The study site is located in the Hongla Mountain Yunnan Snub-nosed Monkey National Nature Reserve in Mangkang County, Tibet, China (29°17′10.78′′ N, 98°41′27.45′′ E), the core area of the watershed of the Three Parallel Rivers (Nujiang River, Lancang River, and Jinsha River). The elevation of the study site is 3755 m. The observation period was from November 2020 to October 2022. The study area experiences large diurnal temperature variations and dry conditions in winter, while the summers are warm and humid. The average daily sunshine duration exceeds 10 h, with an annual average temperature of 5° and an average annual precipitation of around 600 mm within a year (Niu et al., 2023). The main tree species in the area include Picea likiangensis var. rubescens, Abies squamata, Sabina tibetica Kom., and Abies ernestii. They are accompanied by the growth of some Quercus aquifolioides, Rhododendron lapponicum, and Potentilla fruticosa shrubs. The average height of the trees is around 30 m, and the forest is in a relatively active growth phase, reaching the state of a mature forest. The vegetation coverage ranges from 70 % to 80 %. The dominant soil type is yellow-brown soil. The mountainous terrain contributes to distinct vertical climate characteristics and significant variations in water and heat conditions, with numerous dry and hot river valleys, widespread canyons, and a clear impact from the southwestern and southeastern monsoons. The varying elevations give rise to diverse ecosystems, transitioning from alpine forests to mountain shrubs. Above 4000 m a.s.l., high alpine grasslands and meadows form a noticeable vegetation transition zone. The mountainous topography results in distinct vertical climate features and significant fluctuations in water and heat conditions, with precipitation showing a markedly uneven distribution throughout the study region (Zemin et al., 2023).

The EC system is deployed at a 35 m high tower located at the study site. At the top of the tower, a 3-D wind velocity (WindMaster, Gill, UK) and an open-path infrared CO2–H2O analyzer (LI-7500DS, Li-Cor, USA) were installed to measure CO2 flux. The instruments had a measurement frequency of 10 Hz. Additionally, micrometeorological sensors were placed at different heights on the tower, including sensors at 35 m for observing air temperature and humidity (HMP155A, Vaisala, Finland), sensors at -5 cm for soil temperature (TEROS 11, Li-Cor, USA), and sensors at 35 m for photosynthetically active radiation (LI-190R, Li-Cor, USA). All data were stored for 30 min in a SmartFlux 3 data logger (Li-Cor, USA) for future download.

When considering only the turbulent transport of matter and energy in the vertical direction, the carbon dioxide flux (Fc) can be represented by the following equation (Yu and Sun, 2006; Monteith et al., 1994): 1FC=W′CO2′‾, where W′ is the vertical component of 3-D wind speed fluctuations (m s-1), and CO2′ represents the fluctuations in the measured CO2 mole concentration. A positive FC indicates carbon emissions, while a negative value represents carbon uptake.

The acquired 10 Hz raw data were processed and corrected using EddyPro software (EddyPro 7.06, Li-Cor, USA). The calibration process involved outlier detection for flux data, lag elimination, coordinate rotation (Jia et al., 2020), ultrasonic temperature correction (Schotanus et al., 1983), frequency correction (Moncrieff et al., 1997), and Webb–Pearman–Leuning (WPL) correction (Leuning and King, 1992). After these controls, the integrity of the effective FC raw data we obtained reached 92.95 %. We removed outliers caused by environmental disturbances such as power outages, rain, snow, and dust particles that interfered with the instrument. Due to the slope of the underlying surface being around 5°, we also corrected for non-uniform and non-flat surfaces using EddyPro for double-coordinate rotation (Cao et al., 2019). As a result, we obtained half-hourly flux data with associated data quality indicators. To evaluate the turbulence steadiness, we employed the “0–1–2” quality assessment method, which classifies flux results into three quality levels: 0 for excellent data quality, 1 for moderate data quality, and 2 for low data quality (Mauder and Foken, 2015; Foken et al., 2005). We removed data points labeled with a quality level of 2. We further eliminated flux data with negative values during nighttime since plants do not perform photosynthesis at night. Additionally, we conducted spectral analysis to identify and remove data points with values significantly deviating from normal. Finally, friction velocities (u*) for each of the 2 years were determined separately using the moving-point method and deleted data recorded during nighttime when u* was less than 0.28 and 0.39 m s-1 (Reichstein et al., 2005). After excluding outliers from the data, the data integrity was 72.67 %. Tovi software (Tovi, Li-Cor, USA) was used in the process.

