Introduction

In light of current climate change challenges, the quantification of Earth’s global carbon (C) cycle is critical for a sustainable future. There is a growing consensus that inland waters may represent a significant carbon dioxide (CO2) source to the atmosphere (Cole et al. 2007; Raymond et al. 2013), and degassing of CO2 from streams and rivers has received significant attention in recent years (e.g., Billett and Moore 2008; Butman and Raymond 2011; Campeau et al. 2014; Hope et al. 2001; Hotchkiss et al. 2015; Jonsson et al. 2007; Richey et al. 2002; Striegl et al. 2012; Teodoru et al. 2015). Despite the increased interest in understanding biogeochemical mechanisms driving fluvial emissions of CO2, the role of old regional groundwater, including its interaction with surface water, has been largely overlooked. Regional groundwater flow is a hydrogeological process by which groundwater moves long distances through deep subsurface pathways beneath surface topographic divides, possibly recharging in one watershed and discharging in another many kilometers away (Tóth 2009; Schaller and Fan 2009; Smerdon et al. 2012; Pacheco 2015), thus creating the potential for long-distance subsurface transport of C between watersheds and ecosystems. However, little is known about CO2 emissions from streams and rivers affected by regional groundwater discharge and the potential effects on C balances of the aquatic and surrounding terrestrial ecosystems.

At La Selva Biological Station in the lowland rainforest of Costa Rica, several streams and wetlands receive significant inputs of old regional groundwater that has high concentrations of major ions and dissolved inorganic carbon (DIC) (Pringle et al. 1993; Genereux and Jordan 2006; Genereux et al. 2009, 2013). This regional groundwater is recharged outside La Selva at higher elevations in the volcanic Cordillera Central, and enters watersheds at La Selva through the process of interbasin groundwater flow (IGF) (Genereux and Jordan 2006; Genereux et al. 2005, 2009; Solomon et al. 2010). Quantifying CO2 emissions from tropical streams with this kind of input may be especially important as they can act as hotspots of CO2 degassing (Genereux et al. 2013; Oviedo-Vargas et al. 2015) and because tropical streams and rivers are emerging as a key component of the global C cycle (Alin et al. 2011; Rasera et al. 2013; Richey et al. 2002).

Evasion of CO2 from flowing waters can be estimated from the gradient in CO2 concentration across the air–water interface and the gas exchange piston velocity (k, units of length time−1) of CO2 (\(k_{{{\text{CO}}_{2} }}\)). Whole-stream tracer additions (e.g., Genereux and Hemond 1992; Kilpatrick et al. 1989; Wallin et al. 2011) are generally considered the most robust and reliable approach for field determination of k in streams and rivers. The method is well-suited for stream reaches on the order of 100 m to a few kilometers, but not for showing dynamics at a small spatial scale (e.g., reaches of ~10 m) or for repeated measurements (e.g., frequent monitoring).

A number of recent studies investigating CO2 emissions from streams and rivers have used floating chambers to measure CO2 fluxes (e.g., Billett et al. 2007; Crawford et al. 2013; Dinsmore et al. 2010; Neu et al. 2011; Rasera et al. 2013; Sand-Jensen and Staehr 2012; Striegl et al. 2012; Teodoru et al. 2015; Wu et al. 2007). In this technique, the CO2 flux across the water surface is measured by monitoring the buildup of CO2 emitted into the chamber. Potential benefits of this technique include the relatively low financial cost and simplified logistics in comparison with tracer injections (which facilitates repeated measurements for monitoring), and the small spatial scale of the measurements, which may allow insights into relationships between gas exchange and potential controlling variables (insights that may not be possible at the scale of whole-stream tracer additions). Also, chambers offer direct measurements of CO2 fluxes, eliminating the need to determine \(k_{{{\text{CO}}_{2} }}\). However, the floating chamber technique has been criticized for introducing artifacts that affect the estimation of gas exchange rate (e.g., Belanger and Korzum 1991; Broecker and Peng 1984; Hartman and Hammond 1984; Matthews et al. 2003; Vachon et al. 2010), by means such as sheltering the water surface beneath the chamber from wind, not sampling riffles or similar high-turbulence areas, inducing turbulence by flow against the bottom edge of chamber, or warming of chambers exposed too long to direct sunlight, with potentially different effects depending on whether the chamber is allowed to travel with the water current (drifting) or is placed in a fixed position (static) (Lorke et al. 2015).

The primary objective of this study was to quantify the magnitude and temporal and spatial variability of CO2 emissions from the Arboleda stream at La Selva Biological Station, a stream receiving significant inputs of DIC-rich regional groundwater. We used both drifting and static floating chambers. A second objective was to assess the feasibility of chamber work on this relatively small rainforest stream (most chamber work has been done in larger rivers). Our third objective was to examine the differences in results between drifting and static chambers, and between chambers and previously-published work that used whole-stream tracer methods to quantify CO2 emissions from the same stream (Oviedo-Vargas et al. 2015).

