Journal of Geophysical Research: Biogeosciences

Seasonal and spatial variability of CO2 emission from a large floodplain lake in the lower Amazon

Authors

Errata

This article is corrected by:

  1. Errata: Correction to “Seasonal and spatial variability of CO2 emission from a large floodplain lake in the lower Amazon” Volume 117, Issue G1, Article first published online: 14 January 2012

Abstract

[1] The inundation status of the Amazon floodplain affects biogenic gas production and evasion. We analyzed spatial variability of dissolved CO2 concentration and gas evasion in a large floodplain lake in the lower reach of the Amazon River in four hydrological phases. We calculated surficial CO2 concentrations from measurements of pH, dissolved inorganic carbon, temperature, and conductivity and used meteorological data to calculate gas transfer coefficients to estimate CO2 evasion. Gas transfer coefficients that take into account both wind and heating and cooling at the lake's surface are on the order of 10 cm hr−1, approximately four times higher than values previously used in regional estimates of gas evasion from lakes on the Amazon floodplain. Supersaturation of CO2 occurred throughout the lake and was higher in the littoral zone and in regions receiving Amazon River inflows. CO2 concentration was reduced in regions with phytoplankton blooms. The range of CO2 concentrations was least at low water, 47 μM to 233 μM, and largest at high water, 1 μM to 656 μM; the average annual value was 125 μM. We estimate mean (±standard deviation) fluxes from open-water in L. Curuai to the atmosphere of 44 ± 15, 348 ± 13, 371 ± 23, and 364 ± 20 mmol CO2 m−2 d−1 during receding, low, rising, and high water, respectively. The error associated with these values reflects, for each hydrological phase, the spatial variation in CO2 concentration in L. Curuai, a likely range in atmospheric CO2 levels and temporal variations in gas transfer coefficient within 10-day periods.

1. Introduction

[2] The Amazon basin stores and cycles a large quantity of carbon, yet Houghton et al. [2009] suggest that the region's carbon balance has been, on average, nearly neutral over the last decade. However, there remain uncertainties regarding changes in land use and in the balance between photosynthesis and respiration due to variations in climate. The fluvial system, recently recognized as an important component of the carbon budget, releases to the atmosphere an amount of CO2 roughly equal to the carbon sequestered by the forest. The basin-wide outgassing of CO2 from rivers and floodplains was estimated by Richey et al. [2002] as 500 Tg C yr−1, but this is considered conservative and is currently under revision [Richey et al., 2009]. Refined analyses of spatial and temporal variations in fluxes are required for uncertainties to be resolved.

[3] Carbon dioxide is supersaturated in most Amazonian rivers and floodplain waters [Richey et al., 1988; Richey et al., 2002] and its exchange with the atmosphere occurs via diffusive and turbulent transfer across the air-water interface [MacIntyre et al., 1995]. On the scale of an individual lake, Melack and Engle [2009] reported that CO2 evasion represented the majority of carbon export. The production of dissolved CO2 in Amazon floodplains and other tropical wetlands includes respiratory releases from aquatic plants and animals, but is dominated by microbial metabolism [Hamilton et al., 1995; Waichman, 1996] which consumes labile organic matter of young age released from seasonally flooded forests, aquatic macrophytes and upland forests [Quay et al., 1992; Mayorga et al., 2005; Engle et al., 2008; Melack et al., 2009; Richey et al., 2009].

[4] The annual flood of the Amazon River influences the ecology and biogeochemistry within its floodplain and plays a major role in retention and transformation of organic matter and exchange of gases between land, water and the atmosphere [Junk, 1997; Melack et al., 2009]. Richey et al. [1988] measured CO2 concentration in lakes as part of cruises along the Amazon River, but within lake spatial and temporal variability has yet to be examined. Hence, we determined spatial distributions of CO2 concentration and efflux from open-water of Lago Grande de Curuai (Lake Curuai) for four hydrological phases. We calculate surficial CO2 concentrations from measurements of pH, dissolved inorganic carbon (DIC), temperature, and conductivity and use meteorological data to calculate gas transfer coefficients to estimate CO2 evasion. The results from our study complement earlier estimates of CO2 evasion from Amazon floodplains in two important ways: (1) the seasonal and spatial variability in the open-water environment of a large shallow lake system was examined in detail; and (2) samples spanning four hydrological phases were combined with meteorological measurements in the same lake to calculate gas exchanges using wind-based approaches, a surface renewal model and a wind and buoyancy flux based model.

2. Site Description

[5] Lago Grande de Curuai lies in the floodplain of the Amazon River to the south of Óbidos, Pará, Brazil (Figure 1). It is composed of several interconnected lakes temporarily or permanently connected to the Amazon R. by channels and overbank flow. The stage of the Amazon R. varies ca. 6 m annually in the nearby reach with a peak in June and minimum in November (Figure 2). The lake's flooded area ranges from about 850 to 2274 km2; at flood peak open-water covers about 65% of flooded area with flooded vegetation covering the remaining area (Table 1). Water movements and mixing in L. Curuai are complex, and the effects of physical forces (e.g., wind, solar radiation, river inflows) are modified by the lake's surface area and depth, varying seasonally and spatially (Figures 3a–3d and Table 1). During low water the lake becomes sufficiently shallow for sediments to be resuspended by wind-induced turbulence.

