Nonmonotonic Elevational Patterns of Soil CO₂ Flux Driven by Temperature Dominance and Moisture Thresholds in the Sejila Mountains, Tibetan Plateau

Abstract

Understanding spatiotemporal variation in soil CO₂ flux (FCO₂) along elevational gradients is essential for predicting carbon–climate feedback in alpine ecosystems. However, how temperature- and moisture-related factors jointly regulate daily-scale FCO₂ and how their contributions vary with elevation remain unclear, particularly in the Sejila Mountains (Southeastern Tibetan Plateau). We conducted continuous in situ measurements of daily-scale FCO₂, air temperature (Ta), relative humidity (RH), soil temperature (ST, 0–10 cm), and volumetric soil water content (SW) across five elevational bands (3000–4200 m) in 2024–2025. Across both years, FCO₂ showed a unimodal seasonal cycle and a robust nonmonotonic spatial pattern, with the highest efflux at 3000 and 4200 m and peak rates exceeding 5.0 µmol CO₂ m⁻² s⁻¹. Cumulative carbon loss at 4200 m (909.90 g C m⁻²) exceeded that at mid-elevation sites. Linear mixed-effects models identified Ta as the most consistent positive predictor; the ST × SW interaction was not significant, indicating that temperature and moisture effects are largely additive at the daily scale. Piecewise regression revealed nonlinear SW thresholds (θ) in the FCO₂ response, with θ varying nonmonotonically with elevation. Multiple linear regression further showed that thermal predictors (Ta, ST) explained substantially more variance than moisture predictors (RH, SW), and the relative importance of thermal drivers increased with elevation. These results challenge the common expectation of a monotonic decline in soil respiration with elevation and suggest that, when SW remains above critical thresholds, warming may amplify soil carbon losses at high elevations on the Tibetan Plateau.

1. Introduction

The Tibetan Plateau (TP), located in the interior of Eurasia, has its highest point exceeding 8000 m, while the northeastern edge and river valley regions descend to below 2200 m [1]. Its average elevation exceeding 4000 m, earning it the ‘Third Pole’ [2,3]. It plays a crucial role in regulating the Asian monsoon, large-scale atmospheric circulation, and downstream water availability [4,5,6], and it is widely regarded as both a hotspot and an amplifier of global climate change [3,7]. In recent decades, the TP has warmed markedly faster than the global mean [8], driving substantial shifts in hydrothermal regimes and, in turn, altering the structure and functioning of regional ecosystems [9,10]. Soil respiration is a major pathway of CO₂ flux from terrestrial ecosystems to the atmosphere [11,12]. Its magnitude and spatiotemporal variability strongly influence ecosystem carbon budgets [13,14,15] and are central to quantifying terrestrial carbon-cycle responses and feedback to climate change [15,16,17]. Therefore, elucidating soil-respiration dynamics and their environmental controls in representative TP ecosystems is essential for advancing process understanding in the Third Pole and for better constraining related model simulations.

FCO₂ is jointly controlled by multiple factors, including temperature, soil moisture, substrate supply, and biotic community composition [18,19,20]. Beyond these biological and climatic drivers, soil compaction, a physical disturbance resulting from human activities like forestry operations, also critically influences CO₂ dynamics. Compaction alters soil structure, reduces pore space, and thus impedes gas diffusion, leading to elevated CO₂ concentrations, even from seemingly mild disturbances with long-term implications [21]. Among these, temperature and moisture are the most immediate and sensitive environmental drivers [22,23]. Extensive evidence shows that soil respiration generally increases with temperature, a relationship commonly described using exponential models such as the Arrhenius equation or the Lloyd–Taylor function [16,24]. Recent global syntheses further support this positive temperature dependence. For example, Bond-Lamberty et al. [25] linked multi-decadal increases in soil respiration to concurrent warming based on global records. However, temperature sensitivity is not constant: long-term warming experiments by Luo et al. [26] documented thermal acclimation, with an initially strong stimulation of soil respiration weakening over time, implying that seasonal to interannual temperature variability can alter patterns of soil carbon flux. Consistently, Reichstein et al. [27] used multifactor models to show that spatiotemporal variation in soil respiration is strongly associated with temperature fluctuations. A meta-analysis of 81 field-warming experiments reported significant short-term (typically 1–3 years) increases in FCO₂—18.0%, 13.1%, and 5.9% in forests, croplands, and grasslands, respectively—while the effect often declined with longer warming duration, consistent with thermal acclimation [28]. The effect of temperature, however, depends strongly on soil moisture [27,29]. Moisture regulates respiration in a nonlinear manner because soil water simultaneously constrains microbial activity and metabolic pathways and governs the diffusion of substrates and gases. Under dry conditions, water limitation suppresses microbial activity and restricts substrate transport. Under excessively wet conditions, water-filled pore space reduces aeration and promotes a shift from aerobic to less efficient anaerobic metabolism [30,31], yielding an intermediate ‘optimal moisture range’ for respiration [32]. Moreover, soil moisture can modify the apparent temperature sensitivity of respiration via strong temperature–moisture interactions. Field manipulations by Davidson et al. [16] showed that temperature and moisture effects may be additive or interactive. Such hydrothermal coupling is especially complex in alpine ecosystems with highly variable precipitation, where single-factor models based solely on temperature or moisture often fail to reproduce the temporal dynamics of FCO₂ [33]. Mountain ecosystems, with strong elevational relief and steep hydrothermal gradients over short horizontal distances, provide a natural laboratory for disentangling coupled temperature and moisture controls on soil respiration. Although elevational-gradient studies have been conducted in montane forests, alpine grasslands, and shrublands, most have targeted limited parts of the growing season or relied on coarse temporal sampling. As a result, daily-scale characterization of FCO₂ and their hydrothermal controls remains insufficient [27,34,35], particularly in montane ecosystems within the humid–subhumid transition zone along the southeastern margin of the Tibetan Plateau, where in situ observations are scarce.

Sejila Mountain lies on the southeastern margin of the TP, within the middle–lower reaches of the Yarlung Tsangpo River. Influenced by the warm, moist Indian Ocean monsoon and complex topography, the region exhibits strong elevational differentiation and well-defined vertical zonation [36]. Air temperature and precipitation vary markedly with elevation, and soils and vegetation are organized into altitudinal belts that exert important controls on soil hydrothermal regimes and carbon cycling [35,37]. Previous studies in this area have primarily focused on vegetation productivity, soil physicochemical properties, and hydrothermal processes [35,38,39]. However, diel FCO₂ and its hydrothermal drivers along elevational gradients remain poorly constrained, limiting efforts to reduce uncertainty in the contribution of southeastern TP montane ecosystems to regional carbon budgets and their responses to climate change.

Building on this context, we established five representative plots on the western slope of Sejila Mountain at 3000, 3300, 3600, 3900, and 4200 m a.s.l., while maintaining comparable topographic conditions (e.g., aspect and slope). From April 2024 to December 2025, we continuously monitored daily air temperature and relative humidity, as well as 0–10 cm soil temperature and volumetric water content, and simultaneously measured daily-scale FCO₂. Specifically, we aimed to (i) characterize seasonal patterns in hydrothermal conditions and daily-scale FCO₂ along the elevational gradient; (ii compare FCO₂ rates and cumulative emissions among elevations; and (iii) quantify the combined effects of soil temperature and moisture on daily-scale FCO₂ and evaluate how these effects vary with elevation. Using these high-temporal-resolution in situ observations, we provide an empirical basis for assessing soil-respiration responses to hydrothermal variability in montane ecosystems of the Tibetan Plateau (‘Third Pole’) and for informing regional carbon-cycle modeling and climate-impact assessments.

