Importance of active layer freeze-thaw cycles on the riverine dissolved carbon export on the Qinghai-Tibet Plateau permafrost region

Study area

This study was conducted in five catchments located in the Fenghuo Mountain region of the QTP, which is covered by continuous permafrost (Fig. 1). Table 1 shows the characteristics of the catchments in detail. Catchment 2 to 5 were nested in Catchment 1. The five catchments constitute an entire watershed, Zuomaokong watershed, which is a tributary of the Tongtian River that represents the headwaters of the Yangtze River. The total watershed area is 117.0 km², and the elevation of the watershed ranges from 4,720 to 5,392 m above sea level.

The study area is characterized by a cold and dry continental alpine climate. The mean annual air temperature is −5.2 °C, and the mean annual precipitation is 328.9 mm. The highest temperature (averagely 11.6 °C) in each year occurs in July or August, and the monthly mean air temperature is above 0 °C from May to September. The relative humidity of the air ranges from 17% to 96% in the winter and summer, respectively. Precipitation events that occur from June to September account for 85% of the total annual rainfall amount, and peak precipitation occurs in July or August. During the freezing season from November to the following April, the total precipitation (snowfall) is usually less than 21 mm (Song et al., 2017). The snow cover during the winter season is sporadic and thin, thus, snowmelt events rarely occur during the spring thaw period (Wang et al., 2017).

Permafrost depths of the study area vary from 80 to 120 m, and active layer depths range from 1.3 to 2.5 m (Wu & Liu, 2004). The soil types in the study area are primarily classified as Mattic-Gelic Cambisols (alpine meadow soil) (Wang, Hu & Li, 2009). The vegetation within the study area is dominated by alpine meadow (70% in our study watershed) and alpine swamp meadow, which are the most widespread types of vegetation across the QTP permafrost region. The dominant plant species are sedges Kobresia pygmaea and K. humilis (Fig. S1). Catchment 5 has the highest vegetation coverage (Vc) (57.3%), whereas Catchment 2 has the lowest Vc (32.3%). Livestock grazing pressure in the study area is light.

The frozen season usually begins in October and ends in the following April. During the frozen season, the river channel contains no open liquid water for sampling. The annual runoff processes out of the frozen season have two flooding periods and two dry periods. These periods include a summer flooding period that starts in July followed by a slight recession in early August, an autumn flooding period in September with baseflow recession from late September to October (Wang, Liu & Liu, 2011).

Field observations

At the outlet of the entire watershed (Catchment 1), flow depths were recorded continuously using a U20 HOBO Water Level Logger (MicroDAQ, Contoocook, NH, USA) at a 1-h interval (0.021 m resolution). The stream velocity was measured using an FP101 Digital Water Velocity Meter (GLOBAL WATER, College Station, TX, USA, 0.03 m s⁻¹ resolution). We used the widely used velocity-area method to calculate river discharge in this study. Soil temperature and moisture of 10 sites at depths of 0.05, 0.2, 0.5, 1, and 1.6 m were measured using temperature and moisture sensors (5TM; Decagon Devices, Pullman, WA, USA) installed inside the Catchment 3 (Fig. 1), and the measurements were recorded by data loggers (EM50; Decagon Devices, Pullman, WA, USA). Soil moisture and temperature were measured simultaneously at 30-min intervals. The daily average soil moisture and temperature data were processed in R (version 3.3.2), using the “xts” package. The thawed depth during the thawing seasons and the frozen depth during the frozen seasons of the soil active layer (Figs. S2 and S3) were determined based on the daily soil temperature (Frauenfeld et al., 2004). We used daily soil temperature measurements at different depths to interpolate the depth of 0 °C isotherm depth. Then the 0 °C isotherm depths were used to determine thaw and freeze depth. Daily mean soil temperature and moisture were displayed in Fig. S4. Meteorological data, including precipitation and air temperature, were collected at the meteorological station located in the Zuomaokong watershed.