Figure 1.

Study site and location of meteorological buoy.

Figure 2.

Seasonal variation of spatially averaged CO2 concentration for four hydrological phases in Lake Curuai. Long-term mean and standard deviation, and 2003–2004 and 2004–2005 hydrographs of the Amazon River at Óbidos, Pará, Brazil.

Figure 3.

(a–d) Lake bathymetry for four hydrological phases. Images are masked (gray) according to inundation extent. Spatial distribution of dissolved concentration of CO2 at (e) receding (Sep 2003), (f) low (Nov 2003), (g) rising (Feb 2004), and (h) high water (Jun 2004). Samples were interpolated using a multiquadratic radial basis function. Images are masked according to open-water extent obtained from Landsat image classification. White regions are unsampled isolated lakes. Dark gray regions are differences between inundation and open-water areas and represent flooded vegetation.

Table 1. Morphological Variables During Hydrological Phases: Flooded Area (Aflooded); Open-Water Area (Aopen-water); Mean Depth (zmean) Within Open-Water; and Maximum Depth (zmax)a
 Hydrological Phase
HighReceding/RisingLow
  • a

    Standard deviations are in parenthesis.

  • b

    Estimated from digital elevation model obtained by integrated interpolation of bathymetry and Shuttle Radar Topography Mission data [Barbosa et al., 2006; C. Barbosa and C. Rudorff, unpublished data, 2010] and water stage gauged at L. Curuai.

  • c

    Estimated from classified Landsat images (Barbosa and Rudorff, unpublished data, 2010).

Afloodedb (km2)22742015850
Aopen-waterc (km2)14791278612
zmean (m)3.2 (2.5)1.6 (1.5)0.6 (0.4)
zmax (m)15.012.49.7

[6] Bonnet et al. [2008] determined the hydrologic balance for the lake, and reported the lake and its catchment to be, on an annual basis, a source of water to the Amazon R., which accounts for about three quarters of annual inputs of water to the lake. Runoff from the local catchment, direct rainfall, and groundwater seepage were estimated to represent ca. 10%, 9%, and 4% of the annual total inputs, respectively. Each year, the storage phase of L. Curuai starts between November and January and lasts until May–June; the drainage phase starts in July and lasts until November.

[7] Barbosa et al. [2010] described large seasonal and spatial variations in total suspended sediment (TSS), chlorophyll-a, and pH throughout L. Curuai. Average surface water concentrations of TSS changed from 14.5 mg L−1 during high water to 463 mg L−1 during low water. Average chlorophyll-a varied from 8 μg L−1 during rising water to 70 μg L−1 during receding water. While pH in the Amazon R. was relatively constant throughout the year (ca. 6.5), pH in the lake varied from an average of 6.7 during low water to 7.7 during receding water. High pH, up to 9.2, was found during receding water in areas with algal blooms of up to 338 μg L−1 of chlorophyll-a.

3. Methods

3.1. Field Measurements

[8] Four campaigns were conducted to sample limnological parameters during the four phases of inundation: receding (September 2003), low (November 2003), rising (February 2004), and high water (June 2004). Sampling stations were distributed in a dispersed spatial pattern (P < 0.01; based on the average nearest neighbor Euclidean distance test) over the main open-water region with average nearest neighbor distance of 3 km (Figure 3 and Table 2). Samples were collected between 1000 and 1400 h over periods of 10 to 16 days. Transparency was measured using a 20-cm diameter Secchi disk. Temperature (accuracy of ±0.3°C), conductivity (accuracy of ±1%), and pH (accuracy of ±0.05) were measured with routinely calibrated Horiba water quality sensors. Water was collected using a cylindrical sampler which integrated the water column from surface to Secchi depth. Water samples were filtered under vacuum through glass fiber filters (Whatman GF/C, 47 mm diameter) and frozen in the field for subsequent laboratory analysis. These manipulations may change DIC concentrations slightly. In the laboratory, concentrations of DIC were determined using a Shimadzu TOC-5000A carbon analyzer calibrated with potassium biphthalate. Further description of sampling strategy and additional parameters measured are provided by Barbosa et al. [2010].

Table 2. Field Campaign Dates and Number of Samplesa
 Hydrological Phase
RecedingLowRisingHigh
  • a

    Mean transparency from Secchi disk (SD), light attenuation coefficient (kd), pH, dissolved inorganic carbon (DIC), dissolved CO2 concentration in surface waters (CO2aq), and dissolved CO2 concentration in equilibrium with the atmosphere (CO2eq). Standard deviations are in parenthesis.

  • b

    Estimated using the conversion factor for turbid lakes kd = 1.3/SD [Koenings and Edmundson, 1991].