Study site

La Selva Biological Station is a 16 km2 tropical rainforest reserve located at the transition between the Caribbean lowland plains and the foothills of the Cordillera Central of Costa Rica (Fig. 1). La Selva ranges in elevation from 35 to 130 m. Average annual rainfall at La Selva from 1963 to 2013 was 4.3 m, with February, March, and April as the driest months and July, November, and December as the wettest (La Selva Biological Station website. Online meteorological data. http://www.ots.ac.cr/meteoro/default.php?pestacion=2). From 1982 to 2013, the mean air temperature was 25.0 °C and diurnal changes averaged 9.5 °C (La Selva Biological Station website. Online meteorological data. http://www.ots.ac.cr/meteoro/default.php?pestacion=2). Annual evapotranspiration is about 2.1 m in a year of typical annual rainfall (Loescher et al. 2005).

Fig. 1
figure 1

Arboleda stream at La Selva Biological Station in Costa Rica. The black arrow in the map showing chamber measurement locations indicates the direction of stream flow. The drifting chamber measurements were conducted in reaches R1–R12 (in green). Static chamber measurements were conducted at three points in each of nine cross-sectional transects of the stream at the locations marked with red asterisks. Repeated measurements during the wet and dry seasons were conducted with drifting chambers in reaches R6, R8, and R9 (color figure online)

Full size image

The Arboleda stream at La Selva Biological Station (Fig. 1) drains a surface area of 46.1 ha and has an average annual streamflow of approximately 13 m (annual water discharge normalized by watershed area; Zanon et al. 2014), of which about 34 % comes from regional groundwater (Genereux et al. 2005). Based on 14C age dating, the regional groundwater is estimated to be much older (2400–4000 years) than the local groundwater recharged in the lowland watersheds (10 years or less), and this is consistent with other geochemical and hydrogeological observations (Genereux et al. 2005, 2009; Solomon et al. 2010). Concentrations of DIC and dissolved aqueous CO2 in the Arboleda stream, [DIC] and [CO2]aq respectively, averaged 4.1 and 2.0 mM in 2010, 4.3 and 2.1 mM in 2011, 4.1 and 2.0 mM in 2013, and 4.2 and 2.0 mM in 2014 (Table 1, Electronic Supplementary Material, ESM).

The regional groundwater input to the Arboleda made it an ideal location for the chamber work; the CO2 degassing flux from the Arboleda was large enough that the deployment time of the chamber could be reduced to a fraction of what is often needed: 90 s rather than 5–60 min (e.g., Matthews et al. 2003; Striegl et al. 2012). This helped limit artifacts from changes in air temperature, pressure, and air–water concentration gradient inside the chamber. Also, the Arboleda has few riffles and few emergent natural debris dams of branches and leaves, likely due in part to the higher stream discharge and depth (about 0.8 m) associated with the large input of regional groundwater. These physical characteristics, together with a short deployment time, facilitated the use of the chamber, particularly for the drifting mode which requires the chamber to float freely along a stream section.

Materials and methods

Discharge and dissolved CO2 concentration

During the chamber experiments, volumetric discharge (Q) of the Arboleda stream was monitored at high frequency (15 min) at a V-notch weir downstream of the chamber measurement sites (Fig. 1). Just upstream of the weir, pH was measured weekly using an Oakton 11 pH meter and weekly water samples collected at the same time as the pH measurements were analyzed for alkalinity (Alk) using a digital titrator (Hach, Inc.) and 1.6 N sulfuric acid (Hach Company 2013). [CO2]aq was calculated from Alk and pH following Stumm and Morgan (1996):

$$[{\text{CO}}_{2} ]_{\text{aq}} = \frac{{\alpha_{0} \left( {{\text{Alk}} - \left[ {{\text{OH}}^{ - } } \right] + \left[ {{\text{H}}^{ + } } \right]} \right)}}{{\alpha_{1} + 2\alpha_{2} }}$$
(1)

where α0, α1, and α2 are the ionization fractions for CO2 (or carbonic acid H2CO3), bicarbonate, and carbonate, respectively. The ionization fractions were calculated as:

$$\alpha_{0} = \left( {1 + \frac{{{\text{K}}_{1} }}{{[{\text{H}}^{ + } ]}} + \frac{{{\text{K}}_{1} {\text{K}}_{2} }}{{[{\text{H}}^{ + } ]^{2} }}} \right)^{ - 1}$$
(2)
$$\alpha_{1} = \left( {1 + \frac{{[{\text{H}}^{ + } ]}}{{{\text{K}}_{1} }} + \frac{{{\text{K}}_{2} }}{{[{\text{H}}^{ + } ]}}} \right)^{ - 1}$$
(3)
$$\alpha_{2} = \left( {1 + \frac{{[{\text{H}}^{ + } ]}}{{{\text{K}}_{2} }} + \frac{{[{\text{H}}^{ + } ]^{2} }}{{{\text{K}}_{1} {\text{K}}_{2} }}} \right)^{ - 1}$$
(4)