When turbulence is weak, a portion of CO2 is stored in the vegetation canopy and the atmosphere below the measurement height. At this time, the NEE is calculated as (Zhang et al., 2018) 2NEE=FC+FS, where NEE represents the net ecosystem exchange of CO2, FC stands for the observed flux during a specific period, and FS represents the CO2 storage in the forest canopy. FS is calculated as (Δc/Δt) h, where Δc is the difference in CO2 concentration between two consecutive measurements, Δt is the time interval between two consecutive measurements, and h is 35 m.

We adopted the following formula as a gap-filling strategy for daytime NEE (NEEday) concerning PAR, aiming to address missing values during the daytime (Falge et al., 2001): 3NEEday=α×PAR×Pmaxα×PAR+Pmax-Rday, where α (µmol CO2/µmol PAR) represents the apparent photosynthetic quantum efficiency, which characterizes the maximum efficiency of converting light energy during photosynthesis; PAR (µmol m-2 s-1) is the photosynthetically active radiation, a measure of the amount of light energy available for photosynthesis; Pmax (µmol CO2 m-2 s-1) is the apparent maximum photosynthetic rate, representing the maximum CO2 uptake rate under optimal conditions; and Rday (µmol CO2 m-2 s-1) is the daytime dark respiration rate, which denotes the rate of CO2 release during daylight hours. We obtain α, Pmax, and Rday through the nonlinear fitting of the Michaelis–Menten model to the observed data.

During nighttime, the NEE is modeled using an exponential function of ecosystem respiration and soil temperature to fill in the missing values of NEE during the night (NEEnight) (Lloyd and Taylor, 1994; Kato et al., 2006): 4NEEnight=a×exp(bt), where a and b are estimated values for the exponential function used in modeling NEEnight, and t represents the soil temperature measured at the depth of 5 cm. Origin 2023 (OriginLab Corporation, USA) is the data processing software used for this analysis. For the missing data, interpolation was performed using Tovi software, allowing for data interpolation to fill in the gaps and ensuring a continuous dataset for further analysis (Reichstein et al., 2005). A total of 27.33 % of missing data were interpolated using Tovi after filtering, resulting in a flux dataset with complete data integrity.

In flux analysis, the significance of source area contributions cannot be overlooked. In this study, the peak distances of the 90 % flux contribution areas averaged over 2 years are 364.2 and 357.1 m. In terms of seasons, the average peak distances of the 90 % flux contribution areas for winter, spring, summer, and autumn over the 2 years are as follows: 353.9, 358.2, 350.05, and 344.34 m, respectively.

Ecosystem respiration (ER) is the sum of plant and heterotrophic respiration in an ecosystem and is obtained by adding the measured nighttime data to the extrapolated daytime data. Gross primary productivity (GPP) is the total amount of organic carbon fixed by green plants through photosynthesis per unit of time and per unit of area: 5ER=Rday+Rnight,6GPP=-NEE+ER. Carbon use efficiency (CUE) is a crucial parameter that reflects the ability of an ecosystem to sequester carbon. It is defined as the ratio of net ecosystem productivity (NEP) to gross primary productivity. CUE can be expressed using the following equation: 7CUE=NEPGPP=-NEEGPP. To study the variation in ecosystem respiration rates with environmental factors, we considered the dependence of nocturnal ecosystem respiration on soil temperature (Pavelka et al., 2007; Mamkin et al., 2023): 8Q10=exp(10×α)9ln(NEEnight)=α×T+γ, where T is the soil temperature (°C), and γ is an empirical parameter of the equation.

To clarify the carbon sink potential of forests in the QTP and to compare it with other ecosystems, a search was conducted in two authoritative databases, the Web of Science and China National Knowledge Infrastructure, for research articles on the current utilization of EC systems in the QTP. A total of 82 research results were collected from 48 studies, and their annual average environmental factors, such as air temperature, precipitation, and altitude, were obtained.

