Spatial and temporal variability of pCO2 and CO2 emissions from the Dongjiang River in South China

CO2 efflux at the water–air interface is an essential component of the riverine carbon cycle. However, the lack of spatially resolved CO2 emission measurement still hinges the accuracy of estimates on global riverine CO2 emissions. By deploying floating chambers, seasonal changes in river water CO2 partial pressure (pCO2) and CO2 evasion from the Dongjiang River in South China were investigated. 15 Spatial and temporal patterns of pCO2Lateral soil CO2 input and dilution effect caused by precipitation played critical roles in controlling riverine pCO2 in small rivers, while the decomposition of allochthonous organic carbon is responsible for pCO2 variability in large rivers. were mainly affected by terrestrial carbon inputs and in-stream metabolism, both of which varied due to differential catchment settings, land cover, and hydrological conditions. Temperature-normalized gas transfer velocity (k600) in 20 small rivers were 8.29 ± 11.29 m d and 4.90 ± 3.82 m d for the wet season and dry season, respectively, which were nearly 70 % higher than that of large rivers (3.90 ± 5.55 m d during the wet season and 2.25 ± 1.61 m d during the dry season). A significant correlation was observed between k600 and flow velocity but not wind speed regardless of river size. Majority of the surveyed rivers were net CO2 source, while exhibiting substantial seasonal variations. The mean CO2 flux was 300.1 and 264.2 mmol m d 25 during the wet season for large and small rivers, respectively, 2-fold larger than that during the dry season. However, no significant difference in CO2 flux was observed between small and large rivers. The absence of commonly observed higher CO2 fluxes in small rivers could be associated with the depletion effect caused by abundant and consistent precipitation in this subtropical monsoon catchment.


Introduction 30
River networks act as a processor that transfers and emits the carbon entering the water, rather than just a passive pipe that transports carbon from the terrestrial ecosystem to the ocean (Cole et al., 2007;Battin et al., 2009;Drake et al., 2018). CO2 emissions at the water-air interface are an essential component of the riverine carbon cycle. CO2 emitted from inland waters to the atmosphere reaches up to 2.9 PgC yr −1 , surpassing that transported from land to ocean through rivers (Sawakuchi et al., 2017;Drake et al., 2018). 35 Understanding the role that rivers play in the global carbon cycle is still hindered by uncertainty on the estimate of CO2 flux outgassing from rivers (Cole et al., 2007;Raymond et al., 2013;Sawakuchi et al., 2017;Drake et al., 2018). Riverine carbon emissions have significant temporal and spatial variations, making it challenging to quantify carbon emissions accurately. In addition, watershed geomorphology, hydrological conditions, climate, and other environmental factors can affect the CO2 efflux in rivers (Alin 40 et al., 2011;Abril et al., 2014;Almeida et al., 2017;Ran et al., 2017a;Borges et al., 2018). Thus, there are substantial differences in CO2 efflux among rivers in different climate regions, or the same river but between different seasons (Denfeld et al., 2013;Rasera et al., 2013). An enhanced understanding of the temporal and spatial characteristics of the water-air CO2 flux will facilitate a more robust estimate.
However, global riverine carbon emission estimates were largely based on data disproportionately 45 focusing on temperate and boreal regions, including North America and Europe (Raymond et al., 2013;Lauerwald et al., 2015;Drake et al., 2018). In light of this data gap, more studies are required in other data-poor regions to achieve a more accurate estimate.
Rivers in tropical and subtropical regions of East Asia and Southeast Asia are among those underrepresented regions that need more attention since they are essential participants in riverine carbon 50 transport (Ran et al., 2015;Ran et al., 2017b;Drake et al., 2018). The high temperature in this region facilitated a high net primary productivity in the terrestrial ecosystem and intense biochemical activities, and both contributed to the carbon input dynamic from soil to rivers . Meanwhile, rivers in this region are under the heavy influence of monsoon, and riverine CO2 emissions vary significantly among seasons due to the changes in temperature and precipitation. In addition, different rivers in this 55 region may have contrasting trends in CO2 dynamic due to different underlying controlling factors. Some rivers have the highest CO2 efflux in the wet season Le et al., 2018;, while others have the highest CO2 efflux in the dry season   Le et al., 2018;, suggesting that an increase in wet season runoff can have two distinct consequences. One possibility is that it increases external carbon inputs and CO2 emissions (Hope et al., 60 2004;Johnson et al., 2008), while the other is that it leads to a dilution of CO2 in rivers and accordingly a reduction in CO2 emissions (Ran et al., 2017b;Li et al., 2018). Since starkly different outcomes can occur, it is important to investigate the processes behind such diverse response of rivers to the monsoon.