  • c

    Estimated from samples.

  • d

    Estimated from interpolated surfaces (see text).

  • e

    Temporal variations in atmospheric CO2 concentrations (S. Miller, unpublished data, 2004) and spatial variation in L. Curuai surface water temperature were used to calculate variability in CO2eq.

Date of Field Campaigns25 Sep to 7 Oct 200322 Nov to 2 Dec 20033–12 Feb 20043–19 Jun 2004
Samples (n)71737274
SD (m)0.23 (0.11)0.08 (0.04)0.10 (0.03)0.65(0.05)
kdb (m−1)6.6 (2.7)18.1 (4)14.7(6.3)2.0 (0.2)
pH7.68 (0.75)6.69 (0.58)7.25 (0.33)7.51 (0.76)
DIC (mg L−1)11.8 (2.7)7.5 (4.0)16.6 (4.1)32.3 (3.6)
CO2aqc (μM)92.2 (114.2)125.2 (34)148 (85.2)133 (150.3)
CO2aqd (μM)81.1 (95.6)116.8 (27.5)141.5 (81.8)131 (139)
CO2eqe (μM)11.9 (1.9)12.2 (1.7)12.6 (1.9)12.7 (1.7)

[9] Time series measurements of meteorological parameters became available from an instrumented buoy moored in the southeastern region of L. Curuai (Figure 1) in April, 2004. Monitored parameters included water temperature at 1 m depth (shielded thermistors: MEAS 44033RC; accuracy of ±0.1°C), air temperature and relative humidity (Rotronic MP103A; accuracies of ±0.3°C and ±1.5%, respectively), down- and upward shortwave radiation (wavelength range of 400–1100 nm; sensitivity of 490 μV/W/m2; accuracy of ±16%), wind speed and direction (R. M. Young 05106; accuracies of ±0.3 m s−1 and ±3 degrees, respectively) at 3 m above the lake, and atmospheric pressure (Vaisala PTB100-A; accuracy of ±0.3 mb at 20°C).

[10] Since the high water campaign in June 2004 was the only one with concurrent meteorological data, to approximate conditions during the three other seasons measurements acquired in the corresponding days of the following year were used. The drought at the end of 2005 did not affect our data since it occurred outside of the period of our measurements; the water levels were in good agreement between the periods from mid-2003 to mid-2004 and from mid-2004 to mid-2005. Measurements from the Vila Franca meteorological station [Fitzjarrald et al., 2009], located 20 km to the southeast of the buoy (2° 20′ 55″ S; 55° 1′ 44″ W), indicated that solar radiation and air temperature were on average 18% and 3% higher, respectively, in the period from mid-2004 to mid-2005 compared to mid-2003 to mid-2004. Small gaps in data from L. Curuai buoy station, due to telemetry signal losses, were filled with data from the Vila Franca meteorological station. Longwave radiation emitted from the lake surface was estimated from water temperature according to the Stefan-Boltzmann law with an emission coefficient of 0.97. The downward flux of longwave radiation was obtained as hourly averages from two years (2002 and 2003) measured at the Tapajos National Forest (3° 1′ 5″ S; 54° 58′ 8″ W) [Miller et al., 2009].

[11] Depth of the actively mixing layer (zAML) is required for computation of the gas transfer coefficient using the surface renewal model. This layer is also known as the surface layer and is the layer in direct contact with the atmosphere and thus where heat and momentum are being exchanged [Imberger, 1985]. It is generally quantified by slight changes in temperature or by near uniform values of turbulence [Shay and Gregg, 1986; MacIntyre et al., 2009]. Values of zAML were assumed to be 0.5 m or the seasonal lake mean depth (zmean) because temperature profiles were not measured. This decision was based on our examination of the frequency and extent of vertical mixing in Lake Calado, determined from multiyear time series [Melack and Fisher, 1983; MacIntyre and Melack, 1984, 1988], and the seasonal morphometric changes of L. Curuai (Figures 3a–3d and Table 1). Seasonal stratification persisted once the depth of L. Calado reached about 6 m. During stratification, the mixed layer deepened at night due to cooling and stratified in the day due to heating. Similar patterns are expected in L. Curuai, although wind induced mixing may cause greater mixed layer depths in the larger lake. However, computed attenuation coefficients reach 18 m−1 during low water (Table 2); hence daily stratification is likely. At high water, regions deeper than 6 m represent only 16% of open-water, and this portion is insignificant during the other hydrological phases. Based on similarities and differences between L. Curuai and L. Calado, the water column in L. Curuai during the low and medium (rising and receding) water stages is likely to be fully mixed at sunrise, develop a shallow thermocline (∼0.5 m) toward mid-day under low wind conditions, and fully mix again at night. During the high water stage, the lake may develop periods of stable stratification under low wind conditions with significant diel variation of the mixed layer depth (1 to 6 m, depth permitting).

