Reliable reference for the methane concentrations in Lake Kivu at the beginning of industrial exploitation

Dissolved methane in Lake Kivu (East Africa) represents a precious energy deposit for the neighbouring countries, but the high gas loads have also been conceived as a threat by the local population. This is especially the case when stratification in the lake is changed during the planned industrial exploitation. Both 15 issues require accurate and reliable measurements of dissolved gases and temporal changes to take responsible action. Previous data fulfilled these requirements only unsatisfactorily. Prior to our measurements, there was considerable disagreement about prognosticated new formation of methane. We show how measurement accuracy could be significantly improved by implementing equipment, which was especially designed and modified for the complex gas conditions in Lake Kivu. From 150m to 430m depth, samples were taken to 20 determine the amount of dissolved methane and dissolved carbon dioxide more reliably and more accurately. Beyond the provision of gas concentration profiles at the beginning of exploitation, this investigation should also provide methods to survey the further evolution of gases over time. The use of gas tight sampling bags produced highly reliable and accurate measurements. Our measurements confirmed the huge amount of stored methane, but they do clearly not support the current believe of a significant recharge beyond diffusive loss. Direct 25 measurements with a custom-made gas pressure sensor indicated no imminent endangerment through limnic eruptions. A further survey of gas pressures, however, is mandatory to detect changing conditions. With sampling bags and gas pressure sensor, we introduced reliable and highly accurate measuring approaches for the survey of the further development of gas concentrations. This equipment only requires little effort for calibration, which can also be accomplished in remote areas of Africa. 30


Introduction
Lake Kivu, located on the border between Rwanda and the Democratic Republic of the Congo in Central and East Africa, contains large amounts of dissolved methane (CH 4 ) in its permanently stratified deep water (Tietze 1978, Tietze et al. 1980, Schmid et al. 2005. It hence represents an important resource for the neighbouring countries, especially Rwanda, which currently has no access to any other hydrocarbon deposit in its territory. The 5 commercial-scale exploitation of this resource was started on 31 December 2015 with the commencement of the KivuWatt power plant operation Phase 1 with an installed capacity of 26 MW. Further power plants are planned on both sides of the border. For a responsible handling, observation and the survey of the management prescription, a reliable reference of gas content and the development of suitable measurement equipment for documenting its temporal evolution are mandatory. 10 In addition to methane, large amounts of carbon dioxide are dissolved in the deep water. Together, these gases potentially pose a risk for the local population, as spontaneous degassing could be feared, where large amounts of gas could be released from the lake with catastrophic consequences for the local population as it happened at Lake Monoun in 1984(Sigurdson et al. 1987 and Lake Nyos in 1986 (Kling et al. 1987, Kusakabe 2015. 15 Previous measurements indicated rising gas concentrations in Lake Kivu and hence an increasing risk over time scales of few decades (Schmid et al. 2005). To avoid any endangerment for the population, a management prescription for withdrawal depth and deposition of partially degassed deep water and wash water needed to be developed to avoid damage to the lake ecology and endangerment of the local population. 20 There have been a number of measurement campaigns dating back as far as 1935 aiming at quantifying the methane deposit in Lake Kivu (Damas 1937, Schmitz and Kufferath 1955, Degens et al. 1971, 1973, Tuttle et al. 1990). Either samples were locked in containers and recovered to the surface or hoses were used to bring deep water in a continuous flow to the surface. All of them, even the more recent measurement trials, had to struggle with the loss of gas and water while recovering samples (Tassi et al. 2009, Pasche et al. 2011. So far, only one 25 published measurement series used in-situ sensors (Schmid et al. 2005).
Currently, the gas content of Lake Kivu is mainly based on the gas concentrations measured in November 2003 by the team of Michel Halbwachs (published in Schmid et al. 2005). The temporal evolution, i.e. a possible recharge has been quantified from the difference to measurements by Klaus Tietze in 1974/5: Schmid et al. (2005 concluded that CH 4 concentrations increased by up to 15% within three decades, potentially leading to an increased risk of a gas eruption. According to this prognostication, action to release the gas pressure would have been required to avoid an endangerment of the local population. However based on a more detailed analysis of the carbon budget of the lake, Pasche et al. (2011) concluded that the concentrations most likely were not increasing as fast.