where K1 and K2 correspond to the first and second acid dissociation constants of the carbonate system (Stumm and Morgan 1996). CO2 concentrations derived from alkalinity can be problematic for systems where pH is very low and/or organic acid concentrations are high (Hunt et al. 2011; Abril et al. 2015), however this does not represent a major concern for the Arboleda stream where pH is close to neutral (the mean of weekly measurements from 2010 to 2014 was 6.4), and DOC concentrations are low (averaging 93 μM; Genereux et al. 2013).

Chamber measurements

Degassing fluxes of CO2 (\({\text{f}}_{{{\text{CO}}_{ 2} }}\)) from the Arboleda stream were measured using floating cylindrical chambers that penetrated about one cm below the stream surface. An infrared CO2 sensor (GMP343, Vaisala, Inc.), a temperature/humidity sensor (ChipCap2™, GE Measurement & Control), and a mixing fan were built inside the chamber. A microcontroller board (Arduino, LLC) with SD storage, a display, and a 12 V battery (1300 mAh) were housed in an enclosure secured on top of the chamber. Two chambers used in this work had different water coverage area and floatation devices. One chamber had an area (A) of 0.126 m2, a headspace volume (V) of 10.9 L, and two rectangular Styrofoam pieces for floatation. The Styrofoam pieces were placed ~10 cm from the outer wall of the chamber on opposite sides of it (Fig. 1 ESM). A smaller chamber (A = 0.041 m2, V = 5.6 L) remained afloat by means of a Styrofoam ring fixed ~5 cm from the outer wall of the chamber (Fig. 2 ESM). The floating devices were not installed directly adjacent to the chambers to ensure that the stream surface area contributing to the measured flux was limited to the footprint of the chamber (Figs. 1 and 2 ESM).

During static deployment the chamber was held in place for 90 s, and during drifting deployment the chamber was allowed to be carried downstream by the water current for 90 s. For each measurement, the drifting chamber was released from the same location three times (or a sufficient number of times to capture three complete replicate runs). If the drifting chamber got caught at the stream edge or on any obstruction, data were discarded and the measurement was repeated. Before static or drifting measurements a 30-s air flush was conducted holding the chamber upside down to remove the chamber air from the previous run with the help of the fan. Observed increases of chamber CO2 concentrations during the last 30 s of chamber deployment were used to calculate values of \({\text{f}}_{{{\text{CO}}_{ 2} }}\), the rate of CO2 flux out of the stream water following Eq. (5):

$${\text{f}}_{{{\text{co}}_{2} }} = \frac{{10{\text{VP}}}}{\text{ART}} \cdot \frac{\text{dC}}{\text{dt}}$$
(5)

where \({\text{f}}_{{{\text{CO}}_{ 2} }}\) has units of μmol CO2 m−2 s−1, V corresponds to the chamber volume (cm3), P is the air pressure in the chamber (assumed equal to 1 atm, or 101.3 kPa), A represents the chamber area (cm2), R is the universal gas constant (8.314 cm3 MPa K−1 mol−1), T is the chamber air temperature (K), dC/dt is the rate of change of gas phase CO2 concentration in the chamber (ppm s−1, i.e., μmol of CO2 per mol of bulk gas, per second), and 10 is a factor that accounts for unit conversions. All chamber measurements had linear increases of CO2 with R2 >0.95 (88 % had R2 >0.99).

The spatial variability of \({\text{f}}_{{{\text{CO}}_{ 2} }}\) in the Arboleda stream was examined using both static and drifting chamber measurements. Drifting chamber measurements with the smaller chamber (Fig. 2 ESM) were conducted on 17–18 August 2014 in 12 reaches (R1 through R12) spaced along a 348-m section of the Arboleda stream (Fig. 1). The length of the reaches varied between 4 and 32 m (the approximate length of reaches R1 through R12 was 10, 13, 8, 32, 9, 20, 12, 7, 10, 12, 4, and 9 m, respectively). Water velocity was measured at three or four locations within the path followed by the drifting chamber in each deployment, using a Flowtracker hand-held acoustic Doppler velocimeter. Measurements with the static chamber were conducted on 31 July 2014 at nine different cross-sections of the stream (three points per cross-section) in the lower part of the 348-m stream section (Fig. 1), using the larger chamber (Fig. 1 ESM). Water velocity was measured at the chamber locations in these nine cross-sections using a Flowtracker. In addition, five static and seven drifting measurements were made within a single small reach of the Arboleda stream near reach R9 within a 35-min period on the morning of 16 August 2014, using the small chamber (Table 2 ESM).

