Estimation of greenhouse gas emissions of a tropical reservoir in Colombia

Tropical reservoirs are generally flooded in soils with a high content of organic matter. This, combined with high temperatures, favors the generation of carbon dioxide (CO2) and methane (CH4) by biological degradation, contributing to the impact on climate change. A tropical reservoir in Colombia was monitored for 7 years in the pre-fill, fill and post-fill stages, for the last of these during the day and night. Emissions from diffusive fluxes at the surface of the water were measured using a floating static chamber, while inverted funnel methodology was used to measure the fluxes by bubbling. The samples collected in the field were analyzed in the laboratory using a gas chromatograph with a mass detector. The results showed average emissions of 70,892.51+ 41,079.16-ton CO2eq/year for pre-filling; 178,254.53+ 105,838.01-ton CO2eq/ year for filling; and 466,946.57-ton CO2eq/year for post-filling (for 5 years), concluding that the weather conditions and the filling percentage (Area surface and volume) had an impact on the generation of greenhouse gases at filling and post-filling stages, as did the organic matter present in the area of influence of the sampling point. Higher greenhouse gas emissions were found during the day compared to the results at night, indicating that temperature affects these processes, especially in tropical reservoirs. This study, currently unique in Colombia, will allow directing efforts towards mitigating the impacts of greenhouse gas emissions in tropical reservoirs.


Study area
The study was carried out in a reservoir located in Colombia, in the department of Santander (Figure 1), 75 km upstream from the mouth of the Magdalena river and 62 km downstream from the confluence of the Suárez and Chicamocha rivers. The average temperature of the area ranges from 24.6 to 32.0°C in the day and 21.4 to 26.7°C at night (IDEAM 2017), with slight increases in dry or transitional seasons. The average windspeed fluctuates between 0.1 and 0.3 m/s and the average humidity percentage is 51% (Ruiz et al. 2019). The reservoir uses the waters of the River Sogamoso in the generation of electricity through a 190 m high dam, a surface of water body of 6,960 ha, a total volume of 4,800 m 3 and an underground powerhouse with three generating units. With 820 MW of installed capacity and an average annual generation of 5,056 GWh-year, it is the hydroelectric plant with the four highest installed capacities in Colombia. In all sampling campaigns, data on temperature, precipitation, filling and height of the water sheet were taken into account in order to correlate the information obtained on greenhouse gas emissions. The height data were converted to area and volume using the reservoir bathymetry.

Greenhouse gas sampling campaigns
To assess the generation of GHG in the reservoir, time monitoring was performed during the pre-filling stage (in the tributaries) and the filling and post-filling stages in the reservoir (Table 1). Before filling the reservoir, nine tributary rivers were selected upstream and one river downstream of the reservoir (Table 2), in order to establish the base line of GHG emissions. For the selection of the sampling points, the recommendations defined in the GHG measurement guidelines for freshwater reserves (IHA 2010) were taken into account.

Uncorrected Proof
Pre-sampling was carried out to adjust the dimensions of the GHG collection equipment (the UNESCO/IHA GHG Measurement Guidelines recommend the type and form of the equipment, but not the dimensions), and five samples were carried out for the GHG determination (Table 1). The samplings were scheduled taking into account the hydrological behavior in R4, the main tributary of the reservoir, in order to have sampling campaigns at different times of the year, according to the annual cycle of multi-year average flows.

Filling the reservoir
Measurements during the filling of the reservoir were carried out downstream and upstream of the tributary rivers and at two points located along the longitudinal axis of the reservoir. In total, four sampling points were selected (Table 3).
The filling evaluation phase was carried out during three sampling campaigns (Table 1), with the filling percentage of the reservoir being the selection criterion for each sample. The aim was to assess how the GHG generation of this body of water varied, taking into account the baseline established in the pre-fill and the characteristics of the tributaries that supply it.

Post-filling of the reservoir
Post-fill measurements were carried out at points located along the longitudinal axis through the reservoir and near the dam (Table 4). In total, eight sampling points were selected.
The post-filling evaluation was carried out during 20 sampling campaigns between the years 2015 and 2019 (Table 1). As in the pre-filling stage, the samplings were scheduled taking into account the hydrological behavior according to the annual cycle of multi-year average flows of the study area, in order to have sampling campaigns at different times of the year. Likewise, the filling percentage was established in each sampling, in the same way as during the filling of the reservoir. Additionally, nictemeral evaluation was carried out, in which gas samples were collected to evaluate the variation of day and night emissions, following the same methodology as for daytime sampling.

