Partitioning inorganic carbon fluxes from paired O2–CO2 gas measurements in a Neotropical headwater stream, Costa Rica

The role of streams and rivers in the global carbon (C) cycle remains unconstrained, especially in headwater streams where CO2 evasion (FCO2) to the atmosphere is high. Stream C cycling is understudied in the tropics compared to temperate streams, and tropical streams may have among the highest FCO2 due to higher temperatures, continuous organic matter inputs, and high respiration rates both in-stream and in surrounding soils. In this paper, we present paired in-stream O2 and CO2 sensor data from a headwater stream in a lowland rainforest in Costa Rica to explore temporal variability in gas concentrations and ecosystem processes. Further, we estimate groundwater CO2 inputs (GWCO2) from riparian well CO2 measurements. Paired O2–CO2 data reveal stream CO2 supersaturation driven by groundwater CO2 inputs and large in-stream production of CO2. At short time scales, CO2 was diluted during storm events, but increased at longer seasonal scales. Areal fluxes in our study reach show that FCO2 is supported by greater in-stream metabolism compared to GWCO2. Our results underscore the importance of tropical headwater streams as large contributors of carbon dioxide to the atmosphere and show evaded C can be derived from both in-stream and terrestrial sources.


Introduction
Inland waters play an important role in the global carbon (C) cycle and estimates of C fluxes between inland waters, the atmosphere, and terrestrial ecosystems are being revised at a rapid rate (Cole et al. 2007;Raymond et al. 2013;Tank et al. 2018;Drake et al. 2018;Gómez-Gener et al. 2021). Streams, wetlands, and lakes transform, export, and store terrestrial C prior to either delivery downstream to the ocean (Cole et al. 2007) or evasion of CO 2 to the atmosphere (Battin et al. 2009). These processes are particularly important in headwaters streams, which comprise 79% of stream network length (Colvin et al. 2019) and disproportionately contribute to evasion of CO 2 to the atmosphere on an areal basis (Raymond et al. 2013;Borges et al. 2019). However, these findings represent the synthesis from predominantly temperate ecosystems, whereas tropical streams have received less study (Aufdenkampe et al. 2011;Oviedo-Vargas et al. 2015). As a result of higher yearround temperatures, high and continuous organic matter inputs, and drainage from soils with high respiration rates, tropical streams can be hotspots of CO 2 evasion (Borges et al. 2019). In-stream production of CO 2 from mineralization of organic matter can further contribute to high CO 2 fluxes (Richey et al. 2002;Mayorga et al. 2005;Hotchkiss et al. 2015).
Streams are generally CO 2 supersaturated with respect to atmosphere (Wetzel and Likens 2000), reflecting large input fluxes of CO 2 of both terrestrially derived and internally produced CO 2 . Terrestrially derived CO 2 originates from soil respiration and CO 2 -rich geologic formations, which is then transported to headwaters via overland, subsurface, and groundwater flows. In contrast, CO 2 from internal processes is largely the result of respiration of organic matter to CO 2 , though other processes, including photooxidation ) and CH 4 oxidation (Lupon et al. 2019), contribute to total internal production. In headwater streams, the influence of terrestrially derived CO 2 is predicted to be greater than internal production and to decrease in magnitude in larger streams and rivers (Hotchkiss et al. 2015).
Losses of CO 2 from streams include evasion to the atmosphere, uptake through photosynthesis, and hydrologic export downstream. Evasion of CO 2 from freshwaters is a large flux of C largely unaccounted for in terrestrial budgets (Genereux et al. 2013) and is important at global scales (Raymond et al. 2013;Liu et al. 2022). Hydrologic export of dissolved inorganic carbon (DIC) comprises aqueous CO 2 , H 2 CO 3 , HCO 3 − , and CO 3 2− -, which speciate according to pH (Stumm and Morgan 1996). High concentrations of CO 2 may reduce pH by an amount that depends on the buffering capacity of the water (Wetzel and Likens 2000;Small et al. 2012) and accelerate dissolution of carbonate minerals (Stoddard et al. 1999).
Advances in sensor technology allow measurement of CO 2 and O 2 at high frequency in freshwaters (Johnson et al. 2010), permitting estimation of fluxes of these gases under different hydrologic conditions (e.g., base or storm flows) and as drivers of stream physicochemistry. Results from discrete sampling of F CO2 have revealed important aspects of underlying ecological and morphological processes, including the importance of gas exchange rates and depth (Rasera et al. 2013;Campeau et al. 2014;Oviedo-Vargas et al. 2015). Aquatic sensors allow differentiating C sources and sinks at higher frequency and under conditions hard to capture using discrete sampling (e.g., floods). Further, coupling net ecosystem productivity (NEP) estimates with CO 2 sensor data allows for finer accounting of CO 2 , variability during base and storm flow events, and evaluation of processes that affect concentrations of both gases together (e.g., gas exchange, respiration) and separately (e.g., anaerobic respiration) (Vachon et al. 2020;Aho et al. 2021b;Haque et al. 2022). The advance in CO 2 sensing demonstrates that temporal variation at short (e.g., storm event) and longer (e.g., growing season) scales across the continental US is multi-faceted, driven by interactions among stream biology, physics, and hydrology (Crawford et al. 2017). While combining CO 2 and O 2 sensor data in headwater streams to partition and account for various sources and sinks has been applied in Arctic and temperate streams (Lupon et al. 2019;Rocher-Ros et al. 2020), these methods have not been applied in tropical streams.
In this paper, we present the results of continuous deployment of in-stream and riparian well sensors to estimate areal and volumetric C fluxes in a headwater Neotropical stream, Costa Rica. We simultaneously estimate reach CO 2 losses as evasion (F CO2 ), and CO 2 inputs as NEP and terrestrial CO 2 from groundwater (GW CO2 ) at hourly intervals for six months, allowing comparison of fluxes at short-(event) and long-term (seasonal) hydrologic conditions. We hypothesized that: (1) O 2 would be undersaturated and CO 2 would be supersaturated relative to the atmosphere due to in-stream respiration, (2) pCO 2 would decrease at short time scales due to dilution (e.g., storm events) and increase at long-term scales from sustained terrestrial inputs (e.g., seasonal), (3) external fluxes of CO 2 would exceed internal CO 2 production following the theoretical predictions for low order streams (Hotchkiss et al. 2015), and (4) stream pH would decrease with elevated GW CO2 as predicted by carbonate equilibrium.