5
In conclusion, the data availability was not sufficient to take responsible measures for the exploitation of the gas resource and the risk assessment during exploitation. Hence, the Rwandan government took action to invite specialist teams to implement their measurement approaches in Lake Kivu to get a data base reliable enough for political decisions (e.g. Wüest et al. 2012). Various approaches hence were modified for the special conditions in 10 Lake Kivu, each requiring considerable effort to meet the expectations (Schmid et al. 2019). In this paper, we present gas measurements from waters that have been collected in sampling bags and an analysis of the gas composition by gas chromatography. An intercomparison with competing approaches (Schmid et al. 2019, Grilli et al. 2014, Bärenbold et al. 2020 will be published in separate where also the complex conversion between gas pressures and gas concentrations will be done. 15 We felt that, for the special case of Lake Kivu, a reliable measuring method must be implemented that is suited for the local scientific staff to document changes in the gas charge at a later time. Only measuring techniques should be included that required few calibration that could be reliably done at a remote location as Lake Kivu. The approach should be comprehensible and avoid any hidden errors. In conclusion, we modified the sampling 20 method from the gas charged mining lake Vollert-Sued (Horn et al. 2017) and Guadiana pit lake (Sanchez-Espana et al. 2014 for the conditions in Lake Kivu. In addition to methane and carbon dioxide concentration, we also included measurements of dissolved solids in this paper. Finally, we measured total dissolved gas pressure (TDGP) using a customized sensor (for details see section 2.4), since TDGP is the appropriate measurement for judging the proximity to spontaneous ebullition.

5
The measurement campaign took place from 9 to 13 March 2018 near Gisenyi/Rubavu at the Northern shore of Lake Kivu in Rwanda. Sampling was mainly accomplished from the convenient sampling platform ("GIS" in were not expected to influence comparability, as horizontal transport was much faster than vertical transport and gas production or consumption, as indicated by Wüest 2012, Ross et al. 2015). Therefore, it was reasonable to assume horizontal and temporal steady-state for the duration of the campaign. In addition, a few samples from an earlier campaign in 2017 were included in this paper.

5
Vertical profiles of temperature, electrical conductivity, oxygen concentration, pH and turbidity were measured at the platform and the deep location using a multiparameter probe CTM1143 (Sea and Sun Technology, Germany).
Sensor properties and description can be found on www.sea-sun-tech.com.
Here and in all following methods, pressure (bar) was converted to depth (m) by dividing through 0.0978 bar m -1 .

10
Temperature compensation of electrical conductivity to 25°C was done using the equation In 2018, the oxygen zero point was slightly off calibration (by +0.37 mg/l), which was corrected after sampling. For gas measurements, gas tight sampling bags (TECOBAG; see Horn et al 2017) were lowered to the investigation depths and partially filled with water by operating a pump for a short period. Enough remaining capacity of the bag was retained to accommodate the total amount of gas when bags were recovered and thus pressure was reduced to atmospheric level. Other than in previous implementations of this technique in Lake Vollert-Sued, Germany (Horn et al. 2017) and Guadiana Pit Lake, Spain (Boehrer et al. 2018), the pumps were 5 switched on and off by a submersible controller (see Fig. 2). For all samplings, a CTD probe (Sea and Sun Technology; CTM102) accompanied the sampling equipment for an accurate sampling depth determination.

Water sampling for gas analysis
After filling, sampling bags were transported to the LKMP laboratory and left there over night to equilibrate gases between gas phase and liquid phase.

10
The volume of water was measured by weighing the bags on the laboratory weight scale of LKMP (subtracting the weight of bags) and dividing by density (e.g. Moreira et al. 2016). The volume of the gas space was measured thereafter by withdrawal through syringes from the bags. Part of the gas withdrawn from the sampling bags was introduced into a gas chromatograph (GC, Perkin-Elmer Clarus 580) in the LKMP laboratory to detect the gas composition quantitatively. Calibration was performed with dry gas standards of composition 20% N 2 , 40% CH 4 ,

15
40% CO 2 at an accuracy of 0.7% or better. In all samples, the sum of all detected gases amounted to around 97%.
The remainder of undetected 3% corresponded very well with the expected moisture at laboratory temperatures and hence were attributed to water vapour. The measured concentrations of CH 4 and CO 2 were multiplied by the gas volume to yield the amount in the gas space. To determine the entire amount of CH 4 in the sample, the residual portions dissolved in the water were calculated assuming equilibrium between the water and gas phase at . (see Sander, 2015; in units (mol/l)/atm; temperature as absolute temperature in Kelvin). As the gas volume was of the same order of magnitude as the water inside the bag, this contributed a few percent to the total amount of CH 4 .