Determination of diffusive emissions
The diffusive fluxes of CO 2 and CH 4 were measured at the air-water interface at the different sampling points, using the floating static chamber method (Figure 2(a)) established by the IHA (2010). The acrylic chamber was designed by the GDCON research group, with dimensions of 0.4 m width Â 0.4 m length Â 0.5 m height, and a capacity to function at a depth of 0.3 m.  The camera was submerged in the water and, by means of lateral floats, its stability was guaranteed and the movement due to the current was reduced. Two quick-closing valves were located in the upper part of the chamber; one as a purge valve to equalize the pressures before starting the measurement, and the other with a flexible hose that was connected to a gas collection pump, which extracted the gas from the acrylic box with a flow rate of 55 mL/min and stored it in Tedlar bags with a capacity of 0.5 L. The chamber had a fan inside, which allowed the air inside the chamber to be homogenized. For the measurement of diffusive fluxes, four samples were taken at each point, every 7 min, for a total of 28 min of monitoring (IHA 2010; Sousa et al. 2019). This methodology was performed in order to provide data to perform linear regressions over time and to quantify the amount of both CO 2 and CH 4 . The results were accepted or rejected following the methodology proposed by Lambert & Fréchette (2005). Finally, the correlation coefficient between the concentration of each gas versus time (R 2 . 0.90) and the analysis of variance (p , 0.05) were determined according to the methodologies proposed by Sousa et al. (2019), Guérin & Abril (2007) and Soumis et al. (2004). The diffusive fluxes in units of g/m 2 · d were determined from the ideal gases equation, using the following equation.
where P is the atmospheric pressure (atm); X is the linear regression slope of the gas collected volume versus time (L/d); V FC is the volume of the floating chamber (L); W 1 is the molecular weight of CO 2 or CH 4 depending on the calculation made (g/mol); V TB is the volume of the tedlar bag (L); R is the universal gas constant (atm L/mol K); T is the temperature (K) and A FC is the area of the floating chamber (m 2 ). For this study, a P of 0.895 atm, V FC of 32 L, W 1 of 46 g CO 2 /mol and 16 g CH 4 /mol, V TB of 0.5 L, R of 0.08206 atm L/mol K and A FC of 0.16 m 2 were considered. In the case of T, the value changed according to weather conditions. As an analytical control of the sampling, for points R1, R5, E1 and E8, samples were taken in duplicate, using two floating chambers of equal size, simultaneously. The samples were collected and taken to the GDCON group laboratory, where they were analyzed in a maximum time of 7 days, as suggested by Lambert & Fréchette (2005), using an Agilent 7890B gas chromatograph with a three-valve system and two detectors with packed 1/8' stainless steel columns (HayeSep Q 80/100), with 5975C mass detector for determination of both CO 2 and CH 4 (Hoyos et al. 2021).

Determination of emissions by bubbling
Bubble methane fluxes were estimated using the inverted funnel method (IHA 2010), with a specific design by the GDCON group. This device consisted of a 1 m diameter cone connected to a silicone hose, which in turn was connected at the upper end to a 0.5 L Tedlar bag in which the sample was collected and, at the lower end, assembled to a rigid rod that allowed the passage of gases. The cone had four weights distributed in its diameter, which allowed it to collect the sample at a depth of 4 m in the water column. Additionally, the system had a buoy that counteracted the weight, and allowed the funnel to float (Figure 2(b)).
The measurement was carried out for four uninterrupted hours, counted from the moment the equipment was placed in the water. After this, the samples were taken to the GDCON laboratory where they were analyzed in a maximum time of 7 days (Lambert & Fréchette 2005), using an Agilent 7890B gas chromatograph, with a 5975C mass detector.
The bubbling fluxes in units of g/m 2 d were determined from the ideal gases equation using the following equation.
where P is the atmospheric pressure (atm); C is the collected volume of the gas in a time of 4 h (L/d); V F is the volume of the inverted funnel (L); W 1 is the molecular weight of CH 4 (g/mol); V TB is the volume of the Tedlar bag (L); R is the universal gas constant (atm L/mol K); T is the temperature (K) and A F is the area of the inverted funnel (m 2 ). For this study, a P of 0.895 atm, V F of 68 L, W 1 of 16 g CH 4 /mol, V TB of 0.5 L, R of 0.08206 atm L/mol K and A F of 2.54 m 2 were considered. In the case of T, the value changed according to weather conditions.