To examine temporal variability of \({\text{f}}_{{{\text{CO}}_{ 2} }}\), we used the smaller chamber in drifting mode to measure \({\text{f}}_{{{\text{CO}}_{ 2} }}\) values in three of the 12 reaches used in the spatial survey (R6, R8, and R9, Fig. 1) on 11 days from October 2014 to November 2014 (during the wet season) and 6 days from March 2015 to April 2015 (during the dry season). In each reach, deployments were conducted five to 15 times on each measuring day.

Statistical analysis

We used t-tests to determine if (1) average \({\text{f}}_{{{\text{CO}}_{ 2} }}\) measured with the smaller chamber in drifting mode was different from that measured with the same chamber in static mode (16 August 2014); (2) average \({\text{f}}_{{{\text{CO}}_{ 2} }}\) measured with the static chamber on 31 July 2014 was significantly different from that measured with the drifting chamber on 17 August 2014 (in the stream section in which both sets of measurements were conducted). A two-way analysis of variance (ANOVA) was used to determine significant differences in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) among reaches and seasons (n = 11 per reach during the wet season and n = 6 per reach during the dry season). Significant ANOVAs were followed by Tukey’s honest significant difference (HSD) for multiple comparisons. Pearson correlation coefficient (r) was used to examine the linear correlation between \({\text{f}}_{{{\text{CO}}_{ 2} }}\) and water velocity during the spatial variation measurements on 31 July and 17–18 August 2014. Statistical analyses were performed in Sigma Plot 11. Significance was tested at the 95 % confidence (P value < 0.05).

Results

Spatial survey (static and drifting chambers, wet season)

Average stream discharge was 10.4 m3 min−1 during the period of the static chamber measurements on 31 July 2014, and 11.0 and 11.5 m3 min−1 during the drifting chamber measurements on 17 and 18 August 2014, respectively (Table 1). [CO2]aq measured weekly at the weir during this part of the study period (27 July 2014 to 24 August 2014) was 1.4 ± 0.2 mM (mean ± standard deviation, SD). Results for \({\text{f}}_{{{\text{CO}}_{ 2} }}\) from the static chamber averaged 72.7 µmol C m−2 s−1 and ranged from 9.9 to 157.6 µmol C m−2 s−1 (n = 27, Table 1, Fig. 2a). Results from the drifting chamber in reaches overlapping with locations of the static chamber (R7–R12 sampled on 17 August 2016) averaged 24.4 µmol C m−2 s−1 and ranged from 14.2 to 43.1 µmol C m−2 s−1 (n = 18, Table 1, Fig. 2a). Mean \({\text{f}}_{{{\text{CO}}_{ 2} }}\) from static chambers was 3.0 times higher than that from drifting chambers and this difference was statistically significant (t-test, t43 = 4.55, P < 0.001). Average \({\text{f}}_{{{\text{CO}}_{ 2} }}\) from the drifting chamber in all 12 reaches was 35.5 µmol C m−2 s−1 (range was 14.2–104.2 µmol C m−2 s−1).

Table 1 Carbon dioxide degassing flux (\({\text{f}}_{{{\text{CO}}_{ 2} }}\)) from the Arboleda stream estimated using static and drifting chambers
Full size table
Fig. 2
figure 2

a Box plot for all 27 static chamber measurements of \({\text{f}}_{{{\text{CO}}_{ 2} }}\) conducted at 9 different cross sections of the stream on 31 July 2014, and 18 drifting chamber measurements made on 17 August 2014 in reaches that overlap spatially with the static measurements (R7 through R12 in Fig. 1, and gray box in Fig. 2c). b \({\text{f}}_{{{\text{CO}}_{ 2} }}\) measured with a static chamber in 9 transects on 31 July 2014. Error bars represent the standard error of the three repetitions made in the same transect. Inset shows stream water velocity versus static chamber \({\text{f}}_{{{\text{CO}}_{ 2} }}\). c \({\text{f}}_{{{\text{CO}}_{ 2} }}\) measured with a drifting chamber in reaches 1 through 12 on 17–18 August 2014. Error bars represent the standard error of the three repetitions made in the same reach. Reaches within the gray box (7 through 12) overlapped spatially with static chamber measurements (Fig. 2b). Inset shows stream water velocity versus drifting chamber \({\text{f}}_{{{\text{CO}}_{ 2} }}\) for all 12 reaches; horizontal and vertical error bars are the standard error of the three drifting repetitions made in each reach, and the three or four measurements of stream velocity made along the path of the chamber, respectively

Full size image

The coefficient of variation (CV) for \({\text{f}}_{{{\text{CO}}_{ 2} }}\) measured with the static chamber was 52 % (Table 1). For the drifting chamber, the CV was 44 % for reaches R7 through R12 (overlapping with static chamber measurements) and 68 % for reaches R1 through R12 (Table 1). Static chamber \({\text{f}}_{{{\text{CO}}_{ 2} }}\) values were positively correlated with stream velocity measured at the location of the chamber deployment (Pearson r = 0.85, P < 0.0001, Fig. 2b inset). No correlation was observed between \({\text{f}}_{{{\text{CO}}_{ 2} }}\) measured with the drifting chamber and water velocity (r = 0.27, P = 0.391, Fig. 2c inset).

