Seasonal Variations in the Relationship Between Canopy Solar-Induced Chlorophyll Fluorescence and Gross Primary Production in a Temperate Evergreen Needleleaf Forest

Abstract

The temperate evergreen needleleaf forest (ENF), primarily composed of Mongolian Scots pine (Pinus sylvestris var. mongolica), plays a pivotal role in the “The Great Green Wall” Shelterbelt Project in northern China as a major species for windbreak and sand fixation. Solar-induced chlorophyll fluorescence (SIF) has emerged as a revolutionary remote sensing signal for quantifying photosynthetic activity and gross primary production (GPP) at the ecosystem scale. Meanwhile, eddy covariance (EC) technology has been widely employed to obtain in situ GPP estimates. Although a linear relationship between SIF and GPP has been reported in various ecosystems, it is mainly derived from satellite SIF products and flux-tower GPP observations, which are often difficult to align due to mismatches in spatial and temporal resolution. In this study, we analyzed synchronous high-frequency SIF and EC-derived GPP measurements from a Mongolian Scots pine plantation during the seasonal transition (August–December). The results revealed the following. (1) The ENF acted as a net carbon sink during the observation period, with a total carbon uptake of 100.875 gC·m⁻². The diurnal dynamics of net ecosystem exchange (NEE) exhibited a “U”-shaped pattern, with peak carbon uptake occurring around midday. As the growing season progressed toward dormancy, the timing of CO₂ uptake and release gradually shifted. (2) Both GPP and SIF peaked in September and declined thereafter. A strong linear relationship between SIF and GPP (R² = 0.678) was observed, consistent across both diurnal and sub-daily scales. SIF demonstrated higher sensitivity to light and environmental changes, particularly during the autumn–winter transition. Cloudy and rainy conditions significantly affect the relationship between SIF and GPP. These findings highlight the potential of canopy SIF observations to capture seasonal photosynthesis dynamics accurately and provide a methodological foundation for regional GPP estimation using remote sensing. This work also contributes scientific insights toward achieving China’s carbon neutrality goals.

1. Introduction

Terrestrial ecosystems can absorb up to 33% of total anthropogenic carbon emissions (from both fossil fuel combustion and land use), contributing to the reduction of atmospheric CO₂ levels [1]. Gross primary production (GPP) refers to the amount of CO₂ fixed by terrestrial vegetation per unit time through the reduction of CO₂ into organic compounds via photosynthesis, representing the primary pathway of carbon input into ecosystems and the fundamental mechanism underlying terrestrial carbon sequestration [2,3]. Accurately characterizing photosynthesis (i.e., GPP) at spatial and temporal scales provides crucial information on the timing, location, and amount of carbon absorbed [4,5]. This is a critical component of terrestrial carbon budgets. Changes in vegetation structure, growing season, and climate-responsive functions across ecosystems indicate that GPP is undergoing global changes [6]. A precise understanding of GPP is fundamental for comprehensively evaluating the role of terrestrial ecosystems in the global carbon cycle and climate change, yet global GPP remains difficult to quantify accurately.

GPP estimation primarily relies on eddy covariance (EC) flux observations and greenness-based remote sensing models. EC flux towers are considered one of the most accurate methods for quantifying canopy-scale biomass productivity [7,8,9]. By measuring the net ecosystem exchange (NEE) of CO₂ between vegetation and the atmosphere, EC reflects the overall carbon exchange processes within the flux footprint, thereby enhancing our understanding of the terrestrial carbon cycle [10,11]. However, due to the limited number of observation sites and their uneven spatial distribution, the representativeness and coverage of EC flux towers at the global scale are constrained. Greenness models based on remote sensing vegetation indices (e.g., NDVI) serve as an important approach for GPP estimation and have been widely applied in monitoring large-scale carbon fluxes [12,13,14]. These models estimate potential photosynthetic activity based on the correlation between vegetation indices and photosynthetic capacity, but they are limited in capturing instantaneous photosynthesis or the dynamic nature of actual GPP [15,16]. Moreover, GPP estimations based on light use efficiency (LUE) models are highly dependent on the accuracy of meteorological input data, and uncertainties in model structure and parameterization further limit their applicability in complex ecosystems [17].

