Stream respiration exceeds CO2 evasion in a low‐energy, oligotrophic tropical stream

Carbon dioxide (CO2) can be either imported to streams through groundwater and subsurface inputs of soil‐respired CO2 or produced internally through stream metabolism. The contribution of each source to the CO2 evasion flux from streams is not well quantified, especially in the tropics, an underrepresented region in carbon (C) cycling studies. We used high‐frequency measurements of dissolved O2 and CO2 concentrations to estimate the potential contribution of stream metabolism to the CO2 evasion flux in a tropical lowland headwater stream. We found that the stream was heterotrophic all year round, with net ecosystem productivity (NEP) values ranging from 0.84 to 4.06 g C m−2 d−1 (median 1.29 g C m−2 d−1; here we expressed gross primary productivity (GPP) as a negative flux and ecosystem respiration (ER) as a positive flux). Positive NEP values were the result of a relatively low and stable GPP through the seasons, compared to a higher and more variable ER favored by the high temperatures and organic matter availability, particularly during the wet season. The CO2 evasion flux was relatively low due to low turbulence (median: 1.09 g C m−2 d−1). As a result, daily NEP rates exceeded the CO2 evasion flux with a potential contribution of 129% (median; 120–175% interquartile range), despite the strong seasonal changes in flow regime and landscape connectivity. The CO2 excess was likely transported downstream, where it was ultimately emitted to the atmosphere. Our results highlight the overwhelming importance of ER to the C cycle of low‐energy, oligotrophic tropical streams.