Sample collection and analyses

We collected stream water, rainwater, and groundwater samples for DOC and DIC analyses. Stream water samples were collected at the outlet of the five catchments from May to October in 2014 and 2016. We sampled about weekly in 2014 and daily in 2016 thaw and freeze seasons. Sampling work was conducted in the middle of the day. Stream water samples were collected at approximately half depth from the water surface to the river bed and center of the transects at the outlets of the sub-catchments (Fig. 1). Rainwater samples were collected near the meteorological station after each day of rain and was assumed to represent the entire watershed precipitation. Groundwater samples were collected in three groundwater observation wells (∼2 m depth below ground, Fig. 1) inside the watershed at a 3-day interval in 2016. Fresh stream water samples were filtered by vacuum filtration method immediately after sampling through 0.45 μm pore size glass fiber filters (Shanghai Xinya, Shanghai, China). After filtration, 100 mL of each water sample were poured into pre-rinsed high-density polyethylene bottles with tight-fitting caps and then immediately stored under frozen and dark conditions. The samples were kept frozen before the analysis. We used an Elementar Vario TOC Select Analyzer (Elementar, Langenselbold, Germany) to analyze the DOC, DIC, and total dissolved nitrogen (TDN) concentrations of the water samples within one week after sampling. Results of triple injections of indicated the analytical precision <3%. Base cations (Na⁺, Mg²⁺, K⁺, Ca²⁺) from river water samples during 2014 were analyzed with ICS-90 Ion Chromatography System (Dionex, Sunnyvale, CA, USA).

Calculations and statistics

We combined dissolved carbon measurements and continuous discharge data and used the USGS load estimator (LOADEST) program (Runkel, Crawford & Cohn, 2004) to derive fluxes of DIC and DOC of our study watershed. LOADEST uses daily flow and carbon concentration data to establish relationships between flow and concentration and applies these relationships to the complete daily discharge record and then derive flux estimates. LOADEST centers the discharge and dissolved carbon concentration data to eliminate multicollinearity and selects one of nine predefined regression models to fit the data, based on the Akaike Information Criterion and Schwarz Posterior Probability Criteria. This method has been widely used in the Arctic river flux calculation (Tank et al., 2012b; Raymond et al., 2007). In our study, the model 9 was selected by the LOADEST for DIC and DOC flux estimation: (1) $$\begin{array}{l}\text{ln}(\text{flux})=a_0+a_1\text{ln}Q+a_2\text{ln}{Q}^2+a_3\text{Sin}(2\pi \text{dtime})+{\text{a}}_4\text{Cos}(2\pi \text{dtime})+a_5\text{dtime}\\ +a_6{\text{dtime}}^2\end{array}$$where flux is provided in kg/day, Q is river discharge (ft³/s), lnQ equals ln(streamflow) minus center of ln(streamflow), dtime equals decimal time minus center of decimal time, other parameters (i.e., a₀, a₁, a₂, a₃, a₄, a₅, a₆) are shown in Table 2. LOADEST produces daily, monthly and annually flux values using maximum likelihood estimation, adjusted maximum likelihood estimation (AMLE), and least absolute deviation statistical approaches. All three output types provided similar results, the AMLE results were used in this study. LOADEST models explained large variability of DIC and DOC, with R-squared values of 0.9836 and 0.9794 for DIC and DOC, respectively.

Baseflow is usually related to subsurface flow and groundwater flow. It is necessary to test the relations between baseflow and riverine carbon since permafrost degradation could enhance soil drainage and recharge, suprapermafrost flow, groundwater-surface water exchange, sub-permafrost flow, and baseflow (Walvoord & Kurylyk, 2016). The full observation of daily river runoff during our study period allows us to use the digital filter method (Lyne & Hollick, 1979) for baseflow separation. To be specific, we generated daily baseflow data in 2014 and 2016 of Catchment 1 with a web-based hydrograph analysis tool, which was built on the BFLOW and Eckhardt filter modules (Lim et al., 2005).