In recent years, solar-induced chlorophyll fluorescence (SIF) has gained increasing attention as a physiological indicator of GPP [18]. Unlike reflectance-based vegetation indices, SIF originates directly from the photosynthetic process and can more accurately reflect plant photosynthetic activity [19]. Therefore, SIF is regarded as a more sensitive indicator of vegetation’s carbon uptake capacity [20]. With the advancement of satellite remote sensing technologies, large-scale retrieval of SIF has become increasingly feasible. Studies have shown that SIF can be integrated into land surface models to improve GPP estimation, offering a novel pathway for remote-sensing-based GPP monitoring. Early studies using low-spatial-resolution satellite SIF data (tens of kilometers) revealed a strong linear relationship between SIF and GPP [21]. The emergence of high-resolution satellite platforms, such as NASA’s Orbiting Carbon Observatory-2 (OCO-2) with a spatial resolution of 1.2 × 2 km, has made it possible to match satellite-derived SIF data with ground-based EC observations of GPP [22]. Research has demonstrated a high degree of consistency between SIF and GPP across diurnal, seasonal, and interannual timescales, supporting the stability and scalability of SIF as an indicator of GPP [23].

To address the scale mismatch between satellite remote sensing and ground-based observations, an increasing number of studies have focused on the relationship between SIF and GPP at the canopy scale. At this scale, numerous studies have explored the response of SIF to photosynthesis in agricultural systems, such as maize (Zea mays L.), rice (Oryza sativa L.), and broadleaf tree species, such as beech (Fagus grandifolia Ehrh) and red oak (Quercus rubra) [24,25,26]. In ENFs, the relationship between SIF and GPP varies depending on the forest type. A linear relationship has been observed in lodgepole pine (P. contorta Douglas ex Loudon) and Engelmann spruce (Picea engelmannii Parry ex Engelm) in Colorado [21], whereas a nonlinear relationship has been reported in Korean pine (Pinus koraiensis Siebold et. Zucc) ecosystems [27]. These contrasting patterns may be driven by differences in canopy structure, leaf area index, and environmental conditions, suggesting that the SIF–GPP relationship in ENFs is modulated by a combination of biophysical and ecological factors [28,29]. The Mongolian Scots pine (Pinus sylvestris var. mongolica), a key afforestation species in China’s “The Great Green Wall” Shelterbelt Program, plays an important ecological role in combating desertification. However, as a representative semi-arid evergreen needleleaf ecosystem, its carbon exchange dynamics and seasonal variations in photosynthetic activity remain unclear, and it is still uncertain whether SIF can accurately reflect the temporal dynamics of its GPP.

In this study, we conducted tower-based synchronous SIF and EC observations in a Mongolian Scots pine plantation in northern China and estimated GPP using EC measurements. Our objectives were to (1) evaluate the carbon sequestration capacity in this ENF ecosystem; (2) investigate the relationship between SIF and GPP, with an emphasis on its seasonal dynamics; and (3) examine the environmental drivers of SIF and GPP during seasonal transitions.

2. Materials and Methods

2.1. Study Location

The study site of the Mongolian Scots pine plantation is located at the border between Jianping County, Chaoyang City, Liaoning Province, and the Inner Mongolia Autonomous Region over the period of 1 September to 30 November 2020 (Figure 1a). The dominant landform is low mountainous hills with an elevation of 871 m above sea level. The site is geographically situated at 41°58′169″ N, 119°25′104″ E and lies within a temperate continental monsoon climate zone. The region is characterized by pronounced monsoon activity and prevailing winds of 2–3 on the Beaufort scale. Spring is dry with little rainfall, while winter lasts approximately 140 days, typically from early November to late March of the following year. Summer lasts for about 90 days, and both spring and autumn are relatively short, each lasting less than two months.