Site description
Our study took place at La Selva Biological Station, Costa Rica (LSBS), a 1600 ha tropical wet forest reserve, with elevation ranging from 22 to 146 m above sea level (Fig. 1). LSBS is located at the transition from the upland foothills of the Cordillera Central of Costa Rica and the Caribbean lowlands, and is the lowland terminus of the altitudinal transect in the Braulio Carrillo National Park (Fig. 1). Mean annual  Figure S1 has additional details on the sensor deployment and reach delineation rainfall is 4300 mm, with a dry season from January to April and a wet season from May to December (Sanford et al. 1994).

Data collection
We collected stream data from April 1, 2013 to September 30, 2013. Discharge (Q, m 3 s −1 ) was continuously measured at a sharp-crested 90-degree V-notch weir constructed near the downstream end of Taconazo (Genereux et al. 2005). Rainfall (mm) was collected from the LSBS weather station, located ~ 900 m from Taconazo. Meteorological data are available at https:// tropi calst udies. org/.
We secured a YSI 600xlm multiprobe sonde 5 cm above the stream bottom at the downstream end of the study reach and just upstream of the pool created at the weir (Fig. S1). The sonde was placed in a PVC tube with holes drilled to allow for exchange of stream water and secured to a rebar stake in the stream bed. The sonde continuously measured temperature (°C), pH, and DO (mg L − 1 and %-saturation) at hourly intervals. The sonde was retrieved every 2 weeks for recalibration and to clean fouling on the sensor heads. Partial pressure of CO 2 (ppmv) was measured using a Vaisala GMT221 infrared gas analyzer (IRGA) in the stream and in a riparian well. The sensors were prepared for submerged deployment as described in Johnson et al. (2010) and pCO 2 was logged hourly using a Campbell CR1000 datalogger. As pCO 2 was measured under water, the data were corrected for depth and temperature (Johnson et al. 2010). We excluded data from all sensors when the water level fell below the height of the sonde (e.g., conductivity < 0.005, DO near air saturation), which was 8.5% (~ 15 days) of the dataset. Sensor data underwent further QAQC following the guidance of Taylor and Loescher (2013) and visual examination using the datacleanr R package v 1.0.3 (Hurley 2021). The sensor schematic and deployment stations in the stream are shown in Fig. S1.
To evaluate the precision of the sensor, we compared the estimates of CO 2 from the in-stream sensor to headspace samples. Headspace samples were collected weekly from Taconazo by collecting 4 mL of stream water into a 10 mL syringe and injecting the sample into an inverted sealed serum vial prefilled with 100 µL of HCl with a 22-gauge needle and a 0.45 μm pore filter. Samples were equilibrated on a shaker table, after which a 250 µL headspace was removed and pCO 2 analyzed on an SRI Instruments gas chromatograph (Las Vegas, Nevada, USA) equipped with a thermal conductivity detector and 3-foot silica gel column (He carrier flow rate 10 mL min − 1 , detector 150 °C, oven 90 °C). For both sensor and headspace measurements of pCO 2 , we calculated [CO 2 ] aq using temperature-dependent Henry's Law constant (Plummer and Busenberg 1982) and For both measured and saturation concentrations of the in-stream gases, we measured departure, CO 2-dep and O 2-dep , as the difference of measured concentration from saturation concentration. Departure concentrations were plotted as a data cloud for each month, O 2-dep vs. CO 2-dep . For each month, we calculated the: (1) location of the cloud centroid, and (2) 1/|slope| the inverse of the slope of a best-fit line through each month's data cloud. The inverse slope shows the efficiency of metabolism, interpreted as the moles of CO 2 produced per moles of O 2 consumed during ecosystem respiration.