25
Due to its high solubility however, a considerable portion of CO 2 remained in solution, and hence required a more accurate determination. The Henry coefficient for freshwater • 0.043 • • . (Sander, 2015;units as above). Only in the case of CO 2 , the dependence of the Henry coefficient on electrical conductivity (of few percent in the considered range) was included in the calculation: 30 based on electrical conductivity C 25 at the sampling depth (see Fig. 2) the Henry coefficient was interpolated between freshwater (C 25 = 0 mS/cm; Bunsen coefficient of 0.7519) and seawater (C 25 = 53 mS/cm; Bunsen coefficient of 0.6406) by implementing the factor C 1 /53 * 1 640.6/751.9 (given by Sander 2015, Murray and Riley 1971, see also Boehrer et al. 2016). In addition, the concentration of bicarbonate in the water phase was calculated based on pH-measurements in the laboratory by inserting a sensor after the volume 5 and GC measurements were completed. A dissociation constant of pK 1 = 6.2 was used (Cai and Wang 1998).
Hence, we could evaluate the dissolved inorganic carbon DIC as the sum of three contributions: CO 2 in gas space, CO 2 dissolved in water, and HCO 3 -; CO 3 2is practically not present at the given pH. CO 2 concentrations in the lake were calculated by adding the CO 2 gas volume in the sampling bag and the dissolved amount. The tiny contribution coming from bicarbonate shifting to dissolved CO 2 could be quantified (relative contribution of 10 -5 ) 10 on the base of pH change from the field (multiparameter profiles) to the laboratory measurement. Hence it could have been neglected without increasing the expected error.
The accuracy of the measurements was estimated from the following contributions: the accuracy of volume measurement of gas was within 4% (above 250 m 6 % due to smaller volume) with a syringe; Henry coefficients 15 for fresh water (and salt water) were known within 5% and temperature fluctuations in laboratory allow for additional but less than 5%. Finally we need to include an unknown variation of the Henry coefficient with Kivusalts of less than 5%. We added the single contributions as independent errors. This resulted in an 8% error for the Henry coefficient. As less than half of the CO 2 remained in solution, this contributed less than 4% to the expected error. In the case of CH 4 , only a small fraction remained in solution, and the corresponding error 20 contribution was less than 1%. The error in the measurement of the mass (and volume) of the sample was smaller than 1%. Altogether, we expected a precision of the measurement of 5% for CH 4 (7% above 250 m depth) and 6% for CO 2 (8% above 250 m). If required, the accuracy could be improved by implementing a better volume measurement of the gas. In the case of CO 2 , a more exact knowledge of the Henry coefficient would help as well.Gas tight sampling bags for CH 4 and CO 2 were a precondition, which was very well accomplished for the 25 investigated gases in the bags used (see Horn et al. 2017).In conclusion, the sampling bags represented a simple approach, which required careful sampling and a good understanding of the solubility of gases for the processing of data. Results, however, were reliable as the approach did not provide much space for hidden errors. 9

Total gas pressure
For gas pressure measurements, we used a Pro-Oceanus probe, which was especially customized for the application in Lake Kivu, as standard sensors for total dissolved gas pressure (TDGP) were not suited for the deployment depths and/or expected gas pressures in Lake Kivu. The pressure sensor contained a small gas 5 measurement volume separated from the lake water by a membrane, which was permeable for dissolved gases such as CH 4 , CO 2 and N 2 . The partial pressures of all gases in the measurement volume adjust to equilibrium with their concentrations in the water. If the total gas pressure inside the gas space equals the outside pressure (hydrostatic plus atmospheric pressure), a virtual gas bubble withstands the pressure at depth and persists long enough to start moving upwards through the water column through its own buoyancy. Hence, the ratio of the total 10 pressure inside the measurement volume compared to absolute pressure represents a quantification of the proximity of lake water to spontaneous ebullition (e.g. Schmid et al. 2004. Before deployment, the sensor needed to be immersed in water for several hours at a total pressure above surrounding gas pressure. Thereafter, a response time of only few minutes was expected. Measurements were 15 performed from the platform at discrete depths where the probe was left for about 20 to 30 min (Fig. 3). Data were recorded continuously, and from the time series at 286.1 m depth, a response time of less than 4 min (halfvalue time t 1/2 of 150 s) was determined by fitting an exponential curve to the observations. As the available time for measurements was limited and also the recovery of the sensor required time, measurements were done at seven discrete depths. The CTM1143 probe accompanied the TDGP sensor for an accurate depth reference.