2. Materials and Methods

2.1. Study Area

The Sejila Mountains (29°10′–30°15′ N, 93°12′–95°35′ E) lie on the southeastern margin of the Tibetan Plateau (TP), northwest of the Yarlung Tsangpo Grand Bend (Figure 1) [40], forming a major alpine canyon landscape at the tectonic junction of the Nyainqêntanglha and Himalayan ranges. The region exhibits a pronounced ecological transition from humid subtropical conditions at lower elevations to cold, semi-humid alpine conditions at higher elevations.

The regional climate is influenced by both the Indian Ocean and plateau monsoon systems and features cool summers, mild winters, and a distinct wet–dry seasonal cycle. Long-term observations at the Sejila Mountain Central Ecological Station (

Table 1. Summary of sampling-site information across the elevational gradient of Sejila Mountain, China.

Sites	Elevation (m)	Level	Lat. (N) (°)	Long.(E) (°)	Ta (°C)	RH (%)	γ (g·cm−3)	θs (%)	SOC (g·kg−1)	Dominant Tree Species	Soil Type [45]	Climate Zones	Vegetation Zones
HB1	3000	Low	29.582	94.462	12.5	57.3	1.24	32.57	29.30	Alpine oak, Alpine pine	Cambisols	Montane warm temperate zone	Coniferous and broad-leaved mixed forest
HB2	3300	Low	29.566	94.548	9.7	57.8	1.14	23.01	6.35	Larix griffithii	Cambisols	Montane temperate zone	Dark-coniferous forest
HB3	3600	Mid	29.561	94.550	8.5	60.4	0.95	24.79	55.22	Picea likiangensis var. linzhiensi	Cambisols	Montane temperate zone	Dark-coniferous forest
HB4	3900	Mid	29.570	94.575	8.7	63.2	1.46	27.67	12.28	Abies georgei var. smithii	Cambisols	Subalpine boreal zone	Dark-coniferous forest
HB5	4200	High	29.608	94.608	8.5	59.5	0.76	30.35	43.82	uniperus saltuaria	Cambisols	Subalpine boreal zone	Dark-coniferous forest

Note: (1) Ta and RH were calculated as daily averages for each day during the observation period (April–December 2024–2025). (2) Soil bulk density (γ) was measured as the mean dry bulk density of the 0–10 cm soil layer using the core method. (3) Soil saturated water content (θ_s) was determined as the mean gravimetric water content at saturation from 0–10 cm core samples measured under laboratory conditions. (4) The Latin names for “coniferous and broad-leaved mixed forest” and “dark-coniferous forest” are Silva conifera et foliosa and Silva conifera umbra, respectively.

) indicate mean annual temperature from 0.5 to 15.8 °C, mean annual evaporation of 308.8 mm, and mean annual precipitation ranges from 677 to 866 mm [36]. This precipitation far exceeds evapotranspiration, indicating that soil moisture in the study area is generally replenished by rainfall. Precipitation is highly seasonal, with most rainfall occurring during the June–September wet season and exhibiting a unimodal annual distribution. Along the elevational gradient, air temperature declines sharply with elevation, while precipitation and snow dynamics vary, producing distinct vertical climatic zones that transition from subtropical or temperate to alpine conditions. This pronounced altitudinal differentiation in hydrothermal regimes provides an ideal setting for examining elevational patterns of FCO₂ and their responses to coupled temperature and moisture constraints [41,42]. According to the World Reference Base for Soil Resources (WRB), the soils in the study area are classified as Cambisols (pH 4–6) [43]. From the summit to the foot of the mountain, soils can be subdivided into five types: alpine tundra soils above 4800 m a.s.l.; alpine meadow soils above the treeline; subalpine shrub meadow soils beneath alpine shrubs; montane drift-gray soils beneath the high-altitude Abies fabri forest; and montane dark-brown soils beneath a conifer–broadleaf mixed forest. The parent material is uniform; the soil-forming material is granite weathering regolith [44,45].

2.2. Site Description and Monitoring of Environmental Variables

To capture key ecotones along the elevational gradient (3000–4200 m a.s.l.), we selected five representative sites spanning montane temperate dark conifer forests, fir–spruce mixed forests, and the treeline ecotone (Table 1, Figure 1) [36,46]. Sites were chosen to maintain comparable slope aspect, slope gradient, and topographic position among plots [47], thereby improving inference on hydrothermal and topographic controls on soil processes across elevations. At each 30 m × 30 m plot, a microclimate monitoring system and soil observation points were installed at the plot center. To characterize in situ meteorological conditions, an automatic weather station (RS-WS-4G-C3; Jianda Renke, Jinan, China) was deployed at each site to continuously record air temperature (Ta) and relative humidity (RH). Ta was measured using a high-precision thermistor sensor (−40 to 80 °C; resolution, 0.1 °C) that was calibrated in the laboratory against multiple reference points, with an absolute error of ±0.2 °C. RH was measured using a capacitive sensor (0%−100%; with an accuracy of ±2%).

To quantify near-surface soil hydrothermal conditions, soil volumetric water content (SW) and soil temperature (ST) were continuously monitored using temperature–moisture sensors (RS-WS-N01-TR-1; Jianda Renke, China) installed at a 10 cm depth within the 0−10 cm soil layer at each site. The sensors measure SW over 0%−100% and temperature over −40 to 80 °C, with a resolution of 0.1% and 0.1 °C, respectively. Before field deployment, sensors were calibrated in the laboratory using standardized soil samples collected from each elevation; the post-calibration absolute error in SW was within ±0.8%. Sensors were installed adjacent to the soil-respiration collars to ensure spatial consistency between hydrothermal monitoring and soil FCO₂ measurements.

All sensors recorded measurements at 4 h intervals, and data were transmitted in near-real time to a cloud-based platform via a 4G network. Sensors were inspected regularly and cross-checked throughout the monitoring period to ensure the continuity and reliability of the time series.

Additionally, to characterize substrate availability across the elevational gradient, a baseline soil organic carbon (SOC) survey was conducted during plot establishment in September 2023, prior to the commencement of continuous FCO₂ monitoring. Given that SOC pools in alpine forests are vast and highly stable, exhibiting negligible overwinter turnover, these baseline data accurately reflect the background soil carbon substrate conditions during the subsequent flux observation period. At each elevational site (HB1–HB5), following the removal of surface litter, soil samples from the 0–20 cm layer were collected using a soil auger (5 cm in diameter). Within each plot, five randomly extracted soil cores were pooled to form a single composite sample. The composite samples were gently broken apart along natural fracture planes and passed through a 6 mm sieve to remove coarse roots, stones, and debris. After removal of coarse materials, the composite samples were thoroughly mixed to ensure homogeneity, then air-dried, gently ground, and sieved to 2 mm for SOC analysis. All samples were transported to the laboratory on the same day. Detailed plot establishment protocols and comprehensive baseline SOC validation are documented in a concurrent publication ([36];

Table 2. Statistical summary of FCO₂ across sample plots along the elevational gradient.

