectance data and LANDVI product at 1.2 km, offering the potential to apply the same SIF-VI model to generate finer-resolution SIF products than currently available from TROPOMI. Finally, reconstructing SIF from reflectance is valuable because SIF remains one of the most direct remote sensing proxies for GPP; producing consistent SIF estimates across sensors, spatial scales, and time periods improves comparability, supports data fusion, and extends the utility of SIF for predicting ecosystem productivity and stress responses.

2 Materials and methods

2.1 Data acquisition and temporal resolution

The data for this study was derived from two primary sources: SIF measurements from the Tropospheric Monitoring Instrument (TROPOMI) sensor, and the PACE mission’s Level 3 mapped LANDVI product, which computes various vegetation indices using the equations in Table 1. To ensure global coverage in each temporal step, 8-day average averages were selected for both datasets. Satellite SIF data from the TROPOMI instrument on board European Space Agency’s Sentinel-5p satellite was gridded using L2B TROPOMI near-infrared SIF into a global product of 0.01-degree resolution and 8-day interval from 1 January 2024, to 31 December 2024. The gridding algorithm was established in a Julia package (https://github.com/cfranken/gridding), which has been used effectively in Cheng et al. (2022a), Cheng et al. (2022b) and Kunik et al. (2023). The analysis period spanned from the earliest available PACE 8-day average March 5–12, 2024, to December 18–25, 2024. The partial period from December 26–31, 2024, was excluded to maintain temporal alignment, as recalculating new 8-day averages would introduce possible inconsistencies between a new averaging method and the method imbedded in the LANDVI product.

TABLE 1. Equations for each vegetation index included in the LANDVI suite (Caplan and Huemmrich, 2025).

IndexEquationReferences
NDVIρ N I R − ρ r e d / ρ N I R + ρ r e dHuete et al. (2002)
EVI2.5×(ρNIR−ρred)/(ρNIR+6.0×ρred−7.5×ρblue+1)Huete et al. (2002); Jiang et al. (2008)
NDWIρ NIR − ρ 1250 / ρ NIR + ρ 1250Gao (1996)
NDIIρ NIR − ρ 1618 / ρ NIR + ρ 1618Hardisky, et al. (1983); Ji et al. (2011)
CCIρ green 1 − ρ red / ρ green 1 + ρ redGamon et al. (2016); Middleton et al. (2016)
NDSIρ green 2 − ρ 1618 / ρ green 2 + ρ 1618Hall, Riggs, and Salomonson (2001)
PRIρ 530 − ρ 570 / ρ 530 + ρ 570Grace et al. (2007); Coops et al. (2010)
CIREρ 800 / ρ 705 − 1Gitelson et al. (2007), Gitelson and Solovchenko (2017)
Car1 / ρ 495 − 1 / ρ 705 × ρ 800Gitelson et al. (2007)
mARI1 / ρ 550 − 1 / ρ 705 × ρ 800Gitelson et al. (2007)

Each vegetation index was calculated from PACE Ocean Color Instrument (OCI) surface reflectance values following the formulations in Table 1 from Caplan and Huemmrich (2025). Each ρ represents the surface reflectance measured by OCI at wavelength or band i. When a specific wavelength is required, single-band reflectance values are used. For indices requiring broad spectral regions (blue, green, red, and near infrared), reflectance values are averaged across all OCI bands within the corresponding wavelength interval according to Equation 1:where N is the number of OCI bands within the wavelength interval used for aggregation from i to n. Each wavelength interval is defined as NIR, 847–877 nm; red, 620–671 nm; green (1), 526–537 nm; green (2), 545–566 nm; and blue, 459–479 nm (Caplan and Huemmrich, 2025).

ρ aggregated = ∑ i n ρ i , OCI N (1)

2.2 Reclassification of biomes and land cover

The coarse spatial resolution of the datasets often encompassed multiple biome and land cover types within a single pixel, potentially confounding analyses of vegetation dynamics. To enhance homogeneity in study areas, biomes and land cover classes were reclassified into broader, generalized categories. Biome classifications were based on the World Wildlife Fund (WWF) Terrestrial Ecoregions of the World dataset (Olson et al., 2001). Land cover data originated from the Copernicus Global Land Service (2022 release), aligned with the United Nations Food and Agriculture Organization’s (FAO) Land Cover Classification System (LCCS).

Biomes were aggregated into six categories: forested tropical (1), forested temperate (2), taiga (3), grasslands (4), Mediterranean (5), and xeric shrublands (6). The Mediterranean category was retained separately due to its heterogeneous land cover mosaic, distinct from other biomes. Tundra was excluded because minimal vegetation cover would not inform photosynthetic proxies like SIF. Mangroves were also omitted, as their fragmented and sparse distribution prevented homogeneous study areas at the spatial scale of the data.

Land cover classes were similarly aggregated into three categories: agriculture (1), forest (2), and grassland (3). Non-vegetated classes (e.g., bare soil, water) were masked out to focus on vegetated areas. Wetlands were ultimately excluded from final analyses due to their limited spatial extent and inability to support homogeneous study regions. The exact reclassification guide for each original land cover and biome classification dataset can be found in the supplemental materials (Supplementary Tables S1, S2).

2.3 Study area selection

To balance data volume with representativeness, study areas were defined as 100 × 100-pixel boxes (approximately 1,100 km2 at 0.1° resolution). Given the challenges of random sampling in achieving homogeneity, particularly across mixed land covers, areas were selected visually using GIS software, prioritizing alignment between reclassified biome and land cover maps. Three areas were intentionally designed to capture mixed cover types (e.g., agriculture, forest, grasslands mosaics), while the remainder targeted near-homogeneous compositions. An R script was employed using the sf 1.0–19 package and the raster 3.6–32 package to quantify the proportional composition of each land cover class within candidate areas. Given coarse spatial resolution, homogeneous sites were prioritized; xeric sites were included to counterbalance productivity bias and widen the range of data available for modelling to avoid a bias towards highly vegetated areas. Homogeneous sites required ≥80% dominance by a single class, while mixed sites targeted 30% (±10%) for primary classes. Multiple rounds were run on each study area, and the study area was relocated or removed until all sites met the standard (Figure 1). This approach yielded diverse, biome-stratified regions suitable for global-scale SIF-EVI modeling (Table 2).

Fifteen ∼1,100-km<sup>2</sup> study sites were strategically selected to span the global range of biome types, productivity levels, and phenological regimes while maintaining high within-site homogeneity (≥80% single land-cover class for most sites). Locations of the 15 study areas (red dots) used for all analyses. Numbered labels correspond to the legend (1–15). Three sites (Eunápolis Mixed Region, Madrid Mixed Region, Central France Forests) were deliberately chosen as heterogeneous mosaics.
Figure 1. Fifteen ∼1,100-km2 study sites were strategically selected to span the global range of biome types, productivity levels, and phenological regimes while maintaining high within-site homogeneity (≥80% single land-cover class for most sites). Locations of the 15 study areas (red dots) used for all analyses. Numbered labels correspond to the legend (1–15). Three sites (Eunápolis Mixed Region, Madrid Mixed Region, Central France Forests) were deliberately chosen as heterogeneous mosaics.