Five static and seven drifting measurements were made with the smaller chamber near reach R9 within a 35-min period on 16 August 2014; the ratio of static to drifting CO2 degassing flux was 3.3 (Table 2 ESM), similar to the ratio of 3.0 from more numerous measurements at more sampling locations when static chamber measurements were done on 31 July 2014 and drifting chamber measurements were done on 17 August 2014.

Repeated measurements (drifting chamber, wet and dry seasons)

Stream discharge at the Arboleda during the study periods included frequent high flow events throughout the wet season (October 2014–December 2014), in particular during December (Fig. 3a), and a steadier baseflow discharge during the dry season (Fig. 3b). [CO2]aq measured at the Arboleda weir during the wet and dry season study periods averaged 2.4 ± 0.3 and 2.4 ± 0.4 mM (Fig. 3a and b), respectively.

Fig. 3
figure 3

Wet season (a) and dry season (b) continuous stream discharge (every 15 min) and weekly dissolved CO2 concentration at the Arboleda weir, and wet season (c) and dry season (d) repeated measurements of \({\text{f}}_{{{\text{CO}}_{ 2} }}\) using the drifting chamber in reaches R6, R8, and R9. Error bars represent standard error (SE) of the 5–15 repetitions made each day at each reach

Full size image

Among the reaches repeatedly sampled to assess temporal variability, \({\text{f}}_{{{\text{CO}}_{ 2} }}\) generally decreased from upstream to downstream (R6 > R8 > R9, Fig. 3c and d): during both wet and dry seasons, mean \({\text{f}}_{{{\text{CO}}_{ 2} }}\) in R6 was significantly higher than the mean values in R8 (Tukey’s HSD, P < 0.001; dry and wet season) and R9 (Tukey’s HSD, P < 0.001; dry and wet season), and mean \({\text{f}}_{{{\text{CO}}_{ 2} }}\) in R8 was significantly higher than that of R9 but only during the wet season (Tukey’s HSD, P < 0.001 and P = 0.78 for the wet and dry seasons, respectively). Through the wet season, \({\text{f}}_{{{\text{CO}}_{ 2} }}\) averaged 38.1, 26.3, and 19.1 µmol C m−2 s−1 in reaches R6, R8 and R9, respectively (Fig. 3c, Table 1). In the dry season, \({\text{f}}_{{{\text{CO}}_{ 2} }}\) averaged 34.3, 17.8, and 16.1 µmol C m−2 s−1 in reaches R6, R8, and R9 respectively (Fig. 3d, Table 1). Although mean \({\text{f}}_{{{\text{CO}}_{ 2} }}\) in all three reaches was lower during the dry season, this difference was only significant in R8 (Tukey’s HSD for R6, R8, and R9 had P values of 0.139, <0.002, and 0.273, respectively). In general, decreased variability in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) measurements was observed during the dry season relative to the wet season (Fig. 3, Table 1). A significant relationship was observed between \({\text{f}}_{{{\text{CO}}_{ 2} }}\) and discharge in reaches R6 and R9, but that was not the case for R8 (Fig. 4). In reaches R6 and R9, discharge explained about 35 % of the variation in \({\text{f}}_{{{\text{CO}}_{ 2} }}\), however these relationships are sensitive to data from two high discharge events in December 2014 (Fig. 4).

Fig. 4
figure 4

Relationship between \({\text{f}}_{{{\text{CO}}_{ 2} }}\) in stream reaches R6, R8, and R9 and stream discharge measured downstream of the reaches at the Arboleda V-notch weir. Triangles represent dry season \({\text{f}}_{{{\text{CO}}_{ 2} }}\) and squares represent wet season \({\text{f}}_{{{\text{CO}}_{ 2} }}\)