Pearson’s correlation analysis was performed to determine potential relationships between hydrological and surface meteorological factors and dissolved carbon concentrations. The correlation result was illustrated as a correlation matrix, which displayed Pearson’s correlation coefficients (r_p) with significance levels and scaled kernel regression smoothing lines. Linear correlation analyses were performed to assess the impact of spatial difference and thaw/freeze depths on DIC and DOC concentrations. Since Catchment 3 has the most intense observations of active layer thaw, we use the data from Catchment 3 to analysis the impact of freeze and thaw depth on dissolved carbon export.

Temporal and spatial variations of dissolved carbon

The DIC and DOC concentrations at the outlet of the Zuomaokong watershed displayed different fluctuation patterns from May to October (Figs. 2A and 2B). The DIC concentrations showed larger fluctuations than those of DOC from May to October in both 2014 and 2016. The coefficients of variation (CVs) of DIC in 2014 and 2016 were 0.39 and 0.19, while the CVs of DOC in 2014 and 2016 were 0.22 and 0.18, respectively. The DIC concentrations ranged from 8.15 to 30.9 mg L⁻¹ and 21.58 to 47.5 mg L⁻¹ in 2014 and 2016, respectively. The DOC concentrations during 2014 and 2016 ranged from 2.33 to 6.38 mg L⁻¹ and 4.99 to 13.79 mg L⁻¹, respectively. Large variations of dissolved carbon concentrations appeared both seasonally and annually. Generally, the DIC concentrations increased with Julian day during the open-water seasons (Fig. 2, slopes of trend lines are 0.11 and 0.09 for 2014 and 2016), while the DOC concentrations were relatively steady. Similar patterns were also found in other four sub-catchments (Fig. S6).

Seasonal variations of the dissolved carbon fluxes from the entire watershed were high in 2014 and 2016 (Fig. 3). The highest DIC and DOC fluxes (DICF and DOCF) occurred in September, whereas the lowest DICF and DOCF occurred in May. Not surprisingly, the DICF and DOCF were closely related to river discharge since fluxes were calculated as the function of concentration and discharge. The largest runoff depths and carbon fluxes appeared simultaneously in September. However, the month with the highest precipitation did not have the highest carbon flux. The fluvial DIC fluxes were 2.76 g m⁻² year⁻¹ in 2014 and 5.14 g m⁻² year⁻¹ in 2016, whereas the DOC fluxes were 0.61 g m⁻² year⁻¹ in 2014 and 1.27 g m⁻² year⁻¹ in 2016. The average riverine DIC and DOC fluxes were 3.95 and 0.94 g m⁻² year⁻¹, respectively.

The average DIC concentration was markedly higher than the average DOC concentration at the watershed outlet and in all of the sub-catchments (Table 3), which demonstrated that DIC was the dominant form of dissolved carbon export in QTP permafrost rivers. In the meantime, dissolved carbon concentrations showed spatial differences among the sub-catchments. The mean DIC and DOC concentrations of the entire watershed measured at the outlet (Catchment 1) were 27.81 ± 9.75 and 6.57 ± 2.24 mg L⁻¹, respectively. Catchment 4 had the highest mean DIC and DOC concentrations, whereas Catchment 3 displayed the lowest DIC concentration, and Catchment 2 displayed the lowest DOC concentration (Table 3).

Additionally, the groundwater samples had DIC and DOC concentrations that were nearly double those of the stream water, which were 59.28 ± 16.58 and 11.59 ± 2.53 mg L⁻¹, respectively. But DIC and DOC concentrations of the rainwater were much lower (3.08 ± 1.93 and 2.66 ± 1.93 mg L⁻¹ for DIC and DOC, respectively) than stream water.

Factors influencing DIC and DOC: from time to space

The correlations between the time series of hydrological and meteorological variables and DIC and DOC concentrations in Catchment 1 showed that river runoff was slightly correlated with the DIC and DOC concentrations, with r_p of 0.30 and 0.32, respectively. Interestingly, DIC and DOC concentrations showed closer correlations with the baseflow (r_p = 0.67 and 0.54, respectively) than total river runoff. Precipitation and air temperature showed weak correlations with DIC and DOC concentrations. Meanwhile, the DIC and DOC concentrations were closely correlated to each other, with an r_p of 0.84 (Fig. 4), indicating close associations between DIC and DOC.