TABLE 2. Study areas with biomes and cover classes.

Study areaBiomeCover
Northeast AppalachiaTemperate forestsForested
Northeast US AgTemperate forestsAgriculture
Canadian taigaBoreal forests/TaigaForested
Southern Mexico AgTropical forestsAgriculture
Yucatan forestsTropical forestsForested
Peru grasslandsGrasslandsGrasslands
Chapada DiamantinaXericAgriculture
Eunápolis mixed regionTropical forestsMixed
Brazilian HighlandsGrasslandsAgriculture
Central France forestsTemperate forestsMixed
Iberian grasslandsMediterraneanGrasslands
Madrid Mixed regionMediterraneanMixed
Congo forestsXericForests
Namibia ScrublandsXericGrasslands
Kazakh steppeGrasslandsMixed

Certain land cover/biome combinations could not be created due to lack of sufficient homogeneity of land cover within the biome. For example, the boreal forest biome was comprised almost entirely of either forests or non-vegetated areas due to its lack of suitability for agriculture and other forms of low-lying vegetation. Agricultural and forested areas were too interspersed with non-vegetated areas in the Mediterranean area due to substantial portions of this area being urbanized. More robust descriptions of each of these study areas can be found in the supplemental materials.

2.4 Multivariate correlation analysis

Pixel-level values for SIF and LANDVI rasters were extracted from the selected study areas for each 8-day timestep and compiled into time-series datasets. Initial exploratory analyses involved simple linear regression between SIF and each vegetation index across the full temporal range. R 2 values were computed for each timestep and visualized as temporal line plots to assess seasonal variations in correlation strength. Average R 2 values were also calculated per index to identify the most robust correlations.

To assess multicollinearity among candidate vegetation indices and guide the selection of indices for multivariate models, variance inflation factors (VIFs) were calculated for all possible combinations of indices. VIF for each index was computed using Equation 2 where R 2 is obtained by regressing the index of interest against all other indices.

V I F = 1 1 − R 2 (2)

This determined which indices had high variance inflation and should not be used (in a multivariate model) as well as which indices could be used together.

2.5 Other indices and SIF

To establish baseline global relationships between SIF and the vegetation indices, two simple linear regression models were fitted using the entire pooled dataset from all study areas and all available dates. The relationship between SIF and EVI was obtained by fitting a linear model of the form SIF ∼ EVI using the lm() function in R version 4.4.2, e.g., lm(SIF ∼ EVI, data = global_dataset). Similarly, the relationship between SIF and CIRE was derived by fitting lm(SIF ∼ CIRE, data = global_dataset). Model coefficients, standard errors, 95% confidence intervals, R 2 values, and p-values reported in Equations 3, 4 come directly from the summary() output of these lm objects. Spatiotemporal autocorrelation could slightly inflate model fit statistics, but the variance across biomes was the major source of variability. Residuals from both global models were subsequently mapped for the period 1–11 August 2024 to visualize spatial patterns in model performance.

To examine how the relationships between SIF and all the vegetation indices vary throughout the growing season, the data for each study area were stratified into three temporal periods. The phenological differences varied between each study site, as addressed in Section 3.3, and were used to inform the results. When creating the seasonal analysis, to ensure that differences among sites and seasons reflected true variation in the relationships between SIF and vegetation indices rather than differences in scale, we normalized all SIF and index values using the pooled distribution of values across all study areas. Specifically, each variable was transformed using min-max normalization calculated from the complete dataset (i.e., each value was rescaled between 0 and 1 based on the global minimum and maximum across all regions and time periods). Normalizing in this way allowed the analysis to place tropical forests, temperate systems, croplands, and arid shrublands on a common scale despite their vastly different absolute reflectance and SIF magnitudes. This step was critical for comparing values across sites, ensuring that differences in correlation strength reflected ecological or phenological variation rather than artifacts of disparate measurement ranges. It also prevented high-SIF or high-reflectance biomes (such as humid forests) from dominating the structure of the models simply because they operate on naturally larger numeric scales. Separate scatterplots with associated linear regression lines and R 2 values were then generated for each seasonal period, biome, and land-cover type (Figure 2). This seasonal partitioning allowed us to identify which vegetation index exhibited the strongest and most consistent correlation with SIF in a given biome or cover type at various stages of the year, revealing important temporal shifts in correlation performance that were masked when using annual or non-stratified data.

EVI and CIRE outperform all other vegetation indices in tracking SIF throughout the year on a global scale, with *R* <sup>2</sup> values exceeding 0.75 for most 8-day periods. Temporal dynamics of linear model performance (*R* <sup>2</sup>) for each LANDVI index when regressed against SIF across all study sites (2024–2025).
Figure 2. EVI and CIRE outperform all other vegetation indices in tracking SIF throughout the year on a global scale, with R 2 values exceeding 0.75 for most 8-day periods. Temporal dynamics of linear model performance (R 2) for each LANDVI index when regressed against SIF across all study sites (2024–2025).

3 Results

3.1 Correlation analysis

EVI and CIRE were the only indices that maintained strong linear relationships with SIF (R 2 > 0.75) through most of the year, confirming their potential superiority as global SIF proxies. Several other indices (NDVI, NDII, NDWI) periodically achieved moderate explanatory power (R 2 > 0.55), particularly during peak growing-season months, suggesting that biome- or season-specific models could occasionally benefit from using alternative indices to augment performance.

All indices exhibited synchronized declines in performance in late March and again in late November - early December, likely reflecting rapid phenological transitions (leaf-out and senescence) that temporarily exclude reflectance-based indices from instantaneous photosynthetic uptake. NDSI performed poorly year-round (R 2 consistently <0.25) because permanent snow or ice was absent from nearly all study sites during the growing season (Figure 3).

A density scatter plot shows the relationship between observed SIF and EVI across all study areas and time steps. Points are binned and colored by density (blue: low; yellow: high) on a log scale. Black lines represent global univariate linear regressions with the equation annotated. The global univariate linear regression line is shown in black with the equation annotated. EVI explains substantial variance in SIF, supporting its potential utility as a simple proxy for solar-induced fluorescence.
Figure 3. A density scatter plot shows the relationship between observed SIF and EVI across all study areas and time steps. Points are binned and colored by density (blue: low; yellow: high) on a log scale. Black lines represent global univariate linear regressions with the equation annotated. The global univariate linear regression line is shown in black with the equation annotated. EVI explains substantial variance in SIF, supporting its potential utility as a simple proxy for solar-induced fluorescence.

Averages were compiled from each index to definitively determine which indices were best. This helped solidify EVI and CIRE were the best indices to use in modeling of SIF, more than just a visual assessment of the line graph (Table 3).