Full size image

Discussion

Magnitude of CO2 emissions from the Arboleda

Regardless of the season or chamber mode (static or drifting), \({\text{f}}_{{{\text{CO}}_{ 2} }}\) values measured in the Arboleda were at least 10 times higher than most average values reported for other streams and rivers in studies using floating chambers, whether tropical, temperate, or polar (Table 2). Emissions reaching magnitudes similar to those from the Arboleda (average \({\text{f}}_{{{\text{CO}}_{ 2} }}\) = 35.5 µmol C m−2 s−1 for drifting chambers) were reported by Neu et al. (2011) from a tropical headwater stream in Brazil (Table 2), apparently linked to stream inputs of groundwater carrying CO2 from microbial respiration of deep soil organic matter (Johnson et al. 2008). Teodoru et al. (2015) measured emissions comparable to those in the Arboleda stream in some locations of the Zambezi River in Africa (Table 2), downstream of wetlands or floodplains. Unlike the systems investigated by Neu et al. (2011) and Teodoru et al. (2015), in which high CO2 emissions were associated with local enrichment of surface water or young groundwater with biogenic CO2, high \({\text{f}}_{{{\text{CO}}_{ 2} }}\) from the Arboleda stream results from a large input of CO2 of deep crustal origin (Genereux et al. 2009) by discharge of old regional groundwater to the stream (Genereux et al. 2009; Oviedo-Vargas et al. 2015). The distinction is important because the processes fueling fluvial CO2 emissions influence how this flux should be accounted for in the C budget of an ecosystem.

Table 2 Summary of studies using chambers to measure CO2 fluxes (\({\text{f}}_{{{\text{CO}}_{ 2} }}\)) from streams and rivers
Full size table

In recent years, it has become apparent that terrestrial respiration can be underestimated if CO2 export by streamflow or stream degassing is ignored, thereby overstating the land as a C sink (Aufdenkampe et al. 2011; Cole et al. 2007). However, that is only true if the exported or degassed CO2 is biogenic. If this CO2 is instead of geological origin, as in the Arboleda stream, then it does not represent ecosystem respiration, and counting it as part of ecosystem respiration in an ecosystem C budget may lead to overestimation of ecosystem respiration and underestimation of the true C sink strength of the ecosystem. Thus, understanding the origin of stream CO2 (which may be connected to regional as well as local hydrogeology) is critical for using field data to address the fundamental question of whether the ecosystem is a net source or sink for CO2. This has recently been stressed for peatlands where there is evidence of CO2 emission of geological origin (Billett et al. 2015), but remains largely overlooked. Results from the Arboleda are among the few that shed light on how old regional groundwater (as opposed to much younger shallower local groundwater) influences our understanding of the C source/sink status of ecosystems. However, the Arboleda is very unlikely to be a unique case, given that the hydrogeological factors resulting in high \({\text{f}}_{{{\text{CO}}_{ 2} }}\) at the Arboleda have been documented not only for the Central American isthmus (e.g., Pringle et al. 1993) but globally (e.g., Genereux et al. 2013 and references therein). Streams like the Arboleda may represent relatively common hotspots for CO2 emission and thus merit a closer look regarding their degassing fluxes.

Spatiotemporal dynamics of CO2 emissions

Floating chambers were both practical and informative in investigating CO2 fluxes from the Arboleda stream at a small spatial scale (at the fixed locations of static chambers and the small reaches defined by chamber drift paths of 5–30 m), and repeatedly through time. Both drifting and static chambers showed high spatial variability in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) within the 348 meters of stream length studied (Fig. 2). Spatial differences appear to be maintained though time, as shown by the repeated measurements at reaches R6, R8, and R9 (Fig. 3); with few exceptions, \({\text{f}}_{{{\text{CO}}_{ 2} }}\) generally decreased from upstream to downstream (from R6 to R8 to R9). These results suggest that small-scale geomorphic characteristics of the stream locations play a significant role in the magnitude of the CO2 fluxes, consistent with results from the spatial survey. For example, in reach R4 where the highest \({\text{f}}_{{{\text{CO}}_{ 2} }}\) was measured (Fig. 2), water spilled over a tree log partially submerged perpendicular to the stream flow, and the log also directed a large part of the water flow into a small cross-section of the stream. Comparing among reaches R6, R8, and R9, R6 (20 m) was straight with relatively shallow and fast-flowing stream water, R8 (7 m) included a bend in the channel and slow flow, and R9 (10 m) was straight with slow flow impounded by a large log near the end (all flow went underneath the log). To obtain representative CO2 fluxes from fluvial systems, it appears critical to deploy chambers at numerous locations of varying hydraulic/geomorphic character throughout the reach of interest.