Streams receive substantial organic and inorganic inputs of terrestrial carbon (C) that can either be transported downstream, released to the atmosphere as carbon dioxide (CO 2 ; often referred to as "CO 2 evasion"), or stored in sediment (Hope et al. 1994;Cole et al. 2007). Evasion often represents the largest of these fluxes; the latest estimates indicate that globally, between 48% (Kirschbaum et al. 2019) and 76% (Drake et al. 2018) of the influx of terrestrial C into inland waters is released back to the atmosphere. The evading CO 2 can originate from two major sources: (1) external inputs of terrestrially derived inorganic C via soil water and/or groundwater inflows and (2) internal production of CO 2 that is a byproduct of stream metabolism. The partitioning between external and internal C sources and the potential contribution of each of these sources to the CO 2 evading flux is not well quantified. Improved understanding of all terms of the C balance is crucial to our understanding of C cycling in freshwater ecosystems.
Stream metabolism represents the balance between gross primary productivity (GPP) and ecosystem respiration (ER) that occur in the water column and streambed. GPP refers to the fixation of CO 2 by primary producers and its conversion to organic C, while ER refers to the aerobic oxidation of organic C to CO 2 , predominantly by microbial activity. The rates of GPP and ER in a stream result from the interaction between biotic and abiotic factors, such as light availability, water temperature, quantity and quality of organic C, and nutrient concentrations (Fisher 1977;Bott 1983;Demars et al. 2011). The stream net ecosystem productivity (NEP) is the net balance between these two processes, and positive NEP (i.e., ER < GPP) indicates that the stream is assimilating more CO 2 than it is producing, and the system is autotrophic. Negative NEP (i.e., ER > GPP) indicates that the stream is heterotrophic.
Autotrophic production is predicted to increase from headwater streams to large rivers (Vannote et al. 1980). As stream size and width increase, GPP increases as more light reaches the water column, often resulting in higher NEP in larger rivers (Hotchkiss et al. 2015). In contrast, both turbulence and the lateral inputs of CO 2 and organic C tend to decrease along the river continuum (Marx et al. 2017;Wallin et al. 2018). Taken together, these spatial patterns will favor lower CO 2 evasion fluxes in large streams and rivers, albeit with a higher contribution of NEP to CO 2 evasion (Hotchkiss et al. 2015).
The partial pressure of CO 2 ( pCO 2 ) in water, along with turbulence at the water-atmosphere interface, are key drivers of CO 2 evasion from aquatic systems (Elsinger and Moore 1983;Wanninkhof et al. 1990;Rocher-Ros et al. 2019). Turbulence determines where the evasion takes place, while pCO 2 regulates the magnitude of CO 2 evasion (Rocher-Ros et al. 2019). In contrast to larger rivers, which have a more laminar flow despite higher discharge, small streams show favorable characteristics for high evasion rates as they are typically shallow and steep, with high streambed roughness that increases their turbulence (Kokic et al. 2018). Global syntheses have highlighted the potential of headwater streams to contribute disproportionately to CO 2 evasion (Raymond et al. 2013;Marx et al. 2017). Evasion is also strongly influenced by changes in flow regimes, with streams and rivers typically releasing more CO 2 at higher flows (Liu and Raymond 2018). In the wet-dry tropics, monsoonal events generate the bulk of CO 2 evasion, due to both higher turbulence and higher landscape connectivity (Geeraert et al. 2017;Duvert et al. 2020).
Additional to size-based differences between streams and rivers, the magnitude of CO 2 evasion and stream metabolism changes latitudinally. CO 2 evasion from tropical streams may be four to five times higher than evasion from temperate streams (Raymond et al. 2013;Lauerwald et al. 2015) and respiration rates higher than the 75 th percentile from their temperate counterparts (Marzolf and Ard on 2021). This difference in metabolic rates between temperate and tropical systems is likely a consequence of the higher productivity of terrestrial vegetation driving high soil respiration and high organic matter inputs to streams, where high temperatures enhance microbial activity. Therefore, we can expect that stream metabolism will have a particularly high contribution to CO 2 evasion in tropical regions, especially in small, predominantly heterotrophic streams. However, most tropical studies have occurred in humid areas such as the Amazon, highlighting the need for more data in seasonal tropical biomes such as the wet-dry tropical savannas.
Among the few studies combining assessments of stream metabolism and CO 2 evasion, most have been conducted at high latitudes (Hotchkiss et al. 2015;Wang et al. 2021;Bernal et al. 2022). These studies showed that most streams are heterotrophic, with potential contributions of NEP to CO 2 evasion ranging from 0% to > 100%. To our knowledge, there is only a handful of studies conducted in the tropics that compared stream metabolism and CO 2 evasion fluxes. In the Australian wet-dry tropics, Duvert et al. (2019) reported a potential contribution of NEP to CO 2 evasion of between 8% and 31% (based on NEP values reported by Townsend et al. 2011). Ellis et al. (2012) calculated that between < 1% (in headwater streams) and 100% (in large rivers) of the CO 2 evasion flux from the Amazon basin can be sourced from internal respiration. These results are in line with findings from Abril et al. (2014) who concluded that the major source of CO 2 evasion in the central Amazon River-floodplain system is root and microbial respiration from wetlands. Marzolf et al. (2022) found that 16% of the CO 2 evasion from a small, steep stream in Costa Rica can be sourced from stream metabolism, similar in magnitude to groundwater and upstream inputs of CO 2 . In tropical Africa, estimations by Borges et al. (2015), Borges et al. (2019), Moustapha et al. (2022), and Teodoru et al. (2015) indicate that stream metabolism contributes little CO 2 to the evasion flux, with contributions ranging from 0% to 33% in individual rivers, and 14% at a continental scale. Overall, these studies illustrate the huge variability in processes across river orders and morphologies and suggest that more work is needed to better understand and quantify the key controls on these fluxes for different tropical climates and landscapes.
Our study focuses on quantifying the potential contribution of stream metabolism to CO 2 evasion in a low-energy (i.e., low topographic gradient) tropical headwater stream in northern Australia. We monitored the O 2 and CO 2 concentrations continuously at the stream outlet for over 2 yr to assess the temporal variability of this contribution. We hypothesized (H1) that the stream is heterotrophic (i.e., higher ER than GPP) regardless of the season, because its oligotrophic nature, high organic matter loading and high year-round temperatures are likely to favor ER over GPP. We further hypothesized (H2) that this internal production of CO 2 contributes significantly to the evasion flux, given the high ER rates, the comparatively low groundwater CO 2 inputs and the relatively low CO 2 evasion rates due to the low-energy nature of the stream. Finally, we hypothesized (H3) that the contribution of NEP to CO 2 evasion is higher during the dry season than during the wet season, because of lower rates of CO 2 evasion at low flow and lower landscape connectivity limiting the external inputs of CO 2 .

Study site
The study was conducted in Manton Creek, a small headwater stream located in the wet-dry tropics of northern  Beck et al. (2018)) with 90% of annual rainfall occurring during the wet season (Cook and Heerdegen 2001). For the observation period (2018-2021), a weather station located 22 km southwest of the monitoring station (http://www.bom. gov.au/climate/, Batchelor weather station, Batchelor Airport, Sta. 014272) reported a mean annual precipitation of 1167 mm (range: 920-1404 mm), while the long-term mean annual precipitation is 1538 mm (period 2002-2021). Manton Creek has a very low topography, with an average stream slope of 0.2%, and is made of a series of connected pools and runs. The stream supports a thick vine forest corridor that limits light exposure along much of its length. The mean daily runoff for 2002-2021 was 0.3 m 3 s À1 , ranging from 0 to 2.4 m 3 s À1 (Northern Territory Government, gauging station G8170075, http://www.water.nt.gov). During the study period, the catchment experienced drier conditions than average, causing the creek to dry completely for at least 2 months at the end of the dry season in 2019 and 2020. The stream at its outlet is a third-order stream during the wet season and a first-order stream during the dry season.