The annual average temperature is 7.6 °C, with extremes ranging from a minimum of −29 °C to a maximum of 42.2 °C. The region receives approximately 2765 h of sunshine annually, and the average daily sunshine duration is approximately 7.5 h per day. The mean annual precipitation is 450 mm, with most rainfall concentrated between June and August. The average frost-free period is 138 days per year.

The ENF in the study area is approximately 30 years old, with an average tree height of 10 meters and virtually no understory vegetation (Figure 1b).

2.2. Eddy Covariance (EC) Flux Measurements

Open-path eddy covariance technology measures CO₂ flux by calculating the covariance between fluctuations in vertical wind speed and CO₂ concentration. A three-dimensional sonic anemometer (CSAT 3, Campbell Scientific, Logan, UT, USA; Figure 1d) and an open-path infrared CO₂/H₂O gas analyzer (IRGA; LI-7500A, LI-Cor., Lincoln, NE, USA) were installed at 15 m height on the observation tower. The LI-7500A was configured with a temperature control mode set to 5 °C. Both instruments operated at a sampling frequency of 10 Hz, with data automatically collected and stored by a datalogger (CR2000, Campbell Scientific, Logan, UT, USA).

2.3. Solar-Induced Chlorophyll Fluorescence (SIF) Observations

Canopy-level SIF was monitored using an automated chlorophyll fluorescence observation system (AUTOSIF, Bergson Spectrum Science Limited., Beijing, China). The AUTOSIF system included a high-sensitivity spectrometer (QEPRO, Ocean Optics, Dunedin, FL, USA) for retrieving SIF in the O₂-A absorption band, featuring a signal-to-noise ratio (SNR) of 1000, a wavelength range of 730–785 nm, and a spectral sampling interval of 0.07 nm. A high-throughput optical fiber was installed 5 m above the forest canopy, targeting a circular observation area with a diameter of 4.4 m. The downwelling irradiance was collected using a cosine-corrected optical configuration, while the target surface was observed via a bare optical fiber (Figure 1c,e). The fiber type used in the system was a 600 μm core fiber (Ocean Optics, Dunedin, FL, USA). Radiometric and wavelength calibrations were performed using a mercury calibration lamp (HL-3 Plus, Ocean Optics Inc., Dunedin, FL, USA) and an integrating sphere (LightFluxColor LFC-illumia Plus2., Labsphere, North Sutton, NH, USA), respectively.

2.4. Meteorological Observations

Meteorological data primarily included air temperature (T_air) and relative humidity (RH) (HMP45C, Vaisala, Helsinki, Finland), and photosynthetically active radiation (PAR) (LI-190, LI-COR., Lincoln, NE, USA). Data from the conventional meteorological observation system were collected at a sampling frequency of 0.5 Hz using a CR2000 datalogger (Campbell Scientific, Logan, UT, USA), which automatically recorded and stored the measurements.

2.5. Data Processing

2.5.1. CO₂ Flux Correction and Gap Filling

The raw data obtained from flux observations are based on ideal conditions (e.g., flat and homogeneous underlying surfaces) and therefore require further processing, including double coordinate rotation and Webb–Pearman–Leuning (WPL) correction. All of these procedures were performed using EddyPro software (v7.0.6, LI-Cor., LI-COR., Lincoln, NE, USA). Outlier detection was conducted separately for daytime and nighttime data within a 13-day moving window. Nighttime periods were identified using a global solar radiation (R_g) threshold of 20 W/m², and day/night segmentation was determined based on R_g.

The R package REddyProc (v1.3.3) [30] was employed to perform u* filtering with 200 bootstrapped resamplings (default setting). The gap-filling procedure was applied to data that had already undergone outlier removal, and the Marginal Distribution Sampling (MDS) method was used for flux data interpolation.

2.5.2. NEE Partitioning and GPP Estimation

In this study, photosynthesis is quantified as GPP, which represents the total amount of CO₂ fixed by vegetation at the ecosystem scale. The estimation process was performed in two steps. First, ecosystem respiration (Reco) was estimated by partitioning the NEE based on environmental drivers. Second, GPP was derived by subtracting NEE from Reco.