Sites	Year	Statistical Analysis of Daily-Scale FCO2						Statistics on Cumulative Emissions and Seasonal Peak Values
		Mean μmol m−2 s−1	Min μmol m−2 s−1	Max μmol m−2 s−1	Q1	Q3	CV	Peak Month	Cumulative Emissions g C m−2
HB1	2024	3.22 ± 1.54	0.31	5.90	2.00	4.35	0.48	7	846.94 ± 19.58
	2025	3.13 ± 1.51	0.31	6.23	1.66	4.88	0.58	7	836.82 ± 5.17
HB2	2024	1.47 ± 0.64	0.47	3.30	1.08	1.98	0.44	8	385.32 ± 2.43
	2025	1.17 ± 0.44	0.33	2.11	0.89	1.39	0.37	8	339.32 ± 12.00
HB3	2024	2.39 ± 1.12	0.37	4.03	1.58	3.60	0.47	8	648.40 ± 3.19
	2025	2.34 ± 1.25	0.11	4.44	1.42	3.43	0.54	8	655.00 ± 3.16
HB4	2024	1.78 ± 0.78	0.23	3.88	1.22	2.32	0.44	7	466.26 ± 8.00
	2025	1.50 ± 0.68	0.22	2.97	1.06	1.92	0.45	7	331.43 ± 5.85
HB5	2024	3.52 ± 1.53	0.23	5.91	2.76	4.70	0.44	8	909.90 ± 11.13
	2025	3.13 ± 1.68	0.04	5.72	1.98	4.52	0.54	8	808.23 ± 9.91

Note: Using the FCO₂ data collected at stations spanning multiple elevations from April to December, we calculated summary statistics (mean, SD, median, and range) for FCO₂. Descriptive statistics are reported as mean ± SD. Min and Max denote the minimum and maximum values, respectively; Q1 and Q3 denote the first and third quartiles, respectively; and C_V denotes the coefficient of variation. Based on weekly observations of CO₂ flux rates, daily emission values were estimated using linear interpolation. After unit conversion to obtain daily carbon emissions, the daily values were summed to yield cumulative CO₂ emissions for each site over the observation period (April to December; g C·m⁻²). Cumulative emissions were then averaged across replicate plots and reported as mean ± SD.

).

2.3. Site Selection and Experimental Layout Along the Elevational Gradient

FCO₂ was measured in situ using a portable soil carbon flux system (PS9000; Lijia United, Beijing, China) equipped with an infrared gas analyzer (IRGA). During each measurement, the instrument continuously recorded CO₂ concentrations inside the chamber; the manufacturer reports an accuracy of approximately ±1.5% of the reading, suitable for ecosystem-scale monitoring of FCO₂. To ensure representativeness and comparability across elevations, measurement locations within each plot were selected in areas with relatively homogeneous vegetation cover, soil surface conditions, and microtopography. In each plot, three permanent PVC collars (20 cm inner diameter; 15 cm height) were installed as bases for FCO₂ measurements. Each collar was inserted 12 cm into the soil, leaving 3 cm above the surface to minimize lateral gas leakage and maintain a stable chamber–collar seal. This insertion depth is particularly important in high-altitude ecosystems where fine roots are predominantly distributed in the upper 0–10 cm soil layer; therefore, it served to physically sever most live roots within the collar, minimizing the contribution of autotrophic (live root) respiration to the measured FCO₂. Thus, following a sufficient stabilization period, the FCO₂ primarily represents heterotrophic respiration from soil organic matter and recently dead root detritus. Collars were installed on 22 March 2024, and initial FCO₂ measurements were conducted on 14 April 2024 to allow disturbances from installation to dissipate.

Field measurements followed a standardized schedule and operating protocol throughout the monitoring period. FCO₂ was measured at 7-day intervals, and all five sites were sampled within a fixed time window on each measurement date to minimize confounding effects of diurnal variation and changes in radiation conditions. For each measurement, an automated chamber was sealed to the collar, and the increase in CO₂ concentration was recorded over a short closure period (typically 5–10 min). Using the system software, the initial post-closure stabilization period and apparent outliers in the CO₂ time series were excluded. FCO₂ was calculated from the linear rate of change in chamber CO₂ concentration, chamber volume, and the soil surface area enclosed by the collar, and is reported in µmol CO₂ m⁻² s⁻¹.

2.4. Data Processing and Statistical Analysis

2.4.1. Microclimate and Soil Variables

Air temperature (Ta), relative humidity (RH), soil temperature (ST), and soil volumetric water content (SW) were recorded continuously at 4 h intervals. Data were screened for completeness; records affected by sensor malfunction, transmission failure, or external disturbance were removed. Potential outliers were identified through a combination of statistical screening and field verification. For each variable, observations deviating from the site-specific mean by more than 3 standard deviations were flagged and excluded [48]. Flagged observations were cross-checked against field inspection notes and instrument maintenance logs to avoid discarding valid extreme conditions.

Short gaps due to temporary interruptions were rare and were filled using linear interpolation to preserve time-series continuity. Quality-controlled 4 h records were aggregated to daily resolution to compute daily means of Ta and RH, as well as daily means of ST and SW for the 0–10 cm soil layer. The resulting daily data were used to characterize elevational differences in microclimate and seasonal variability and to serve as candidate predictors in analyses of FCO₂.

2.4.2. Processing and Quality Control of Soil CO₂ Flux Data

FCO₂ measurements underwent multi-level quality control. At the instrument level, the PS9000 system automatically excluded the initial stabilization period immediately after chamber closure and removed clear outliers in the CO₂ concentration time series that were inconsistent with stable accumulation during the closure period. During field campaigns, flux readings were monitored in real time using the PS9000 mobile application (v1.0.5). If anomalous values were detected (e.g., negative or zero fluxes, or values that deviated markedly from the previous observation at the same collar), the measurement was repeated immediately to minimize the influence of occasional operational errors.

After instrument-based screening, an additional statistical check was applied to the flux data. For the three collars per plot on a given sampling date, any replicate deviating from the plot mean by more than 3 standard deviations was classified as an outlier and removed [49]. Field notes documenting rainfall, strong winds, or other disturbances were consulted to avoid excluding rare but ecologically meaningful events. Following these procedures, the FCO₂ dataset was deemed suitable for comparisons across elevations and for analyses of environmental responses.

2.4.3. Statistical Analysis

Analyses were conducted at time scales aligned with each objective. Seasonal dynamics of FCO₂ were assessed from daily-scale flux data paired with daily mean microclimate variables. Elevational patterns were evaluated using plot-level means of FCO₂ and environmental variables across the full observation period. Environmental response relationships were examined using paired daily observations FCO₂ and environmental variables, with primary emphasis on soil temperature and soil water content in the 0–10 cm layer.

Differences in FCO₂ and environmental variables among elevational sites were tested in SPSS (version 31). Assumptions of normality and homogeneity of variance were assessed before conducting parametric tests. When the assumptions were satisfied, one-way ANOVA was used to evaluate site effects; otherwise, nonparametric alternatives (e.g., Kruskal–Wallis tests) were applied. Statistical significance was defined as p < 0.05. Correlation and regression analyses were performed in MATLAB (2022) to quantify relationships between FCO₂ and environmental variables. Data visualization was conducted in Origin (2025), including time-series plots, elevational contrasts, and response relationships.

3. Results

3.1. Seasonal Dynamics of Hydrothermal Conditions Along the Elevational Gradient

Seasonal variation in Ta, ST, RH, and soil SW across the five sites is shown in Figure 2 and Figure 3. All hydrothermal variables exhibited pronounced seasonality at each elevation, but their absolute levels and the magnitude of seasonal variability varied along the elevational gradient.