Response of CO 2 to storm events
To evaluate the role of storms on pCO 2 during the monitoring period, we compared gas concentrations during high and low flow events. We used a hydrograph separation approach to determine baseflows and storm flow discharges for each hour in our time series. We used a modified Lyne-Hollick filter, which uses a recursive data filtering approach to separate baseflow for a given hydrograph (Ladson et al. 2013). The approach relies on a filter parameter, α, to determine the volume of baseflow and volume of storm flow. We assessed a range of α values from 0.9 to 0.99 for Taconazo (Fig. S2) and visually determined α of 0.96. The separation filter identified each discharge measurement as baseflow when the fraction of baseflow ≥ 0.5 or stormflow when the fraction < 0.5.
To evaluate pCO 2 during storm and baseflow conditions, we compared pCO 2 in each month and in each flow condition. We used a non-parametric twoway Aligned Rank Transform (ART) test to determine which months and flow conditions had greater pCO 2 . Hydrograph separation was calculated in the hydrostats R package v 0.2.8 using the baseflows() function (Bond 2021) and ART in the ARTool v 0.11.1 R package (Kay and Wobbrock 2020).

Reach-scale CO 2 fluxes
The sensor deployment in the stream and riparian well allowed us to calculate reach-scale areal fluxes within the focal reach. We simultaneously estimated CO 2 inputs through groundwater (GW CO2 ) and net ecosystem production (NEP), and outputs as evasion to the atmosphere (F CO2 ).
Estimates of GW CO2 were calculated as the product of groundwater discharge into the study reach and [CO 2 ] measured in the well. Groundwater discharge into the lower Taconazo reach had been determined using both instantaneous and continuous conservative tracer injections to the stream (Ardón et al. 2013;Oviedo-Vargas et al. 2015). In a 132 m reach, groundwater discharge was measured as 17.8% (dry season) and 7.2% (wet season) of stream discharge (Oviedo-Vargas et al. 2015). In the same 75 m reach in this study during the dry season, groundwater discharge was 9% of stream discharge (Ardón et al. 2013). We favor this approach, which uses empirical field data, to alternatives (e.g., hydrograph separation, see above), which introduce uncertainty (Fig. S2, S5-S11) into our process-based flux estimates.
We estimated groundwater discharge (Q GW ) into the 75 m study reach upstream of the weir every hour as where f GW is the percent change in stream discharge from conservative tracer injections, equal to 17.8% for the dry season (April) and 7.2% for the wet season (May-September), and Q is total hourly discharge (m 3 h − 1 ). The seasonal approximation of Q GW as a (1) Q GW = f GW * Q fraction of Q assumes no temporal variation in f GW into the reach. The lack of variation is unlikely given the seasonality of rainfall, but the data to determine the variability could not be collected to the same accuracy as the other parameters and the percentages used reflect discrete sampling with high accuracy (Genereux et al. 2005;Ardón et al. 2013;Oviedo-Vargas et al. 2015).
Groundwater CO 2 flux (mol CO 2 m − 2 h − 1 ) was calculated as input of CO 2 in groundwater that enters the stream bed area in the study reach where L is the length of the reach (75 m), w is the wetted width (1.5 m in the dry season and 2.8 m in the wet season), and Q GW (m 3 h − 1 ) is estimated from Eq. 1.