Field measurements
Dissolved oxygen (DO; miniDOT, PME), pCO 2 (eosGP, Eosense, Canada) and barometric pressure (CS100, Campbell Scientific) sensors were installed at the outlet of the catchment where discharge measurements were taken. The pCO 2 and pressure sensors were connected to a Campbell Scientific logger (CR1000) and measurements were recorded every 10 min between April 2018 and March 2021. The DO sensor was calibrated in June 2020 and March 2021 using a zero solution (HI7040L, Hanna Instruments), and the pCO 2 sensors were calibrated before deployment using four gas standards (0, 574, 5000, 10,000 ppm; CALGAZ).
The monitoring station was visited for maintenance every 3-4 weeks during the wet season and every 2-3 months during the dry season. During these visits, the DO sensor was cleaned and its battery was changed, while the pCO 2 sensor was tested for failure and replaced if needed. There was no difference in the pre-and post-cleaning DO readings, a sign that biofouling had little effect. DO values were obtained for a total of 748 d and pCO 2 for 579 d.
Water samples were taken at the monitoring station during each visit from November 2017 until March 2021 for dissolved organic carbon (DOC) concentrations. Samples were collected in pre-acidified 40-mL amber glass vials and kept refrigerated for later analysis. In addition, historic data of nitrogen as NO x -N (i.e., the sum of nitrate and nitrite), total phosphorous, and suspended sediments were extracted from the Northern Territory Government database for the period of 1966-2011. The water quality data obtained consist of discrete measurements unequally distributed through the years, with two gaps from 1994 to 2005, and from 2007 to 2010.
Discharge and water-level data were obtained from continuous (10-min frequency) stream height measurements at the gauging station (G8170075). A linear relationship between stream width and discharge, and a power relationship between flow velocity and discharge (Supporting Information Fig. S3), were derived from measurements from 2016 to 2019 provided by the Northern Territory Government. These relationships were then used to estimate the flow velocity and stream width throughout the study period.

Assessment of stream reach homogeneity
Our metabolism estimates (see "Stream Metabolism Estimation" section) assume that the stream is relatively homogenous over a certain length (Chapra and Di Toro 1991;Hall and Hotchkiss 2017). To assess stream reach homogeneity, we measured the geometry of several cross sections along the reach immediately upstream of our DO measurement point. We found a $ 160-m long reach relatively homogeneous in terms of both average depth, width, and canopy cover (Supporting Information Fig. S2a). Upstream of the 160-m reach lies a $ 15-m long pool with slightly lower flow velocities. The small pool (depth < 1 m) receives water from a $ 300-m reach with again similar morphology, slope and canopy cover to the downstream 160-m reach. Although the occurrence of the small pool between the two stream reaches creates some heterogeneity, key drivers of stream metabolism such as temperature, light availability, and slope (Supporting Information Fig. S2b) were consistent throughout the 475-m reach-pool-reach sequence.
Assuming an average flow velocity of 0.1 m s À1 and an average depth-normalized reaeration coefficient K of 24 d À1 , the distance required to turn over 75% of O 2 ($ 1.4 v/K; Chapra and Di Toro 1991) would equal 504 m. This turnover distance is comparable to the length of our reach-pool-reach sequence. Given these considerations, and while we acknowledge the potential uncertainties resulting from the occurrence of the small pool, we can reasonably conclude that the stream homogeneity assumption is met.

Data preprocessing
DO concentrations were corrected for sensor drift assuming a linear change between sensor calibrations. The pCO 2 series were corrected using a calibration equation obtained in the laboratory before deployment. Values were later corrected for hydrostatic pressure following the manufacturer's procedure and were transformed to CO 2 aqueous concentrations according to Henry's law as per Weiss (1974), assuming a salinity of 0 ppt. Incomplete days, periods with no flow or when the sensors malfunctioned were discarded from both datasets.
The data were divided into four seasons based on rainfall and flow regime thresholds. We considered early wet season storms ("early storms" in the following) as the period from the first runoff events in December until the first monsoonal rains, usually around mid-January to late January. The "wet season" included the large, monsoon-driven events and flooding from January or February until April; the "recession period" occurred from May to June; and the "dry season" from July until September or October. To estimate the start of the dry season, we used a baseflow separation algorithm (EcoHydRology package in R) and a threshold of 98% baseflow contribution.
The relationship between different variables was evaluated using the Spearman correlation coefficient. A nonparametric method was preferred given that not all variables were normally distributed. All data were analyzed using R (R Core Team, 2017; version 3.6.0).