TABLE 3. EVI and CIRE are the clearest winners: only they achieve average R2 > 0.75 (shown as bolded average R2 values) with SIF. Mean R2 values from linear regressions of each vegetation index against SIF across all 8-day periods and study sites in 2024–2025.

IndexNDVIEVINDWINDIICCINDSIPRICIRECarmARI
Average R 20.680.800.560.600.560.040.230.770.410.23

EVI and CIRE are the clearest winners: only they achieve average R2 > 0.75 (shown as bolded average R2 values) with SIF.

Multicollinearity among the candidate vegetation indices was pronounced. Pairwise variance inflation factors exceeded 5 and in several cases 8 for many structurally similar indices, most notably between EVI and NDVI (VIF = 9.4), NDII and NDWI (VIF = 9.6), EVI and CIRE (VIF = 8.2), NDVI and CIRE (VIF = 6.6), and Car and mARI (VIF = 5.7). These high values indicate substantial redundancy in the physiological and structural information captured by these indices. EVI and NDVI are both capturing structural information from the canopy while NDII and NDWI are both capturing moisture sensitive properties (Table 4).

TABLE 4. Strong multicollinearity prevents meaningful multivariate models: EVI and CIRE, despite being the two best individual indices, cannot be combined because of high mutual collinearity (VIF >6.5). Pairwise variance inflation factors (VIF) calculated among all candidate vegetation indices and SIF. Values >5 (red shading) indicate severe collinearity; values <2 (green shading) indicate negligible collinearity.

IndexSIFNDVIEVINDWINDIICCINDSIPRICIRECarmARI
SIF​3.05.11.62.12.21.01.34.31.61.3
NDVI3.0​9.41.42.03.01.01.36.62.01.7
EVI5.19.4​1.52.23.11.01.38.21.81.4
NDWI1.61.41.5​9.62.01.81.51.61.11.0
NDII2.12.02.29.6​3.41.41.72.41.31.1
CCI2.23.03.12.03.4​1.13.63.01.21.0
NDSI1.01.01.01.81.41.1​1.31.01.11.4
PRI1.31.31.31.51.73.61.3​1.31.01.1
CIRE4.36.68.21.62.43.01.01.3​2.11.6
Car1.62.01.81.11.31.21.11.02.1​5.7
mARI1.31.71.41.01.11.01.41.11.65.7​

Although EVI and CIRE individually emerged as having the strongest univariate correlations with SIF across the global dataset, their strong mutual collinearity rendered them unsuitable for inclusion in the same multivariate model. Attempts to construct multivariate models using two non-collinear index combinations yielded adverse changes in explanatory power relative to the best univariate models. All models with two or more independent variables had lower adjusted R2s than the best univariate models. It is possible that having more than one independent variable could improve season- or biome-specific models but checking each combination was beyond our study scope. For this reason, multivariate modeling was not pursued further, and subsequent analyses relied on the two best-performing univariate models: SIF ∼ EVI and SIF ∼ CIRE.

3.2 SIF modeling

These scatter plots (Figures 4, 5) illustrate the relationship between SIF and EVI. The density scatter plot of SIF versus EVI reveals a strong positive linear relationship (Figure 4), with the fitted global univariate model yielding Equation 3. The second shows the relation of the modeled and actual SIF values across various study areas for all time steps, with data points color-coded by region (Figure 5). We attempted to add confidence intervals, but the confidence intervals of 95% around the fitted relationship would be too narrow to be visually distinguishable from the regression line.

S I F = − 0.535 + 3.722 E V I (3)

A simple global linear model using only EVI explains 80% of the variance in SIF across all biomes and seasons. Observed versus modeled SIF from the global univariate linear regression SIF ∼ EVI (Equation 3). Each point represents an 8-day pixel-level observation from one of the 15 study areas, colored by site. The red dashed line is the 1:1 line; the black line is the fitted regression. *R* <sup>2</sup> = 0.80, RMSE = 0.33 W m<sup>-2</sup> sr<sup>−1</sup> µm<sup>-1</sup>.
Figure 4. A simple global linear model using only EVI explains 80% of the variance in SIF across all biomes and seasons. Observed versus modeled SIF from the global univariate linear regression SIF ∼ EVI (Equation 3). Each point represents an 8-day pixel-level observation from one of the 15 study areas, colored by site. The red dashed line is the 1:1 line; the black line is the fitted regression. R 2 = 0.80, RMSE = 0.33 W m-2 sr−1 µm-1.
The global SIF ∼ EVI model shows almost no bias in most biomes but consistently underestimates SIF in boreal forests and parts of the Amazon while slightly overestimating in a few high-latitude and island regions. Global map of residuals (observed - modeled SIF) from the univariate SIF ∼ EVI model (Equation 3) for 4–11 August 2024. Color scale ranges from −1.5 (blue, strong underestimation) to +1.5 (red, strong overestimation).
Figure 5. The global SIF ∼ EVI model shows almost no bias in most biomes but consistently underestimates SIF in boreal forests and parts of the Amazon while slightly overestimating in a few high-latitude and island regions. Global map of residuals (observed - modeled SIF) from the univariate SIF ∼ EVI model (Equation 3) for 4–11 August 2024. Color scale ranges from −1.5 (blue, strong underestimation) to +1.5 (red, strong overestimation).

This indicated a positive correlation between the two variables. The model explains 80% of the variance (R 2 = 0.80) with a root mean square error (RMSE) of 0.33W m-2 sr−1 µm-1), suggesting a reasonably good fit despite some scatter in the data across all time steps for our study areas (Figure 1). The quantity of the data points used in this analysis >55,000 coupled with a significant R 2 value led to an extremely small p value (2.2e-16). The shape of this graph does suggest the possibility of a non-linear fit, but quadratic models were tried and only resulted in a <0.02 improvement in the R 2 value (0.82) and a non-statistically significant difference in RMSE (0.33 W m-2 sr−1 µm-1). Towards the origin of the graph there is a cluster mostly consisting of points from the Kazakh Steppe (light grey) where persistent year-long snow cover on steppes decouples EVI and SIF (Qiu et al., 2019).

In Figure 6 we extrapolate this model to a single timestep for the global scale to show the spatial distribution of residuals (observed SIF - modeled SIF) from the EVI-based model for 04–11 August 2024. The model performs well over most tropical and temperate vegetated surfaces, with residuals near zero (pale yellow to light green) prevailing across croplands, grasslands, and seasonal forests. For this current time-step the model is underestimating in boreal areas with the exception of Greenland and the tropical forests of South America. Overestimation seems to be much less of an issue being mostly relegated to tropical islands, Greenland, and some parts of the Northeastern United States. This pattern was fairly consistent across all time steps. The distribution of residuals had a standard deviation being ∼0.40 W m-2 sr−1 µm-1. The other 36 time-steps had roughly similar means and standard deviations with some variation of 0 ± 0.05 W m-2 sr−1 µm-1 for the mean and ±0.10 W m-2 sr−1 µm-1 for the standard deviation.