Temporal variation in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) was lower than spatial variation. The coefficient of variation (CV) in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) was about 10–30 % for each individual reach (R6, R8, or R9) within an individual season, wet or dry (Table 1), which is lower than the CV of 50–70 % obtained from the wet season spatial surveys (Table 1). Repeated drifting chamber measurements showed that overall, differences in CO2 fluxes between seasons were small and that higher variability in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) occurred during the wet season (Fig. 3). Also, in reaches R6 and R9, \({\text{f}}_{{{\text{CO}}_{ 2} }}\) and stream discharge at the weir were positively related (Fig. 4) and \({\text{f}}_{{{\text{CO}}_{ 2} }}\) was not significantly different between the wet and dry seasons, while reach R8, located between R6 and R9, had the opposite behavior: no statistically significant relationship between \({\text{f}}_{{{\text{CO}}_{ 2} }}\) and stream discharge (Fig. 4), and significantly different \({\text{f}}_{{{\text{CO}}_{ 2} }}\) between seasons. These differences between reach R8 and reaches just upstream (R6) and downstream (R9) may lie in some as-yet unrecognized connection between CO2 emissions and small geomorphic or hydraulic variations along the channel such as those mentioned for R6, R8, and R9 in the previous paragraph. The variation we have observed in CO2 fluxes and their linkages to season and to stream discharge suggests the usefulness of chambers in revealing new insights regarding stream CO2 emissions, but also highlights the underlying complexity of those emissions, with possible biological, hydrological, hydraulic, and other controls.

Static and drifting chambers

Measurements made within a 35-min period on the morning of 16 August 2014 showed the ratio of static to drifting \({\text{f}}_{{{\text{CO}}_{ 2} }}\) was 3.3 (Table 2 ESM), similar to the ratio of 3.0 found with more numerous measurements at more sampling locations when static chamber measurements were done on 31 July 2014 and drifting chamber measurements were done on 17 August 2014. This is within the range of static to drifting \({\text{f}}_{{{\text{CO}}_{ 2} }}\) ratios (2.0–5.5) found recently by Lorke et al. (2015) for three data sets representing eight different streams and one river in Germany and Poland (Table 2). Of course, it is possible that the static/drifting ratio of 3.0 from our Arboleda data was determined in part by an actual difference in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) from 31 July to 17 August (i.e., not solely the difference between static and drifting chambers).

CO2 degassing flux from a stream can be expressed as: \({\text{f}}_{{{\text{CO}}_{ 2} }}\) = \(k_{{{\text{CO}}_{2} }}\)(C–Csat) = λCO2D(C–Csat), where \(k_{{{\text{CO}}_{2} }}\) is the gas exchange piston velocity for CO2 (length time−1), C is the aqueous CO2 concentration in the stream water, Csat is the aqueous CO2 concentration that would be in equilibrium with atmospheric CO2, λCO2 is the first-order gas exchange rate constant for CO2 (time−1), and D is the stream depth (Rathbun and Tai 1982; Duran and Hemond 1984; Parker and Gay 1987). Given an atmospheric CO2 concentration, there are three stream controls on \({\text{f}}_{{{\text{CO}}_{ 2} }}\): λCO2, D, and C. The first-order gas exchange rate constant has been observed to be relatively steady over at least an order of magnitude range in stream discharge (Genereux and Hemond 1992), supporting the idea that it is unlikely there was significant change in λCO2 from 31 July to 17 August (stream discharge increased by only about 6 %; Table 1). Measurements at the Arboleda weir indicate that from 27 July to 24 August 2014, C increased by about 26 % (Table 1 ESM) while D increased by about 3 % (from 35 to 36 cm), suggesting a potential increase in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) of about 29 % during this time, corresponding to a hypothetical static (31 July) to drifting (17 August) ratio of 0.78. The observed ratio of 3.0 is much closer to the estimate of 3.3 from the 16 August 2014 measurements. Thus, while some change in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) from 31 July to 17 August cannot be ruled out, the ratio of \({\text{f}}_{{{\text{CO}}_{ 2} }}\) between those dates seems to be controlled more by differences between the static and drifting chambers, and the ratio of 3.3 from 16 August (when static and drifting measurements were made within 35 min) indicates that static chambers yielded \({\text{f}}_{{{\text{CO}}_{ 2} }}\) values about 3 times those from drifting chambers on the Arboleda.

We found that stream water velocity in the Arboleda was highly correlated to \({\text{f}}_{{{\text{CO}}_{ 2} }}\) measured with the static chamber but not the drifting chamber (Fig. 2). Teodoru et al. (2015) also report a strong linear correlation between \({\text{f}}_{{{\text{CO}}_{ 2} }}\) and stream water velocity beneath static chambers. In Danish lowland streams, Sand-Jensen and Staehr (2012) found log\(k_{{{\text{CO}}_{2} }}\) determined using static chambers strongly correlated to the logarithm of stream velocity (their Table 2), and discussed the potential benefits of such a relationship for scaling up (i.e., estimating values of \(k_{{{\text{CO}}_{2} }}\) for a stream system based on measured or modeled stream velocity). However, Teodoru et al. (2015) viewed their correlation as an artifact of induced turbulence from the disturbance of stream flow by the lower part of the chamber walls, and \({\text{f}}_{{{\text{CO}}_{ 2} }}\) values measured with static chambers were excluded from their results because of this. Lorke et al. (2015) attributed elevated gas fluxes beneath static chambers (relative to drifting chambers) to the increased turbulence found beneath, and caused by, static chambers. In our results from the Arboleda stream, the positive correlation of \({\text{f}}_{{{\text{CO}}_{ 2} }}\) with stream velocity for static chambers and lack of correlation for drifting chambers are consistent with the static chamber introducing artificial turbulence as stream water flowed past the lower edge of the chamber, enhancing stream degassing and resulting in overestimation of \({\text{f}}_{{{\text{CO}}_{ 2} }}\).