Gas transfer velocity and CO 2 evasion estimates
The gas transfer velocity (k) was estimated using the seven empirical equations from Raymond et al. (2012). We obtained seven 10-min series of k estimates based on 10-min series of flow velocity, stream width, water depth and discharge, and the stream slope obtained through GIS. The reliability of these empirically-derived estimates was assessed against k values derived from discrete measurements undertaken using a custom-made floating chamber connected to a LI-7810 gas analyzer (LI-COR; Supporting Information Fig. S4). Equation 6 (Raymond et al. 2012) was the model that yielded the lowest root mean squared error between modeled and chambermeasured k estimates (see Supporting Information Table S1). This model was selected as the most accurate prediction of k for the study reach. To account for uncertainties in our k estimates, we used 10,000 Monte-Carlo simulations for each 10-min time step, with randomized values of model parameters as per uncertainties reported in Raymond et al. (2012), and 20% errors on flow velocity and stream slope. Modeled daily k data with associated uncertainties are presented in Supporting Information Fig. S5. Specific gas transfer velocity for CO 2 (k CO2 ) and O 2 (k O2 ) were calculated based on Wanninkhof (1992) using the Schmidt coefficients provided by Raymond et al. (2012).
A time series of the evading CO 2 flux (F CO2 ) was calculated using the estimated k CO2 and pCO 2 values measured in the stream according to Fick's law of gas diffusion (Raymond et al. 2012): where CO 2wat (μmol L À1 ) is the concentration of CO 2 in the water, CO 2equil (μmol L À1 ) is the concentration of CO 2 in equilibrium with the atmosphere, and k (m d À1 ) is the gas transfer velocity. We assumed an atmospheric CO 2 mixing ratio of 410 ppm (range 408-411 ppm over the observation period, Cape Grim CO 2 station).

Stream metabolism estimation
Conventionally, ER is expressed as a negative flux (Hall and Hotchkiss 2017). However, from a CO 2 perspective, ER adds to the stream CO 2 pool, while GPP removes CO 2 from it. To emphasize this notion, and following others before us (Bernal et al. 2022), we chose to express ER as a positive flux and GPP as a negative flux. As a result, we obtained positive NEP values when ER exceeded GPP and negative NEP values when GPP exceeded ER.
Daily ER and GPP rates were estimated using a Bayesian model (BASE; Grace et al. 2015) based on the one-station technique (Odum 1956). BASE can successfully model stream metabolism rates in a stream reach where the three underlying assumptions of the one-station technique are met. In addition to (1) the stream reach homogeneity assumption (discussed in "Assessment of Stream Reach Homogeneity" section), (2) the discharge must be constant throughout the day, and (3) there must be no other physical processes, such as water inflows, that can affect the DO diel cycle. To fulfill the last two assumptions, we (2) focused our analysis on days for which the discharge was relatively constant (daily change < j5j%) and (3)  The model was set to estimate four parameters: the coefficient of light saturation, temperature dependence, ER and GPP. The metabolic rates (ER and GPP) were obtained from hourly inputs of DO, DO saturation (calculated according to Garcia and Gordon 1992), water temperature, photosynthetically active radiation and barometric pressure. By default, BASE uses noninformative priors for both ER and GPP, where the priors follow a truncated normal (TN) distribution such as ER $ TN (0, 0.25, 0) and GPP $ TN (0, 0.25, 0). Daily reaeration coefficients were given as a fixed prior, based on our daily k estimates from Eq. 6 by Raymond et al. (2012) (see "Gas transfer velocity and CO 2 evasion estimates" section) and daily water depth series. To avoid the issue of using potentially different k for our evasion vs metabolism estimates, we forced the model to adjust k within AE 0.001 d À1 from its prior. The model reliability was assessed according to Grace et al. (2015). Briefly, we discarded all days for which modeled GPP was negative (i.e., positive according to our sign convention), modeled ER was positive (i.e., negative according to our sign convention), as well as days with poor convergence (Ȓ > 1.1) or poor model fit (posterior predictive check PPfit < 0.1 or > 0.9; effective number of parameters pD < 0).
The modeled GPP and ER are expressed as O 2 fluxes, which we converted into CO 2 fluxes using a range of photosynthetic (PQ) and respiratory (RQ) quotients. Because both quotients can be highly variable (Bott 1983;Bott 2006;Berggren et al. 2012), we considered the uncertainties on GPP and ER estimates due to variations in PQ and RQ between 0.8 and 1.2. We propagated these uncertainties, as well as the uncertainties Solano et al.
Stream respiration exceeds CO 2 evasion on GPP and ER from BASE to the daily NEP estimates using 10,000 Monte-Carlo simulations for each daily time-step. We then extracted the daily interquartile range as a measure of uncertainty (see Supporting Information Fig. S7). Daily metabolism rates were estimated based on 322 d of data between December 2019 and March 2021, distributed unequally across the seasons (early storms 36 d, wet season 83 d, recession period 88 d, dry season 115 d; Fig. 2). A total of 356 d were excluded from the calculations, 89% of them due to high discharge changes during the day or no flow, 9% due to sensor cleaning and maintenance, and 2% due to sensor failure. In addition, there were 60 daily occurrences for which the model did not converge and/or showed a poor fit. This is most likely a result of extreme low GPP rates in the stream during those days.