The Dongjiang River (DJR), located in the subtropical region of South China, is one of the three tributaries of the Pearl River. Previous studies on riverine carbon transportation and emissions in the 65 Pearl rivers mainly focused on the Xijiang River, which was characterized by widely distributed carbonate rocks, and the estuary area of the Pearl River Delta (Yao et al., 2007;Zhang et al., 2015;Zhang et al., 2019;Liang et al., 2020). Though some studies have been conducted in the Dongjiang River basin (DJRB) focusing on carbon transport and the carbon sink effect of chemical weathering (Tao et al., 2011;Fu et al., 2014), there is still a lack of understanding of the characteristics of catchment-wide CO2 70 emissions in DJRB. Furthermore, a predominantly hilly landscape combined with abundant precipitation favors the formation of a great number of small rivers in DJRB (Ding et al., 2015). However, the current estimate of basin-wide CO2 emission from the river network was mostly based on the data from large rivers, and small rivers are heavily underrepresented (Raymond et al., 2013;Drake et al., 2018). Because the controlling factors and the input of carbon could be significantly different between large and small 75 rivers (Johnson et al., 2008;Dinsmore et al., 2013;Hotchkiss et al., 2015;Marx et al., 2017), which can lead to very distinctive pattern of carbon dioxide evasion, More more comprehensive quantification of CO2 evasion from small headwater streams is necessary. Therefore, studies on the characteristics of riverine CO2 emission in DJRB should be conducted among river size spectrums, and the impact of monsoon ought to be considered. 80 By using directly measured river water CO2 partial pressure (pCO2) and CO2 efflux data from DJRB, and in conjunction with hydrological and physicochemical data, the objectives of this study were to 1) investigate the spatial and temporal pattern of pCO2 and CO2 emission along stream size spectrum, 2) examine the differences in hydrological and physicochemical controls of pCO2 and the CO2 evasion between small headwater streams and large rivers. The results of this study could shed light on the 85 underlying controls of the spatial and temporal distribution of riverine pCO2 and support a refined estimate of regional and global carbon budgets.

Site Description
The DJR in South China is one of the three major tributaries of the Pearl River system (Figure 1). It has 90 a 562 km long mainstem channel and a drainage area of 35,340 km 2 (Chen et al., 2011). Due to its subtropical monsoon climate, precipitation in DJRB exhibits significant seasonal variability (Figure 2a).
The multi-annual average precipitation is about 1800 mm, 80 % of which is concentrated during the wet season from April to September. The Boluo Hydrological Gauge is the lowermost gauge of the Dongjiang River mainstem channel, controlling a drainage area of ~23,000 km 2 . The multi-annual average water 95 discharge at Boluo Hydrological Gauge is 23.7 km 3 (Zhang et al., 2008). About 80-90 % of this discharge is transported during the wet season ( Figure 2b). The landscape is characterized by plains and hills, accounting for 87.3 % of the river basin area (Ding et al., 2015), and The the dominant land use of the catchment is highly diverse evergreen forests of broad-leaved and needle-leaved species (Ran et al., 2012;Chen et al., 2013). The impacts of human activities on land use vary among three regions in the DJRB. 100 Urban expansion and agricultural activities have substantially altered the land use in Lower and Middle Dongjiang River Basin (LDJRB and MDJRB), respectively, while the Upper Dongjiang River Basin (UDJRB) is less affected by human activities (Figure 1)., and the landscape is characterized by plains and hills, accounting for 87.3 % of the river basin area (Ding et al., 2015).

Field Measurement and Analysis
In total, there were 43 sampling sites from spanning seven Strahler stream orders. Fourth to seven order streams were mainstem and major tributaries, while first to third order streams were small tributaries. River widths were measured by a laser rangefinder. Sampled rivers were categorized, according to their stream orders, into small rivers (first to third order streams, SR) and large rivers (fourth to seventh order 120 streams, LR). The small rivers had an average width of 15.4 ± 10.2 m (4.8 ± 2.3 m, 10.4 ± 5.6 m, 22.9 ± 8.1 m for first to third order streams, respectively), while large rivers have an average width of 180.8±156.0180.3 ± 159.3 m (Table S1) (75.2 ± 51.0 m, 168.0 ± 48.6 m, 235.7 ± 29.6 m, 433.4 ± 178.0 m for fourth to seventh order streams, respectively). Those sampling sites were widely distributed in the mainstem and nine major subcatchments among three regions with different topographic features and 125 land cover ( Figure 1).
In order to investigate CO2 emissions during different hydrological conditions, we performed five fieldwork campaigns from December 2018 to October 2019, including three in the wet season (early wet season -late April, middle wet season -early July, and late wet season -late August) and two in the dry season (middle dry season -December 2018 to early January 2019 and early dry season -late October 130 2019. Sample sites were measured in the daytime over two weeks for each field trip. Three rounds of campaigns in the wet season allow each sample site to be measured under different hydrological conditions, and the two-week duration of each campaign allowed streams with different orders and sizes to be measured under various discharges. As for the dry season, the hydrological condition was relatively stable due to low precipitation. However, field measurements conducted during the daytime could lead 135 to an underestimate in pCO2 and CO2 emission (Reiman and Xu, 2019a).In order to investigate CO2 emissions during different hydrological conditions, we performed five fieldwork campaigns from December 2018 to October 2019, including late December 2018 to early January 2019 (middle dry season), April (early wet season), early July (middle wet season), late August (late wet season) and late October 2019 (early dry season). Nocturnal CO2 emission rates in rivers could be 27% greater than the 140 daytime rates (Gómez-Gener et al., 2021).