The NEE data were partitioned based on the light response curve of NEE to R_g, the effect of T_air on Reco, and the influence of R_g and vapor pressure deficit (VPD) on GPP. The temperature sensitivity parameter E₀ was fitted using nighttime NEE and T_air data within a 15-day window, with T₀ fixed at −46.02 °C and T_Ref at 15 °C. The optimal E₀ was determined as the value corresponding to the smallest standard deviation and used as the annual E₀. Based on the fitted E₀, the reference respiration rate R_Ref was calculated using a 7-day moving window with a 4-day step size (3-day overlap). Curve fitting and linear interpolation were used to derive a unique R_Ref for each sample [31].

$$NEE={\displaystyle \frac{\alpha \beta R_g}{\alpha R_g+\beta }}+R_{Ref}exp\left[E_0\left({\displaystyle \frac{1}{T_{Ref}-T_0}}-{\displaystyle \frac{1}{T-T_0}}\right)\right]$$

(1)

$$\beta =\left\{\begin{array}{ll}{\beta }_0\mathit{exp}\left(-k\left(VPD-{VPD}_0\right)\right)& VPD>{VPD}_0\\ \beta ={\beta }_0& VPD<{VPD}_0\end{array}\right.$$

(2)

In these equations, R_Ref is the baseline ecosystem respiration rate at the reference temperature, E₀ is the temperature sensitivity parameter, T_Ref is the reference temperature, and T₀ is a constant. α is the canopy light use efficiency (μmolC·m⁻²·s⁻¹), β is the maximum CO₂ uptake rate (μmolC·m⁻²·s⁻¹), and the parameter k is fitted using a 4-day window, assuming VPD₀ = 0.

Following the estimation of Reco, GPP was calculated based on the following equation [30]:

$$GPP=Reco-NEE$$

(3)

where GPP is gross primary production (gC·m⁻²·day⁻¹), Reco is ecosystem respiration (gC·m⁻²·day⁻¹), and NEE is net ecosystem carbon exchange (gC·m⁻²·day⁻¹).

All flux data processing was performed using the REddyProc package in the R environment (v4.5.0) [31].

2.5.3. Solar-Induced Chlorophyll Fluorescence (SIF) Data Processing

The retrieval of SIF signals is based on algorithms derived from the classical Fraunhofer Line Discrimination (FLD) method [32]. In this study, we employed the Spectral Fitting Model (SFM) algorithm. This method performs least-squares fitting using multiple spectral bands over a relatively wide wavelength window based on the natural spectral characteristics of both SIF and surface reflectance to reconstruct a continuous SIF spectral curve. It assumes that surface reflectance and the fluorescence spectrum conform to specific mathematical functions (e.g., linear, polynomial, or Gaussian functions), enabling spectral fitting and inversion of the SIF signal [30,33]):

$$L\left(\lambda \right)={\displaystyle \frac{r_{MOD}\left(\lambda \right)·E\left(\lambda \right)}{\pi }}+F_{MOD}+\epsilon \left(\lambda \right)=L_{MOD}\left(\lambda \right)+\epsilon (\lambda )$$

(4)

where r_MOD(λ) is the modeled reflectance, F_MOD(λ) is the modeled fluorescence, L_MOD(λ) is the modeled apparent radiance, and ε(λ) is the residual between the observed and modeled values at each wavelength, representing the model fitting error.

The SFM algorithm was adopted as the standard fluorescence retrieval algorithm for the FLEX mission. It was widely used in top-of-atmosphere SIF retrievals in the O₂-A and O₂-B bands, and its performance and accuracy were extensively evaluated. Compared to the FLD family of algorithms, the SFM method had higher spectral resolution requirements but offered the advantage of retrieving full-spectrum fluorescence within the fitting window and greater resistance to noise.