20
In order to compensate for the measured response time, the total gas pressure TDGP at time t was calculated from the measured pressures p meas at the times t and t -t 1/2 .

TDGP t 2 /
(2) TDGP approached the final value much faster than the original readings p meas . The expected error of the 25 measurement was estimated from calibration error ( <0.5% of actual measurement) plus 0.04 bar uncertainty from reading.

Multiparameter profiles
The profile on 13 th March 2018 showed typical conditions for the wet season, where the surface layer of the lake (i.e. the top 60 m, which can undergo seasonal mixing during the dry season) is thermally stratified (Fig. 3).

5
There was a steep oxycline with strongly decreasing oxygen concentrations between 25 and 40 m depth, and below about 45 m depth, the water column was anoxic. Below 60 m depth in the monimolimnion of this meromictic lake (Boehrer and Schultze 2008, Gulati et al 2017, the profile showed the usual stepwise increase in temperature and conductivity and decrease in pH as documented in earlier measurements. In the deep water, the biggest gradient in temperature and electrical conductivity at 260m represents the upper limit of the resource zone. Below, concentrations are high enough for industrial exploitation. pH values, which are crucial for DIC calculations, are nearly identical to values shown in the work by Schmid et al. (2005).  Table 1.
Here: Table 1 and Table 2 20

Total gas pressure
Total dissolved gas pressures increase with depth, reaching more than 17 bars at depths greater than 400m (Table   2 and Fig. 4). Parallel to the measured gas concentrations, the major gradient lay between 230m and 290 m, i.e.
between the potential resource zone and the resource zone and hence followed the shape of the methane and the 5 carbon dioxide curve. At any depth, TDGP was much lower than absolute pressure. Within the resource zone, any water parcel needs to be lifted vertically by at least 150m that its gas pressure would overcome absolute pressure to start spontaneous ebullition. At about 350 m depth, gas pressure reached about 17 bar, which was about half the absolute pressure at this depth. Gas pressures would need to double to form bubbles spontaneously.

Stratification
The profile measurements of temperature, electrical conductivity, oxygen and pH confirmed the picture that was known from previous measurements. Seasonal variability extended down to about 60m depth, which could be 15 seen in the differences of the temperature profiles. Down to this depth, the recirculation also supplied dissolved oxygen, while below, a very constant picture was sustained: the stepwise structure of the deep water (monimolimnion), which could be clearly recognized by sharp changes in temperature and electrical conductivity. This concurred with our expectations from the estimated renewal time of close to 1000 years in the deeper monimolimnion . Below 50m depth, the measurements showed no temporal variability in temperature, electrical conductivity in 25 neither pH nor oxygen. Small changes in the temperature profile appeared at around 180m. Beyond this, the measurements confirmed the situation as it was known from earlier campaigns.

Comparison with earlier gas measurements
Measured concentrations of CH 4 and CO 2 were compared with previous observations by Schmitz  , and by Schmid et al. 2004(Published in Schmid et al. 2005. For the management of the lake, both considering commercial CH 4 extraction and the risk of a gas outburst, the most important depth range was the resource zone below the main gradient. In this range, the concentrations from our campaign were between those measured by K. Tietze in 1974/5 and those measured by M. Halbwachs and J.-C. Tochon in 2003 (see Fig. 5).

10
For CO 2 , the concentrations agreed well with previous measurements. Above 260 m and below 400m, the measurements lay very close to the Tietze and the Halbwachs data; only between 260m and 380m our data showed lower values (Fig. 6) with differences marginally above the expected accuracy of this campaign.  Schmid 2004Halbwachs 2003Tietze 1974/5 Schmitz 1952 production in the resource zone of 93 g C m -2 yr -1 (0.18 km 3 yr -1 ) as postulated by Pasche et al. (2011), which would correspond to a growth in concentrations by approximately 0.3% per year. In conclusion, temporal changes in gas concentrations lay below the detection limit at the accuracy of the current and the previous gas measurement methods.