During the field trips, water temperature, pH, and dissolved oxygen (DO) were measured with a portable multiparameter probe (Multi 3430, WTW GmbH, Germany). The pH probe was calibrated before each field trip with standard pH buffers (4.01 and 7.00). Measurements were conducted 10 cm below the water surface. To evaluate the contribution of metabolism on DO changes, ΔCO2 and ΔO2 were calculated as 145 described by Stets et al. (2017) using: and Where CO2w and O2w are measured concentrations of CO2 and O2 in water sample, while CO2a and O2a 150 are the equilibrium CO2 and O2 concentrations (μmol L −1 ).
Flow velocity was determined by using a Global Water Flow Probe FP111 with a precision of 0.1 m s -1 Flow velocity was determined using a flow meter, while wind speed at 1.5 m above the water surface was measured with a Kestrel 2500 handheld anemometer and normalized to a height of 10 m (U10) using the equation from Alin et al. (2011). As the flow velocity was measured near the riverbanks, an 155 underestimation of the flow velocity is possible. Flow velocity measured near the riverbanks is only about 40% of the maximum flow velocity at the cross-section (Moramarco et al., 2004;Le Coz et al., 2008).
We also collected water for analyzing total alkalinity (TA) and dissolved organic carbon (DOC). Firstly, 100 ml of water samples were filtered through a pre-combusted glass fiber filter (pore size: 0.47 µm, 160 Whatman GF/F, GE Healthcare Life Sciences, USA). Then, 50 ml of water used for TA analysis was titrated with 0.1 mol L −1 HCl at on the same day of sampling. The remaining 50 ml of water for DOC analysis was poisoned with concentrated H2SO4 to pH < 2 and preserved in a cooler with ice bags before analysis. DOC was determined by the high-temperature combustion method using a TOC Analyzer (Elementar Analysensysteme GmbH, Langenselbold, Germany) that has a precision better than 3 %. 165

Calculation of pCO2 and CO2 emission flux
The surface water pCO2 was determined using the headspace equilibrium method, which could avoid the possible overestimation of using TA and pH to calculate pCO2 in rivers with a relatively low pH (Abril et al., 2015). We used a 625 mL reagent bottle to collect 400 mL of water from ~10 cm below the surface, leaving 225 mL of space filled with ambient air as headspace. The bottle was then immediately capped 170 and shaken vigorously for at least 1 min to achieve an equilibrium between the water and the CO2 in the headspace (Hope et al., 1994). Then, the bottle was connected to the calibrated Li-850 CO2/H2O gas analyzer (Li-Cor, Inc, USA), and the equilibrated gas in this closed loop was measured. The measurements at each site were repeated twice, and the average was then calculated. The variation between the two measurements was less than 5%, and the accuracy of Li-850 is within 1.5% of the 175 reading. The ambient air pCO2 ( 2 )was measured before the headspace measurements and the chamber deployments.The 2 value varied between 380 and 450 μatm. The ambient air pCO2 ( 2 )was measured before the chamber deployments and varied between 380 and 450 μatm. The measurements at each site were repeated three times, and the average was then calculated. The original surface water pCO2 ( 2 , ) was finally calculated by using solubility constants (K0) for CO2 from 180 Weiss (1974), Carbonate constants (K1, K2) from (Millero et al., 2006), and the volume of the flask, headspace, and residual system (line and gas analyzer) (Dickson et al., 2007;Ran et al., 2017a; al., 2019) using: Where ℎ , and , are the headspace volume, residence system volume, and water volume, 185 respectively. R is the universal gas constant (8.314 J mol −1 K −1 ), T is the water temperature in Kelvin (K), and [H + ] is the concentration of hydrogen ion.
For measuring , We filled the headspace with gas, which had a known pCO2, and measured the pCO2 in the closed loop. was then estimated according to equation (23). A comparative analysis of the syringe and bottle headspace method has been conducted to evaluate the accuracy of the headspace extraction method used in this study (Table S2 and Figure S2). Overall, our method could cause a 1-5% 195 underestimation in pCO2.
To reduce the artificial turbulence induced by anchored chambers, we used a small unmanned boat in the measurement, which allowed us to deploy drifting chambers freely in rivers deeper than 0.2 m and with a high flow velocity up to 2 m s −1 . During the deployment, CO2 emission was determined using a circular, 8.5 L floating chamber with a water surface area of 0.113 m 2 . The chamber walls were lowered about 2 200 cm into the water and mounted with a pneumatic rubber tire. The chamber was connected to an infrared Li-850 CO2/H2O gas analyzer (Li-Cor, Inc, USA) in a floating storage box through Polyurethane tubes for CO2 analysis. An unmanned boat connected to both the chamber and box with ropes was used to deploy them near the central line of the river. Once the entire setup reached its designated location, the readings on the Li-850 were recorded at 0.5 s intervals. During the entire measurement process, the box 205 drifted freely with the current. The Li-850 was calibrated by the manufacturer before field trips. The rate of CO2 efflux (FCO2 in mmol m −2 d −1 ) was calculated from the observed change rate of the mole fraction S (ppm s −1 ) using: Where S is the slope of CO2 accumulation in the chamber (μatm s −1 ), V is chamber gas volume (m 3 ), A 210 is the chamber area (m 2 ), t1 =8.64·10 4 s d −1 is the conversion factor from seconds to days, and t2 is a conversion factor from mole fraction (ppm) to concentration (mmol m −3 ) at in situ temperature (T in K) and atmospheric pressure (p in Pa), according to the ideal gas law: The gas transfer velocity (k) was calculated from FCO2 and pCO2 in both water and ambient air using: 215 To compare gas transfer velocity values among different sites, k was standardized to k600 as described by Alin et al. (2011) using: Where, is the Schmidt number, which is dependent on temperature (T) in degree Celsius (Wanninkhof, 220 1992): In total, 196 chamber measurements were made. In 19 out of 215 sample sites, the drifting chamber was unable to deploy due to shallow water or high flow velocity. Meanwhile, 8 out of 196 k600 data with the air-water pCO2 gradient less than 200 μatm were also excluded, as the error in these calculations could 225 be considerable. (Borges et al., 2004).