Statistical analysis
To observe the spatial and temporal variation of GHG emissions at the different points and during the different samplings, an exploratory analysis of the data, box and whisker plots, and a frequency histogram were performed. Additionally, the correlations between the studied variables were determined. Statistical analysis was performed using R software version 3.5.1.

ANALYSIS AND DISCUSSION
3.1. Emissions in pre-filling  The multidimensional analysis that showed the Pearson product moment correlations between each pair of variables allowed us to statistically demonstrate the similarity between the pairs M1-M4 and M2-M5 (p , 0.05), that is, they were correlated with each other, M1-M4 being in the rainy season and M2-M5 in the dry season. In addition, in order to determine the means that were significantly different from each other, a multiple range contrast test was applied, finding an important difference in CO 2eq between the sampling pairs: M1-M2, M1-M5, M2-M3, M2-M4, M3-M5 and M4-M5, which demonstrates an influence of the time of year on the variation of CO 2eq emissions in rivers. It is important to highlight that the variability in rainfall can generate different seasonal patterns, modifying the circulation time of the water and producing variations in the intensity of the environmental processes that occur in aquatic systems. On the other hand, the external load of organic matter and other compounds in the system increases with episodes of intense rainfall (Armengol et al. 1999), due to the associated runoff. Additionally, precipitation affects CO 2 transfer, causing turbulence and thus inducing CO 2 flow through the air-water interface (Takagaki & Komori 2007).
With respect to each sampling point (Figure 4), R6 presented, on average, the highest CO 2 flow of all the points (8.050 g CO 2 /m 2 d) at the pre-filling stage. This was associated with the proximity of R6 to cattle grazing areas, which increases the concentrations of organic matter, and therefore CO 2 . On the contrary, R8 was the station with the lowest emissions (2.019 g CO 2 /m 2 d), while R4, R5 and R10 presented the greatest variability in data. Of these sites, R4 and R10 are in areas of high mining exploitation, and R5 corresponds to the main river of a municipality, to which domestic wastewater discharges are discharged. For the stations R1, R7 and R9, some atypical data were presented, which is to be expected due to the variability of the weather periods of each sampling.
The data found for the 10 rivers in this study were found in a range of 2.048 to 8.050 g CO 2 /m 2 d. When comparing these results with studies carried out by other authors (Table 5), a high variation is found in the information, particularly for tropical climates, where authors such as Guérin et al. (2006) found CO 2 values of the order of 37.796 g CO 2 /m 2 d, while Richey et al. (2002) reported an average of 8.316 g CO 2 /m 2 d, the latter value being much more in line with what was found in this study. Other investigations carried out in rivers in boreal climates have reported values in the range of 0.36 g CO 2 /m 2 d (Wang et al.  (Humborg et al. 2010), which shows that, in the case of rivers, gas emissions may be more associated with the presence of organic matter or with land uses near the areas of influence of the rivers than with the associated climate regime. Table 6 shows the CO 2 fluxes emitted for each of the filling percentages of the monitored reservoir. In this case, no diffusive fluxes of CH 4 were found, that is, during this stage, strictly anaerobic conditions were not established at the bottom of the reservoir, which would give rise to emissions of this gas.