3. Results

3.1. Seasonal Variation in GPP and SIF Under Changing Conditions

In the analysis of meteorological factors, GPP, SIF, PAR, and evapotranspiration (ET) in the study area, we observed distinct seasonal variation patterns in all of these factors. The daily average of R_g closely followed changes in PAR, but, unlike other factors, both R_g and PAR exhibited a steady decline throughout the observation period, reaching approximately half of their values in winter compared to other seasons. Combined with precipitation data, this suggests a higher frequency of cloudy and rainy days during late summer and early autumn (Figure 2b,e), contributing to greater fluctuations prior to the onset of winter. The daily average T_air ranged from −13.8 °C to 30.0 °C and dropped below 0 °C on Julian day (DOY) 322 (where DOY represents the day of the year starting from January 1st as DOY 1) (Figure 2a). Similarly, the soil water content (SWC) at a depth of 10 cm exhibited seasonal trends similar to those of radiation and temperature, with significant fluctuations in late summer and autumn, and it remained low and stable during winter (Figure 2h). Distinct from other meteorological variables, VPD showed substantial fluctuations until approximately DOY 320, after which it rapidly decreased and stabilized (Figure 2g). Overall, GPP, SIF, and all meteorological factors exhibited a declining trend from DOY 230 to DOY 360 (Figure 2). During late summer and autumn (DOY 230–270), GPP and SIF fluctuated sharply. Further analysis of rainfall, radiation, soil moisture, and VPD indicated similar fluctuations during the same period. Particularly noteworthy is the transitional period between autumn and winter (DOY 290–310), during which temperature dropped below 0 °C and VPD plummeted from around 10 kPa to near 0, accompanied by sharp declines in SIF, GPP, and ET. Compared to other meteorological factors, ET and SIF exhibited the most synchronized temporal response to seasonal changes (Figure 2d,f), both approaching zero around DOY 293. Notably, SIF declined more rapidly than GPP, leading by approximately 20 days (DOY 285–310), and it approached zero earlier.

In addition to overall seasonal patterns, the diel dynamics of CO₂ fluxes revealed distinct seasonal differences (Figure 3). Because plant respiration emits CO₂, CO₂ fluxes are positive during the day, indicating a carbon source, and negative at night, indicating a carbon sink. From August to December, NEE in the study area was 100.875 gC·m⁻², suggesting that the ecosystem functioned predominantly as a carbon sink. The only exception occurred in November, where it acted as a carbon source, releasing 0.989 gC·m⁻².

Trends in GPP and Reco exhibited similar temporal trends, characterized by an initial increase followed by a subsequent decline. The diurnal variation in NEE exhibited a “U-shaped” pattern (Figure 3), with negative values during daylight hours and positive values at night. In August, the minimum NEE reached −7.0 μmol·m⁻²·s⁻¹ at approximately 12:00, while in December the minimum dropped only to −3.62 μmol·m⁻²·s⁻¹ around 10:00, indicating a significant seasonal decline in photosynthetic capacity.

From August to December, the timing of the minimum (most negative) NEE advanced from around 12:00 to 10:00, indicating a seasonal decline in photosynthetic capacity. Similarly, the transition from a carbon sink back to a source occurred between 16:30 and 17:30, with later transitions observed closer to the peak of the growing season (Figure 3).

3.2. Variations in the Correlation Between GPP and SIF Under Seasonal Transitions

At the half-hourly scale, both SIF and GPP exhibited strong temporal correspondence across different months (Figure 4). The hourly R² values for each month were August—0.698; September—0.748; October—0.884; November—0.452; and December—0.293. Both SIF and GPP exhibited diurnal patterns with values increasing from morning to midday, peaking around 12:00, and declining in the afternoon. However, GPP exhibited a more stable temporal pattern compared to SIF. From August to December, the SIF value at 12:00 decreased from 0.42 to 0.003 μW·cm⁻²·nm⁻¹·sr⁻¹, while GPP decreased from 14.35 gC·m⁻²·day⁻¹ to 2.84 gC·m⁻²·day⁻¹. Notably, SIF responded earlier than GPP in terms of both its daily onset and decline. At sunrise (around 6:30), while GPP remained minimal, SIF had already responded to early light conditions. In the afternoon (around 17:00), SIF had nearly dropped to zero, while GPP continued to decline. These observations suggest that weak photosynthetic activity persisted in the ENF during the winter months of November and December.