Physical and Biochemical Characteristics
The Dongjiang River was characterized by substantial seasonal variations in hydrologic regimes ( Table   Figure 21). Stream width in the wet season was 17.0 % and 45.6 % larger than that in the dry season for 230 small and large rivers, respectively (Table S1). The Q discharge ranged 5 4 orders of magnitude from 0. 01 m 3 s −1 in the small headwater streams during the dry season to 6690 m 3 s −1 in the main stem during the wet season ( Figure S1). Water temperature was higher in July and August (21.4-33 and 21-33.4 ℃, respectively) than that in January (8.1-22.2 ℃), April (16.5-26.9 ℃), and October (17.4-29.7 ℃). pH varied from 6.38 to 8.14, with a mean of 7.08. There was no significant (independent sample t test, p > 235 0.05) change in pH between wet and dry seasons. U10 based on all stream sites was higher in large rivers (0.86 ± 0.91 and 1.43 ± 1.58 m s −1 in wet and dry season, respectively) than in small rivers (0.62 ± 0.61 and 0.76 ± 0.73 m s −1 in wet and dry season, respectively).
Similarly, the alkalinity in large rivers was 790 ± 402 μmol L −1 in wet season, 14.5 % lower than 924 ± 411 μmol L −1 in dry season. However, the lowest value of alkalinity in large rivers was observed in August (739 ± 312 μmol L −1 ) instead of April in small rivers. 245 Spatial and seasonal changes in DOC concentration were also observed in the surveyed rivers (Table 1).
DOC concentration in larges rivers (1.94 ± 1.52 mg L −1 ) was 41.6 % higher than that in small rivers (1.37 ± 0.72 mg L −1 ). Meanwhile, DOC concentrations in the wet season were 2.22 ± 1.82 mg L −1 and 1.54 ± 0.72 mg L −1 for large and small rivers, respectively, which were 45.1 % and 54 % higher than that in the dry season (1.53 ± 0.72 and 1.11 ± 0.63 mg L −1 for large and small rivers, respectively). 250
Meanwhile, there was also an increasing trend of pCO2 from rivers in UDJRB to those in LDJRB. The pCO2 values were 2105 ± 959 and 2487 ± 1276 μatm for small and large rivers respectively in LDJRB, which were 146.7% and 70% higher than that in UDJRB, respectively (Figure 3b). 265 Seasonal variations of pCO2 differ across the stream size spectrum (Figure 3b4). In small rivers, the highest pCO2 was observed in April (1506 ± 880 μatm), which was 50.3 % higher compared to with January (1002 ± 660 μatm). pCO2 then decreased in July (1131 ± 589 μatm) and increased in August 275 (1325 ± 863 μatm) and October (1414 ± 900 μatm). Compared to with small rivers, the peak of pCO2 in large rivers occurred later but persisted for a longer period of time. In large rivers, an increase in pCO2 was not observed until July. pCO2 in April was 1831 ± 793 μatm, which was similar to 1805 ± 1010 μatm in January, and it increased 39.3 % to 2550 ± 1210 μatm in July. pCO2 peaked in August (2885 ± 1351 μatm) and then decreased to 2176 ± 1166 in October. Overall, pCO2 was 9.3 % and 21.7 % higher 280 in wet season than in dry season for small and large rivers, respectively. ± 511.7 mmol m −2 d −1 respectively during the wet season, which was 87.2 % and 123.1 % higher 295 compared tothan that in the dry season (141.1 ± 188.7 and 134.5 ± 129.5 mmol m −2 d −1 for small and large rivers respectively). No significant (independent sample t test, p > 0.05) difference in FCO2 was observed between small and large rivers.
k600 differs greatly between river size classes and among hydrological periods (Figure 5b4b). k600 values in small rivers were significantly (independent sample t test, p < 0.001) higher on average than in large 300 rivers. The mean values of k600 in small rivers were 8.29 ± 11.29 m d −1 and 4.90 ± 3.82 m d −1 for the wet season and dry season, respectively, which were 112.6 % and 70 % higher than that of large rivers (3.90 ± 5.55 m d −1 in the wet season and 2.25 ± 1.61 m d −1 in the dry season). k600 during the wet season were also significantly (independent sample t test, p < 0.05) higher than the dry season. k600 increased 112.7 % and 118.2 % from dry season to wet season in small and large rivers, respectively. However, comparisons 305 between different phases in the same hydrological period (e.g. early, middle, and late wet season) did not differ significantly (paired sample t test, p > 0.05) for both river size classes.