3.2. Calculation of CO2 Concentrations

[12] Dissociation constants from Plummer and Busenberg [1982] and inorganic carbon species as a function of pH from Stumm and Morgan [1996] were used for conversion of DIC to dissolved carbon dioxide concentrations in surface waters (CO2aq). Adjustments were made for ionic strength, but without complete major ion data, we estimated ionic strength from conductivity [Butler, 1992]; these adjustments were very small because the waters are dilute (mean conductivity of 39 μS with a standard deviation of 10 μS). Atmospheric partial pressures of CO2 of 430 ± 50 μatm were used in our calculations. Measurements of atmospheric partial pressure of CO2 over Amazon floodplain lakes are scarce, and we based our values on recent observations by S. Miller (unpublished data, 2004) obtained above the surface of Lago Verde, located 50 km southeast of L. Curuai during high water. The concentration of dissolved CO2 in equilibrium with the atmosphere (CO2eq) was calculated using Henry's law. For comparison in our discussion, concentrations presented in μatm in the study by Richey et al. [2002] were converted to μM, using Henry's constant for CO2 at 25°C (0.0339 M atm−1) and correcting for the partial pressure of water (0.0313 atm).

3.3. Gas Exchange at the Air-Water Interface

[13] The magnitude and direction of the gas flux is dependent on the concentration difference between the atmosphere and water and upon transport processes at the interface. CO2 fluxes were calculated using the dissolved CO2 concentration departure from equilibrium and a gas transfer coefficient (k).

equation image

Due to the uncertainty in parameterizing the gas transfer coefficient, we used three models to determine a range of k values.

[14] The first model (C&C) used is the wind-based approach of Cole and Caraco [1998]:

equation image

where k600 is the gas exchange coefficient normalized to the Schmidt number of CO2 at 20°C in freshwater (Sc = 600) and U10 is the wind speed at a 10 m height above the water surface. Wind measured at 3 m was converted to U10 assuming law of the wall scaling and taking into account atmospheric stability.

[15] The second model (SR) used is a small eddy version of the surface renewal model, which takes into account turbulence generated by wind action and heat losses that cause convective motions in the water column [MacIntyre et al., 1995].

equation image

where ν is the kinematic viscosity. We let c1 = 0.56 [Isenogle, 1985]. The coefficient has not yet been fully established and current values range from 0.419 to 1.2 [Zappa et al., 2007; MacIntyre et al., 2010]. The dissipation of turbulent kinetic energy (ɛ) in the active mixed layer was computed from the water friction velocity (u*w) and buoyancy flux (β).

equation image

where the coefficients were determined empirically [Lombardo and Gregg, 1989] and κ is the von Karman constant. u*w is computed from shear stress at the air-water interface assuming it is equal on either side of the interface, giving τw = ρwu*w2 = τa = ρaCdU2 where τa and τw are shear stress on air and water side of the air-water interface; U is wind velocity at 3 m height and Cd is a drag coefficient computed for that height following Amorocho and DeVries [1980]; ρa and ρw are density of air and water, respectively. The buoyancy flux is calculated as

equation image

where g is gravity, α is the coefficient of thermal expansion, Cp is heat capacity, and H is the effective heat flux.

equation image

where S = ρwCpwT′(0)〉 is the surface heat flux, obtained by the sum of latent heat flux, sensible heat flux, and longwave back radiation; q is shortwave radiation; and z is depth. Light attenuation coefficient (kd) was estimated based on Secchi disk (SD) measurements using the conversion factor for turbid lakes kd = 1.3/SD suggested by Koenings and Edmundson [1991].

[16] The penetration of radiation into the water column was parameterized using Beer's Law following Jellison and Melack [1993]; angle of refraction was included in the calculation.

[17] Following MacIntyre et al. [2010], a damping coefficient (c2/zAML) was introduced to equation (4) to account for the effect of dampening of turbulence during periods of heat gain. Dissipation (equation (4)) was then computed as ɛ = 0.84[0.58(−β) + (c2/zAML)1.76u*w3zAML] when Lw/zAML ≥ 0 and U10 < 3 m s−1; where Lw is the water-side Monin-Obukhov length scale. Since data are lacking to determine zAML, the computation was run twice with zAML fixed at 0.5 m and at the seasonal lake mean depth (zmean) which represent the approximate range of diurnal variation expected for L. Curuai (as described above).

[18] The third model (EC) used is a wind-based model that includes diel heating and cooling [MacIntyre et al., 2010].

equation image
equation image

CO2 exchange coefficients (k) were converted as a function of field conditions based on the Schmidt number (Sc) dependency, assuming n = 0.5 which corresponds to the case of turbulent flow below a free surface [Jähne et al., 1987].

equation image

where i is an index representing the different models (C&C; SR; EC).