Potential contribution of NEP to CO 2 evasion
The potential contribution of NEP to the CO 2 evading flux was calculated as a percentage for days when both NEP and CO 2 evasion were estimated (154 d across the entire monitoring period). The uncertainties on NEP and F CO2 were propagated to the percentage estimates using 10,000 Monte-Carlo simulations for each daily time step. Values equal to or higher than 100% indicate that the CO 2 evading from the stream has the potential to be sourced solely from NEP.

O 2 and CO 2 departure concentration
To gain further insight into the drivers of stream metabolism and CO 2 evasion, we also assessed the departure concentrations of CO 2 and O 2 from atmospheric equilibrium on a daily basis (Vachon et al. 2020). Departure concentrations dominated by stream metabolism will display a 1 : À1 relationship, whereas chemical (e.g., carbonate dissolution and precipitation, photo-oxidation of organic matter), physical (e.g., water temperature change), and hydrological processes (e.g., groundwater inflows) will all create an offset from the 1 : À1 line. Given that the O 2 -CO 2 relationship is symmetric, ellipses grouping the departure concentrations during different seasons can give information about the temporal dynamics in the stream (Vachon et al. 2020).
The CO 2 atmospheric partial pressure was assumed as 410 ppm (see above) while the O 2 atmospheric equilibrium was calculated for each day according to Garcia and Gordon (1992). Departure data were plotted against each other as per  Vachon et al. (2020) and seasonal cloud metrics such as centroid coordinates, shifts from the 1 : À1 line and the inverse of the slopes from a linear regression fitted to each cloud were obtained using the code provided by Vachon et al. (2020). Visualization plots were created using the function stat_ellipse from the ggplot2 package in R Studio.

Daily CO 2 and O 2 concentrations
Daily CO 2 and O 2 concentrations were inversely correlated through the hydrological year (Fig. 1). Higher flow conditions during the wet season were associated with low CO 2 values and high O 2 values, while the lower discharges during the late dry season and early storms were related to higher CO 2 values and lower O 2 values. Overall, CO 2 concentration had a negative correlation with discharge (Spearman correlation: À0.48, p < 0.001) and was positively correlated to temperature (Spearman correlation: 0.25, p < 0.001). DO, on the other hand, showed a negative correlation with temperature (Spearman correlation: À0.34, p < 0.001) and a positive correlation with discharge (Spearman correlation: 0.62, p < 0.001).

Water quality
Total phosphorous and nitrogen (NO x -N) concentrations, DOC and suspended sediments in Manton Creek were low throughout the year when compared to other streams in tropical Australia (Table 1; Lintern et al. 2021). Apart from nitrogen, all the parameters exhibited the same seasonal behavior, with concentrations peaking during the early storms, then decreasing during the wet season and the recession period and increasing slightly during the dry season. Nitrogen, on the other hand, reached its maximum during the recession period and its minimum during the dry season.

Stream metabolism estimates
Daily ER estimates were 10-100 times higher than daily GPP estimates. Daily ER rates ranged between 0.85 and 4.29 g C m À2 d À1 (median 1.42 g C m À2 d À1 ), while daily GPP ranged from < À0.01 to À0.94 g C m À2 d À1 (median À0.11 g C m À2 d À1 ). The resulting daily NEP values were always positive and ranged from 0.84 to 4.06 g C m À2 d À1 , with a median of 1.29 g C m À2 d À1 . NEP mirrored the seasonal variability of ER, with its highest values during the wet season and lowest values during flow recession and the dry season (Fig. 2). These results were affected little by variations in the PQ and RQ (Supporting Information Fig. S7).