The different characteristics of the sub-catchments allow us to analyze the potential factors affecting the spatial variations of dissolved carbon concentrations with correlation analyses. The results showed that the dissolved carbon concentrations were closely associated with the land cover. DIC and DOC concentrations were positively correlated with Vc (r_p = 0.82 and 0.91, respectively), but weakly correlated with wetland cover (r_p = 0.21 and 0.32, respectively) within the five catchments (Fig. 5). The bare land coverage showed negative relationships with DIC and DOC (r_p = −0.88 and −0.78, respectively). The dissolved carbon concentrations decreased as mean elevation increased since vegetation cover decreased and bare land cover increased in higher elevation catchments. Catchment size showed a limited impact on the dissolved carbon concentrations in our study watershed.

Effects of active layer freeze-thaw cycles

Our results revealed that the thawing process had a significant impact on dissolved carbon concentrations and fluxes. The linear relationships between the DIC concentrations and the thawed depths during the 2016 thawing periods were statistically significant (R² = 0.289, p = 0.007, Fig. 6A). As the active layer thawed at greater depths, the DIC concentrations in the river increased. The relationship between the DOC concentration and the thawed depth was weaker but also statistically significant with positive relations (R² = 0.162, p = 0.05, Fig. 6C). In this period, the DIC and DOC concentration was 19.47 and 5.56 mg/L, respectively. The DIC and DOC fluxes showed strong positive relationships with the thawed depth. The DIC and DOC fluxes increased as the thawed depth increased in both 2014 and 2016 thawing seasons. The R² values and p-values of these relationships were in the range of 0.42–0.49 and <0.001, respectively.

As temperature decreased in October, the soil active layer started to freeze two-sided, from the surface and the bottom of the soil active layer (Woo, 2012). The results showed that the DIC and DOC concentrations increased as the frozen depth increased during the 2016 freezing period (Figs. 7A and 7C). In this period, the DIC and DOC concentration was 35.91 and 7.28 mg/L, respectively. As the freezing process proceeded and the frozen depth became deeper, the DIC and DOC fluxes decreased (R² = 0.5546 and 0.5147, p < 0.001, and p < 0.001, for DIC fluxes and DOC fluxes, respectively).

Patterns in stream water DIC and DOC

The mean DOC concentration in our study area was much lower than many comparable small Arctic streams, where these rivers showed tens to hundreds milligram per liter of DOC concentrations (Frey, Sobczak & Mann, 2016; Tank et al., 2012a; Vonk et al., 2013). The values of SOC density measured in Tibetan Plateau alpine meadows (9.05 kg C m⁻² in 0–100 cm depth) (Yang et al., 2008) were much lower than those measured in Arctic permafrost regions (32.2–69.6 kg C m⁻²) (Tarnocai et al., 2009), which may explain the low DOC concentrations compared to most Arctic rivers. The rapid utilization of DOC after thaw (Drake et al., 2015; Vonk et al., 2015a) may also result in the low DOC in permafrost river water. The mean DIC concentrations of stream water within our field area were higher than those of Arctic permafrost rivers (Prokushkin et al., 2011; Tank et al., 2012b). One possible reason may be the high SIC density (11.87 kg C m⁻² in 0–100 cm depth) in the QTP alpine meadow (Yang et al., 2010a) which produce high DIC concentrations. Also, the large proportion of subsurface interflow, which is averagely 25.2% of total runoff (Wang, Hu & Li, 2009), could possibly introduce much soil carbon into the river channels. The dissolved carbon concentrations were higher than non-permafrost rivers (Ran et al., 2013; Vihermaa et al., 2016). The relatively high total dissolved carbon concentrations at our site indicated that the active layer in such high-elevation permafrost is a hotspot for promoting dissolved carbon loss.