Emissions in the filling stage
Points R1 and R4 correspond to rivers downstream and upstream of the reservoir, respectively. For point R1, the CO 2 emissions in the three samplings were very similar, indicating that the filling percentage of the reservoir had no impact on the generation of CO 2 at this point. This was to be expected at this stage, in which there was not yet energy generation that could give rise to the degassing phenomenon, which tends to increase gases downstream from the dam. On the other Uncorrected Proof hand, comparing the range of CO 2 emitted at point R1 during the pre-filling (0.555-8.163 g CO 2 /m 2 d) with the range in the filling (1.019-1.300 g CO 2 /m 2 d), we find that the values are within the same order of magnitude; therefore, there is no significant difference. Regarding point R4, located at the tail of the reservoir, a slight increase in CO 2 emissions was evidenced with the filling percentage of the reservoir, with a minimum value of 2.724 g CO 2 /m 2 d and a maximum of 4.690 g CO 2 /m 2 d for filling percentages of 25.00 and 83.40%, respectively. This is due to the increase in the water column over the plant material, which accelerates the degradation processes of organic matter and gives rise to the release of more CO 2 . During filling, the maximum average value of CO 2 emissions was at point E8 in the reservoir (10.806 g CO 2 /m 2 d), and the minimum was found at point R1 (1.175 g CO 2 /m 2 d). For all points, the highest emission values were reported for the highest filling percentage of the reservoir (83.40%), with 1.300 g CO 2 /m 2 d for point R1, 4.690 g CO 2 /m 2 d for R4, 9.195 g CO 2 /m 2 d for E3 and 15.533 g CO 2 /m 2 d for R8. The minimum CO 2 emission values were reported for the lowest percentage of reservoir filling (25.00%) in each of the points, indicating that the higher the water volume, the more flooded area and, therefore, the more decomposing vegetation generating more CO 2 . This is confirmed by the statistical analysis that showed a correlation between the filling percentage and CO 2 emissions (p , 0.05). At this stage, it was not possible to correlate the climatic variation with the CO 2 emissions, since the filling of the reservoir was carried out in 3 months, so a high seasonal variation did not occur between each sampling.
The point that presented the highest emissions in the reservoir during the three samplings carried out was station E8. This point is influenced by the discharge of organic matter from the R5 river, which is the recipient of wastewater from a nearby municipality, so the increase in CO 2 emissions is due not only to the decomposition of the flooded organic matter but also to the contribution of organic matter that enters through the R5 river. Another factor that probably had an impact on the generation of gases between one point and another in the reservoir is the use that the soil had before the flood since there were areas where there had been intensive use of livestock, which generates a greater contribution of organic matter in the area, thus increasing GHG production.
Comparing the results obtained from the CO 2 fluxes at points E3 and E8 of the reservoir, with the data reported in the literature, it is found that UNESCO/IHA (2008) and Integrated Environments (2006) report average diffusive fluxes of 9.086 g CO 2 /m 2 d in reservoirs with a tropical climate, with minimum values of 0.836 g CO 2 /m 2 d and maximum values of 19.008 g CO 2 /m 2 d. This suggests that the data found for this study (on average 8.250 g CO 2 /m 2 d for E3 and 10.806 g CO 2 /m 2 d for E8) are within this range. Figure 5 shows the results of both diffusive and bubbling fluxes during post-filling. The highest CO 2 emissions occurred at points E1 and E8, with average values of 20.508 and 22.405 g CO 2 /m 2 · d, respectively ( Figure 5(a)). E1 is the point in the dam where the gases begin to accumulate until they are finally released by the degassing process. E8 is the furthest point from the reservoir, at the point of discharge of the R5 river which, as previously mentioned, crosses a municipality where discharges of domestic wastewater occur. With regard to the lowest CO 2 emission, station E3 obtained an average value of 16.843 g CO 2 /m 2 · d, this point being on the longitudinal axis of the reservoir. Comparing these results with the literature, it is found that the values in this study are slightly higher than those reported by UNESCO/IHA (2008) andIntegrated Environments (2006) for diffusive fluxes in tropical reservoirs, which show an average of 9.086 g CO 2 /m 2 · d with a minimum of 0.836 g CO 2 /m 2 · d and a maximum of 19.008 g CO 2 /m 2 · d. In this case, it is important to highlight that the age of the reservoir plays a fundamental role in GHGs, since very young reservoirs, like the one in this study, tend to have higher emissions due to the degradation process of flooded organic matter which, over time, tend to decrease or stabilize. This phenomenon has been reported by authors such as Valle & Kaplan (2019) and Campo & Sancholuz (1998), who found that the decomposition of easily degradable organic matter occurred over a period of 10 years, after which time, decomposition was slower, with a loss of 43% of totally flooded organic matter after 37 years.