At the daily scale, our results indicated a significant linear relationship between GPP and SIF in the Pinus sylvestris var. mongolica plantation from August to December, with an overall coefficient of determination (R²) of 0.6784 (Figure 5). This relationship was consistently strong at both daily and sub-daily scales (Figure 5). Specifically, the monthly correlations (R²) between GPP and SIF were August—0.626; September—0.249; October—0.619; November—0.301; and December—0.086. During August and September, the scatter points were densely distributed along the regression line, with daily GPP values ranging from 1.65 to 12.25 gC·m⁻²·day⁻¹ and SIF values from 0.003 to 0.045 μW·cm⁻²·nm⁻¹·sr⁻¹, indicating both high photosynthetic activity and strong SIF sensitivity. In contrast, October and November were characterized by low solar radiation and frequent cloudy days, leading to larger scatter in the data and a weakened correlation, with SIF below 0.004 μW·cm⁻²·nm⁻¹·sr⁻¹ and GPP rarely exceeding 2.0 gC·m⁻²·day⁻¹.

The correlation peaked in October during the seasonal transition from autumn to winter (DOY 290–310), a period of the most pronounced changes in both variables. In contrast, during September—characterized by low radiation and frequent cloudy or rainy days—the SIF–GPP correlation weakened. As winter approached and photosynthetic activity nearly ceased, the relationship became negligible by December.

We analyzed the sensitivity of SIF and GPP to key meteorological factors, including PAR, VPD, ET, R_g, T_air, and SWC. Although both SIF and GPP were significantly influenced by environmental factors, their correlation patterns exhibited notable differences (Figure 6).

Both SIF and GPP showed strong positive correlations with most environmental factors, especially with ET, PAR, and SWC. However, there were differences in the strength of these correlations. For instance, the correlation between SIF and ET was 0.78, whereas GPP and ET showed a higher correlation of 0.84, indicating that ET has a more significant impact on GPP. Conversely, SIF was slightly more sensitive to PAR (R² = 0.79) compared to GPP (R² = 0.72).

Furthermore, the two indicators responded differently to temperature and radiation. SIF exhibited stronger correlations with T_air and R_g (R² = 0.64 and 0.71), while the correlations between GPP and these factors were slightly lower (R² = 0.77 and 0.66). This suggests that SIF, as a proxy for chlorophyll fluorescence, responds more acutely to light and thermal conditions, while GPP—representing broader canopy-level carbon assimilation—is more regulated by the combined effects of temperature and water availability.

In summary, although SIF and GPP are both modulated by common environmental factors, SIF appears more sensitive to light-related variables, while GPP is more dependent on moisture and thermal conditions. The differences in the correlation strengths between SIF and GPP provide valuable insights for further understanding how environmental factors influence plant photosynthesis and ecosystem carbon dynamics.

4. Discussion

4.1. Carbon Sequestration Capacity Across Different Ecosystems in Semi-Arid Regions

The NEE of the ENF (Pinus sylvestris var. mongolica) from August to December was −100.875 gC·m⁻², indicating that the ecosystem functioned as a carbon sink. Previous studies have reported relatively high carbon sink capacities in comparable arid ecosystems. For example, the Mojave Desert in the United States exhibited a total NEE ranging from −127 to −102 gC·m⁻² [34,35], which is higher than that reported for the Tengger Desert in China (−23.4 to −13.9 gC·m⁻²) [36] and the Fukang Desert ecosystem (−49 gC·m⁻²) [37]. Other studies have also reported significant variability in the carbon source–sink status, even within the same desert ecosystem. For instance observed that the NEE in the Baja California region of Mexico ranged from −52 to −9 gC·m⁻² [38].