Dissolved inorganic carbon was the dominant component of dissolved carbon exported to river water. The main reason for the higher DIC than DOC may be that the SIC stock is as large as 2.1 times of SOC stock in the Tibetan Plateau ecosystem (Yang et al., 2010a). The dominant throughflow leached inorganic material from the deep soil layer, which has higher SIC density than the upper soil layer (Yang et al., 2010a). According to the cation measurements of our site (Fig. S5), the dominant cations are Ca and Na+K (Na⁺ > Ca²⁺ > Mg²⁺ > K⁺), which indicate an important role of mineral weathering and evaporates salt dissolution in the study watershed. A previous study showed widespread carbonate weathering processes (CaCO₃ + H₂O + CO₂ → Ca²⁺ + 2HCO₃⁻) across the Yangtze River source region (Wu et al., 2008), which can consume CO₂ and produce DIC. Besides, the average DOC and total dissolved nitrogen ratio (DOC/TDN, C/N) in our watershed were ∼4.32 (Table S1), which is much lower than the global average C/N value of 22.1 (Meybeck, 1982). Low C/N ratio implies an abundant supply of nitrogen that could enhance the biological metabolism and DOC microbe decomposition (Wiegner et al., 2006). Another evidence is that a previous study at this site showed considerable amounts of old water (which is dominant in groundwater) constituted the stream flow (Song et al., 2017). The long retention time of old water within the basin could increase the contact of DOC to subsurface microbes, which may promote the decomposition of DOC in the active layer (Lyon et al., 2010; Striegl et al., 2005, 2012). The higher DIC and lower DOC concentrations also indicated large proportions of subsurface flow and deeper groundwater recharge to streamflow in permafrost watersheds (Giesler et al., 2014; Walvoord & Striegl, 2007). High DIC and low DOC were also found in Arctic rivers including Tanana and Porcupine rivers (Striegl et al., 2007), Yenisey River (Prokushkin et al., 2011), and the Yukon River (Striegl et al., 2005) due to enhanced organic carbon mineralization.

There were increasing trends of DIC concentrations during the thawed periods of 2014 and 2016, but the DOC concentration remained relatively steady (Fig. 2). The DIC:DOC ratios increased during the thawed periods of 2014 and 2016 (Fig. S7). Walvoord & Striegl (2007) also found an increasing trend of the groundwater contribution and adding of DIC to streamflow in the Yukon River basin. The similar pattern of linear increase through the year in DIC was also found in Sweden rivers (Giesler et al., 2014). A possible explanation is that both DIC concentration and DOC concentration increased because the groundwater flow and flow path lengths increased as thawing progressed to greater depths (Vonk et al., 2015a), but the SIC density increased with soil layer depth (Zhang et al., 2016), which could result in higher DIC export when active layer thaw. The increased DIC concentration with Julian day in our study indicated deeper thawed depth and larger groundwater contribution proportion to total runoff from spring to autumn (Wang et al., 2017). In the autumn, the groundwater discharge contributed more than 75% of the total river runoff (Wang et al., 2017), which brought both high DIC concentration and DOC concentration to stream water. Future degradation of permafrost may lead to groundwater-dominated hydrologic systems among permafrost regions (Frey & McClelland, 2009), which will cause more carbon export through groundwater.

The correlation relationships of DIC and DOC concentration and discharge were limited compared to those of previous studies (Li & Bush, 2015; Song et al., 2016; Tank et al., 2012b; Tian et al., 2015). The dilution of dissolved carbon during high flow periods and the large proportion of groundwater recharge during low flow periods may weaken the relationship between dissolved carbon concentrations and discharge (Tank et al., 2012b). Baseflow was more closely related to dissolved carbon concentrations than total runoff. Baseflow contains shallow subsurface flow and groundwater flow, which could directly leach soil carbon to water channel when active layer thaw. As a result, more riverine carbon originated from the baseflow rather than the overland flow. The role of baseflow for permafrost riverine carbon export will become increasingly important in a warming climate. As permafrost degrades, enhanced baseflow and subsurface hydrological activities (Frey et al., 2007; Walvoord & Kurylyk, 2016) could bring more permafrost carbon to aquatic systems. Our results highlight the importance of dissolved carbon supply from baseflow to river flow.