CO 2 emissions in post-filling stage
Meanwhile, regarding methane emissions, Figure 5(b) shows the variation of diffusive fluxes. In this case, as with the CO 2 flow, the maximum value was presented in E8 (0.455 g CH 4 /m 2 · d) and the minimum in E2 (0.037 g CH 4 /m 2 · d), located on the transverse axis of the reservoir. These results are of the same order of magnitude as reported by UNESCO/IHA (2008) and Integrated Environments (2006) with an average of 0.410 g CH 4 /m 2 · d and a minimum and maximum of 0.0048 and 0.816 g CH 4 /m 2 · d, respectively. Comparing these results with studies of other reservoirs, both tropical and boreal (Table 7), it is found that the maximum value in this study is much higher than those reported elsewhere, with the exception of the Balbina reservoir, where a value of 1.128 g CH 4 /m 2 · d was reported. Finally, CH 4 emissions by bubbling (Figure 5(c)) were higher at points E1, E7 and E8, with values of 0.107, 0.264 and 0.274 g CH 4 /m 2 · d, respectively. These correspond to the point of the dam (E1) and the furthest points in the longitudinal axis of the reservoir (E7 and E8), where in addition to the contribution of organic matter that reaches them through the rivers, there are also laminar flows conditions, with few intrusive currents and a prevalence of advective phenomena, which favors anaerobic conditions.
Regarding the generation of GHG from the reservoir according to the level of filling (Figure 6), it was found that there is a linear relationship (p , 0.05) with a correlation coefficient (R 2 ) of 0.8343. Therefore, it can be inferred that the higher the filling level of the reservoir, the greater the emissions generated by the system. This is also correlated with the weather season, since the highest filling percentages occurred in the rainy and transition season, with an average of 81.56%, and  Uncorrected Proof the lowest (60.33% average) in the dry season. Likewise, the water level in the reservoir is a determining variable in the GHG emission processes, and it is recommended that the filling percentage of the reservoir should remain below 75% in order to minimize gas generation.
In any type of reservoir, during the first years after filling, GHG emissions tend to increase due to the rapid decomposition of flooded organic matter, mainly plant material. In this study, emissions increased from 2015 to 2019 (Figure 7), but these increases were less in 2015 and 2016, compared to 2017 to 2019. This indicates that, during the first two years of operation of the reservoir, the decomposition process of the plant biomass was slow, which is associated with the breaking of the polymer chains present in the plant material. According to García et al. (2010), the plant material has heterogeneous chemical compounds that include structural polymers of the cell wall (proteins, cellulose, hemicellulose and lignin) and soluble materials (simple sugars, amino acids, oils, waxes and phenols). Therefore, much of the organic matter present in plant biomass has high molecular weights. Organic matter of this nature generally has a first stage of microbial decomposition, called enzymatic hydrolysis, which occurs before biodegradation to low-intermediate organic metabolites of low molecular weight begins. These metabolites then produce CO 2 -CH 4 (anaerobic processes) or CO 2 (aerobic processes). In 2015 and 2016, intermediate metabolites could have been formed, so the increase in emissions from 2017 to 2019 could be due to the decomposition of low molecular weight metabolites, such as amino acids, disaccharides, that can degrade faster.