The spatial and temporal variation of CO2 efflux generally coincided with the changes in pCO2 and k600 since high FCO2 occurred when k600 or pCO2 were elevated. In small rivers, the highest CO2 effluxes were 346.8 ± 625.2 mmol m −2 d −1 during April, consistent with the high k600 and pCO2 in this period. In 310 large rivers, high CO2 effluxes were observed in both April (339.9 ± 828.6 mmol m −2 d −1 ) and August (329.9.0 ± 270.0 mmol m −2 d −1 ), which were attributed to high k600 in April and high pCO2 in August.

Underlying Processes of pCO2 dynamics
Previous studies show that riverine CO2 originated from both lateral soil CO2 input and in-stream 320 metabolism (Yao et al., 2007;Li et al., 2013;Abril et al., 2014). The river water pCO2 was positively related to DOC and negatively related to DO (Figure 5), indicating that decomposition of terrestrial organic carbon is an important source for pCO2 (Stets et al., 2017;Liang et al., 2020). To compare the contribution of internal metabolism on pCO2 in small and large rivers, ΔCO2: ΔO2 stoichiometry was used to evaluate the impact of respiration and photosynthesis processes on the concentration of O2 and 325 CO2 in water bodies (Stets et al., 2017). The inverse relation between ΔCO2 and ΔO2 ( Figure 6) demonstrated that metabolic processes are important for CO2 variation (Amaral et al., 2020). However, the imbalanced ΔCO2:ΔO2 stoichiometry ( Figure 6) indicates that, in addition to in-stream metabolic processes, other factors also affect the CO2 and O2 in the water (Stets et al., 2017). For example, 183 out of 215 observations are above the 1:1 ΔCO2:ΔO2 line, suggesting additional sources of carbon input. The 330 difference in the ΔCO2:ΔO2 stoichiometry between small and large rivers reflects their differences in the controlling processes (Rasera et al., 2013). In large rivers, the ΔCO2:ΔO2 stoichiometry is closer to the 1:1 line than in small rivers, suggesting large rivers are more affected by the metabolic processes (Jeffrey et al., 2018;Amaral et al., 2020). In comparison, the deviation from the 1:1 line in small rivers indicates a stronger impact of additional carbon sources (Abril et al., 2014;Amaral et al., 2020). 335 The spatial pattern of pCO2 in the DJRB is likely resulting from changes in the intensity of in-stream metabolism. Our data showed that river water pCO2 was negatively related to DO and positively related to DOC (Figure 6), suggesting that metabolic processes are important for CO2 variation (Amaral et al., 2020). High pCO2 and low DO in large rivers could result from more favorable conditions for OC composition. Terrestrial organic carbons are difficult to convert into CO2 in small rivers due to the high 340 flow velocity and short water residence time (Hotchkiss et al., 2015). Conversely, a greater fraction of OC could be transported and fuel the heterotrophic respiration in large rivers, where low flow velocity and long water residence time facilitated the decomposition of organic carbon within the water column (Denfeld et al., 2013).  Differences in seasonal changes of pCO2 between small and large rivers also suggest various primary controlling processes. pCO2 in small rivers are mainly controlled by changes in lateral soil CO2 input. The highest value of pCO2 observed in April could be attributed to a rapid surge of additional soil CO2 input caused by increasing 355 precipitation (Figure 7). In spring, warming temperatures increase the net primary productivity of the terrestrial ecosystem, with a corresponding increase in soil carbon content. Meanwhile, increased precipitation in April facilitates the transportation of the soil carbon from land to the river system (Rasera et al., 2013). Thus, the temperature and precipitation in April dominantly control the soil CO2 concentration, and hence mediate aqueous pCO2 (Hope et al., 2004;Yao et al., 2007;Johnson et al., 2008). In contrast, a decrease of pCO2 in July was observed, 360 and it was likely the result of the CO2 depletion effect in the soil combined with the dilution effect of precipitation. The soil carbon has experienced a depletion effect due to the continuous precipitation and soil erosion since April, limiting the supply of terrestrial carbon input for rivers in July (Hope et al., 2004;Johnson et al., 2007;Dinsmore et al., 2013). Meanwhile, the increase in precipitation and runoff can also cause a dilution effect, which leads to a decrease of pCO2 (Ran et al., 2017b;Li et al., 2018). (Hope et al., 2004)Seasonal variations in alkalinity substantiate 365 the dilution effect and the depletion effect in July. Although the lowest alkalinity in small rivers was recorded in April, the highest pCO2 values in small rivers were recorded in that month. It suggests that the effect of increased soil CO2 input outweighs the dilution effects, both of which are caused by precipitation increase. In contrast, the synchronous upward trend of the alkalinity and pCO2 in the later months of the year implies that the rise in pCO2 results from weakened dilution effect . Moreover, low pCO2 during dry season demonstrates 370 inorganic carbon input via groundwater plays a minor role. Therefore, the variation of soil CO2 input and dilution effect caused by precipitation are the main controlling factors of seasonal changes in riverine CO2 among small rivers.