The mean river runoff in 2016 was 20% larger than 2014 (1.27 vs. 1.06 m³/s of daily mean discharge). This may explain the dissolved carbon concentrations and fluxes in 2016 were larger than those of 2014. The carbon fluxes results of QTP differed from measurements taken in the Yenisey basin, where carbon export was dominated by DOC (Prokushkin et al., 2011). The DIC flux constituted 81% of the total dissolved carbon flux, and this proportion was close to Yukon rivers at Eagle (80%) and Pilot Station (75%) (Striegl et al., 2007). In 2014 and 2016, the total riverine dissolved carbon exported from the Zuomaokong watershed were 431 and 818 t C year⁻¹, respectively. The higher dissolved carbon flux in 2016 was due to the higher dissolved carbon concentrations and larger river runoff in 2016.

The spatial variations in the dissolved carbon concentrations of the five catchments showed both the DIC and DOC concentrations are higher in catchments with high Vc values (Fig. 5). The catchments with larger Vc will have higher ecosystem productivity and SOC density (Wang et al., 2002, 2008) and thus contain more abundant sources for riverine carbon. Previous studies at this site showed that the higher the Vc is, the higher the soil temperature during the thawing and freezing period will be (Wang et al., 2012). High vegetation covers also led to earlier thaw-rise times (Wang et al., 2012). Thus, the mean dissolved carbon concentration increased because the higher soil temperatures and earlier thaw-rise times make carbon from the thawed soil available to the water flow. The coupling between vegetation and soil thermal conditions caused the differences in dissolved carbon concentrations under different vegetation cover. Conversely, catchments with larger bare land cover could have lower soil carbon density and later thaw-rise time, which decrease the dissolved carbon concentrations. Dissolved carbon concentrations decreased with the mean elevation of the catchment may be due to the high-elevation catchments have greater areas of bare land and snow cover. Wetland distribution can enhance the production of organic carbon (Huntington & Aiken, 2013) and have high DOC concentration in the soil profile (Chen et al., 2017), which explained the positive correlation between DOC and wetland coverage.

Active layer dynamics and riverine carbon export

During the thawing period, the DIC concentration increased as the thawing processes progressed (Fig. 6). Thawed permafrost could enhance the connection between water and soil horizons and increase carbon concentrations (Vonk et al., 2015a). During the thawing period in spring, as air and soil temperatures rose, the surface soil layer started to thaw. The dominant runoff component in this period was subsurface interflow (Wang et al., 2017). According to the previous investigation in the QTP grassland, SIC density at different soil depths showed a pattern of top 30 cm < top 50 cm < top 100 cm (Yang et al., 2010a). Thus, when the active layer gradually thawed to the deeper layer, more SIC was export by the lateral flow. The higher the air temperature, the greater thickness of the active layer (Wu & Zhang, 2010), which could result in more soil interflow and deeper flow pathways. As soil thaw depth increased, the leaching potential of dissolved carbon from the soil also increased. Besides, the soil respired CO₂ also rise as active layer thaw (Chen et al., 2017). More soil CO₂ and soil water can form carbonic acid, which may enhance the mineral weathering and produce more DIC (Zolkos, Tank & Kokelj, 2018). Consequently, more DIC leached from the active layer to the water flow, and the DIC concentrations of river water increased.