Uncorrected Proof
On the other hand, comparing the baseline (pre-fill) with the post-fill (Table 8), it is found that the CO 2eq emissions in the pre-fill stage were of the order of 70,892.51-ton CO 2eq /year, and in the post-fill stage 278,111.43-ton CO 2eq /year on average. This shows the incidence of flooded plant material and organic matter present in the water body, which contribute to the formation of GHG. It is important to highlight that the deviation of the data is relatively high, considering that different sources were monitored in the pre-filling stage, which contained variable concentrations of organic matter, total solids and nutrients, factors that affect microbial dynamics and in turn accelerate or decrease the gas emission process. In addition, all the monitoring was carried out in different climatic regimes, which has an impact on the rate of degradation of organic matter and on the transfer of gases in the air-water interface.
Taking stock of the emissions in the pre-fill (baseline) and the post-fill stages during the years 2015 to 2019 (Table 8), average net emissions were established in the post-fill stage of 396,054.058-ton CO 2eq /year. Considering that in the pre-fill stage this figure was 70,892.510-ton CO 2eq /year, there was an increase in gas emissions over the years in the reservoir due to the degradation of the flooded plant biomass and the organic load present in the tributary rivers.
To evaluate the mitigation potential of the buffer zone with respect to GHG emissions from the reservoir in 2019 (Table 9), the carbon balance of the system was calculated. On the one hand, the emissions of the reservoir correspond to the average emission of the reservoir in 2019 (654,642.72-ton CO 2 eq/year) minus the emissions of bodies of water in the pre-filling (year 2013; 70,892.59-ton CO 2 eq/year). While the mitigation potential of the buffer zone by aerial biomass corresponds to the amount of CO 2 eq captured by aerial biomass (137,101.26-ton CO 2 eq/year) minus soil emissions (19,680.33-ton CO 2 eq/ year) in 2019. In this way, the balance is 466,328.28-ton CO 2 eq/year, which indicates that the protection band counteracts approximately 20% of the emissions produced in the reservoir.
Studies carried out by various authors have shown the high variability of CO 2eq emissions in tropical reservoirs (Figure 8), where factors such as the area and age of the reservoir constitute an additional variable in the process. Authors such as dos Santos et al. (2017), Demarty & Bastien (2011), Kemenes et al. (2007) and Rosa et al. (2002) carried out the analysis of nine tropical reservoirs with areas from 60 to 2,360 km 2 and ages from 1 to 36 years located in Brazil and reported GHG emissions of the order of 87,000.00-ton CO 2eq /year for the Segredo reservoir -2,433,000.00-ton CO 2eq /year for the Balbina reservoir. On the other hand, in the case of Colombia, where there are no reports of measurements in the reservoir, authors such as Cuadros et al. (2019) applied a model to predict emissions in 12 reservoirs with areas from 3.4 to 150 km 2 and ages between  (Figure 8). However, compared to other reservoirs with similar areas, the reservoir in this study has higher emissions. This is attributable to the fact that this reservoir is relatively new, so the plant matter is in a high state of degradation. It would be expected that, with increasing age, emissions from the reservoir will decrease in line with the same reduction in the organic matter content and with respect to the other reservoirs  Uncorrected Proof in Colombia, it is important to highlight that in the study of Cuadros et al. (2019) only the diffusive emissions were taken into account, without including those of bubbling and, furthermore, they constitute purely theoretical results. Finally, comparing the day and night samples, it is concluded that gas emissions are greater during the day than at night, a situation that occurred in the samples of the years 2016, 2017, 2018 and 2019 (Figure 9). Author as Herrera et al. (2013) found variations around 21-52% in gas emissions between day and night in a tropical reservoir, with those in the day being greater than those at night, and the highest emissions being specifically at 12 pm. They associated this situation with the difference between day and night temperatures since lower temperatures affected the metabolism of the microorganisms and therefore slowed down the degradation process. In this regard, in the case of this reservoir, the differences in day and night emissions were in the order of 6.3, 2.2, 14.8 and 16.4% for the years 2016, 2017, 2018 and 2019, respectively, percentages that are in the ranges found by Herrera et al. (2013). Additionally, in the study area, the average temperatures in the day range from 24.6 to 32.0°C and at night between 21.4 and 26.7°C (IDEAM 2017), which indicates that the temperature has a direct incidence on these processes.

CONCLUSIONS
This study represents the first greenhouse gas inventory available in a reservoir in Colombia, from the beginning of its operation to its start-up and will serve as a contribution to future research in a country where the energy transition to other renewable sources is find study stage. According to the results, in the reservoir, there was a high spatial variability in GHG emissions, which was mainly related to the presence of organic matter at the sampled points of the reservoir, that is, this spatial variation is influenced by the location of each sampling point and the tributaries that reach the reservoir. There is also temporal variability in the emissions, which is related to the level of the reservoir's water column (area, volume and percentage of filling, according to the bathymetry of the reservoir), which favors processes that enable CO 2 emissions. However, it was possible to demonstrate that the registered emissions are in the ranges established for tropical climates. The fact that the results obtained did not present a constant trend could be related to what was recorded in the bibliographic review, whereby in the first years of the reservoir, emissions tend to increase and be more significant, before stabilizing in later years. However, to verify what is described in the literature regarding CO 2 emissions during the first years after filling the reservoir, it is necessary to continue taking measurements for a longer period of time, enabling the respective comparisons with other studies and the creation of predictive models of CO 2 fluxes in the reservoir.