A density scatter plot shows the relationship between observed SIF and CIRE across all study areas and time steps. Points are binned and colored by density (blue: low; yellow: high) on a log scale. The global univariate linear regression line is shown in black with the equation annotated. CIRE explains substantial variance in SIF, supporting its potential utility as a simple proxy for solar-induced fluorescence.
Figure 6. A density scatter plot shows the relationship between observed SIF and CIRE across all study areas and time steps. Points are binned and colored by density (blue: low; yellow: high) on a log scale. The global univariate linear regression line is shown in black with the equation annotated. CIRE explains substantial variance in SIF, supporting its potential utility as a simple proxy for solar-induced fluorescence.

These scatter plots illustrate the relationship between SIF and CIRE across various study areas for all time steps. The density scatter plot for SIF versus CIRE shows a clear positive correlation, described by the linear equation Equation 4. Dense point clusters align well with the regression line, particularly at low to mid CIRE values (Figure 7). For the second graph with data points color coded by region to show how the different study areas cluster depending on the physiology of the site. (Figure 8). A linear regression model in R version 4.4.2 yielded the equation represented in Equation 4

S I F = − 0.047 + 0.801 C I R E (4)

A simple global linear model using only CIRE explains 74.8% of the variance in SIF across all biomes and seasons. Observed versus modeled SIF from the global univariate linear regression SIF ∼ CIRE (Equation 4). Each point represents an 8-day pixel-level observation from one of the 15 study areas, colored by site. The black dashed line is the fitted regression. *R* <sup>2</sup> = 0.77, RMSE = 0.36 W m<sup>-2</sup> sr<sup>−1</sup> µm<sup>-1</sup>.
Figure 7. A simple global linear model using only CIRE explains 74.8% of the variance in SIF across all biomes and seasons. Observed versus modeled SIF from the global univariate linear regression SIF ∼ CIRE (Equation 4). Each point represents an 8-day pixel-level observation from one of the 15 study areas, colored by site. The black dashed line is the fitted regression. R 2 = 0.77, RMSE = 0.36 W m-2 sr−1 µm-1.
The global SIF ∼ CIRE model is essentially unbiased over most of Earth’s vegetated land surface but exhibits stronger and more widespread overestimation near the equator and in mountainous regions compared to the EVI model. Global map of residuals (observed–modeled SIF) from the univariate SIF ∼ CIRE model (Equation 4) for 4–11 August 2024. Color scale ranges from −3.0 (blue, underestimation) to +3.1 (red, overestimation).
Figure 8. The global SIF ∼ CIRE model is essentially unbiased over most of Earth’s vegetated land surface but exhibits stronger and more widespread overestimation near the equator and in mountainous regions compared to the EVI model. Global map of residuals (observed–modeled SIF) from the univariate SIF ∼ CIRE model (Equation 4) for 4–11 August 2024. Color scale ranges from −3.0 (blue, underestimation) to +3.1 (red, overestimation).

CIRE is strongly correlated with SIF. The model explains 77% of the variance (R 2 = 0.77) with a RMSE of 0.37 W m-2 sr−1 µm-1 across all time steps for the study areas presented in Figure 1, suggesting a slightly poorer fit than the model utilizing EVI. Once again for the CIRE model, the quantity of the data points used coupled with the R 2 value led to an extremely small p value (2.2e-16). Once again there is some clustering of points toward the origin within the Kazakh Steppe (light grey).

In Figure 9 once again we extrapolate the localized model to a single timestep for the global dataset to show the spatial distribution of residuals (observed SIF–modeled SIF) from the CIRE-based model for 04–11 August 2024. The model performs similarly to the EVI model over most tropical and temperate vegetated surfaces, with residuals near zero (pale yellow to light green) prevailing across croplands, grasslands, and seasonal forests. For this current time-step the model is once again underestimating in boreal areas, more heavily in Russia, and the tropical forests of South America. Overestimation seems to be more of an issue with the CIRE model around equatorial latitudes, Asiatic mountains, and some select spots in Ontario and Quebec. This pattern was fairly consistent across all time steps. The distribution of this data had a mean that showed a slightly positive bias, with a standard deviation of ∼0.37 W m-2 sr−1 µm-1. The other 36 time-steps had roughly similar means and standard deviations with some variation of ±0.10 W m-2 sr−1 µm-1for the mean and ±0.07 W m-2 sr−1 µm-1 for the standard deviation.

EVI and CIRE dominate with the highest SIF correlations in almost all biomes and seasons, but water- and pigment-based indices (NDWI, NDII, CCI) occasionally outperform them during dry-season or early-senescence periods in moisture-limited ecosystems. Normalized SIF plotted against the single best-performing vegetation index for each site and phenological season (early-, peak-, and late-year). Points are colored by month; black line = linear fit; red dashed line = 1:1. *R* <sup>2</sup> and identity of the best index are shown in each panel.
Figure 9. EVI and CIRE dominate with the highest SIF correlations in almost all biomes and seasons, but water- and pigment-based indices (NDWI, NDII, CCI) occasionally outperform them during dry-season or early-senescence periods in moisture-limited ecosystems. Normalized SIF plotted against the single best-performing vegetation index for each site and phenological season (early-, peak-, and late-year). Points are colored by month; black line = linear fit; red dashed line = 1:1. R 2 and identity of the best index are shown in each panel.

3.3 Seasonal analysis

While EVI and CIRE remained strongly correlated in many scenarios for the first grouping NDWI outperformed CIRE significantly in Yucatan Forest for the dry season (CIRE R 2 = 0.40) and marginally in the summer months of the Appalachians (CIRE R 2 = 0.55). All indices had poor correlations in the xeric Namibia Scrublands likely due to a lack of vegetation in the area. NDII also marginally outperformed EVI and CIRE in the spring and summer for the mixed cover in Central France (spring EVI R 2 = 0.61, summer CIRE R 2 = 0.32). The strongest correlation of any seasonal grouping can also be observed on this graph during the spring months in the Appalachians showing an R 2 of 0.94 for a normalized SIF ∼ EVI.

EVI and CIRE remained strongly correlated in many of the study areas for the second grouping, yet we still see NDWI outperforming in certain tropical areas. For all seasons in the Eunápolis Mixed Region NDWI was marginally higher than EVI, however much of the wet season (November-April) was excluded due to availability of data (EVI R 2 = 0.26, 0.65, and 0.63 respectively). NDWI also marginally outperformed in the xeric forests of the Congo after the end of the wet season in April into the early portions of the dry season of May through September (CIRE R 2 = 0.79). In the Brazilian Highlands CCI marginally outperformed other indices (EVI R 2 = 0.64) which is the only substantial instance of CCI correlating better than the rest. NDII also marginally outperformed other indices once again in the rain shadow of the Andes for the early portions of the dry season in the Peru Grasslands (EVI R 2 = 0.58) and the early portions of the dry season in the Chapada Diamantina (EVI R 2 = 0.54) where herbaceous plants and crops are beginning to enter dormancy.