During the observational period, the environmental conditions exhibited significant fluctuations. The winter and spring seasons were characterized by cold and dry conditions, while the summer and autumn seasons were warm and humid. The daily maximum AT recorded was 15.87° (on 15 June 2021), and the minimum temperature was -9.88° (on 17 January 2022), with a mean annual average of 5.5° over the 2 years. RH is averaged at 55.89 %, and VPD is averaged at 4.46 hPa. ST exhibited a similar trend to air temperature. The highest observed soil temperature was 13.53° (on 27 June 2021), while the minimum was -3.78° (on 18 January 2022), with an annual average of 6.11°. PAR is averaged at 447.24 mol m-2 s-1.

The observations from the forest ecosystem indicate distinct diurnal and seasonal variations in NEE and GPP (Fig. 3). The NEE and GPP exhibit a pronounced U-shaped curve, with significant seasonal differences. Summer and autumn are characterized by peak carbon uptake, with the maximum NEE reaching 10.70±5.19 µmol CO2 m-2 s-1 (autumn, 12:30 Beijing time, UTC+8:00). During nighttime, the ecosystem generally releases carbon, while during favorable daytime meteorological conditions, it demonstrates a carbon uptake capacity. Peak carbon absorption of the forest ecosystem occurs from 12:00 to 15:00 (Beijing time, UTC+8:00). The daily carbon sequestration in summer and autumn is 1.5–3 h longer than in winter. The timing of maximum carbon sequestration capacity changes with each season. In winter, the transition from nighttime carbon release to daytime carbon uptake occurs around 08:30, which is approximately 1 h later than in summer. GPP characterizes the forest's carbon sequestration capacity, and since photosynthesis does not occur at night, GPP is zero during nighttime. The maximum daily total productivity is recorded at 14.76±7.34 µmol CO2 m-2 s-1 during the summer of the second year, with a standard deviation indicating greater variability in GPP and NEE during summer and autumn compared to winter and spring. Although diurnal variations in ER are relatively small, there are significant seasonal differences. During the night, when only respiration occurs, ER equals NEE. However, as photosynthesis becomes active during the day, ER gradually increases and stabilizes. The respiratory rate of the coniferous forest is the highest in autumn, being 8 times greater than in winter.

Principal component analysis (PCA) of NEE and environmental factors (Fig. 4) indicates that the explanations for the first principal component (PC1) and the second principal component (PC2) are essentially the same between the 2 years. The total contributions of PC1 and PC2 are 87.7 % and 87.5 %, respectively, with PC1 accounting for 64.0 % and 64.6 % individually. The angle between photosynthetically active radiation (PAR) and PC1 is minimal, suggesting a strong correlation between PAR and PC1. Additionally, PAR and VPD contribute the most to PC1, while AT and RH contribute the most to PC2. The analysis results reveal a significant positive correlation between NEE and RH, while a significant negative correlation is observed with AT, VPD, and PAR. Increased RH is detrimental to forest carbon dioxide absorption. Excessively high relative humidity causes plant leaf stomata to close, reducing the amount of carbon dioxide available to the plant. This, in turn, leads to a decrease in the efficiency of carbon fixation through photosynthesis. Among these environmental factors, PAR plays a dominant role. Furthermore, Fig. 4 illustrates the relationships between environmental factors, showing a positive correlation between RH and TA and a negative correlation with VPD and APR. The indicators exhibit some seasonality, with notable differences between the winter–spring and summer–autumn seasons, indicating limited similarity between seasons.

NEE did not show significant interseasonal differences (Fig. 5). However, data distribution indicates that the variability in the NEE rate differs across different seasons, particularly between summer–autumn and winter–spring. The changes in GPP over the 2 years were similar, with significant differences observed between summer and winter (P<0.05). The ER was higher during summer–autumn compared to winter–spring. The highest ecosystem respiration occurred in the first year during autumn, while in the second year, it was the highest during summer. Within the same year, summer and autumn exhibited significant differences (P<0.05), while between the same seasons in different years, notable distinctions were not observed. This pattern is also reflected in GPP and NEE.

The cumulative fluxes over the 2 years for the forest ecosystem are shown in Fig. 6. NEE indicates the net carbon sequestration in each month. The cumulative respiration reached its highest value of 361 g C m-2 in the summer of 2022. The total NEE, GPP, and ER for the first year were -332, 1121, and 788 g C m-2, respectively, and -351, 1199, and 847 g C m-2 for the second year, respectively. CUE was higher during spring and lower during autumn, with a maximum value of 0.74 and a minimum value of 0.07. The average CUE over the 2 years was 0.40 and 0.35.