Solano et al. Stream respiration exceeds CO 2 evasion
Contribution of NEP to CO 2 evasion CO 2 evasion (F CO2 ) was estimated for 572 d between 2018 and 2021; F CO2 ranged from 0 to 6.70 g C m À2 d À1 over the observation period (median 1.05 g C m À2 d À1 ; Supporting Information Fig. S5). Fluxes varied seasonally, the highest evasion rates occurred during the wet season and recession periods, which coincided with low pCO 2 and high discharge. Conversely, the lowest evasion rates occurred during the late dry season, when streamflow was at its lowest (Supporting Information Fig. S8; Fig. 1a).
The contribution of NEP to CO 2 evasion was estimated using a subset of 154 d from the study period, when both estimates were available ( Fig. 3; Supporting Information Figs. S7, S8, S9). This subset of CO 2 evasion and NEP estimates followed the same seasonal pattern as the whole dataset. However, NEP estimates were calculated only under stable streamflow (< j5j% daily change), so that high discharge events (i.e., highest evasion rates) were not considered in this comparative exercise. CO 2 evasion for this subset of data ranged from 0.43 to 4.77 g C m À2 d À1 (median 1.01 g C m À2 d À1 ); NEP, on the other hand, ranged from 0.99 to 4.06 g C m À2 d À1 with a median of 1.37 g C m À2 d À1 .
The median proportion of the CO 2 evasion flux potentially contributed by NEP was 129% across the study period (120%-175% IQR). This percentage was almost always > 100% and exhibited substantial variations throughout the year, with values between 160% and 171% (IQR, median 169%) during the early storms, 117% and 127% (IQR, median 122%) during the wet season, 128% and 175% (IQR, median 135%) during the recession period, and 106% and 195% (IQR, median 185%) during the dry season. The drier and cooler periods (recession and dry season) showed the highest variability in the daily percentages of NEP to F CO2 (difference of more than 125% between minimum and maximum daily data), while during the wet season and early storms, these percentages were more stable (difference of less than 60% between minimum and maximum daily data).
Relationship between dissolved CO 2 , CO 2 evasion, and discharge Dissolved CO 2 was negatively correlated with discharge (Spearman's coefficient À0.5, p < 0.001; Fig. 4a). This was likely a result of dilution of the stream CO 2 pool by rainfall and runoff during high flow events. Evading CO 2 fluxes, on the other hand, increased with increasing discharge (Spearman's coefficient 0.93, p < 0.001; Fig. 4b), a process that can further decrease the CO 2 concentration in the water column. The increase in F CO2 with discharge can be attributed to the increase in flow velocity and turbulence, which caused an increase in k (Supporting Information Figs. S6, S8). However, the relatively low slope of the relationship in Fig. 4b suggests that large increases in discharge resulted in modest increases in F CO2 . This attenuation effect was likely related to (1) the channel morphology, with the stream overflowing its banks at very high flow causing limited changes in flow velocity and turbulence; and (2) the concomitant decrease in the stream CO 2 pool. Yet, the modest increase in k always outweighed the decrease in stream CO 2 .

CO 2 and O 2 departures
During the study period, Manton Creek was always undersaturated in O 2 and oversaturated in CO 2 relative to the atmosphere. When grouped by season, the data deviated from the 1 : À1 line (Fig. 5), with mostly negative offsets in the recession and dry season (À5 and À23) and positive offsets in the early storms and wet season (< 1 and 32), indicating that different biogeochemical and physical processes occurred in different seasons. The inverse of the slope ranged between 0 and 2 for three seasons and was much higher for the recession period (inverse of the slope: 47). However, the recession period captured two distinct years, each with a slope $ 1.5.

Discussion
We estimated NEP and CO 2 evasion rates at a daily timestep and for a 2-yr period in a lowland headwater stream of  Year-round dominance of respiration over primary productivity Manton Creek was heterotrophic throughout the year. The consistently low GPP rates led to positive and particularly high NEP rates all year round, which supports our first hypothesis (H1). This result is in line with previous work by Townsend et al. (2011) in the Daly River catchment, 100 km south of Manton Creek, who found that the Daly River and two of its tributaries (Hayes Creek and Douglas River, 5 th to 7 th orders) were characterized by net heterotrophy all year round. There seems to be large cross-system variability in the region though, as Garcia et al. (2015) found that the nearby Edith and Fergusson Rivers (4 th to 5 th orders) were slightly autotrophic. We believe much of this variability can be explained by differences in GPP rates. GPP in Manton Creek (median À0.11 g C m À2 d À1 ) was two to seven times lower than that measured in other northern Australian rivers (Webster et al. 2005;Townsend et al. 2011;Garcia et al. 2015), while our estimates of stream ER (median 1.42 g C m À2 d À1 ) were in the same order of magnitude as those measured in previous studies across the region. For instance, the Daly River and tributaries had ER rates ranging between 1.1 and 2.0 g C m À2 d À1 (Townsend et al. 2011), while the Edith and Fergusson Rivers had slightly lower benthic ER rates of around 0.4 g C m À2 d À1 (Garcia et al. 2015). When compared to other tropical streams globally, both our ER and GPP estimates were in the lower range of expected values (Marzolf and Ard on 2021).
Stream respiration varied throughout the year, with slightly higher ER during the wet season and slightly lower ER during the dry season. This seasonal pattern suggests that ER was at least partly driven by changes in flow regime and stream physicochemical conditions. During the wet season, higher water temperatures (Fig. 1a) were likely to stimulate microbial activity within the water column and hyporheic zone, leading to high ER rates. Bernot et al. (2010) and Yvon-Durocher et al. (2012) have shown the dependence of aquatic respiration to increases in temperature. In addition, the increased connectivity between seasonal wetlands and the stream network during the wet season may have resulted in inputs of fresh organic matter to the stream (Table 1). In a nearby river, Birkel et al. (2020) found that seasonal wetlands were a major source of DOC, with up to 90% of the annual DOC load contributed through wetland drainage. The availability of highly labile DOC in the wet season, combined with high temperatures, may have led to the high ER rates observed during that period.
Low flow conditions during the dry season likely led to hypoxic or even anoxic conditions (Fig. 1), an effect that is commonly observed in streams at very low flow (Blaszczak et al. 2023). Oxygen depletion can affect the nature and rate of microbial metabolism (Ponnamperuma 1972;Hamilton et al. 1995;Pardo and García 2016) and cause changes in the structure of decomposer communities, often reducing ER (Medeiros et al. 2009). However, it is important to note that