It is in group 3 where there is little deviation from what we expect, where EVI and CIRE remain the two best indicators of SIF. For many of the groupings, we see R 2 values >0.70 during the growing seasons, while in the drier environments like Spain and Mexico, we see a drop in the correlations of all indices. Even though there is a drop in the correlation for Southern Mexico and Madrid, the variation in the data seems limited as many of the points remain in tight clusters. Across all 15 study sites and three phenological seasons, EVI or CIRE showed the highest correlations from the univariate model for normalized SIF in 81% of cases (37 of 46 site–season combinations), routinely achieving R 2 > 0.75 and often >0.90 during peak growing season (Figure 2).

Because EVI consistently emerged with the strongest global correlations to SIF, we also present a separate figure that focuses exclusively on the normalized SIF ∼ EVI relationships across all study areas and seasons. While the full set of indices provides important insight into local deviations and biome-specific behavior, isolating EVI allows a clearer visualization of its dominant global performance. (Figure 10). A similar breakdown of the SIF ∼ CIRE relationship was provided in the suplmental materials (Supplementary Figure S1).

While some Vis improve correlations slightly from EVI, it still maintains a strong correlation throughout most sites and times. Normalized SIF plotted against the Normalized EVI for each site and phenological season (early-, peak-, and late-year). Points are colored by month; black line = linear fit; red dashed line = 1:1.
Figure 10. While some Vis improve correlations slightly from EVI, it still maintains a strong correlation throughout most sites and times. Normalized SIF plotted against the Normalized EVI for each site and phenological season (early-, peak-, and late-year). Points are colored by month; black line = linear fit; red dashed line = 1:1.

4 Discussion

This study demonstrates the utility of vegetation indices for modeling SIF across diverse global biomes. The overall strong correlations observed between SIF and both EVI and CIRE align with previous work showing that greenness and red-edge based vegetation indices capture meaningful variation in photosynthetic activity (Huete et al., 2002; Gitelson et al., 2007; Neuwirthová et al., 2026), as well as a prior analysis showing strong correlation (r = 0.89) between OCO-2 SIF and MODIS EVI (Merrick et al., 2023). The strength of the SIF ∼ EVI relationship (R 2 = 0.80) is consistent with studies linking EVI to canopy structure and APAR, which together support stable correlation across ecosystems (Sims et al., 2008). We did not normalize SIF by PAR because, while this can provide an estimate of fluorescence yield, the precalculated PACE PAR product was optimized for flat, dark ocean surfaces and is not well-suited for terrestrial applications. More importantly, at seasonal to annual scales across biomes, variance in SIF is driven primarily by the fraction of absorbed PAR, which is already highly correlated with EVI, rather than by incident PAR, which exerts stronger influence at diurnal scales. Similarly, the performance of CIRE reflects its sensitivity to chlorophyll content and canopy photosynthetic capacity, echoing findings from crop and forest studies where red-edge indices have proven to correlate highly with SIF and GPP (Gitelson and Solovchenko, 2017; Wu et al., 2024).

The performance of EVI and CIRE can be directly linked to the radiative transfer processes that govern SIF emission and escape. SIF originates from the re-emission of absorbed photons by chlorophyll molecules, meaning that the absorption features surrounding the red edge (∼700 nm) exert strong links with fluorescence level (Meroni et al., 2009). Because CIRE responds to subtle shifts in chlorophyll absorption and leaf biochemical status, it is highly sensitive to the same pigment-driven processes that modulate fluorescence yield. By contrast, EVI is primarily sensitive to canopy structure, leaf area index, and APAR, which together determine both the total pool of absorbed energy and the efficiency with which fluorescence can escape through the canopy. These layered radiative effects explain why both indices track SIF robustly at global scales. Each represents a distinct but complementary component of the fluorescence signal, i.e., biochemical absorption at the leaf level and radiative transport at the canopy level. However, because SIF is highly related to escape (Yang et al., 2021), it may be the case that the areas in which the model is overpredicting or underpredicting residuals, the inclusion of a variable that tracks escape probability may further improve residuals.

The use of 8-day composite data from PACE and TROPOMI provided a robust temporal framework for capturing vegetation seasonality. This approach offers clear advantages over MODIS vegetation index products. Both Terra and Aqua MODIS instruments are experiencing orbital drift and are scheduled for decommissioning in late 2026 (Aqua) to early 2027 (Terra), limiting their long-term utility. In contrast, PACE provides a forward-looking hyperspectral data stream. As a spectrometer, PACE’s Ocean Color Instrument enables the calculation of narrow-band vegetation indices, such as the Chlorophyll Index Red Edge, Photochemical Reflectance Index, modified Anthocyanin Reflectance Index, and narrow-band versions of the Carotenoid Chlorophyll Index, that cannot be accurately derived from MODIS’s broader multispectral bands. Additionally, the native 8-day temporal resolution of PACE aligns directly with our TROPOMI SIF dataset, eliminating the need for manual temporal aggregation and reducing potential resampling artifacts. While 8-day MODIS composites can be generated from daily reflectance data, they lack the narrow-band spectral fidelity critical for capturing subtle physiological signals. The improved temporal pairing between PACE and TROPOMI thus allows short-term radiative and physiological changes in fluorescence emission to be captured more reliably.

The reclassification of biomes and land cover into generalized categories allowed for controlled sampling within the constraints of the 0.1° resolution. Although this approach ensured homogeneity in most locations, it may also have introduced bias toward productive, vegetated regions, which is why the xeric biomes were included. The exclusion of tundra and mangroves removes biomes known to exhibit unique photosynthetic phenology. The tundra, for example, often displays short but intense seasonality detectable by SIF even when greenness indices underperform (Magney et al., 2019). Thus, future work may benefit from incorporating alternative biome classifications or higher-resolution sensors to include these regions more effectively. This may be difficult however as the orbit path of PACE can cause seasonal data gaps for extreme latitudes.

Study area selection likely influenced model performance. Xeric regions exhibited weaker correlations, consistent with low canopy cover and high soil background influence, which is known to degrade vegetation index ∼ photosynthesis relationships (Gao, 1996; Hardisky et al., 1983). In these open-canopy systems, the fluorescence escape fraction is high, but the total amount of absorbed energy is low, making the correlation between structural indices and SIF less reliable. This may help explain the comparatively weak EVI correlations in xeric regions. The reliance on visually selected regions introduces potential observer bias, although this was mitigated through the ≥80% homogeneity/∼30% heterogeneity thresholds and R-based compositional checks. Even so, automated sampling approaches, such as stratified random selection or object-based segmentation, could improve reproducibility and reduce bias in future studies.