375
The spatial pattern of pCO2 was also related to the variation in carbon input due to different land cover (Borges et al., 2018). The higher pCO2 in large rivers than small rivers was associated with a higher percentage of urban and cropland cover and lower forest cover ( Figure S3). Compared with the forest, cropland could provide a more favorable condition for soil erosion and the transfer of organic matter from land to rivers, contributing to a higher pCO2. Intensification of agricultural practices could promote 380 the decomposition of soil organic matter (Borges et al., 2018) and increase the concentration of liable DOC, which is more sensitive to in-stream metabolism after entering the rivers (Lambert et al., 2017;. Meanwhile, the input of wastewater with high organic matter concentration from the urban area could also contribute to an increase in riverine pCO2 (Xuan et al., 2020;Zhang et al., 2021). Moreover, our result showed increasing pCO2 from forest-dominated streams in UDJRB to those in 385 agricultural and urban impacted catchments in MDJRB and LDJRB (Figure 3b). Over 70% of forest cover in UDJRB (Figure 1) can reduce the soil erosion associated with precipitation (Ran et al., 2018).
Meanwhile, the organic matter from forest tend to be more aromatic, thus more capable of surviving biodegradation (Kalbitz and Kaiser, 2008), leading to a relatively low riverine pCO2 value. In contrast, cropland, occupying about 49% of the land cover (Figure 1), was the primary land use type in the MDJRB 390 substituting forest, and urban areas accounting for about 17% of the land cover in the LDJRB. The higher pCO2 in the MDJRB and LDJRB is likely under the influence of agricultural practices and wastewater input. Overall, land use mainly affects the spatial distribution of pCO2 by altering the amount and lability of carbon inputs to the rivers.
However, DOC concentration is not likely the primary control of different in-stream metabolism 395 intensities in small and large rivers. Our result showed that large rivers had similar DOC concentration but higher pCO2 compared with small rivers with similar land cover (Figure 7) when the percentage of forest area was over 65% or the percentage of combined cropland and urban area was less than 30%.
This suggested that large rivers have more intense OC decomposition than small rivers with similar DOC concentrations. Therefore, favorable conditions for OC decomposition were more likely to be responsible 400 for the spatial pattern. Another possible carbon source of river water CO2 is direct soil CO2 input.
However, it is unlikely the major contributor of CO2 for large rivers in the DJRB, since the contribution of soil CO2 tends to decrease with the increased stream order and leads to higher pCO2 in small rivers (Marx et al., 2017), which contradicted with the spatial pattern in this study. area combined (b) the relationship between yearly average pCO2 at each site and the percentage of forest area (c) the relationship between yearly average DOC at each site and the percentage of cropland and urban area combined (d) the relationship between yearly average DOC at each site and the percentage of forest area.On the other hand, high pCO2 in large rivers is mainly a consequence of decomposition of organic carbon. Relatively low pCO2 in April indicates a carbon source other than soil CO2 input. When soil carbon dioxide enters river systems, it is readily 410 emitted from the rivers into the air, with little reaching the larger rivers downstream (Denfeld et al., 2013;Drake et al., 2018). The contribution of soil CO2 input to pCO2 could only be secondary. In large rivers, pCO2 increased by 39.3 % from 1831 ± 793 μatm in April to 2550 ± 1210 μatm in July. The rise in temperature from April to July promoted a substantial increase in the net primary productivity of the terrestrial ecosystem and the content of terrestrial organic carbon entering the river (Borges et al., 2018). (Vonk et al., 2013;Dean et al., 2019)Yet, those 415 terrestrial organic carbons are difficult to convert into CO2 in small rivers due to the high flow velocity and short water residence time (Hotchkiss et al., 2015). Thus, a possible explanation of increasing pCO2 in large river is that a greater fraction of OC could be transported and fuel the heterotrophic respiration in large rivers, where long water residence time combined with the high temperature in July facilitate OC decomposition (Denfeld et al., 2013). For large rivers, recent studies have shown that the biological decomposition of allochthonous organic carbon caused by 420 energetic microbial metabolism is the primary source of riverine CO2 (Amaral et al., 2018;Jeffrey et al., 2018). (Borges et al., 2018) (Ran et al., 2018) (Borges et al., 2018) (Lambert et al., 2017; (Xuan et al., 2020;Zhang et al., 2021)  On the other hand, the temporal pattern is likely the consequence of changes in terrestrial carbon input and in-stream metabolism intensity. Our result showed that higher pCO2 occurred in the wet season than 425 the dry season for both small and large rivers (Figure 4). The elevated temperature in the wet season could promote a substantial increase in the net primary productivity of the terrestrial ecosystem, while increased precipitation facilitated the transfer of terrestrial carbon (Rasera et al., 2013), including both soil CO2 and OC, from land to rivers. This could either enhance riverine pCO2 directly or by fuelling OC decomposition (Borges et al., 2018). However, differences in seasonal changes of pCO2 between small 430 and large rivers (Figure 4) also suggested that their primary controlling process could be different. For small rivers, the highest value of pCO2 was observed in April (Figure 4), which is consistent with the rapid surge of terrestrial C input, usually occurring at the beginning of the wet season (Hope et al., 2004;Yao et al., 2007;Johnson et al., 2008). However, such an increase in pCO2 was not observed in large rivers (Figure 4), even though DOC in large rivers, increased during the same period, similar to small 435 rivers (Table 1). A possible