4. Results

4.1. CO2 Concentrations and Spatial Patterns

[19] The four hydrological phases had different statistical distributions of CO2 (Table 3). For low water, the mean and median were nearly equal, the skewness and kurtosis statistics were low, the Kolmogorov-Smirnov test was not significant (P > 0.01), and the data fit a normal curve well. For the other three hydrological phases the mean was higher than the median, skewness and kurtosis statistics indicated a disproportionate number of values on the upper tail of the distributions, the Kolmogorov-Smirnov test was significant (P < 0.01), and the data fit a normal curve poorly. Hence, the mean was not a good measure of central tendency for these data. Additionally, there were statistical outliers with large values in all four data sets; receding and rising water had extreme CO2 values 4.5 and 4 standard deviations above the mean, respectively. For all hydrological phases, extreme observations were located in the littoral zone on the perimeter of the sampling stations.

Table 3. Descriptive Statistics for Dissolved CO2 Concentration in Surface Waters (CO2aq)
 Hydrological Phase
RecedingLowRisingHigh
Mean92.2125.2148.0133.0
Median61.1121.5128.981.7
Standard Deviation114.234.085.2150.3
Minimum0.347.136.01.3
Maximum608.0233.0490.2656.0
Range607.7186.0454.2654.7
Interquartile Range121.334.999.0163.9
Skewness2.3520.6572.121.529
Kurtosis7.3391.4536.381.888

[20] Median values are close to robust estimates of central tendency and indicate receding water as the period of lowest CO2 concentration (61 μM) and rising water as the highest (129 μM). The median test (Chi-square) was used to determine if CO2 levels differ among the four hydrological phases. Across all 286 samples, the median CO2 was 113 μM. Low and rising water had a larger proportion of samples with CO2 greater than the overall median, while receding and high water had a larger proportion of samples with CO2 smaller than the median. The Chi-square value was 26.2, and the asymptotic significance was less than 0.01, indicating the existence of an overall effect of seasonality. The Mann-Whitney test was then used to identify seasonal gradients in levels of CO2. Low water CO2 levels were significantly greater than the levels at receding water (P < 0.01). The transition between low and rising water did not change CO2 levels significantly. CO2 concentrations became lower (P < 0.01) as water level rose to high water. The change in CO2 levels was not significant between high and receding water.

[21] Radial basis functions (subtype multiquadratic) were used to interpolate CO2 spatially within the main open-water body of L. Curuai, as it makes no assumptions about the normality of the data [Buhmann, 2003]. Moran's index, I [Moran, 1948; Ripley, 1981], was used to identify the overall spatial patterns of CO2 levels. Spatial autocorrelation was significant for all hydrological phases with strong tendency of clustering during receding and rising water (P < 0.01) and slightly less for low and high water (P < 0.10). Local Moran statistics [Anselin, 1995], were then used to identify where high or low values cluster spatially, and occurrences of spatial outliers. CO2 concentrations in the lake ranged from 0.3 μM to 608 μM during receding water (Figure 3e). Concentration in the Amazon R. (225 μM) did not influence lake waters during this phase of floodplain drainage. The two samples to the west with high values of CO2 surrounded by low values are spatial outliers (P < 0.01). The clustering of high values observed in the northernmost littoral region is statistically significant (P < 0.01). At low water, CO2 concentrations ranged from 47 μM to 233 μM (Figure 3f). CO2 concentrations increased toward the eastern channels possibly due to the influence of Amazon R. inflow with CO2 concentration estimated as 495 μM. Two points with indication of high CO2 clustering (P < 0.01) and six points of low CO2 clustering were observed, and there were no significant spatial outliers. During rising water, CO2 concentrations ranged from 36 μM to 560 μM (Figure 3g). A cluster of high CO2 observed in the northernmost littoral region is statistically significant (P < 0.01). Clusters of low values were observed in eastern main body of water (P < 0.01); the sample furthest to the east with a high value was identified as a spatial outlier (P < 0.01). However, Amazon R. inflow from the two main eastern channels, with CO2 concentration estimated as 595 μM, is likely to be the cause of the high CO2 concentration observed at this station. During high water, surface concentrations of dissolved CO2 ranged from 1 μM to 656 μM (Figure 3h). Based on two samples, surface concentrations in the Amazon R. were 372 μM and 425 μM. Amazon R. inflows from western channels and northern overbank flow influenced CO2 concentration in receiving waters of the floodplain. Statistically significant (P < 0.01) clusters of high CO2 concentrations were observed in the western littoral regions. Significant clusters of low CO2 were observed in interior pelagic regions. The sample to the east with a high concentration surrounded by low concentrations is a spatial outlier (P < 0.01). In this case the high concentration may not be explained by Amazon R. inflow from the eastern channels since at high water the flows in these channels are outward.