In comparison, the total NEE of the ENF (Pinus sylvestris var. mongolica) was higher than that of other semi−arid ecosystems in northern China, such as the desert steppe in Inner Mongolia (−69 to −7.2 gC·m⁻²) [39] and the semi-arid degraded grassland in Tongyu, Jilin (−73 to 0.32 gC·m⁻²) [40]. Overall, these results suggest that the ENF, as one of the key afforestation species in China’s “The Great Green Wall” Shelterbelt Project, possesses a strong carbon sequestration capacity, highlighting its important role in sand fixation and ecological protection.

4.2. SIF and GPP Relationship and Environmental Drivers

Previous studies have consistently demonstrated a significant linear relationship between GPP and SIF across a variety of ecosystem types. Strong correlations have been observed in deciduous broadleaf forests, such as beech (Fagus grandifolia Ehrh) and red oak (Quercus rubra), at the Harvard Forest site in the United States [24], as well as in key agricultural systems, including soybean (Glycine max L. Merr) fields in Iowa [19], rice (Oryza sativa L.) paddies in Jurong, China [26], and maize (Zea mays L.) fields in both Maryland, USA [25], and Zhangye, China [41].

In agreement with these findings, our study revealed a significant linear correlation between SIF and GPP in the temperate evergreen needleleaf forest (ENF), with a coefficient of determination (R²) of 0.678. In the ENF, the relationship between SIF and GPP is jointly influenced by absorbed photosynthetically active radiation (APAR), as well as by the distinct stability of the canopy structure and the seasonal dynamics of leaf-level physiology [27]. Compared to crops and deciduous forests, ENFs exhibit minimal seasonal variation in canopy structural parameters, such as leaf area index (LAI), leaf angle distribution, and leaf clumping, leading to a relatively stable canopy escape fraction (f_esc) [42]. However, this relationship was not temporally uniform. Seasonal changes in environmental conditions, particularly radiation and precipitation, were found to influence the strength of the SIF–GPP coupling. For example, due to more frequent cloudy and rainy days, the correlation in August was markedly weaker than that observed in September and October, which were characterized by higher radiation levels. In ENFs, seasonal variation in SIF is primarily driven by changes in the chlorophyll fluorescence yield (Φ_F), which is regulated by physiological adjustments in response to environmental stresses. For instance, during winter, ENF trees activate photoprotective mechanisms that increase the carotenoid/chlorophyll ratio, enhancing sustained non-photochemical quenching (Φ_sNPQ) to prevent photodamage [43]. As environmental conditions improve in spring, this mechanism is reversed, allowing for higher Φ_PS(II) and reversible NPQ (Φ_rNPQ) to dominate energy partitioning [44]. In contrast, GPP is mainly regulated by canopy light use efficiency (LUE_P) and leaf gas exchange. Thus, the SIF–GPP relationship in ENFs reflects a unique coupling between stable canopy structure and seasonally dynamic leaf physiology [21].

These results suggest that the SIF–GPP relationship is modulated by environmental drivers, with SIF and GPP exhibiting different sensitivities to microclimatic factors. While GPP reflects net carbon assimilation and is influenced by both radiative and physiological constraints, SIF is more directly linked to absorbed light and its subsequent dissipation. Our continuous ground-based observations during autumn and early winter provided detailed insight into canopy SIF responses under variable weather conditions. The diurnal patterns of SIF differed seasonally, with stronger signals in autumn and near-zero levels in winter. This is consistent with observations in coniferous forests in Colorado [21].

Notably, SIF exhibited a continuous increase with PAR throughout the growing season, with no evidence of midday depression or saturation. In contrast, GPP was primarily regulated by SWC. This partial decoupling of SIF and GPP under certain conditions is consistent with findings from rice paddies, where SIF showed a stronger correlation with absorbed light than GPP did [45]. The divergence arises from energy allocation mechanisms, as SIF serves as an important non-photochemical dissipation pathway that protects the photosynthetic apparatus under excess light, whereas GPP depends on additional biochemical and water-related constraints.