Solano et al.
Stream respiration exceeds CO 2 evasion low O 2 conditions can lead to an underestimation of ER when calculated using solely O 2 -based methods, due to unaccounted anaerobic respiration. The lower DOC concentrations in the stream during the dry season (Fig. 1), likely due to the disconnection between the stream and wetland-derived C stocks, may have also contributed to the slight decrease in ER rates (Table 1). Studies have shown that stream ER can be limited by the (un)availability of labile DOC (e.g., Burrows et al. 2017), and decreases in labile DOC have been reported in rivers of the region during the recession period (Webster et al. 2005).
In contrast to the high ER rates, we found that GPP was low all year round. We believe that the low GPP rates in Manton Creek are due to its oligotrophic nature and low light environment. First, closed canopies drive low GPP in tropical streams (Marzolf and Ard on 2021), and the vine forest thicket present along the stream corridor tended to inhibit the penetration of solar radiation into Manton Creek. Second, soils of the Australian wet-dry tropics are highly weathered and nutrient-poor (Prescott 1941;Warfe et al. 2011). Limited nutrient availability (Table 1) affects the growth of primary producer biomass and in turn reduces the GPP potential, as was observed in the Daly River during the dry season (Townsend et al. 2011). There was a slight increase in GPP during early storms, which may have been caused by an increase in phosphorous availability (Table 1), potentially linked to the hydrological activation of wetlands and their connection to the stream network. Conversely, wetter conditions led to more turbid and faster flows that tended to limit GPP even further. A similar pattern was reported for the Daly River, with a substantial decrease in photosynthesis at high flow (Webster et al. 2005). More generally, the low median annual GPP found in Manton Creek is consistent with the findings of Bernhardt et al. (2022) who showed that rivers with more variable flow regimes tend to have lower GPP than more stable rivers.