Air and soil temperatures followed consistent seasonal trajectories across sites, rising from the early growing season to a mid-season maximum and declining toward the end of the season (Figure 2). Both Ta and ST decreased with increasing elevation, though the overall seasonal pattern was similar among sites. Relative to Ta, ST exhibited a smaller seasonal amplitude.

RH was lowest in the early growing season and increased thereafter, remaining high through the middle to late growing season. Higher-elevation sites generally exhibited higher RH (Figure 3a–e). SW varied more strongly than RH, and its seasonal pattern differed among elevations (Figure 3f–j), indicating spatial heterogeneity in near-surface moisture dynamics along the slope.

3.2. Multiscale Variation in Soil CO₂ Flux and Elevational Differences

3.2.1. Dynamics of Soil CO₂ Flux

Daily-scale FCO₂ during the 2024 and 2025 growing seasons is shown in Figure 4. Across all sites and both years, FCO₂ exhibited a pronounced unimodal seasonal trajectory: fluxes rose from the early growing season, peaked mid-season, and declined toward the season’s end.

Daily-scale FCO₂ differed significantly among sites along the elevational gradient (Figure 4; Table 2). The low-elevation site HB1 and the high-elevation site HB5 exhibited the highest growing-season mean FCO₂ in both years. In 2024, mean FCO₂ was 3.22 and 3.52 µmol CO₂ m⁻² s⁻¹ at HB1 and HB5, respectively; in 2025, both sites averaged 3.13 µmol CO₂ m⁻² s⁻¹. Maximum Daily-scale FCO₂ was also highest at HB1 and HB5, ranging from 5.90 to 6.23 µmol CO₂ m⁻² s⁻¹ at HB1 and 5.72–5.91 µmol CO₂ m⁻² s⁻¹ at HB5 (Table 2). In contrast, HB2 and HB4 showed lower FCO₂, with growing-season means < 2.0 µmol CO₂ m⁻² s⁻¹ in both years and maxima generally < 4.0 µmol CO₂ m⁻² s⁻¹ (Figure 4; Table 2).

Across sites, the coefficient of variation (C_V) of Daily-scale FCO₂ ranged from 0.37 to 0.58 (Table 2), indicating substantial within-season temporal variability. In 2025, C_V was higher at HB1 and HB5, consistent with greater day-to-day fluctuations at these sites.

Seasonal trajectories were broadly similar between 2024 and 2025, although peak magnitude and flux levels during certain periods differed (Figure 4; Table 2). These interannual differences did not alter the overall elevational pattern of FCO₂.

3.2.2. Monthly Dynamics of Soil CO₂ Flux

Monthly dynamics of FCO₂ at the five sites across 2024 and 2025 are shown in Figure 5. Across HB1–HB5, monthly FCO₂ exhibited a consistent unimodal seasonal trajectory in both years: fluxes rose from April to June, peaked in July or August, and then declined from September to December to low levels.

Monthly FCO₂ differed significantly among sites along the elevational gradient (Figure 5; Table 2). HB1 and HB5 showed higher monthly fluxes in both years, with monthly peaks approaching or exceeding 5.00 µmol CO₂ m⁻² s⁻¹. Over the observation period, mean FCO₂ at HB1 and HB5 was 3.22 and 3.54 µmol CO₂ m⁻² s⁻¹ in 2024 and 3.13 and 3.18 µmol CO₂ m⁻² s⁻¹ in 2025 (Table 2). Maximum daily-scale FCO₂ was also highest at HB1 and HB5, ranging from 5.90–6.23 µmol CO₂ m⁻² s⁻¹ at HB1 and 5.72–5.91 µmol CO₂ m⁻² s⁻¹ at HB5 (Table 2). In contrast, HB2 and HB4 showed lower FCO₂, with mean FCO₂ < 1.50 µmol CO₂ m⁻² s⁻¹ in both years and maxima generally < 4.0 µmol CO₂ m⁻² s⁻¹ (Figure 5; Table 2). HB3 and HB4 were intermediate, remaining consistently lower than HB1 and HB5.

Across sites, monthly peaks generally occurred in July and August, with small differences in peak timing; however, peak magnitude varied substantially among elevations (Figure 5). Seasonal patterns were broadly similar between 2024 and 2025, and site rankings were maintained. Interannual differences primarily reflected modest changes in peak magnitude and did not alter the overall elevational pattern (Figure 5).

3.2.3. Elevational Differences in Soil CO₂ Flux

FCO₂ differed significantly among the five elevational sites in both 2024 and 2025 (Figure 6). Within each year, a one-way ANOVA indicated a significant site effect (p < 0.05), and Tukey’s post hoc tests identified significant pairwise differences among sites.

In 2024, HB5 (4200 m) and HB1 (3000 m) had higher FCO₂ than the other sites (Figure 6). HB5 showed the highest central tendency, with both the median and upper quartile exceeding those of the other sites. In contrast, HB2 (3300 m) consistently exhibited the lowest FCO₂ distribution. HB3 (3600 m) and HB4 (3900 m) were intermediate, with FCO₂ at HB3 higher than at HB4. Most site comparisons showed significant differences (p < 0.05).

In 2025, the distributional pattern closely matched that in 2024, and the relative ranking among sites was unchanged (Figure 6). HB5 and HB1 again formed the high-flux group, HB2 remained the lowest-flux site, and HB3 and HB4 remained intermediate. Although the median and dispersion of FCO₂ differed between years, significant among-site differences persisted, indicating a consistent elevational contrast in soil CO₂ flux over the study period.

3.2.4. Elevational Differences in Growing-Season Cumulative Soil CO₂ Flux

Cumulative emissions of FCO₂ from April to December for 2024 and 2025 are presented in Figure 7. These emissions varied among elevational sites in both years, following a consistent nonmonotonic pattern along the elevational gradient (Figure 7).

Across both years, HB1 and HB5 exhibited higher cumulative emissions than the mid-elevation sites. In 2024, cumulative emissions at HB5 reached 909.90 g C m⁻², which was significantly higher than those at HB2 (385.32 g C m⁻²) (Figure 7; Table 2). Similarly, in 2025, HB1 and HB5 again showed higher cumulative emissions, whereas HB2 and HB4 remained relatively low and HB3 was intermediate (Table 2).

A two-way analysis of variance revealed significant main effects of elevation (p = 0.003) and year (p = 0.012), although the elevation × year interaction was not significant (p = 0.100). These results indicate interannual differences in cumulative emissions, yet the site ranking along the elevational gradient remained broadly consistent. Furthermore, within each year, cumulative emissions were not significantly associated with elevation according to linear regression (2024: R² = 0.020, p = 0.819; 2025: R² = 0.002, p = 0.947; Figure 7), which supports a nonlinear elevational pattern in cumulative FCO₂ emissions.

3.3. Responses of Daily-Scale Soil CO₂ Flux to Hydrothermal Factors

3.3.1. Correlation Between Daily-Scale Soil CO₂ Flux and Hydrothermal Factors Across Elevations

Correlations between daily-scale FCO₂ and hydrothermal variables across HB1–HB5 are shown in Figure 8. Across elevations and years, daily-scale FCO₂ was consistently and positively correlated with ST, and this relationship was significant at most sites (p < 0.05). The direction of the FCO₂–ST relationship was conserved along the gradient, indicating a robust linkage between near-surface soil thermal conditions and daily-scale FCO₂.

Daily-scale FCO₂ was also positively correlated with Ta in most cases, although the strength of the association varied substantially among sites (Figure 8). In some plots, correlations with Ta were strong and significant, whereas in others they were weaker; however, the direction of the FCO₂–Ta relationship remained consistent across years.