explanation is that observed pCO2 rise was mainly originated from soil CO2, which was readily emitted from the small rivers into the air, with little reaching the larger rivers downstream (Denfeld et al., 2013;Drake et al., 2018). Differences in pCO2 dynamic in July and August also reflected differential controlling processes in small and large rivers. A decline in pCO2 in July in small rivers suggested that it might have experienced the depletion effect occurring at middle and late 440 wet season (Hope et al., 2004), during which soil CO2 decreased due to the continual precipitation. In contrast, the increase in pCO2 occurring in large rivers in July indicated that the decrease in soil CO2 input could hardly affect the pCO2 in large rivers during this period. Instead, stronger in-stream metabolism caused by OC input and favorable conditions for OC decomposition is more likely to be responsible for the rising pCO2. 445 To compare the contribution of internal metabolism on pCO2 in small and large rivers, ΔCO2: ΔO2 stoichiometry was used to evaluate the impact of respiration and photosynthesis processes on the concentration of O2 and CO2 in water bodies (Stets et al., 2017). The inverse relation between ΔCO2 and ΔO2 ( Figure 8) demonstrated that metabolic processes are important for CO2 variation (Amaral et al., 2020). It is also supported by the positive relation between river water pCO2 and DOC and the negative 450 relation between pCO2 and DO ( Figure 6). However, the imbalanced ΔCO2:ΔO2 stoichiometry ( Figure   7) indicates that, in addition to in-stream metabolic processes, other factors also affect the CO2 and O2 in the water (Stets et al., 2017). For example, 183 out of 215 observations were above the 1:1 ΔCO2:ΔO2 line, suggesting additional sources of carbon input. The difference in the ΔCO2:ΔO2 stoichiometry between small and large rivers reflects their differences in the controlling processes (Rasera et al., 2013). 455 In large rivers, the ΔCO2:ΔO2 stoichiometry is closer to the 1:1 line than in small rivers, suggesting large rivers are more affected by the metabolic processes (Jeffrey et al., 2018;Amaral et al., 2020). In comparison, the deviation from the 1:1 line in small rivers indicates a stronger impact of external carbon sources (Abril et al., 2014;Amaral et al., 2020), which substantiates our finding that pCO2 of small rivers are more likely affected by soil CO2 input. Furthermore, there were other processes that could affect the 460 riverine pCO2. For example, stronger solar radiation during summer could increase photo-oxidation in rivers. However, commonly observed lower daytime CO2 emission rates than nocturnal rates (Gómez-Gener et al., 2021) suggests that photosynthesis overrides photo-oxidation in CO2 dynamics. Nonetheless, the low DO concentration observed in the surveyed rivers ( Figure 8) suggested that photosynthesis is not the primary control of the seasonal variation of pCO2. 465

Figure 8
The relationship between ΔCO2 and ΔO2. Points greater than zero are oversaturated, and less than zero are undersaturated. Points above the 1:1 line indicate the existence of additional carbon sources, apart from in-stream metabolic processes.

Environmental Control of k600 variation 470
Environmental factors, including wind speed and hydrological variables, could affect the gas exchange at the water-air interface and were typically used to explain the variance in k600 (Alin et al., 2011;Raymond et al., 2012). Flow velocity generally determine the k600 in rivers, while wind speed becomes a more important factor in controlling the k600 in large rivers, reservoirs and estuary (Gué rin et al., 2007;Rasera et al., 2013;Amaral et al., 2020). In our surveyed rivers, k600 displayed a significant linear 475 correlation (Pearson correlation, p < 0.001) with the flow velocity. Our k600 model (Figure 8) base on 188 field measurement data is similar to that developed by Alin et al. (2011) (k600 = 13.82 + 0.35v).
However, in our studied rivers, no significant correlation (Pearson correlation， p > 0.05) was found between wind speed and k600 regardless of stream size. This could be explained by the lower wind speed (Table 2, 0.68 ± 0.66 m s −1 and 1.09 ± 1.06 m s −1 for small and large rivers, respectively) (Gué rin et al., 480 2007). As the wind speed decreases, the impact of flow velocity on k600 will increase considerably (Borges et al., 2004). Therefore, the accuracy of k600 estimation based on wind speed in nearby regions should be examined using measurement data (Yao et al., 2007;Li et al., 2018). The temporal heterogeneities of k600 between small and large rivers reveal the differences in flow regime. k600 in small rivers are significantly (independent sample t test, p < 0.001) higher than in large rivers, which could be 485 explained by higher flow velocity in small rivers due to a higher gradient. Meanwhile, significantly higher k600 (independent sample t test, p < 0.05) was also observed in the wet season compared to with the dry season, which is the result of increasing flow velocity and turbulence due to plentiful monsooninduced precipitation during wet season (Gué rin et al., 2007;Alin et al., 2011;Ho et al., 2018).  Exceptionally high k600 values were observed in the surveyed rivers ( Figure 89). The highest k600 in large and small rivers were 41.83 and 79.97 m d −1 , which were 5-fold and 3-fold larger than calculated k600, respectively. This is the result of the exponential increase in k600 due to extreme flood events. Generally, flood events associated with heavy rainfall during the wet season can increase flow velocity and turbulence at the water-air interface (Almeida et al., 2017;Geeraert et al., 2017), leading to substantially 500 higher k600. Yet, neither our model nor the one from Alin et al. (2011) was suitable for the estimation of k600 during extreme flood events because the calculated k600 could deviate far from the measured k600 when they occurred. Therefore, the extent to which flood events affect k600 and riverine CO2 emission is still uncertain and warrant continued research (Drake et al., 2018).