Internal metabolism exceeds CO 2 evasion
We found that CO 2 evasion was transfer-limited (Rocher-Ros et al. 2019) all year round. At the event scale, increased flow and turbulence caused a dilution of CO 2 in stream water, but there was a concomitant increase in evasion rates (Fig. 4). Transfer limitation was most prominent during the dry season, when the stream CO 2 pool was at its highest ( Fig. 1) and CO 2 evasion at its lowest (Fig. 3). This notion of transfer limitation echoes the findings from other low-energy systems such as peatland draining streams (Schneider et al. 2020;Taillardat et al. 2022) and contrasts with high-energy, mountainous streams (Crawford et al. 2015;Horgby et al. 2019;Clow et al. 2021), where evasion tends to be limited by C availability. More generally, our results indicate that CO 2 evasion changes as a function of the interplay between dissolved CO 2 availability and flow turbulence. This is in line with previous work that showed the importance of event-based and seasonal changes in streamflow in regulating both the stream CO 2 pool and CO 2 evasion rates (Dinsmore et al. 2013;Liu and Raymond 2018;G omez-Gener et al. 2021).
The contribution of NEP to the CO 2 evasion flux was almost always > 100% (median 129%), a result that supports our second hypothesis (H2). This NEP contribution is the second highest of all values previously reported for streams and rivers < 10,000 km 2 ( Fig. 6; Supporting Information Table S2; overall median 29%). Hotchkiss et al. (2015) showed that the NEP contribution to CO 2 evasion is typically 10%-19% for low-order streams across the United States. Our results also largely exceed previous estimates from other tropical systems ( Fig. 6; Supporting Information Table S2; median for tropical streams and rivers < 10,000 km 2 : 5%). Marzolf et al. (2022) estimated a potential contribution of 16% in a headwater stream in Costa Rica, while Ellis et al. (2012) estimated contributions between 0.1% and 7% in headwater catchments of the Amazon basin. We believe that our considerably higher estimate for Manton Creek is a direct consequence of the local climate and geomorphology of the catchment, where the very low topography (average stream slope 0.2%) may have both limited the CO 2 evasion flux (low turbulence) and ensured long transit times, even at high flow, which, combined with high temperatures and availability of organic matter, was more favorable to stream metabolic activity. These interpretations are in full agreement with those of Carter et al. (2022), who found that the long residence times, limited turbulence, and substantial organic matter inputs of a low-gradient temperate stream led to CO 2 accumulation from microbial respiration and an overwhelming contribution of this CO 2 pool to the evasion flux. Importantly, our results suggest that this excess CO 2 is transported downstream, where it is likely to be ultimately emitted to the atmosphere, a pathway whose significance will increase with high flow conditions. Another likely fate for some of the excess stream CO 2 is the potential for carbonate dissolution within the stream (Stets et al. 2017), as suggested by the high pH values during the dry season (Supporting Information Fig. S10).
We also note that due to differences in methodological approaches, comparisons should be made with caution. The estimates of Ellis et al. (2012) are based on direct measurements taken over a day or less, Marzolf et al. (2022) measured paired O 2 and CO 2 dissolved concentrations hourly for 6 months, while Hotchkiss et al. (2015) used historical data and modeling to obtain a single NEP value for each stream size. More importantly, the CO 2 and NEP estimates in Hotchkiss et al. (2015) came from two distinct groups of streams. In light of our results, we suggest that no conclusive trends on the contribution of NEP to F CO2 should be drawn unless paired observations of both NEP and F CO2 that capture both event-based and seasonal trends are available. Furthermore, due to methodological constraints, estimations of the potential contribution of NEP to F CO2 usually exclude storms events, when evasion is likely to be high and NEP might be low, which could lead to a slight overestimation.
Although the seasonal differences in CO 2 evasion were more pronounced than the seasonal differences in NEP (Fig. 3), NEP rates were always higher than F CO2 , with potential contributions of NEP to F CO2 ranging from 122% during the wet season to up to 184% during the dry season. A comparison between the departure concentrations of CO 2 and O 2 (Fig. 5) helps further discern seasonal differences in the potential sources contributing to the stream CO 2 pool. During the early storms and wet season, the stream CO 2 pool originated predominantly from external CO 2 inputs (Fig. 5), as supported by the positive offsets in the O 2 -CO 2 stoichiometry in those periods (Fig. 5). External CO 2 sources in this case may have comprised soil water and groundwater inputs of terrestrially derived CO 2 (Johnson et al. 2008;Duvert et al. 2018), a pathway that can be facilitated by the high connectivity between wetland soils, riparian areas and the stream network at high flow (Birkel et al. 2020;Moustapha et al. 2022). Given the concomitantly high internal CO 2 production via NEP during the early storms and wet season (Fig. 3), and the comparatively low F CO2 rates, we can reasonably conclude that the stream was in a state of transfer limitation during these high flow periods.
In contrast, the negative O 2 -CO 2 offsets observed during the recession and dry season (Fig. 5) indicate the dominance of internal processes during those periods. We expect these offsets to be the result of carbonate dissolution within the stream, which may have reduced the CO 2 pool due to increased alkalinity (Vachon et al. 2020). Changes in carbonate equilibria were a likely process in the dry season given the almost exclusive contribution of deep groundwater from a dolostone aquifer to streamflow at low flow, as suggested by seasonal changes in electrical conductivity and pH (Supporting Information Fig. S10). The effect of carbonate-rich groundwater inputs on in-stream carbonate buffering and CO 2 evasion has previously been documented in the nearby Howard River (Duvert et al. 2019) and in other streams globally (Ran et al. 2021;Wang et al. 2021). Despite the importance of internal biogeochemical processes during flow recession and the dry season, the contribution of NEP to F CO2 did not increase significantly during those periods (Fig. 3), which does not support our 3 rd hypothesis (H3).
It is also worth noting that our results may not hold in a different geomorphological context. For instance, shallow riffles can maintain high gas exchanges with the atmosphere, even under low flow conditions, due to stream bed roughness, while more open pools are characterized by laminar flow even under wet conditions, thus limiting the gas exchange potential. Further work is required to refine our understanding of *Contribution of ER, the study does not provide GPP estimates. a Small streams. b Large rivers. c Regional scale. RN, River Network. Further information about the data in this figure can be found in the Supporting Information (Table S2).

Conclusions
In summary, we have quantified the seasonal differences in stream metabolism and its contribution to CO 2 evasion in a low-energy headwater stream of the wet-dry tropics. We have shown that metabolic rates are relatively stable throughout the year, with slight changes driven by changes in landscape connectivity and flow conditions. Evasion fluxes are more variable seasonally, but never as high as the CO 2 produced via stream metabolism. Our results suggest that in low-energy, oligotrophic streams of the wet-dry tropics, stream metabolism can be a major contributor to CO 2 evasion. Our work also highlights the importance of high-resolution, long-term, and paired O 2 -CO 2 datasets to assess the impact of seasonality in the biogeochemical cycles of streams and rivers. Understanding these processes is essential for the accuracy of future climate projections and comprehensive C accounting inventories.

Data availability statement
The data used in this paper are available in the Supporting Information.