Surprisingly, DOC concentrations were positively correlated with thaw depth (Fig. 6C). Study in a northern QTP stream found DOC concentrations negatively correlated with thaw depth due to less organic carbon source in the deeper soil layer (Mu et al., 2017). While in our study area, the SOC is 27.39, 33.15, and 28.69 mg/g for 0–5, 5–20, and 20–40 cm, respectively (Chen et al., 2017), which partly explained why the riverine DOC increased with thawed depth during spring thaw period. When the active layer thawed to the deeper soil layer where SOC is lower (Ding et al., 2016; Mu et al., 2017; Yang et al., 2008), the groundwater contribution rate increased as the thawing process proceeded (Wang et al., 2017). Therefore, the increased groundwater, which contains a high concentration of DOC (11.59 mg/L), overriding the potential effects of lower SOC supply in the deeper soil layer and OC decomposition (Drake et al., 2015; Striegl et al., 2005, 2007). As a result, riverine DOC increased. Even so, whether such pattern is widespread across QTP permafrost rivers need further study. The dissolved carbon fluxes also increased with thawed depth as more river runoff and higher carbon concentrations during in deep thawed period.

During the freezing period, the active layer displayed a pattern of two-sided freezing (Woo, 2012). As the air temperature decreased, the surface layer began to freeze and cool downward. The frozen surface layer stopped the escape of heat from the middle layer. The bottom of the active layer also started to freeze at the same time. Therefore, the middle layer was still unfrozen during the early stage of the freezing process. Meanwhile, precipitation decreased rapidly during the freezing period. The frozen impermeable surface and bottom layer obstructed the infiltration of precipitation and groundwater to deeper layers (Wang, Hu & Li, 2009). As a result, the proportion of groundwater supply in river runoff increased when the frozen depth increased (Wang et al., 2017). The groundwater is characterized by high dissolved carbon concentrations (59.28 mg L⁻¹ of DIC and 11.59 mg L⁻¹ of DOC) than the river water and precipitation (Table 3). Under these circumstances, the DIC concentration and DOC concentration increase as the frozen depth increases. The decrease in dissolved carbon fluxes as the frozen depth increased was because of the rapidly decreased river discharge, which caused by reductions in precipitation and the gradual freezing of subsurface flow. Through the above analyses, we can see that active layer soil carbon exported to rivers was closely linked with thawing and freezing period as illustrated in the conceptual diagram (Fig. 8). Although our measurements of active layer thaw/freeze depths may have uncertainties, the pattern of active layer impacts still existed in 2 years of different meteorological and hydrological characteristics.

We consider the soil active layer freeze-thaw cycles to be the most important factors affecting the riverine dissolved carbon export in our site for the following four reasons. First, the thawing and freezing processes controlled the thaw-released soil carbon from the active layer. As shown before, riverine dissolved carbon concentrations were closely related to active layer thaw and freeze depth. Second, permafrost river baseflow, which closely related to riverine carbon, was largely regulated by active layer thaw (Smith et al., 2007). Third, the widely distributed freeze-thaw erosion in the QTP, which can affect the lateral soil carbon export, is controlled by freeze-thaw cycles (Zhang, Liu & Yang, 2007; Wang et al., 2017). Fourth, the dissolved carbon fluxes were determined by discharge volumes, while surface runoff processes were primarily controlled by active layer freeze-thaw cycles rather than precipitation (Wang, Hu & Li, 2009). The systematic shifts of DIC and DOC we observed during the flow season mainly reflected the seasonal change of the active layer.

The active layer was predicted to become thicker in the future warming scenario (Wu & Zhang, 2010; Yang et al., 2010b; Zhang, Frauenfeld & Serreze, 2005). This prediction suggests earlier spring thaw and later autumn freeze up due to the increases in the length and depth of subsurface flow paths that will occur under climate warming (Frampton et al., 2011). The period during which lateral carbon losses occur in each year will be extended. Increased thaw depth could possibly extend the residence time of dissolved carbon in the flow path, which could affect the biodegradability of DOC (Vonk et al., 2015b). It then becomes inevitable that the active layer change will change the carbon export patterns. Additionally, changes in soil water content could affect the active layer thermal state and the thawing and freezing processes of the soil active layer (Wang et al., 2012), which may also accelerate the lateral carbon export. A warming climate and the corresponding increase in the thickness of the active layer (Yang et al., 2010b) may enhance this carbon loss through permafrost degradation and active layer thaw, which will eventually affect the regional carbon cycling.