A key insight from the seasonal stratification analysis is that NDWI often outperformed EVI and CIRE in seasonally dry regions. NDII also showed improved correlations in transitions to the dry season in several grassland and xeric sites, consistent with its established sensitivity to leaf and canopy water content (Hardisky et al., 1983). This aligns with prior work showing strong coupling between water status and photosynthetic efficiency under drought conditions (Li et al., 2022; Wu et al., 2024). NDWI’s superior performance in tropical dry season forests and semi-arid landscapes suggests that moisture sensitive indices may provide critical complementary information where guard cell metabolism and stomatal conductance limit fluorescence emission (Flexas and Medrano, 2002). It has been shown that water stress reduces stomatal conductance and CO2 diffusion into the leaf, which downregulates photosynthetic electron transport and increases non-photochemical quenching, thereby suppressing chlorophyll fluorescence (Porcar-Castell et al., 2014). Biochemical limitations, including reduced Rubisco activity in C3 plants or PEP carboxylase capacity in C4 plants, further decrease electron transport demand, diminishing fluorescence yield under drought or thermal stress (Cornic and Massacci, 1996; von Caemmerer, 2000; Sun et al., 2018). SIF represents the fraction of absorbed energy not used for photochemistry or dissipated through heat, any environmental factor that alters the balance between these pathways such as pigment content, water limitation, or thermal stress, directly modulates fluorescence yield. Vegetation indices that track these same processes therefore provide physically meaningful proxies for SIF variation. It is also worth noting that both our results (Figure 7) and those of Uuh-Sonda et al. (2021) show a change in the sign of the slope in the relationship between EVI and SIF (this study) or EVI and GPP (Uuh-Sonda et al., 2021) between the wet and dry seasons in Yucatan forests. These findings demonstrate that a single global predictor may not fully capture the biome-specific drivers of SIF, reinforcing the need for tailored or multivariate models. The focused breakdown of EVI based on separate spatial and temporal distributions highlights how reliably EVI tracks SIF during periods of active vegetation growth and how its correlation strength remains comparatively robust even in regions where other indices may outperform. Presenting the EVI-only results separately also helps emphasize the agreement with prior studies (Zhang et al., 2023): despite regional nuances, EVI remains the most stable and widely applicable univariate correlation with SIF on a global scale.

The spatial residual analysis further highlighted systematic underestimation in boreal forests and some tropical regions. Boreal underestimation has been documented in prior SIF ∼ VI studies due to evergreen phenology, photoprotective pigment dynamics, and snow related reflectance effects that weaken vegetation index sensitivity to physiological changes (Magney et al., 2019; Middleton et al., 2016). Conversely, overestimation in some tropical regions may relate to subcanopy shading or cloud contamination, both of which have been shown to influence SIF retrievals and reflectance indices differently (Joiner et al., 2011; Köhler et al., 2018).

Another key factor influencing these relationships is the fluorescence escape fraction, which varies strongly by canopy architecture, leaf angle distribution, and background reflectance. Dense forests, particularly evergreen canopies, exhibit low escape fractions due to strong multiple scattering and reabsorption of emitted fluorescence, which can dampen the apparent SIF signal even when photosynthetic activity remains high (Mohammed et al., 2019). Greenness indices such as EVI may saturate under these conditions, while red-edge indices partially compensate due to their sensitivity to optical depth. These radiative constraints likely contributed to the residual underestimation observed in boreal biomes, where fluorescence escape is known to be seasonally modulated by pigment composition and snow-affected reflectance.

Importantly, even with strong overall performance, global linear models are limited in capturing nonlinear or saturation effects, especially in dense tropical forests where red-edge indices may be more informative than greenness metrics alone (Gamon et al., 2016). Although quadratic models only marginally improved fit (<0.02 increase in R 2), nonparametric or machine-learning approaches such as those used by Gentine and Alemohammad (2018) have shown substantial gains in reconstructing SIF from multispectral data and may prove beneficial when PACE datasets extend beyond a single year.

This study underscores the potential of vegetation indices to contribute meaningfully to global SIF modeling. Taken together, these findings indicate that the robust performance of EVI and CIRE is mechanistically grounded rather than purely empirical. CIRE responds to physiological and biochemical properties (chlorophyll content), while EVI is primarily sensitive to canopy structure and the fraction of APAR (though is also often strongly associated with leaf chlorophyll content; Gao et al., 2024), both of which govern SIF emission and escape probability. Red-edge indices succeed because they track chlorophyll absorption, the biochemical gateway for both photochemistry and fluorescence, while greenness indices succeed because they capture canopy structure and light interception, the physical determinants of photon availability and radiative transfer. By linking these indices to the underlying processes that regulate fluorescence emission, this study identifies measurable proxies that stand in for the causal drivers of SIF at global scales. With its hyperspectral measurements, consistent temporal resolution, and global coverage, PACE provides a valuable platform for advancing our understanding of terrestrial photosynthesis. However, limitations such as land cover misclassification or change from the 2022 Copernicus dataset, cloud impacts, and partial year coverage in some regions highlight the need for continued refinement. Incorporating multiple years of PACE data, integrating higher resolution reflectance products, and exploring multivariate or machine-learning models will likely improve global SIF predictions even further and expand the usage cases for this dataset.

5 Summary

This study demonstrates the potential of using other vegetation indices to enhance global SIF predictions, leveraging their associations with photosynthetic activity. The methodology provides a replicable framework for integrating remote sensing data across biomes, with applications in environmental monitoring and climate adaptation strategies. While this study has a global focus, future studies that narrow down the exact land cover/biome combinations could be even more effective at accurately predicting SIF. Future research should focus on expanding temporal coverage to multiple years, incorporating advanced machine learning models, including escape probabilities, and addressing data gaps to further refine global SIF estimation.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frsen.2026.1819365/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Sawaid Abbas, University of the Punjab, Pakistan

Reviewed by: Xiuqing Hu, China Meteorological Administration, China

Jiaochan Hu, Dalian Maritime University, China

Statements

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

The author(s) declared that financial support was received for this work and/or its publication. RC acknowledges the computing resources from the infrastructure LAGUNA HPC cluster supported by the SoCal Reserch Computing Alliance, a collaborative effort between the University of Southern California and the Los Nettos Consortium. This work was also partially supported by a USDA McIntire-Stennis Award to Virginia Tech entitled “Precision Forestry for Southern Pine Carbon Monitoring” (Number: VA-136647). KFH funded by the NASA PACE Science and Applications Team through GESTAR-II Cooperative Agreement [Award # 80NSSC22M0001].

Data availability

The PACE datasets analyzed for this study can be found in the OB.DAAC File Search https://oceandata.sci.gsfc.nasa.gov/api/file_search/. The raw SIF datasets processed and analyzed for this study can be found in the Radiant Earth Stac Browser at https://radiantearth.github.io/stac-browser/#/external/data-portal.s5p-pal.com/api/s5p-l3/sif.

Author contributions

CH: Formal Analysis, Methodology, Visualization, Writing – original draft. KH: Conceptualization, Writing – review and editing. RW: Funding acquisition, Supervision, Writing – review and editing. VT: Funding acquisition, Supervision, Writing – review and editing. RC: Data curation, Software, Writing – review and editing.