Except for HB2, correlations between FCO₂ and SW were generally positive, but both magnitude and consistency differed among sites (Figure 8). In some plots, FCO₂ was more strongly associated with SW, elsewhere the association was weak, indicating nonuniform moisture sensitivity along the elevational gradient.

In contrast, correlations between RH and FCO₂ were weak and variable (Figure 8). Correlation coefficients were generally small and varied in sign among sites, suggesting RH was not a stable descriptor of daily-scale FCO₂ across the gradient. Overall, the correlation structure of daily-scale FCO₂ varied with elevation: ST showed the most consistent association with FCO₂, whereas associations with SW and RH were more site-dependent.

3.3.2. The Interactive Regulation of Soil CO₂ Flux by Soil Temperature and Moisture

To test whether the effects of ST and SW on daily-scale FCO₂ were nonadditive, we evaluated the ST × SW interaction using response-surface analysis and linear mixed-effects models (Figure 9;

Table 3. Linear mixed-effects model results for daily-scale FCO₂ as a function of ST, SW, and their interaction.

Variable	Regression Coefficient β (Estimate)	Standard Error (SE)	t	Degrees of Freedom (DF)	p	95% CI (Lower Limit)	95% CI (Upper Limit)
Intercept	0.605	0.201	3	774	0.003	0.211	0.999
Ta	0.522	0.036	14	774	0.001	0.451	0.593
RH	0.131	0.018	7.2	774	0.001	0.096	0.167
ST	0.038	0.037	1.0	774	0.306	−0.035	0.11
SW	0.049	0.023	2.1	774	0.034	0.004	0.094
ST × SW	0.017	0.016	1.1	774	0.290	−0.015	0.049

Note: Table 3 reports results from a linear mixed-effects model testing the interaction between ST and SW. Bold values indicate p < 0.001, and italicized values indicate p < 0.05. Daily-scale FCO₂ was modeled as the response variable, with Ta, RH, ST, SW, and ST × SW included as fixed effects, and elevational zone and year included as random intercepts. Results are reported as standardized regression coefficients ± SE, with t statistics, degrees of freedom (DF), p values, and 95% confidence intervals (CI). Although the response-surface analysis (Figure 9) suggests that the temperature–efflux relationship varies with moisture conditions, the ST × SW interaction was not statistically significant, indicating limited evidence for a nonadditive interaction at the daily-scale scale.

) [50]. The response surface indicated that FCO₂ generally increased with increasing ST and SW, and that the shape of this response depended on the background level of the other factor, consistent with context-dependent effects.

In the response-surface analysis, FCO₂ increased only slightly with ST under low SW, yielding a shallow temperature gradient. As SW increased, the apparent sensitivity of FCO₂ to ST strengthened, producing a steeper temperature response. A similar interaction was observed for SW: under low ST, increases in SW were associated with minimal changes in FCO₂, whereas under higher ST, FCO₂ increased more strongly with SW (Figure 9). Together, these results indicate that the effect of each driver on FCO₂ depended on the concurrent level of the other.

However, linear mixed-effects modeling showed that the ST × SW interaction was not statistically significant after controlling for Ta and RH and including random effects of elevation group and year (p = 0.29; Table 3). In contrast, the main effects of Ta and RH were significant (p < 0.001), while the main effects of ST and SW and their interaction were weaker (Table 3). Although Figure 9 shows a continuous response of FCO₂ to concurrent variation in ST and SW, the model did not provide statistical evidence for a nonadditive interaction. Thus, at the daily scale, variation in FCO₂ was dominated by additive effects, with context dependence expressed as changes in response magnitude rather than a significant interaction term (Figure 9; Table 3).

3.3.3. Nonlinear Response of Soil CO₂ Flux to Soil Water Content and Threshold Behavior

Further analyses indicated that the daily-scale FCO₂ response to SW was nonlinear. Piecewise regression identified a site-specific moisture threshold (θ) at each elevation (Figure 10). Across HB1–HB5, the FCO₂–SW relationship exhibits a breakpoint that divides a low-sensitivity regime from a higher-sensitivity regime, suggesting that θ is a common moisture-control feature of daily-scale FCO₂.

Estimated θ differed among sites and shows a nonmonotonic pattern along the elevational gradient (Figure 10). At the low-elevation site HB1, θ is relatively low; the point estimate and the bootstrap 95% CI indicate that FCO₂ sensitivity to SW emerged under comparatively dry conditions. In contrast, θ is generally higher at the mid-and high-elevation sites, with clear inter-site differentiation. The highest thresholds occur at HB2 and HB5 relative to the other sites. Uncertainty in θ also varies with elevation: bootstrap 95% CIs are narrower at low elevations but widen at higher elevations, particularly at HB5, indicating greater uncertainty in threshold estimation under high-elevation conditions.

3.3.4. Direction of Overall Effects of Hydrothermal Factors on Soil CO₂ Flux

To quantify the direction and relative importance of hydrothermal predictors while accounting for spatial and temporal structure, we fitted linear mixed-effects models with elevation band and year as random intercepts. We compared candidate fixed-effects structures to identify the best-supported model (Figure 11;

Table 4. Comparison of linear mixed-effects models (LMMs) and standardized fixed-effect estimates for hydrothermal factors.

Model	Fixed Effects	AIC	ΔAIC	BIC	Ta (β ± SE)	ST (β ± SE)	SW (β ± SE)	RH (β ± SE)	LRT
M1	Ta + SW + RH	1356.95	6.81	1389.57	0.58 ± 0.02 ***	—	0.11 ± 0.03 ***	0.20 ± 0.02 ***	χ2(8) = 8.81, p = 0.003
M2	ST + SW + RH	1445.39	95.25	1478.00	—	0.56 ± 0.02 ***	0.14 ± 0.03 ***	0.10 ± 0.02 ***	χ2(8) = 97.25, p ≤ 0.001
M3	Ta + ST + SW + RH	1350.14	0	1387.41	0.47 ± 0.05 ***	0.14 ± 0.05 **	0.11 ± 0.03 ***	0.17 ± 0.03 ***	—

Note: AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion) are model-selection metrics; ΔAIC represents the difference between the AIC of a candidate model and that of the best-fitting model. Table entries report standardized fixed-effects regression coefficients (β) with their SE. **, and *** denote p-values less than 0.01, and 0.001, respectively.

). After accounting for random effects, all hydrothermal predictors had positive fixed-effect estimates, although effect sizes and statistical support differed among variables.

Standardized fixed-effect estimates revealed a clear gradient in predictor importance (Figure 11). Ta had the largest standardized coefficient, with a bootstrap 95% CI entirely above zero, indicating the strongest and most robust positive association with FCO₂. RH also showed a significant positive effect, although its magnitude was smaller than that of Ta. In contrast, the standardized effects of ST and SW were small, and their CIs were near zero, suggesting weaker fixed effects and greater uncertainty after accounting for Ta, RH, and the random-effects structure.

Model comparison further indicated that explanatory performance differed among predictor combinations (Table 4). The model including Ta, ST, SW, and RH (M3) provided the best fit based on information criteria (ΔAIC = 0), and improvements relative to reduced models were supported by likelihood-ratio tests. Overall, these results suggest that daily-scale FCO₂ is associated with multiple hydrothermal drivers, with Ta contributing most strongly and the remaining variables providing additional explanatory power not captured by Ta alone.