To clarify the carbon sequestration contribution of the subalpine forests found in the QTP, we compared these research results (Fig. 7). We found that ecosystems with high vegetation cover exhibited higher annual cumulative carbon sequestration. Among these ecosystems, the subalpine forests in the QTP showed the highest carbon sequestration potential, reaching an average of 368 g C m-2 yr-1. The carbon sequestration potential of different ecosystems ranked as follows: forest > meadow > steppe > shrub. The average value for wetlands indicated that they are a significant source of CO2, releasing 57 g C m-2 into the atmosphere annually. We also analyzed the influence of altitude, mean annual air temperature, and precipitation on NEE at these sites in the QTP. It has been observed that these sites cover a wide range of altitudes, ranging from 1977 to 4800 m. According to existing results, an increase in elevation may lead to a reduction in carbon uptake, while the range of mean annual temperature varies between -14.8 and 15.1°, and higher mean annual temperatures significantly increase carbon uptake. Forests exhibit the highest mean annual precipitation, averaging 827 mm, with mean annual precipitation having a relatively weak impact on NEE. These findings highlight the important role of subalpine forests in carbon sequestration in the QTP and provide insights into the factors that affect carbon exchange in the QTP, such as altitude, temperature, and precipitation.

Climate change significantly affects the vegetation's carbon sequestration capacity, particularly at the seasonal scale due to phenological changes (Acosta-Hernández et al., 2020). In the short term, PAR, AT, RH, and VPD play important roles in regulating vegetation photosynthesis and, consequently, carbon uptake. For instance, PAR represents the portion of solar energy that can be utilized by plants and is an essential component in chloroplast reactions. PAR drives a nonlinear response of GPP to solar-induced fluorescence (SIF) across different seasons, resulting in a strong positive correlation between GPP and SIF (Wang et al., 2023b). VPD affects photosynthesis and transpiration of leaves, with stomata serving as tiny pores mediating carbon dioxide uptake. Research has demonstrated that excessive increases in VPD are detrimental to photosynthesis. For instance, a moderate increase in VPD significantly reduces photosynthetic efficiency under light fluctuations due to changes in RH and/or AT often accompanying fluctuations in light, and studies also indicate that the impact of VPD on sunlight utilization efficiency is primarily determined by RH rather than AT (Liu et al., 2024). In different seasons, the same influencing factors exhibit varying degrees of contribution to NEE. For example, during winter, when the climatic conditions are relatively harsh with low air temperature and humidity, the forest maintains a low level of carbon uptake. On longer timescales, such as annual and decadal variations, the inherent changes in forest NEE may be attributed to disturbances and recovery (Hayek et al., 2018). In this study, significant differences in ecosystem respiration were observed during summer and autumn in different years. Previous studies have suggested that due to leaf aging or water stress, the photosynthetic light use efficiency of the ecosystem peaks after spring leaf expansion and gradually declines (Wehr et al., 2016). This implies a peak in carbon exchange during summer, followed by higher productivity and ecosystem respiration in the following seasons. The variation in different years may be attributed to rainfall regulating the availability of natural resources such as water, biomass, litter, and soil nutrients (Schwinning and Sala, 2004). For instance, in temperate forests, when microbial biomass undergoes seasonal changes, microbial activity exhibits a seasonal lag in response to temperature variation, resulting in a seasonally delayed effect between litter heterotrophic respiration and temperature (Ataka et al., 2020). Whether such differences persist between different years on longer timescales remains to be demonstrated through more sustained and detailed research in the future. Ecosystem respiration sensitivity to temperature is represented by the Q10 coefficient. In this study, seasonal variations influenced the magnitude of Q10 (as shown in Fig. 8). The calculated Q10 for each season is as follows: 9.03, 2.22, 2.71, and 4.48. The winter season exhibited the highest sensitivity of forest ecosystem respiration to temperature, indicating that respiration rates in winter are more responsive to changes in temperature compared to other seasons. The main reason for such differences is that ecosystem respiration consists of heterotrophic respiration and autotrophic respiration, which are typically governed by different factors (Edwards, 1975). For instance, the high activity of soil microbes contributes to heterotrophic respiration, a process dominated by soil temperature and moisture conditions, which are severely restricted during the cold and dry conditions of winter (Falge et al., 2002). Simultaneously, due to the changing relative roles of growth and maintenance respiration, the allocation of autotrophic respiration varies seasonally. In winter, soil CO2 emissions constitute a significant portion of ecosystem CO2 emissions, and in some boreal forests, the ratio between the two can reach 0.6 or even higher (Davidson et al., 2006). In winter, under the frequent coverage of snow, cold-adapted microorganisms thriving in a relatively narrow sub-zero temperature range engage in respiration and exhibit relatively high sensitivity to warming or cooling beyond this range (Monson et al., 2006). The seasonal patterns of the Q10 value are jointly determined by the variation in the ratio of soil respiration to ecosystem respiration, reflecting these seasonal changes.