Areal CO 2 inputs were also quantified through NEP, or the net internal production of CO 2 through aerobic processes. We estimated NEP as where NEP i is hourly instantaneous metabolism (mol O 2 m − 2 h − 1 ), O i is the O 2 concentration in the stream at each timepoint, i (mol m − 3 ), O i−Δt is O 2 at the preceding timepoint, O sat,i is the concentration of O 2 at saturation for the same timepoint (mol m − 3 ), − z is mean reach depth (m), and K O2 is the gas exchange coefficient for O 2 (h − 1 ). We used values of K O2 measured from dry (19.5 d − 1 ) and wet (7.3 d − 1 ) season tracer gas injections of propane (Oviedo-Vargas et al. 2015), converted in terms of O 2 using Schmidt scaling; see Appendix 2 for calculations. NEP was converted to mol C m − 2 h − 1 assuming a 1:1 respiratory quotient between O 2 and CO 2 (Rocher-Ros et al. 2020). We selected the direct metabolism method for determining NEP because the lack of a diel O 2 signal (Fig. S3) and relatively high reaeration make Taconazo ill-suited to alternative stream metabolism methods (Appling et al. 2018;Hall and Ulseth 2020). Further, the direct method allows for estimation of NEP at finer temporal compared to alternative approaches, which are resolved at the daily scale.
Reach-scale losses of CO 2 were estimated as evasion to the atmosphere. F CO2 was estimated as: where [CO 2 ] aq-str and [CO 2 ] sat (mol m − 3 ) are concentrations described in the above text, K CO2 is the gas exchange coefficient for CO 2 (h − 1 ) and − z is mean reach depth (m). As with K O2 above, we used seasonal measurements of dry (20.3 d − 1 ) and wet (7.6 d − 1 ) season gas exchange from propane injections and converted to K CO2 using Schmidt scaling (Raymond et al. 2012).
Reach-scale fluxes were aggregated to daily time steps by summing hourly fluxes. Negative values of NEP (mol C m − 2 d − 1 ) indicate net daily consumption of CO 2 , likely through photosynthesis, whereas positive NEP indicate net production of CO 2 through aerobic respiration. Daily areal inputs (GW CO2 and NEP) were evaluated as drivers of outputs (F CO2 ) using linear regression. Finally, we evaluated the influence of daily GW CO2 on mean daily pH, hypothesizing greater GW CO2 would decrease pH. We regressed log 10 GW CO2 against mean daily pH, removing 4 days of mean daily pH outliers (pH < 4.8 and pH > 5.9). All calculations and statistics were completed in R v 4.1.2 (R Core Team 2022) and are available in Appendix 2.

Stream data
Total rainfall during the 180-day monitoring period was 2067 mm, with median daily rain of 2.29 mm (range: 0-107 mm) and 49 days with no rainfall recorded (Fig. 2a). April had the lowest rainfall (104.1 mm) and June had the most rainfall (498.6 mm). Median hourly stream discharge was 0.004 m 3 s −1 (0-1.378 m 3 s −1 range). Median discharge was lowest in April (0.0002 m 3 s −1 ) and greatest in July (0.0160 m 3 s −1 ). Median estimated groundwater flux into the reach was 0.0003 m 3 s −1 (0-0.0346 m 3 s −1 range) (Fig. 2b). Median pCO 2 in the stream was 6343.6 ppmv (range 773-11,994 ppmv), lower than that measured in the riparian well (median pCO 2 46,924 ppmv, range 28,663-48,683 ppmv). Over the monitoring period, 14% of hourly stream pCO 2 measurements were missed, the largest missing section corresponding with the highest flows in late July (4) F CO2 = ([CO 2 ] aq−str − [CO 2 ] sat ) * K CO2 * − z and early August (Fig. 2d). In the riparian well, 27% of hourly measurements were missing, including the same period missing from the stream time-series and a period in May, during the transition from dry to wet season (Fig. 2e). There was no pattern between stream pCO 2 and well pCO 2 , reflecting the sustained high well pCO 2 (Fig. 3). Estimates of DIC from both discrete sampling (mean 0.23 ± 0.06 mmol DIC L − 1 ) and from the sensor (mean = 0.25 ± 0.04 mmol DIC L − 1 ) were similar (p = 0.34).