4.2. Meteorological Conditions

[22] Ten-day time series of meteorological measurements and computed energy fluxes for the four hydrological phases are shown in Figure 4. Wind generally intensified during daytime and weakened during nighttime, reaching speeds above 5 m s−1 on most days (Figures 4a–4d). Ranges of diel variation of water temperature measured at 1 m varied seasonally (Figure 4e–4h). Greater diel variation of water temperature was observed at low water. Toward the end of the rising water period and during high water, air temperature remained lower than water temperature. Maximum net shortwave radiation typically exceeded 1000 W m−2 each day, except during June (Figures 4i–4l). Sensible heat flux remained negative during June (Figure 4m), and ranged from about −15 W m−2 during nighttime to 20 W m−2 during daytime in September (Figure 4n); −15 W m−2 during nighttime to 50 W m−2 during daytime in November (Figure 4o); and −30 W m−2 during nighttime to 40 W m−2 during daytime in February (Figure 4p). Net longwave radiation ranged typically from −30 W m−2 during nighttime to −70 W m−2 during daytime in all periods. Evaporative cooling was substantial, and values of latent heat flux were lower than −200 W m−2 during windy periods and when surface water temperatures were higher than the air temperatures.

Figure 4.

Hourly average values of (a–d) wind speed at 10 m, (e–h) air and 1 m water temperatures, (i–l) net shortwave radiation (400–1100 nm), (m–p) sensible heat, latent heat, and longwave radiation fluxes, (q–t) surface heat flux and effective heat flux, assuming active mixing layer depth equivalent to 0.5 m (zAML = 0.5 m) and the seasonal lake mean depth (zAML = zmean).

[23] The effective heat flux, which indicates whether the actively mixing layer will gain or lose heat to the atmosphere, was positive during daytime, indicating that the upper water column tended to stratify at those times, and negative at night indicating mixed layer deepening would occur. The effective heat flux was slightly higher during daytime when zAML was assumed to be zmean, compared to when zAML = z0.5 m, due to the greater volume present to absorb and distribute the heat (Figures 4q–4t). However, the timing of the shift from positive to negative values was independent of the value of zAML, hence the sign of buoyancy flux, which is computed from effective heat flux and is needed for equation (4), was independent of zAML.

4.3. Gas Transfer Coefficients

[24] The lowest estimates of k600 were obtained using Cole and Caraco's [1998] wind based model, k600, C&C (Figure 5). Values were on average 85% higher when computed using the regression equations based on wind and buoyancy flux, k600, EC. Estimates of k600, SR with the surface renewal model depended upon the assumed zAML. When using zAML = z0.5 m, k600, SR was statistically higher than k600, EC (P < 0.01; Mann-Whitney test). The slight seasonality observed in both estimates of k600 was significant (P < 0.05; median test) (Figure 6a). When assuming zAML = zmean, seasonality was more pronounced (P < 0.01; median test) with higher estimates obtained during low water and lower estimates during high water.

Figure 5.

Gas transfer coefficients for (a) high, (b) receding, (c) low, and (d) rising water, displayed as 3 h averages, calculated using: wind-based model following Cole and Caraco [1998] (C & C), a wind and buoyancy flux based model (EC), and surface renewal model following MacIntyre et al. [2010] assuming active mixing layer depth equivalent to 0.5 m (SR (zAML = 0.5 m)) and the seasonal lake mean depth (SR(zAML = zmean)).

Figure 6.

(a) Gas exchange coefficients for CO2 in Lake Curuai. Mean and standard deviation based on 10-day time series shown in Figure 5. (b) Estimates of CO2 evasion from lake computed using mean values of CO2 departure from saturation and gas transfer coefficient. (c) Mean lake seasonal fluxes accounting for the variation in open-water surface area for the four hydrological phases. Legends for Figures 6a and 6c are the same as presented in Figure 6b.

4.4. Evasion of CO2

[25] CO2 evasion from open-water was estimated for each hydrological phase (Figure 6b) using the average values obtained from CO2 departure from saturation (Table 2; CO2aqCO2eq), and gas transfer coefficients averaged over the 10-day time series (Figure 6a) of each field campaign. Since the averaged gas transfer coefficients computed with the wind based models (C&C and EC) had almost no seasonal variation, their correspondent results of CO2 evasion shown in Figure 6b reflect the proportions of seasonal average CO2 concentration (Figure 2). The only clear exception was the estimate of k600,SR assuming zAML = zmean; in this situation estimates of areal evasion rate were higher at low water (Figure 6a).

[26] By combining CO2 flux with the total areal extent of the main body of open-water for each phase (Table 1), the highest rates of evasion, obtained using the wind-based models (C&C and EC), were observed during high water (Figure 6c). In these cases, variation in lake areal extent was the principal determinant of seasonal differences, followed by CO2 concentration. Seasonal variation caused by changes in gas transfer velocities was important in determining rates of evasion only when using the surface renewal model and assuming zAML = zmean. The effect of this model was to reduce overall differences in rate of evasion.