Our integrated analysis (as shown in Fig. 7) reveals that despite the high elevation of the Third Pole, the topographic factor of elevation does not have a significant impact on carbon uptake. Instead, NEE gradually increases with a steep rise in elevation. Research conducted by Wang et al. (2023c) has indicated that mean annual average temperature and precipitation are the main driving factors of interannual variations in NEE in alpine meadows and alpine steppes. Decreased precipitation resulted in a transition into carbon sources in some regions with high precipitation-dependent alpine grasslands. It is worth noting that, among all data collection sites, alpine wetlands show an average carbon source trend. Due to prolonged flooding and low temperatures, microbial activity in alpine wetlands is hindered, and the accumulation of organic carbon from plant litter decomposition is substantial. As a result, approximately 57 g C m-2 is emitted into the atmosphere annually. Previous studies have indicated that NEE in alpine wetlands is increasing with global warming (Yasin et al., 2022).

Subalpine forests are integral components of global alpine ecosystems and play crucial roles in the global carbon cycle. Our study on subalpine forests demonstrates a continuous absorption of carbon dioxide even during winter, which aligns well with measurements taken in the vicinity of Mount Fuji in Japan (Mizoguchi et al., 2012). The age of subalpine forests is a crucial factor influencing sustained carbon sequestration. Based on NPP simulations of natural subalpine forests in the Northern Rockies, Carey (2001) found that aboveground net primary productivity reaches its maximum after approximately 250 years, followed by a decline. This challenges the previous view that forests older than 100 years are generally considered to be unimportant carbon sinks. Compared to the (mature) forest of Mount Gongga in the QTP (e.g., Zhang et al., 2018), the subalpine forest in this study exhibits a stronger carbon sequestration capacity. However, its carbon sequestration ability is slightly weaker than that of the Qilian Mountain high-mountain forests (approximately 60–70 years old) in the QTP (Zhang et al., 2018; Du et al., 2022b). Although existing flux monitoring results of high-altitude forests in the QTP indicate that these forest ecosystems act as carbon sinks, it is important to consider that globally there are still many cold regions with coniferous forests serving as carbon sources. For example, continuous CO2 flux monitoring from native boreal forests in Sweden for over 10 years indicates that they are a net carbon source, which is attributed to the contribution of woody debris to ER due to disturbances such as extreme weather events, fires, insect infestations, and pathogen attacks (Hadden and Grelle, 2017). In the summer of 2018, Europe experienced a heat wave that affected the carbon cycling in forests. The mixed coniferous–deciduous forest in southern Estonia, under the influence of the heat wave, transitioned from a net carbon sink to a net carbon source in 2018 (Krasnova et al., 2022). Particular attention should be paid to the long-term monitoring in high-altitude environments of the impact of disturbances on forest carbon sequestration capacity. Our study has shown that forests in the QTP have the strongest carbon sink capacity, indicating that alpine forests will have an important sustained effect on carbon reduction in the QTP in the context of future climate change, but whether this sustained effect will be longer than other ecosystems is still unknown. However, a modeling experiment in a large semi-arid area of California predicted that grasslands are more resilient carbon sinks than forests in responding to climate change in the 21st century (Dass et al., 2018). In terms of carbon sequestration rate, forests in the QTP were significantly stronger than other ecosystems, followed by grasslands, while alpine deserts and alpine grasslands in the northwestern and southern regions were the main carbon sources (Wu et al., 2022). Forests are mostly distributed in the southeastern margin of the QTP and the mid-altitude area near 3000 m in the Sichuan–Tibet alpine gorge area, with an area of 19.3×104 km2 (Z. Y. Wang et al., 2022b). Based on the average value of current carbon flux monitoring, the forest in the QTP will absorb about 71×106 Mg C yr-1.