A Comparison of CO2 Emissions to Other Rivers 505
The mean CO2 fluxes of 225.2 mmol m −2 d −1 in DJRB is comparable to those observed in tropical and subtropical rivers in the Americas, Africa, and Southeast Asia (Table 3). Although the magnitude of the CO2 evasion of these river basins is similar, the seasonal variations and drivers behind them could differ.
The higher CO2 emission in the Dongjiang Basin was observed in the wet season compared to the dry season, and this seasonal pattern is similar to that observed in the Xijiang and Daning rivers (Yao et al., 510 2007; but different from the one from Jinshui River in the upper reaches of the Yangtze River, where pCO2 is high in winter and low in summer , even though all four rivers are in the East Asia Monsoon climate region. The difference in seasonal pattern can be explained by the drivers of pCO2 variability as the seasonal variation of riverine pCO2 is the likely resulting fromresult of the increase changes of external CO2 inputcarbon input, internal production of CO2 (Yao et al., 2007), 515 and the dilution effect caused by precipitation (Johnson et al., 2007). For rivers where pCO2 is lower in summer than in winter, the dilution effect overrides the effect of increased carbon inputs and internal CO2 production . In contrast, for rivers like the Dongjiang river, although the dilution effect remains, increased CO2 input and metabolism are more significant factors in controlling pCO2, thus leading to higher summer pCO2. In addition, the controlling processes of the Dongjiang River may 520 be different even when compared to with rivers with similar seasonal variations in the same climatic zone. For instance, DO in the Xijiang river was supersaturated, indicating that photosynthetic activities in the water body mainly reduce the CO2 concentration in the rivers (Yao et al., 2007). Therefore, other carbon sources like soil respiration and carbonate weathering should be responsible for high pCO2 in summer (Zhang et al., 2019). In contrast, low DO value and a negative correlation between DO and pCO2 525 have been observed in the Dongjiang River, indicating that photosynthesis is relatively weak compared with the respiration in the water body , and the latter one is an essential source of riverine CO2 (Stets et al., 2017) and results resulting in higher pCO2 in summer.

530
The CO2 fluxes in small rivers are similar to that in large rivers, which is contradictory to the finding in previous studies that CO2 effluxes should be higher in small rivers compared tthan ino large rivers due to the input of CO2-rich groundwater (Duvert et al., 2018). The depletion and diffusion effect may be responsible for the discrepancy (Johnson et al., 2007;Dinsmore et al., 2013). In the Dongjiang River 535 Basin, groundwater could be easily diluted due to ample monsoon-induced precipitation, preventing it from supplying the small rivers with high concentrations of carbon dioxide. However, we recognize that the impact of groundwater on pCO2 in small rivers may be overlooked in our sampling process since the CO2 carried by groundwater can emit into the atmosphere within a very short distance (Duvert et al., 2018). In view of the above, it is recommended that further studies targeting the release of groundwater 540 CO2 to the atmosphere be carried out in the future.

Conclusion
Studying CO2 emissions from subtropical rivers is an essential step toward more accurate estimates of global CO2 evasion from river systems. By deploying floating chambers, seasonal changes in riverine pCO2 and CO2 evasion in the Dongjiang river catchment were investigated. Spatial and temporal patterns 545 of pCO2 were mainly affected by terrestrial carbon inputs and in-stream metabolism, both of which varied due to differential catchment settings, land cover, and hydrological conditions.Lateral soil CO2 input and dilution effect caused by precipitation played critical roles in controlling riverine pCO2 in small rivers, while the decomposition of allochthonous organic carbon is responsible for pCO2 changes in large rivers as suggested by the ΔCO2: ΔO2 stoichiometry line. k600 was higher in small rivers than large rivers and 550 higher during the wet season than the dry season, both of which can be explained by the observed significant correlation between k600 and the flow velocity. In contrast to previous studies, similar CO2 fluxes were observed among small and large rivers in the DJRB. It is suggested that the absence of commonly observed higher CO2 fluxes in small rivers could be associated with the depletion effect caused by abundant and persistent precipitation in this subtropical monsoon catchment. There is no doubt 555 that the spatial and temporal variation of CO2 evasion in the DJRB reflected the complexity and diversity of controlling factors. As a step towards a more accurate estimate of the carbon budget in the catchment, comprehensive and systematic measurements of CO2 evasion covering a broad range of stream sizes and seasons are of paramount importance.
Author contributions. BL and LR conceived the study. BL, MT, CC, XY, and LR carried out the fieldwork. BL, MT, and KS designed and performed the laboratory analysis. BL composed the manuscript with contributions from all authors. 565 Competing interests. The authors declare that they have no conflict of interest.