References

  1. Caplan S., Huemmrich K. F. (2025). Unveiling PACE OCI’s hyperspectral terrestrial data products. Remote Sens. Lett. 16 (4), 422 – 433. doi:10.1080/2150704X.2025.2470905
  2. Cheng R., Magney T. S., Orcutt E. L., Pierrat Z., Köhler P., Bowling D. R. (2022a). Evaluating photosynthetic activity across arctic-boreal land cover types using solar-induced fluorescence. Environ. Res. Lett. 17 (11), 115009. doi:10.1088/1748-9326/ac9dae
  3. Cheng R., Köhler P., Frankenberg C. (2022b). Impact of radiation variations on temporal upscaling of instantaneous solar-induced chlorophyll fluorescence. Agric. For. Meteorology 327, 109197. doi:10.1016/j.agrformet.2022.109197
  4. Coops N. C., Hilker T., Hall F. G., Nichol C. J., Drolet G. G. (2010). Estimation of light-use efficiency of terrestrial ecosystems from space: a status report. BioScience 60 (10), 788 – 797. doi:10.1525/bio.2010.60.10.5
  5. Cornic G., Massacci A. (1996). “Leaf photosynthesis under drought stress,” In Photosynthesis and the Environment, Advances in Photosynthesis and Respiration. Editors Baker, N. R. (Dordrecht: Springer), 5. doi:10.1007/0-306-48135-9_14
  6. Flexas J., Medrano H. (2002). Drought-inhibition of photosynthesis in C 3 plants: stomatal and non-stomatal limitations revisited. Ann. Bot. 89 (2), 183 – 189. doi:10.1093/aob/mcf027
  7. Frankenberg C., Köhler P., Magney T. S., Geier S., Lawson P., Schwochert M. (2018). The chlorophyll fluorescence imaging spectrometer (CFIS), mapping far-red fluorescence from aircraft. Remote Sens. Environ. 217, 523 – 536. doi:10.1016/j.rse.2018.08.032
  8. Gamon J. A., Huemmrich K. F., Wong C. Y. S., Ensminger I., Garrity S., Hollinger D. Y. (2016). A remotely sensed pigment index reveals photosynthetic phenology in evergreen conifers. Proc. Natl. Acad. Sci. 113 (48), 13087 – 13092. doi:10.1073/pnas.1606162113
  9. Gao B.-C. (1996). NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens. Environ. 58 (3), 257 – 266. doi:10.1016/S0034-4257(96)00067-3
  10. Gao S., Yan K., Liu J., Pu J., Zou D., Qi J. (2024). Assessment of remote-sensed vegetation indices for estimating forest chlorophyll concentration. Ecol. Indic. 162, 112001. doi:10.1016/j.ecolind.2024.112001
  11. Gentine P., Alemohammad S. (2018). Reconstructed solar-induced fluorescence: a machine learning vegetation product based on MODIS surface reflectance to reproduce GOME-2 solar-induced fluorescence. Geophys. Res. Lett. 45 (7), 3136 – 3146. doi:10.1002/2017gl076294
  12. Gitelson A. A., Viña A., Ciganda V., Rundquist D. C., Arkebauer T. J. (2007). Remote estimation of canopy chlorophyll content in crops. Geophys. Res. Lett. 32 (8), L08403. doi:10.1029/2005GL022688
  13. Gitelson A. A., Solovchenko A. (2017). Generic algorithms for estimating foliar pigment content. Geophys. Res. Lett. 44 (18), 9293 – 9298. doi:10.1002/2017GL074799
  14. Grace J., Nichol C., Disney M., Lewis P., Quaife T., Bowyer P. (2007). Can we measure terrestrial photosynthesis from space directly, using spectral reflectance and fluorescence? Glob. Change Biol. 13 (7), 1484 – 1497. doi:10.1111/j.1365-2486.2007.01352.x
  15. Hall D. K., Riggs G. A., Salomonson V. V. (2001). Algorithm Theoretical Basis Document (ATBD) for the MODIS Snow and Sea ice-mapping Algorithms. Greenbelt, MD: NASA Goddard Space Flight Center.
  16. Hardisky M. A., Klemas V., Smart R. M. (1983). The influence of soil salinity, growth form, and leaf moisture on the spectral radiance of Spartina alterniflora. Photogrammetric Eng. Remote Sens. 49 (1), 77 – 83.
  17. Huete A., Didan K., Miura T., Rodriguez E., Gao X., Ferreira L. (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 83 (1–2), 195 – 213. doi:10.1016/S0034-4257(02)00096-2
  18. Ji L., Zhang L., Wylie B., Rover J. (2011). On the terminology of the spectral vegetation index (NIR−SWIR)/(NIR+SWIR). Int. J. Remote Sens. 32 (21), 6901 – 6909. doi:10.1080/01431161.2010.510485
  19. Jiang Z., Huete A. R., Didan K., Miura T. (2008). Development of a two-band enhanced vegetation index without a blue band. Remote Sens. Environ. 112 (10), 3833 – 3845. doi:10.1016/j.rse.2008.06.006
  20. Joiner J., Yoshida Y., Vasilkov A. P., Yoshida Y., Corp L. A., Middleton E. M. (2011). First observations of global and seasonal terrestrial chlorophyll fluorescence from space. Biogeosciences 8 (3), 637 – 651. doi:10.5194/bg-8-637-2011
  21. Köhler P., Frankenberg C., Magney T. S., Guanter L., Joiner J., Landgraf J. (2018). Global retrievals of solar-induced chlorophyll fluorescence with TROPOMI: first results and intersensor comparison to OCO-2. Geophys. Res. Lett. 45 (19), 10456 – 10463. doi:10.1029/2018gl079031
  22. Kunik L., Bowling D. R., Raczka B., Frankenberg C., Köhler P., Cheng R. (2023). Satellite-based solar-induced fluorescence tracks seasonal and elevational patterns of photosynthesis in California’s sierra Nevada Mountains. Environ. Res. Lett. 19 (1), 014008. doi:10.1088/1748-9326/ad07b4
  23. Li C., Peng L., Zhou M., Wei Y., Liu L., Li L. (2022). SIF-Based GPP as a useful index for assessing impacts of drought on vegetation: an example of a mega-drought in Yunnan Province, China. Remote Sens. 14 (6), 1509. doi:10.3390/rs14061509
  24. Liu F., Wang C., Wang X. (2021). Can vegetation index track the interannual variation in gross primary production of temperate deciduous forests? Ecol. Process. 10 (1), 51. doi:10.1186/s13717-021-00324-2
  25. Magney T. S., Bowling D. R., Logan B. A., Grossmann K., Stutz J., Blanken P. D. (2019). Mechanistic evidence for tracking the seasonality of photosynthesis with solar-induced fluorescence. PNAS 116 (24), 11640 – 11645. doi:10.1073/pnas.1900278116
  26. Meroni M., Rossini M., Guanter L., Alonso L., Rascher U., Colombo R. (2009). Remote sensing of solar-induced chlorophyll fluorescence: review of methods and applications. Remote Sens. Environ. 113 (10), 2037 – 2051. doi:10.1016/j.rse.2009.05.003
  27. Merrick T., Bennartz R., Jorge M. L. S. P., Merrick C., Bohlman S. A., Silva C. A. (2023). Comparing phenology of a temperate deciduous forest captured by solar-induced fluorescence and vegetation indices. Remote Sens. 15 (21), 5101. doi:10.3390/rs15215101
  28. Middleton E. M., Huemmrich K. F., Landis D. R., Black T. A., Barr A. G., McCaughey H. (2016). Photosynthetic efficiency of northern forest ecosystems using a MODIS‐derived photochemical reflectance Index (PRI). Remote Sens. Environ. 187, 345 – 366. doi:10.1016/j.rse.2016.10.021
  29. Mohammed G. H., Colombo R., Middleton E. M., Rascher U., van der Tol C., Nedbal L. (2019). Remote sensing of solar-induced chlorophyll fluorescence (SIF) in vegetation: 50 years of progress. Remote Sens. Environ. 231, 111177. doi:10.1016/j.rse.2019.04.030
  30. Neuwirthová E., García-Martínez C., Pescador-Dionisio S., Moncholí-Estornell A., Čepl J., Pikl M. (2026). Actual photosynthetic yield retrieval based on canopy reflectance: the role of co-varying, red-shifted absorbance behaviour. Remote Sens. Environ. 343, 115484. doi:10.1016/j.rse.2026.115484
  31. Olson D. M., Dinerstein E., Wikramanayake E. D., Burgess N. D., Powell G. V. N., Underwood E. C. (2001). Terrestrial ecoregions of the world: a new map of life on Earth. BioScience 51 (11), 933. doi:10.1641/0006-3568(2001)051[0933:TEOTWA]2.0.CO;2
  32. Porcar-Castell A., Tyystjärvi E., Atherton J., van der Tol C., Flexas J., Pfündel E. E. (2014). Linking chlorophyll a fluorescence to photosynthesis for remote sensing applications: mechanisms and challenges. J. Exp. Bot. 65 (15), 4065 – 4095. doi:10.1093/jxb/eru191
  33. Qiu B., Li W., Wang X., Shang L., Song C., Guo W. (2019). Satellite-observed solar-induced chlorophyll fluorescence reveals higher sensitivity of alpine ecosystems to snow cover on the Tibetan Plateau. Agric. For. Meteorology 271, 126 – 134. doi:10.1016/j.agrformet.2019.02.045
  34. Sims D., Rahman A., Cordova V., Elmasri B., Baldocchi D., Bolstad P. (2008). A new model of gross primary productivity for North American ecosystems based solely on the enhanced vegetation index and land surface temperature from MODIS. Remote Sens. Environ. 112 (4), 1633 – 1646. doi:10.1016/j.rse.2007.08.004
  35. Sun Y., Frankenberg C., Jung M., Joiner J., Guanter L., Köhler P. (2018). Overview of solar-induced chlorophyll fluorescence (SIF) from the orbiting carbon observatory-2: retrieval, cross-mission comparison, and global monitoring for GPP. Remote Sens. Environ. 209, 808 – 823. doi:10.1016/j.rse.2018.02.016
  36. Tao S., Chen J. M., Zhang Z., Zhang Y., Ju W., Zhu T. (2024). A high-resolution satellite-based solar-induced chlorophyll fluorescence dataset for China from 2000 to 2022. Sci. Data 11 (1), 1286. doi:10.1038/s41597-024-04101-6
  37. Uuh-Sonda J. M., Figueroa-Espinoza B., Gutiérrez-Jurado H. A., Méndez-Barroso L. A. (2021). Ecosystem productivity and evapotranspiration dynamics of a seasonally dry tropical forest of the Yucatan peninsula. J. Geophys. Res. Biogeosciences 127 (1), e2019JG005629. doi:10.1029/2019JG005629
  38. von Caemmerer S. (2000). Biochemical models of C 4 photosynthesis. Collingwood, VIC: CSIRO Publishing.
  39. Werdell P. J., Behrenfeld M. J., Bontempi P. S., Boss E., Cairns B., Davis G. T. (2019). The plankton, Aerosol, Cloud, Ocean Ecosystem mission: status, science, advances. Bull. Am. Meteorological Soc. 100 (9), 1775 – 1794. doi:10.1175/BAMS-D-18-0056.1
  40. Wu G., Jiang C., Kimm H., Wang S., Bernacchi C., Moore C. E. (2022). Difference in seasonal peak timing of soybean far-red SIF and GPP explained by canopy structure and chlorophyll content. Remote Sens. Environ. 279, 113104. doi:10.1016/j.rse.2022.113104
  41. Wu Y., Zhang Z., Wu L., Zhang Y. (2024). Solar-induced chlorophyll fluorescence tracks canopy photosynthesis under dry conditions in a semi-arid grassland. Agric. For. Meteorology 356, 110174. doi:10.1016/j.agrformet.2024.110174
  42. Yang P., van der Tol C., Campbell P. K. E., Middleton E. M. (2021). Unraveling the physical and physiological basis for the solar-induced chlorophyll fluorescence and photosynthesis relationship using continuous leaf and canopy measurements of a corn crop. Biogeosciences 18 (2), 441 – 465. doi:10.5194/bg-18-441-2021
  43. Zhang P., Liu H., Li H., Yao J., Chen X., Feng J. (2023). Using enhanced vegetation index and land surface temperature to reconstruct the solar-induced chlorophyll fluorescence of forests and grasslands across latitude and phenology. Front. For. Glob. Change 6, 1257287. doi:10.3389/ffgc.2023.1257287