3.3.5. Seasonal Heterogeneity in Hydrothermal Controls on Soil CO₂ Flux

Season-specific correlation analyses revealed pronounced seasonal heterogeneity in associations between daily-scale FCO₂ and ST and SW (Figure 12). In spring and summer, correlations between FCO₂ and ST were generally stronger, with coefficients ranging from 0.3 to 0.8 across most sites. In 2025, the 95% CI excluded zero for most sites during this period, indicating a consistently positive association between ST and FCO₂. In autumn, correlations weakened (approximately 0.1–0.5) and confidence intervals broadened. In winter, correlations are near zero or negative at many sites (−0.6–0.2), with confidence intervals often spanning zero.

In contrast, correlations between daily-scale FCO₂ and SW were generally weaker and more variable among sites (Figure 12b). In spring and summer, correlations are predominantly positive at most sites, but smaller (0.1–0.6) and with confidence intervals wider than those for the ST relationship. In autumn, among-site divergence increased, and correlations at some low- and mid-elevation sites approached zero or became weakly negative. In winter, correlations remained weak (−0.5–0.3), with confidence intervals broadly overlapping zero.

At the spatial scale, mid- to high-elevation sites (HB3–HB5) exhibited more consistently positive correlations in spring and summer, particularly for the FCO₂–ST relationship, with coefficients remaining above 0.4 across multiple seasons. Interannual comparisons show consistent correlation directions in 2024 and 2025, although correlation strength varies, with generally higher coefficients during the 2025 growing season and fewer confidence intervals overlapping zero.

Overall, the correlation structure between hydrothermal variables and FCO₂ varies systematically across seasons. During the growing season, associations are more consistent in direction and magnitude; in winter, they weaken and become less stable, indicating pronounced seasonal heterogeneity. Although this analysis characterizes seasonal shifts in the direction and stability of relationships between FCO₂ and hydrothermal factors across elevations, it does not quantify the relative explanatory contributions of individual predictors. Accordingly, the next section applies multivariate regression models to compare the relative importance of hydrothermal predictors for daily-scale variation in FCO₂.

3.4. Explanatory Power of Temperature and Moisture for Daily-Scale Soil CO₂ Flux

3.4.1. Regression Relationships and Elevational Differences

To quantify the combined influence of hydrothermal variables on daily-scale FCO₂, we fitted separate multiple linear regression models for each elevational site (

Table 5. Multivariate regression results assessing the effects of hydrothermal factors on daily-scale FCO₂ across plots along an elevational gradient.

Sites	Ta (βstd)	RH (βstd)	ST (βstd)	SW (βstd)	R2	Adjusted R2
HB1	0.19	0.32	1.03	0.20	0.75	0.74
HB2	−0.01	−0.08	0.38	−0.14	0.40	0.39
HB3	0.44	0.26	0.38	0.02	0.63	0.62
HB4	0.16	0.16	0.33	0.05	0.54	0.52
HB5	1.07	0.02	0.32	0.07	0.72	0.71

Note: Multiple linear regression models were fitted for plots at different elevations, with daily-scale FCO₂ as the dependent variable and Ta, RH, ST, and SW as predictors. All predictors were standardized before analysis; therefore, reported coefficients are standardized regression coefficients (β_std). Model fit was evaluated using R² and adjusted R². Multicollinearity was assessed using variance inflation factors (VIF); all predictors had VIF < 10, indicating no problematic multicollinearity.

). Model performance varied across sites: R² ranged from 0.40 to 0.75, and adjusted R² ranged from 0.39 to 0.74. Explanatory power was highest for HB1 and HB5 (R² = 0.74 and 0.71, respectively), followed by HB3 and HB4 (R² = 0.62 and 0.52), while HB2 showed the lowest explanatory power (R² = 0.39). Overall, these results indicate that the explanatory power of hydrothermal variables to explain daily-scale variation in FCO₂ varies along the elevational gradient.

Standardized regression coefficients indicated site-specific differences in the relative contributions of individual predictors (Table 5). ST was positively associated with FCO₂ at all sites, with standardized coefficients ranging from 0.32 to 1.03, making ST the most consistent predictor across elevations. The contribution of Ta increased with elevation, with standardized coefficients rising from 0.19 at HB1 to 1.07 at HB5, indicating a stronger role for Ta at higher elevations. In contrast, the effects of RH and SW were smaller and less consistent across sites: RH coefficients ranged from −0.08 to 0.32 and were near zero at HB5 (0.02), whereas SW coefficients ranged from −0.14 to 0.20, with direction varying among sites.

Overall, regression relationships and the relative importance of hydrothermal predictors exhibit pronounced spatial heterogeneity across elevations. Nevertheless, all site-specific models retained moderate to high explanatory power (R² ≥ 0.39), highlighting elevational variation in the statistical controls on daily-scale FCO₂.

3.4.2. Relative Contributions of Temperature and Moisture and Elevational Variation

To quantify the combined influence of hydrothermal variables on daily-scale FCO₂, we fitted site-specific multiple linear regression models across elevations HB1–HB5, with daily-scale FCO₂ as the dependent variable and Ta, RH, ST, and SW as predictors. All predictors were standardized prior to analysis; thus, the reported coefficients are standardized regression coefficients (β_std). Model fit was assessed with R² and adjusted R². Across HB1–HB5, temperature-related predictors consistently contributed more to daily-scale FCO₂ than moisture-related predictors. The sum of absolute standardized coefficients for temperature ranged from 0.40 to 1.40, whereas the corresponding sum for moisture ranged from 0.10 to 0.50, indicating that temperature accounts for a larger share of daily-scale variation in FCO₂ than moisture along the elevational gradient (Figure 13).

The relative contribution of temperature exhibited a nonmonotonic pattern along the elevational gradient: it was high at HB1 (1.23), declined to a minimum at HB2 (0.39), and then increased toward higher elevations, reaching a maximum at HB5 (1.39). In contrast, the relative contribution of moisture remained low and decreased with elevation, from 0.53 at HB1 to 0.09 at HB5. Consequently, the disparity between temperature and moisture contributions widened with elevation; at the highest site (HB5), temperature-related predictors contributed roughly 14-fold more than moisture-related predictors (Figure 13). These results indicate an elevational shift in the dominant environmental controls on daily-scale FCO₂.

3.4.3. Comparative Analysis of Model Interpretability and Identification of Critical Control Factors

To compare the explanatory power of temperature- and moisture-related predictors for daily-scale FCO₂, we fitted site-specific multiple linear regression models at each elevation (HB1–HB5): (i) a temperature-only model including Ta and ST, (ii) a moisture-only model including RH and SW, and (iii) a full model including all four predictors (

Table 6. Comparison of the explanatory power of temperature-only, moisture-only, and full models for daily-scale FCO₂ across sites at different elevations.

Sites	Temperature Model		Moisture Model		Comprehensive Model
	R2	Adjusted R2	R2	Adjusted R2	R2	Adjusted R2
HB1	0.70	0.69	0.31	0.30	0.75	0.74
HB2	0.33	0.32	0.03	0.02	0.40	0.39
HB3	0.59	0.58	0.24	0.23	0.63	0.62
HB4	0.49	0.48	0.18	0.17	0.54	0.52
HB5	0.72	0.71	0.09	0.08	0.72	0.71

Note: The coefficients of determination (R²) and adjusted coefficients of determination (adjusted R²) from site-specific multiple linear regression models were used to compare each model’s ability to explain variation in daily-scale FCO₂ across elevations HB1–HB5. The temperature-only model includes Ta and ST; the moisture-only model includes SW and RH; and the full model includes Ta, ST, SW, and RH. Models were fitted separately for each elevation site (HB1–HB5), and thus the reported R² values reflect the extent to which each predictor set accounts for variation in daily-scale FCO₂.