Evaluating aqueous concentrations of CO 2 and O 2
In comparing O 2 and pCO 2 to their atmospheric saturation, most timepoints showed CO 2 supersaturation and O 2 undersaturation (Fig. 4). CO 2 departure was ~ 6.6-times greater than O 2 departure, on average. Ellipse centroid location varied little over time ( Table 2). The value of 1/|slope| was greatest in September (467.9) and lowest in August (0.7). The general position of the cloud away from the 1:-1 line and closer to the x-axis indicates aerobic metabolic processes are not responsible for controlling CO 2 and O 2 , and the higher concentration of CO 2 than O 2 suggests CO 2 in excess due to either anaerobic respiration or external inputs (e.g., groundwater).

Discussion
In this study, we combined stream and riparian well aquatic gas sensors to quantify temporal variation of pCO 2 and O 2 , and CO 2 input and output fluxes in a Neotropical headwater stream. The results show sustained high pCO 2 in the stream relative to the atmosphere, and pCO 2 was greater during baseflow than during storms. Elevated pCO 2 led to high F CO2 , sustained by higher CO 2 inputs from both NEP and GW CO2 , in contrast to theoretical predictions that suggest greater groundwater contributions in headwater streams. Last, GW CO2 contributes to lower pH in the stream and defines a possible mechanism for understanding drivers of episodic and seasonal acidification events (Small et al. 2012;Ganong et al. 2021). Our analysis highlights the use of combined sensor arrays in estimating multiple reach-scale CO 2 fluxes and posits a method to estimate terrestrial CO 2 losses to the hydrologic cycle. We found F CO2 to be a major loss of C from the Taconazo, underscoring the importance of considering fluxes from headwater streams in complete C assessments and budgets.

Evaluating aqueous concentrations of CO 2 and O 2
The monitoring period included one month of the dry season with the remaining five months during the wet season (Sanford et al. 1994). The transition from dry to wet season is reflected by the increased rainfall in mid-May (Fig. 2a). The dry to wet transition coincides with a seasonal increase  in stream pCO 2 (Figs. 2d and 5b), though there was little increase in riparian well pCO 2 (Fig. 3). pCO 2 measured in Taconazo (median pCO 2 = 6343 ppmv) was high relative to similarly sized streams in a monitoring study across headwater streams in North America. Taconazo pCO 2 was higher than 6 of the 7 streams monitored, only less than a site in a southern hardwood forest and greater than a highland tropical stream in Puerto Rico (Crawford et al. 2017). We observed little diel variability in pCO 2 or O 2 (Fig. S2), indicating gross primary productivity (GPP) is low in Taconazo, and supported by the NEP estimates > 0 mol C m − 2 h − 1 (Fig. 6b). The paired CO 2 -O 2 cloud (Fig. 4) is located far from the 1:-1 line, with CO 2 in excess of O 2 . This reflects CO 2 supersaturation and O 2 undersaturation, and the deviance from the 1:-1 line indicates aerobic ecosystem processes do not regulate CO 2 and O 2 . In tundra headwater streams, the location of the CO 2 -O 2 departure cloud was closer to the 1:-1 line and reflected the greater influence of in-stream Fig. 5 a Separated hydrograph, with baseflow (blue line) and storm flow (yellow line), summing to total discharge. b pCO 2 measured in each month during each of the flow conditions, baseflow (blue boxplots) or storm flow (yellow boxplots). Boxplots show median, 25th and 75th quartiles, and outliers > 10th and 90th percentiles NEP on F CO2 (Rocher-Ros et al. 2020). The location of the data cloud away from the 1:-1 line suggests that aerobic metabolism is not a dominant process in Taconazo and is not responsible for simultaneously driving O 2 and pCO 2 . In contrast, anaerobic respiration may be an important contribution of CO 2 to the stream from riparian soils or in-stream sediments, to account for the supersaturation not attributed to aerobic respiration.