Citation

Harmon, C. S., et al. (2026). Hyperspectral indices and global fluorescence: how PACE vegetation indices correlate with terrestrial photosynthesis. Frontiers in Remote Sensing, Volume 7. https://doi.org/
Received
Accepted
Published
Volume
7 - 2026

Copyright

Copyright © 2026 Harmon, Huemmrich, Wynne, Thomas and Cheng.

License: https://creativecommons.org/licenses/by/4.0/

*Correspondence: Caleb S. Harmon

Article metrics

Views
0
Downloads
0
Citations
0
Shares
0

Related articles

Original ResearchQuantifying the accuracy of estimating harmful algal bloom cell concentrations using OLCI imagery in the tributaries of lower Chesapeake BayR. Wayne Litaker, Michelle C. Tomlinson, William C. Holland, Kimberly S. Reece, Todd A. Egerton, Travis O. Briggs, Gail P. Scott, Hamish J. Small, Sylvain Gaillard, Savannah A. MapesOriginal ResearchAssessing urban development trajectories using landscape expansion metrics and multivariate statistical approaches: evidences from rohtak city, NCR, IndiaBharat Bhushan, Surender Singh, Somnath Mandal, Vipasha Sharma, Pradeep Kumar, Attila Fülöp, Sara ElhdadEditorialEditorial: Instruments and Technologies for Earth Observation Satellite MissionsOliver Lux, Gennadi Milinevsky