). Model performance varied across predictor sets and elevations. The coefficient of determination, R², ranged from 0.33 to 0.72 for temperature-only models, 0.03 to 0.31 for moisture-only models, and 0.40 to 0.75 for full models, indicating pronounced spatial variation in explanatory power along the elevational gradient.

Across all sites, temperature-only models explained substantially more variation in daily-scale FCO₂ than moisture-only models, indicating stronger statistical control by temperature. By contrast, moisture-only models had limited explanatory power, particularly at mid and high elevations (e.g., R² = 0.03 at HB2 and R² = 0.09 at HB5; Table 6). Full models achieved the highest explanatory power at all elevations (R² = 0.39–0.74), but improvements over temperature-only models were generally modest. This pattern was most evident at HB5, where the full and temperature-only models had nearly identical explanatory power (R² = 0.72).

Overall, while incorporating both temperature and moisture predictors improved model fit, explanatory power along the elevational gradient was driven primarily by temperature. These results identify temperature as the dominant statistical control on daily-scale FCO₂; conversely, the additional contribution of moisture was elevation-dependent and notably limited at higher elevations.

4. Discussion

4.1. Elevational Variation in Soil CO₂ Flux Across the Sejila Mountains

Across the five forest sites (HB1–HB5), FCO₂ exhibited a consistent unimodal seasonal pattern in both years, increasing from the early growing season, peaking in midsummer, and declining toward the season’s end. This pattern aligns with expectations for cold-region montane ecosystems, where biological activity is concentrated during the warmest and wettest period of the year, when temperature and moisture constraints are relaxed and substrate turnover is fastest [25,50,51,52]. Similar seasonal concentration of forest soil respiration has been reported for subalpine forests across the eastern Tibetan Plateau, where environmental sensitivities differ markedly between active and dormant periods [18,52,53]. In Sejila, however, the elevational pattern was nonmonotonic: efflux was higher at HB1 and HB5 than at mid-elevation sites, and HB2 consistently exhibited the lowest values. Although many elevational-gradient studies report a monotonic decline in soil respiration with elevation, often attributed to temperature limitation [54], the U-shaped pattern observed here indicates that temperature alone cannot explain spatial variation in FCO₂ along this monsoon-influenced mountain slope [51,55,56,57]. Instead, our results suggest that the relative importance of thermal limitation and water-related constraints shifts with elevation, producing nonmonotonic responses even over a modest elevational range [58,59,60]. This interpretation is consistent with evidence from Sejila showing that soil respiration varies among forest settings and covaries with both soil temperature and soil moisture, implying joint temperature–moisture control of spatial differences in FCO₂ within this mountain system [39,61,62]. However, hydrothermal conditions alone remain insufficient to fully account for the anomalously high FCO₂ fluxes observed at the highest-elevation site (HB5). Rather, the emergence of this nonmonotonic (U-shaped) pattern is likely driven by the spatial distribution of substrate availability. Baseline survey results (Table 1) indicate that surface (0–20 cm) soil organic carbon (SOC) content at HB5 reaches 43.82 g·kg⁻¹, which is significantly higher than the levels recorded at the mid-elevation site HB4 (3900 m; 12.28 g·kg⁻¹) and the lower-elevation site HB2 (3300 m; 6.35 g·kg⁻¹). In cold alpine and subalpine environments, prolonged low temperatures severely constrain microbial decomposition, thereby promoting the accumulation of substantial organic matter pools [63]. This vast, carbon-rich substrate pool provides an abundant precursor for soil respiration; once hydrothermal conditions become favorable during the growing season, the alleviation of substrate limitation can trigger intense carbon mineralization. Beyond mere substrate availability, recent regional evidence suggests that this 4200 m treeline ecotone functions as a multidimensional carbon-cycling hotspot driven by non-linear biological and physical feedback. First, microbial communities in the Sejila Mountains exhibit a prominent “U-shaped” rebound in diversity and cold-adapted enzymatic activity above 4100 m [64,65], priming the system for rapid heterotrophic responses to summer thermal pulses. Second, severe environmental and nutrient constraints at the treeline drive woody plants to allocate disproportionate carbon belowground, developing dense fine-root networks that enhance both autotrophic respiration and strong rhizosphere priming effects [66]. Finally, the combination of winter snowpack buffering and the highly porous, low-bulk-density soil structure (γ = 0.76 g·cm⁻³) characteristic of this site effectively decouples high moisture availability from gas diffusion resistance [67]. Consequently, this unconstrained hydrothermal-aeration configuration, coupled with highly reactive microbial and root engines, collectively dictates the anomalously high FCO₂ observed at the upper elevational limit.

4.2. Hydrothermal Controls Across Sites and Seasons: Dominance, Thresholds, and Conditionality

Across HB1–HB5, soil temperature showed the strongest and most consistent association with FCO₂, remaining positive and often significant in both years. This finding aligns with evidence that temperature is a primary proximate control of short-term variability in soil respiration, particularly during the growing season when microbial and root activity is strongly temperature-sensitive [25,68,69]. However, several lines of evidence from this study indicate that temperature effects are not spatially uniform and are moderated by water availability.

First, the relative importance of temperature- and water-related predictors shifted along the elevational gradient. In site-specific multiple regression models, temperature variables consistently accounted for most of the explained variance [55], whereas the additional explanatory power of relative humidity and soil water content was smaller and more elevation-dependent [70]. Moreover, the standardized contribution of Ta increased with elevation, suggesting that aboveground thermal conditions became a stronger predictor of day-to-day variability in FCO₂ at higher sites [71]. This pattern is consistent with evidence that, in colder environments, modest changes in temperature can produce disproportionate changes in ecosystem respiration [25]; however, the magnitude of this response may still be moderated by moisture conditions [72,73].

Second, the effects of SW were distinctly nonlinear. Piecewise regression identified site-specific soil moisture thresholds (θ) that separate weak responses from stronger increases in FCO₂ with increasing SW. These thresholds are ecologically meaningful because they suggest that water availability functions less as a linear covariate and more as a switch-like constraint that enables or amplifies temperature-driven carbon losses once minimum moisture conditions are met [74,75,76]. The nonmonotonic variation in θ across elevations, together with greater uncertainty at the highest site, further suggests that effective moisture control of respiration is not determined by elevation per se but by elevation-linked shifts in local microclimate and soil physical conditions that influence CO₂ production and transport [77,78,79].

Third, the interaction between ST and SW highlights an important distinction between descriptive patterns and inferential evidence. Response-surface analyses suggested conditional effects, with greater apparent temperature sensitivity under wetter conditions and greater apparent moisture sensitivity under warmer conditions. However, the mixed-effects model—which accounted for covariation among predictors and included random effects for year and elevation group—did not detect a significant ST × SW interaction. Collectively, these results suggest that day-scale variation is explained primarily by additive effects, although the apparent strength of the temperature response may still depend on background moisture in ways that are not recovered as a robust interaction coefficient after controlling for shared variance among predictors [80,81]. In practice, explicitly representing moisture thresholds and seasonal shifts in sensitivity may be more informative than relying on a single global interaction parameter [82].