5. Discussion

[27] Open-water surficial concentrations of CO2, and hence gas evasion, vary in L. Curuai according to the hydrological phase of the Amazon River and distance from shore. Strong gradients of higher CO2 concentrations were observed toward littoral regions at receding, rising and high water. Significant relations between CO2 and depth, with Pearson's correlation coefficients of −0.25 (P < 0.05), −0.42 (P < 0.01), and −0.23 (P < 0.05), were observed for these phases, respectively. High CO2 levels in littoral regions were likely due to increased root respiration associated with floating macrophytes [Hamilton et al., 1995] and decomposition of plant material in the aquatic-terrestrial transition zone during drying and wetting phases. While in some littoral regions CO2 concentrations reached levels over 600 μM, pelagic regions fell to close to 0 μM during receding and high water when phytoplankton were abundant [Barbosa et al., 2010]. As a result of increased photosynthesis, pH increased with concomitant decreases in CO2. For example, during receding waters at one offshore station, CO2 concentration was 0.3 μM, chlorophyll-a concentration was 338 μg L−1, and pH was 9.2. Significant correlations between CO2 and chlorophyll-a of −0.39 (P < 0.01) and −0.53 (P < 0.01) were observed during receding and high water, respectively. Furthermore, diel stratification is likely to develop in pelagic regions and seasonal stratification may develop in regions deeper that 6 m [Melack and Fisher, 1983; MacIntyre and Melack, 1984, 1988] and limit mixing of water enriched in CO2 with depleted water in the upper mixed layer. Since samples were collected from surficial water during mid-day heating, values are likely to be lower than at other times of day [Crill et al., 1988]. In particular, higher surficial CO2 concentrations would be expected during nighttime due to respiration and deep convective mixing.

[28] Spatial pattern of CO2 concentration at low water was different from the other hydrological phases and likely due to (1) shallow depth (zmean = 0.6 m) and frequent mixing from surface to bottom; (2) light for photosynthesis being limited in the water column; and (3) macrophyte beds composed of dense stands of emergent grasses that retard lateral exchanges.

[29] CO2 concentrations in the Amazon R. remained at least twofold higher than lake averages in each of the hydrological phases. During rising water, concentrations in L. Curuai increased with river inflow from the main eastern channels. However, during high water, western inflowing channels discharged water from the Amazon R. into littoral regions with CO2 levels higher than the Amazon R. In this case, the biological processes associated with littoral regions likely influenced high CO2 levels.

[30] Richey et al. [2002] observed CO2 concentrations in the Amazon R. floodplain ranging from 97 to 1,445 μM. Concentration of CO2 in the floodplain lakes decreased in downstream reaches, averaging 463 ± 335 μM in the upstream reach between Vargen Grande and Itapeua, where a scroll-bar topography produces mostly long narrow lakes, to 207 ± 138 μM in the downstream reach between São José do Amatari and Óbidos, where a wide floodplain allows the dominance of large shallow lakes.

[31] Our annual average estimate for L. Curuai obtained from all samples was 125 ± 107 μM and falls within the range reported for the downstream reach [Richey et al., 2002]. Longer residence times in larger lakes increases the chances of a more complete decay of organic matter within the lake before carbon is lost by outflow to the main stem. Additionally, shallow lakes with larger open-water areas exposed to wind represent greater potential for aquatic CO2 evasion. These attributes apply to large floodplain lakes that are more abundant in the downstream reach [Sippel et al., 1992] and may explain their lower mean CO2 concentration. Given the high spatial variability in these downstream lakes, adequate spatial and temporal sampling is required to determine representative CO2 concentrations.

[32] Measurements of dissolved CO2 available from other inland waters in tropical South America are based on regional surveys or occasionally on seasonal or multiyear sampling. In shallow lakes of the Pantanal wetland, median dissolved CO2 concentrations are about 100 μM with a range from near 0 to 280 μM [Hamilton et al., 1995]. Most of the tropical lakes surveyed by Kosten et al. [2010] were supersaturated with values up to about 164 μM. In a shallow interfluvial wetland near the upper Negro River, dissolved CO2 concentrations obtained over a year averaged 390 μM with standard deviation of 213 μM [Belger et al., 2010]. In Balbina Reservoir on the Uatumã R. in central Brazil, concentrations of dissolved CO2 in surface waters ranged from 42 to 180 μM during a year of sampling [Kemenes et al., 2011]. Roland et al. [2010] reported CO2 concentrations from Manso and Funil reservoirs, located in the cerrado of Brazil, ranged from 13 to 90 μM in the dry season and from 10 to 100 μM in the rainy season. Collections from the epilimnion at a representative station in Petit Saut Reservoir (French Guiana) ranged from 4 to 773 μM over a multiyear period [Abril et al., 2005].

[33] Based on eddy covariance (EC) measurements Jonsson et al. [2008] have demonstrated that conservative estimates are produced using the widely adopted Cole and Caraco [1998] model and suggested another wind-based regression. In a reanalysis of the Jonsson et al. [2008] data, MacIntyre et al. [2010] demonstrated the additional dependence of k600 on buoyancy flux and the applicability of the surface renewal model as a mechanistic approach to use in lakes. k600, EC computed in this study are almost twice as high as those computed following Cole and Caraco [1998]. Estimates of k600, SR(zAML = z 0.5 m) were essentially identical to k600, EC, which implies that the coefficient c1 we used applies for the case of a shallow mixing depth. In the absence of actual zAML measurements, our analysis demonstrates the approximate seasonal ranges for k600, SR are likely to vary for L. Curuai: 50% during high water, when zAML undergoes larger diel variations; 30% during midstage; and 4% during low water. However, additional effort is required to understand controls on the coefficient, c1.