Response of CO 2 to storm events
The hydrograph separation shows that storm flows dilute pCO 2 , but increasing pCO 2 occurs at longer, seasonal scales. Dilution during high flows is likely, as median discharge increases by a factor of 7.5 from base (0.22 m 3 h − 1 ) to storm (1.62 m 3 h − 1 ) flows. Dilution in CO 2 concentration during storm events indicates pCO 2 of surface flows and shallow subsurface flows are lower than pCO 2 measured in deep soils and in-stream, though total flux during the  (Osburn et al. 2018). Additionally, decreased pCO 2 during storm flows may result from increased evasion during high flow events as storms can increase turbulence and increase gas exchange during the event (Raymond et al. 2012;Hall and Ulseth 2020), which reduces pCO 2 during storm events. Increased pCO 2 during baseflow conditions suggest reach scale input fluxes predominate during baseflow periods. While NEP estimates across ranges of discharge are uncommon, elevated pCO 2 during baseflows reflect greater mineralization of CO 2 through NEP. Increased dissolved organic C (DOC) fluxes in storm events fueled in-stream respiration in peat streams (Demars 2019), suggesting similar storm flow related pulses of DOC (Osburn et al. 2018) could fuel increased CO 2 mineralization. Seasonal increases in pCO 2 from the dry to the wet season reflect the increase in GW CO2 in June and July (Fig. 6a).

Reach-scale CO 2 fluxes
Headwater streams are predicted to receive a greater fraction of C from external sources relative to instream production due to contributing drainage area compared to stream-bed area. However, our data show a larger contribution of internally-produced CO 2 . For example, models for temperate streams indicate external inputs of C (i.e. GW CO2 ) to be greater than internal production (i.e. NEP) in low-order streams (Hotchkiss et al. 2015). Our results do not entirely support this hypothesis, as GW CO2 was often a lower contributor of CO 2 to the total flux than NEP. In part, this could be to underestimates of GW CO2 that derive from lack of contribuus estimates of f GW . In addition, our methods do not capture shallow subsurface flow and overland flow, which could be contributing to CO 2 inputs in the reach. Estimates of GW CO2 could be improved with greater understanding of the temporal variation of f GW , both between and within the dry and wet seasons and using alternative approaches to minimize uncertainty at the scale of our data collection. Beyond the uncertainty in hydrologic approaches, our deployment of sensors in a riparian well represents a method to estimate C losses to the hydrologic cycle from terrestrial systems. Given the sustained CO 2 measured in the riparian well (Fig. 3), which are similar to soil CO 2 measured in forested headwater streams in the Amazon (Johnson et al. 2008), we highlight the need to understand variation in groundwater inputs, as a large fraction of soil-derived CO 2 is quickly evaded to the atmosphere.
Estimates of NEP reveal aerobic processes were a source of C to the stream and were often larger in magnitude than GW CO2 . Estimates of NEP showed little temporal variation (Fig. 6b) and are derived from undersaturated DO measurements throughout the period (Figs. 2c and 4) and lack diel variation (Fig. S3). The magnitude of NEP was not affected by variability in gas exchange rates (Fig. S4, b), which confirms the consistent undersaturation of O 2 relative to the atmosphere (Fig. 2c). The NEP estimates were similar to streams and rivers in the tropics (median NEP = 0.9 mol C m − 2 d − 1 , compare to NEP = 1.5 mol C m − 2 d − 1 (Marzolf and Ardón 2021)), and NEP was a stronger contributor to F CO2 than GW CO2 (Fig. 6e). In-stream contribution to F CO2 was also shown in tundra streams with higher primary production (Rocher-Ros et al. 2020). In-stream gross primary production is likely low due to light limitation exerted by the forest canopy, resulting in C cycling in the stream driven by allochthonous organic matter.