The mean \({\text{f}}_{{{\text{CO}}_{ 2} }}\) values from the static and drifting chambers (73 and 36 µmol C m−2 s−1, respectively) overestimate and underestimate, respectively, what is likely the best estimate of overall CO2 emissions: the mean \({\text{f}}_{{{\text{CO}}_{ 2} }}\) of 56 µmol C m−2 s−1 determined using tracer injections of propane and chloride to estimate \(k_{{{\text{CO}}_{2} }}\) (Oviedo-Vargas et al. 2015) in the same section of the Arboleda where the chamber measurements were conducted (R1–R12 in Fig. 1). The tracer injections were conducted on August 4 and 6, 2014, in between the measurements with static chambers (31 July) and drifting chambers (17–18 August). To our knowledge, only one other direct comparison between chamber and tracer methods at approximately the same time and place has been published for a stream (Crawford 2012; Crawford et al. 2013). In an Alaskan stream reach, Crawford (2012) measured \(k_{{{\text{CO}}_{2} }}\) five times using propane injections (reach length ≤400 m), followed immediately by the deployment of static chambers at the downstream end of the reach; \(k_{{{\text{CO}}_{2} }}\) based on the propane work was 1.9× higher than that from the static chambers. Crawford’s results contrast with our results that showed higher \({\text{f}}_{{{\text{CO}}_{ 2} }}\) values from static chambers than from injected tracer work. It is possible that in the study by Crawford (2012), the CO2 fluxes at the downstream end of the reach were not representative of the reach used for the tracer injections.

Assuming the tracer-based value of \({\text{f}}_{{{\text{CO}}_{ 2} }}\) is the best estimate, our results suggest that the drifting chamber underestimated \({\text{f}}_{{{\text{CO}}_{ 2} }}\). The paths of the drifting chamber may have been biased in favor of obstacle-free areas in the thalweg where the chamber was generally released. If there was higher CO2 flux near natural debris dams of logs, branches, and leaves and/or near the stream edge where water was shallower, then the under-sampling of these areas by the drifting chamber may explain the lower \({\text{f}}_{{{\text{CO}}_{ 2} }}\) from the drifting chamber. Assessment of lateral variation in \({\text{f}}_{{{\text{CO}}_{ 2} }}\) across the channel should be a focus in future work with drifting chambers.

Doyle and Ensign (2009) suggest different potential benefits for environmental measurement reference frames that are fixed in place (Eulerian) or following individual water parcels (Lagrangian). In the case of estimating stream water degassing with floating flux chambers, our results and those of Lorke et al. (2015) suggest a systematic bias (overestimation of reach-scale degassing flux) from fixed (static) chambers in streams, and our results further caution that even drifting chambers may contain bias (in our case, underestimation of reach-scale degassing flux), perhaps arising from the drifting chamber preferentially following paths that are not fully representative of gas exchange within the reach.

Models of stream gas exchange based on correlations with stream hydraulic parameters have been and remain problematic as predictive tools (Genereux and Hemond 1992; Raymond et al. 2012; Wallin et al. 2011). For instance, using hydraulic parameters from the Arboleda in empirical equations presented by Raymond et al. (2012) to determine k, we estimated that \({\text{f}}_{{{\text{CO}}_{ 2} }}\) at the Arboleda spanned almost a factor of 3 (58–155 µmol C m−2 s−1,Table S3 ESM), and the equations presented by Raymond et al. as having the highest predictive power (Eqs. 1 and 7 in their Table 2) gave the poorest agreement with Arboleda \({\text{f}}_{{{\text{CO}}_{ 2} }}\) results from the chamber and tracer injection techniques. With the complexity and incomplete understanding of the dynamics of CO2 sources and sinks in streams, reliable prediction of stream CO2 emissions remains an elusive goal. Creative and sound application of chamber, tracer, and other methods is critical, given the emerging climatic and biogeochemical importance of CO2 emissions from streams and rivers and the little-explored role of old regional groundwater in driving high emissions. Drifting chambers can be an important tool, especially if they can be released in adequate numbers and from different points across the width of the channel (not all in the center or thalweg) to allow some to drift near the shoreline where gas fluxes might differ from those in the thalweg.