Finally, hydrothermal regulation was strongly seasonal. The correlation between temperature and FCO₂ was strongest in spring and summer but weakened markedly in autumn, whereas winter correlations were weak and occasionally negative. This seasonal divergence indicates that coupling among near-surface temperature, moisture, and efflux is not stationary over the year. Accordingly, extrapolating growing-season relationships to colder periods using a single parameter set—even within the same site—should be avoided [83]. Season-dependent responses of soil respiration to environmental drivers have also been reported in subalpine forests, underscoring the need to account for seasonal regimes when attributing controls and projecting flux responses [83,84].

4.3. Implications for Carbon–Climate Feedbacks on the Tibetan Plateau Under Warming Scenarios

This study demonstrates that Ta is the most consistent primary driver along the elevational gradient, whereas SW exhibits pronounced nonlinear, threshold responses. Together, these findings provide a useful framework for interpreting how soil carbon stocks on the Tibetan Plateau may respond to future climate change.

A substantial body of meteorological observations and multi-model simulations has established an elevation-dependent warming (EDW) signal over the Tibetan Plateau, in which warming rates at high elevations substantially exceed those at lower elevations [85,86,87]. Consistent with this context, soils at the highest site (4200 m) exhibited persistently high daily-scale FCO₂, with temperature-related factors accounting for more than 50% of the standardized contribution. This indicates that high-elevation soil metabolism is not dormant, rather, it may be primed for rapid carbon release. As corroborated by the baseline substrate data in this study (Table 1), historically low temperatures have constrained microbial activity and decomposition [88], leading to the accumulation of abundant soil organic carbon (SOC up to 43.82 g·kg⁻¹) at the highest-elevation site, HB5, within this cold ecotonal environment [89], thereby constituting a substantial potential carbon source. Consequently, climate warming may alleviate this thermal constraint, accelerating the decomposition of this vast substrate pool and triggering carbon losses that exceed those observed at mid-elevation sites [90]. This inference accords with recent large-scale assessments [91,92], suggesting that high-elevation and high-latitude ecosystems are increasingly at risk of shifting from carbon sinks to carbon sources.

Moreover, CMIP6-based projections indicate a marked warming–wetting tendency in southeastern Tibet over coming decades [93], potentially driven by enhanced water vapor transport associated with a strengthened Indian summer monsoon and its interaction with the mid-latitude westerlies [94,95]. Under such conditions, if SW remains above the threshold (θ) identified here, the warming effect may proceed largely unconstrained by moisture, thereby enhancing soil respiration [96]. This supports the view that increasing moisture availability can substantially amplify the temperature sensitivity (Q₁₀) of respiration in cold ecosystems [97].

By contrast, if warming is accompanied by extreme drought such that SW falls below θ, the temperature-driven positive warming–carbon emission feedback may be transiently attenuated by moisture limitation [98]. As shown in Figure 10, when SW is below θ, FCO₂ declines sharply as moisture decreases, indicating that moisture rather than temperature becomes the dominant constraint. This threshold behavior acts as a hydrothermal valve that regulates FCO₂ by shifting the system between energy limitation and water limitation. Importantly, drought-induced suppression of respiration should not be interpreted as beneficial self-regulation, because severe water deficits often reduce gross primary productivity (GPP) more than ecosystem respiration (Re), thereby decreasing net primary productivity (NPP) [99,100] and increasing the likelihood of a sink-to-source transition driven by reduced carbon inputs rather than by enhanced respiration [101]. Accordingly, under both warm–wet (high respiratory losses) and warm–dry (low productivity) futures, cold-region ecosystems may face a dual risk of carbon-sink weakening.

4.4. Limitations and Future Directions

Although this study employed a high-precision portable carbon-flux system to record high-quality instantaneous soil CO₂ flux rates during each campaign, the logistical constraints of in situ measurements across an elevational gradient meant that weekly discrete sampling might not have fully captured short-lived pulse events triggered by rainfall or soil rewetting [102]. Furthermore, owing to the inherent constraints of manual fieldwork and adverse weather conditions, periodic measurements often provide limited coverage during or immediately adjacent to rainfall events. This limitation is common to manual or periodic soil-respiration observations [103,104]. Nevertheless, methodological comparisons indicate that weekly sampling combined with linear interpolation yields robust estimates of seasonal cumulative CO₂ fluxes and adequately characterizes growing-season dynamics [105]. The resulting season-integrated totals are comparable to those derived from high-frequency automated measurements [104]. Therefore, while weekly discrete observations may underestimate the peak magnitude or immediate contribution of individual short-term pulses, this uncertainty primarily affects the fine-scale partitioning of event-driven processes. It is highly unlikely to alter our overarching conclusions regarding the nonmonotonic spatial distribution of CO₂ emissions and its interannual robustness. Future research should consider deploying continuous automated monitoring systems at key elevations (e.g., low- and high-elevation sites exhibiting high emissions) or implementing intensified sampling protocols following heavy rainfall to more precisely quantify the instantaneous contributions of precipitation pulses to the carbon balance of alpine ecosystems.

Furthermore, although this study clarifies hydrothermal controls on FCO₂, future work should incorporate biological processes (e.g., separating autotrophic and heterotrophic respiration) and non-growing-season freeze–thaw dynamics to more fully assess the climate sensitivity of high-elevation ecosystems on the Tibetan Plateau.

5. Conclusions

Drawing on continuous in situ measurements across five elevational bands (3000–4200 m) in the Sejila Mountains of the southeastern Tibetan Plateau, this study evaluated the spatiotemporal variability and hydrothermal environmental controls of daily-scale FCO₂. Employing linear mixed-effects models, piecewise regression, and multiple linear regression, we identify four main conclusions:

(1) Nonmonotonic elevational pattern. FCO₂ exhibited a robust nonmonotonic (U-shaped) distribution along the elevational gradient, with efflux highest at the gradient extremes (3000 and 4200 m). Peak rates exceeded 5.0 µmol CO₂ m⁻² s⁻¹, and cumulative emissions at the extremes exceed those at mid-elevations. This pattern challenges the expectation that soil respiration decreases monotonically with elevation on temperature lapse rates and supports a nonlinear elevational distribution of FCO₂.

(2) Hydrothermal differentiation with elevation. The hydrothermal environment varied systematically with elevation: air temperature (Ta) and soil temperature (ST) declined roughly linearly, whereas soil water content (SW) and relative humidity displayed nonlinear spatial variation. The combination of a linear thermal gradient and nonlinear moisture variation provides a plausible environmental basis for the observed departure from strictly temperature-driven patterns in FCO₂.

(3) Shift in dominant controls. The contribution of environmental drivers changed along the gradient. Ta was the strongest and most consistent positive predictor across sites. The ST × SW interaction was not significant (p = 0.29), indicating additive temperature and moisture effects at the daily scale. With increasing elevation, regulation shifted from joint temperature–moisture control at lower elevations to predominantly thermal limitation at higher elevations; at 4200 m, the relative contribution of thermal factors (Ta, ST) exceeded that of moisture factors (RH, SW) by more than tenfold. Piecewise regression identified site-specific nonlinear moisture thresholds (θ), indicating threshold-dependent moisture regulation of soil respiration.

(4) Implications for warming and upscaling. Given the high maximum FCO₂ and the dominant role of Ta at 4200 m, warming could disproportionately enhance soil carbon losses at high elevations, provided that SW remains above critical thresholds. The strong predictive performance of Ta also suggests that gridded meteorological products may be used to estimate soil respiration dynamics in remote, data-scarce alpine regions.