Evasion estimates were high and of similar magnitude to previous work in Taconazo. We estimate a median F CO2 0.18 ± 0.14 mol C m − 2 d − 1 , which are similar to fluxes calculated from headspace samples from the dry (0.9 ± 0.4 mol C m − 2 d − 1 ) and Fig. 7 Groundwater CO 2 flux as a driver of the mean stream pH. Black line is line-of-best fit. Point color indicates the month and vertical bars for each point represent the pH daily standard deviation wet (0.6 ± 0.3 mol C m − 2 d − 1 ) season in the same stream (Oviedo-Vargas et al. 2015). Our estimates rely on seasonal gas exchange measurements from Oviedo-Vargas et al. (2015), though the similarity in F CO2 estimates support the sensor measurements of pCO 2 . Further, in our study F CO2 is likely underestimated. Our estimates rely on seasonal measurements of gas exchange, and improved understanding of temporal variation and relationship of gas exchange with stream morphology and hydrology in Taconazo would increase the precision of our estimates. Scaling equation estimates of gas exchange that rely on morphologic parameters, like depth, width, and discharge, are ill-suited to process based C estimates (Raymond et al. 2012). Inverse models rely on sufficient GPP to estimate gas exchange from a diel O 2 curve (Hall and Ulseth 2020), which are absent in our data (Fig. S3). Last, we highlight the sensitivity of F CO2 to variation in gas exchange (Fig. S4), which indicate transport limitation, or F CO2 controlled by flow and morphologic characteristics rather than C supply (Schneider et al. 2020).
For many headwater streams, F CO2 is the dominant fate of aqueous CO 2 , but comparing this flux to other CO 2 losses is an important consideration to understand the stream C cycle. We compared F CO2 from Taconazo to hydrologic export by upscaling the reach-scale estimates of F CO2 to the entire length of the stream. Using total stream bed area estimates (dry season = 3008 m 2 , wet season = 5022 m 2 ; Oviedo-Vargas et al. 2015) and assuming F CO2 in our reach is consistent throughout the entire length of the stream, we estimate total stream mean daily F CO2 of 1499 ± 1038 mol CO 2 d − 1 . In contrast, we multiply aqueous CO 2 by hourly discharge and estimate a mean daily CO 2 export of 8.2 ± 10.0 mol CO 2 d − 1 from Taconazo. This rough comparison shows F CO2 is a 182-times greater fate of CO 2 compared to downstream export and further highlighting the importance of studying F CO2 in tropical streams (Cole et al. 2007;Raymond et al. 2013).
Influence on pH Small et al. (2012) hypothesized that CO 2 from terrestrial sources enter streams with low solutes and reduce pH through carbonic acid creation, primarily during the early wet season in streams at La Selva. Our results provide some support for this hypothesis.
We show GW CO2 has a negative correlation with stream pH (Fig. 7). In the late dry and early wet season (April-May, Fig. 2), small inputs of GW CO2 can reduce mean daily pH to as low as 4.5. During the wet season (June onward), greater GW CO2 reduce pH to between ~ 5. Taconazo has low alkalinity (< 10 µeq L − 1 , Ardón et al. 2013), therefore the acid neutralizing capacity is overcome by small inputs of CO 2 . CO 2 is supplemented by increased DOC during storm events (mean stormflow DOC = 2.25 mg L − 1 , Ganong et al. 2015;Osburn et al. 2018) and redox reactions (e.g., Fe 2+ oxidation) which can collectively reduce pH. During the wet season, DOC inputs to Taconazo are disproportionately organic acids (Osburn et al. 2018) which have a pK a that contributes to lower pH. Decreased pH concurrent with higher CO 2 fluxes has been shown in temperate (Aho et al. 2021a) and boreal streams (Wallin et al. 2010), showing the importance of CO 2 fluxes on stream chemistry.

Conclusions
To our knowledge, ours is the first empirical study to investigate relative contributions of internal production and terrestrial sources of CO 2 to tropical streams. We found in-stream metabolism to be a net contributor to the CO 2 pool, with similar estimates of NEP to streams in across the tropics (Marzolf and Ardón 2021). We also found F CO2 to be consistent with previous measurements based on grab samples (Oviedo-Vargas et al. 2015), rather than using sensors, and comparable to headwater streams. Interestingly, internal production of CO 2 as NEP appears to be a more important contributor to F CO2 than were terrestrial inputs, based on estimates of GW CO2 . This finding is somewhat in contrast to models from temperate stream ecosystems (Hotchkiss et al. 2015), and raises questions on the role of sediment and anaerobic respiration in lowland tropical streams, like Taconazo.
In a global synthesis of efflux from inland waters, both Cole et al. (2007) and Raymond et al. (2013) stress the importance of evasion estimates from headwater tropical streams, which exhibit higher reaeration velocities and higher pCO 2 . We document sustained high pCO 2 leads to higher F CO2 (Raymond et al. 2013). Our study provides estimates of C fluxes from a stream type with disproportionate influence among inland waters, contributing to a major knowledge gap in the terrestrial C cycle.