Furthermore, Chen et al. [46] reported that GPP and SIF tend to decline along a climatic gradient from cold and humid regions to hot and arid regions. Their study also found that under equivalent temperature conditions, increased precipitation can elevate the GPP/SIF ratio. This is consistent with our observations, where frequent precipitation in August likely disrupted direct coupling between SIF and GPP, leading to a lower correlation compared to the drier months of September and October.

4.3. Limitations and Uncertainties

Although this study benefits from continuous, ground-based observations of both SIF and GPP in a semi-arid evergreen needleleaf forest, several limitations and uncertainties remain that should be addressed in future work.

First, the issue of footprint mismatch between flux and spectral observations may affect the accuracy of the observed SIF–GPP relationship. While both the EC system and the automated tower-based spectral system were installed on the same tower and the SIF field of view was largely within the EC footprint, this does not fully resolve the potential discrepancy in spatial representativeness. In heterogeneous or semi-structured ecosystems, such as semi-arid woodlands, even slight misalignments between the observation footprints can introduce inconsistencies in data interpretation [47,48]. A more detailed analysis of spatial matching, including high-resolution footprint modeling, is recommended to further constrain this source of uncertainty.

Second, this study focused on a single vegetation type within a semi-arid biome. While the ENF ecosystem is a representative afforestation type in northern China, the applicability of our findings to other semi-arid ecosystems, such as desert steppe, shrubland, or mixed savanna, remains unclear. Future studies should consider the deployment of coordinated multi-site observational networks that encompass diverse ecosystem types, thereby enabling cross-site comparisons of SIF–GPP relationships under different structural and climatic regimes.

Third, our results highlight the high sensitivity of SIF to incident radiation, particularly under varying sky conditions. SIF values measured only on clear days may overestimate the seasonal mean value, especially during periods with frequent cloudy or rainy weather. This suggests that solar radiation conditions play a critical role in shaping the observed SIF–GPP relationship, at both diurnal and seasonal scales. As such, future studies should incorporate the classification of sky conditions (e.g., clear vs. overcast) into the analysis framework to better interpret the SIF response. Distinguishing between the influence of diffuse versus direct radiation on SIF could also help reveal mechanisms underlying its variability and decoupling from GPP [49].

Taken together, these limitations suggest that although SIF holds strong potential as a proxy for photosynthetic activity, its interpretation requires careful consideration of the observational scale, ecosystem diversity, and the environmental context. Addressing these uncertainties will be essential for improving the robustness of SIF-based GPP estimation models, particularly in complex, semi-arid landscapes.

5. Conclusions

Based on the study conducted in the ENF (Pinus sylvestris var. mongolica) plantation, we found the following. (1) From August to December, the total net carbon uptake of the Mongolian Scots pine plantation was 100.875 gC·m⁻², indicating that the ecosystem functioned overall as a carbon sink. The diurnal variation of CO₂ flux exhibited a “U-shaped” pattern, with the strongest carbon sequestration occurring at 12:00 noon. Under seasonal transition conditions, both GPP and SIF exhibited a gradual decline, with greater fluctuations during autumn and particularly strong responses during the transition from autumn to winter. (2) Meteorological factors in the plantation showed distinct seasonal variation patterns. A clear linear relationship was observed between SIF and GPP (R² = 0.678), with high consistency at both daily and sub-daily time scales. The correlation between SIF and GPP was especially strong at the sub-daily scale, and SIF was found to respond to photosynthetic activity earlier than GPP. Their correlation was strongest during the autumn–winter transition period. Both SIF and GPP reached their peak around 12:00 noon, although GPP exhibited a more stable diurnal pattern than SIF. Notably, even during winter months (November and December), weak photosynthetic activity was still detected through both SIF and GPP. Furthermore, both SIF and GPP were closely related to soil water content and total radiation, and their relationship was significantly affected by cloudy and rainy weather conditions.