[34] During periods with winds sustained above 5 m s−1, with significant wave heights exceeding ∼30 cm, Wanninkhof [1992] provides better estimates of k600 than estimates based on computing turbulence with a surface energy budget (S. MacIntyre, unpublished data, 2010). However, his equation underestimates k600 for winds less than 5 m s−1. Given the large size of L. Curuai and the potential for significant wave development, we have computed k600 using Wanninkhof [1992] for the periods with winds sustained above 5 m s−1 for longer than one hour. Such situations occurred about 24% of the time in the four hydrological phases, and k600 estimates during those periods were on average 26, 47, and 29% greater than estimates of k600, SR(zAML = z mean) during receding, high, and rising water, respectively. At low water the active mixing layer remained shallow, due to constraints of reduced lake depth and high turbidity, and estimates of k600 following Wanninkhof [1992] were on average 5% lower than estimates of k600, SR(zAML = z mean).

[35] The gas exchange coefficients based on Cole and Caraco's [1998] wind-based model and those computed following MacIntyre et al. [2010] are two and four times higher, respectively, than the conservative value (k = 2.7 ± 1.0 cm hr−1) used for a regional assessment of rivers and floodplain environments [Richey et al., 2002]. Therefore, it is likely that outgassing of CO2 from lakes in the lower Amazon floodplain is at least two to four times higher than previously estimated. Furthermore, Alin et al. [2011] reported gas exchange coefficients for flowing waters, including large rivers, in the Amazon basin that are similar to those we calculated.

[36] Richey et al. [2002] estimated a rate of outgassing of CO2 from rivers and wetlands of the central Amazon basin, when converted to a daily rate, of 190 ± 55 mmol CO2 m−2 d−1. We estimate mean (±standard deviation) fluxes from open-water in L. Curuai to the atmosphere of 244 ± 15, 348 ± 13, 371 ± 23, and 364 ± 20 mmol CO2 m−2 d−1 during receding, low, rising, and high water, respectively. The error associated with these values reflects, for each hydrological phase, the spatial variation in CO2 concentration in L. Curuai, a likely range in atmospheric CO2 levels and temporal variations in kEC within the 10-day periods. We used these variations in a Monte Carlo error analysis to determine the variability in areal evasion estimates. These values would be higher if the likely increased concentrations at the air-water interface resulting from nocturnal respiration and mixing were included [Crill et al., 1988; Åberg et al., 2010; Laurion et al., 2010].

[37] The annual water level oscillations affect the dynamics of CO2 evasion from Amazon floodplains, and predictable biological response to fairly regular hydrological seasonality is expected. However, inter-annual variation in river discharge and annual variability in seasonal timing of peak and minimum stages may cause variations not captured by a one year study. The use of autonomous CO2 sensors [e.g., Lynch et al., 2010] and thermistor chains would produce valuable time series and detect important episodic events for further understanding of CO2 emission from the Amazon floodplain. Additionally, this study focused on the open-water environments of L. Curuai, which account for about 72%, 63%, and 65% of the total flooded area of this floodplain during low, medium (receding and rising), and high water stages, respectively. In order to appraise total aquatic CO2 evasion additional measurements would be needed from flooded vegetated environments.

6. Conclusions

[38] The open-water environment of large floodplain lakes has high spatial variability in CO2 concentration with higher concentrations in the littoral and at locations influenced by biological processes and incoming Amazon R. water. Based on results from this study, CO2 evasion from large lakes in the lower Amazon floodplain will be at least two to four times higher than estimated in current regional calculations. The effects of turbulence induced by wind-mixing and convective cooling, morphology, water level change, habitat type and habitat inundation status, and lake metabolism are all important factors that determine the variations on biogenic gas production and emission rates. Ongoing development of coupled models of lake inundation hydrology, hydrodynamics and biogeochemistry will offer the spatial-temporal resolution needed for resolving carbon dioxide emissions, and the other fluxes of carbon within floodplain ecosystems.

Acknowledgments

[39] We are grateful for the funds for the field campaigns granted by Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) (2003/06999–8). Meteorological buoy was funded by Conselho Nacional de Desenvolvimento Científico e Tecnológico (55.0301/02–0) and FAPESP (2002/09911–1). We thank the support provided by the NASA LBA-ECO and Rede GEOMA programs. During the period of manuscript preparation, C. Rudorff was funded by the Ministry of Education of Brazil through a CAPES/Fulbright Award (1705-07-5). Funding was also provided by U.S. National Science Foundation Division of Environmental Biology (grants 0640953 and 0919603). We thank Waldeir Amaral Vilela for his technical assistance.

Ancillary