10
However, a certain amount of methane had to be produced to balance the continuous loss from the lake by upward transport and subsequent oxidation in the surface layer or (to a small extent) emission to the atmosphere (Pasche et al. 2011, Borges et al. 2011, Roland et al. 2018 to 32 g C m -2 yr -1 (0.06 km 3 yr -1 in the resource zone and 0.013 km 3 yr -1 in the potential resource zone) for steadystate. Pasche et al. (2011) estimated it marginally higher at about 35 g C m -2 yr -1 . 5

Total gas pressure
The in-situ measurements of TDGP indicated that spontaneous ebullition from the water will not happen in near future: the difference from the measured gas pressure to the absolute pressure at any depth showed a big safety margin. As also the gas concentrations had not changed significantly over the last decades, an abrupt increase 10 towards dangerous conditions was not indicated by our data.
The Oceanus-Pro custom-made sensor for Lake Kivu showed a convincing performance. After proper preconditioning, measured response time based on in-situ data from Lake Kivu lay just below 4 min (half-value time was 150 s). This was faster than any previously published similar measurement, but still too slow for straight 15 profiling. However, we could prove that sufficient data points could be measured in the resource zone (and the potential resource zone) with acceptable effort. Such a spatially better resolved profile would be feasible and hence are mandatory as reference for the temporal evolution.
In addition, the accuracy of the TDGP measurement only depended on the calibration of the pressure sensor, 20 which could be accomplished at high accuracy. Hence this direct in-situ measurement would probably be the first to indicate changes in gas concentrations, if proper measurements were done at regular intervals. In addition, such gas pressure measurements are the proper indicator for endangerment by limnic eruptions and hence could serve both purposes.

25
Gas pressure had to be attributed to volatile solutes. The major contributors were methane, carbon dioxide and nitrogen. While we had good measurements of methane and carbon dioxide, we missed data of similar quality for nitrogen: hence, we could not give a proper calculation of gas pressure from gas concentrations. In addition, gas pressures at extreme concentrations as in Lake Kivu and high absolute pressures would show a complex behaviour (see fugacities), and an accurate conversion of gas concentrations to gas pressures would go beyond 30 the purpose of this paper.
A quick calculation, based on Henry coefficients at normal pressure (see above section 2.3), attributed about 12 bar gas pressure to methane at 400m depth and about 2.6 bar to carbon dioxide; nitrogen was unknown, but

Conclusions
We could present a sampling method for dissolved gases in Lake Kivu, which outperformed all other previously 15 applied techniques in relation to accuracy, applicability and reliability. The required equipment could be cheaply purchased and implemented by the local scientific personnel in the existing facilities. The current accuracy (of 5 % for methane and 6% for carbon dioxide in the deep water) was sufficient for most purposes, but could be easily and significantly improved by a better volume measurement.

20
The new measurements of dissolved gases fell within the range of previously published data. The methane measurements showed slightly higher concentrations than what was measured in 1974 (Tietze 1978, Tietze et al. 1980) but significantly lower than measurements from 2003 by Halbwachs and 2004 by Schmid et al. (2005).
The stratification of deep water showed no significant changes since scientific investigations of corresponding accuracy have been documented. Temperature, electrical conductivity, oxygen and pH in the deep water showed 25 only little temporal variability. Only the upper 50 m showed dynamic seasonal behaviour.
Rising methane concentrations as postulated on the base of earlier measurements were clearly not happening at rates estimated previously. Hence once exploited, a further recharge at a considerable rate is questionable on the base of the current data. If concentrations are rising, the rates are so low that they cannot be quantified due to the limited accuracy of previous (and the current) gas measurements. A comparison with earlier data suggests that Lake Kivu has been close to a dynamic equilibrium, where methane from decomposing organic material and reduced carbon dioxide replaces the amount continuously lost by diffusion (and mixing) to shallower layers of the lake. The current measurements do not indicate the necessity of expanding exploitation to prevent limnic 5 eruptions.
For safety assessment, direct measurements of gas pressure are feasible with the used prototype or instruments of similar design. These data state quantitatively how large the safety margin to possible spontaneous ebullition is.
The presented sensor is fast enough that high resolution profiles are feasible in an acceptable time frame. As 10 pressure sensors can be calibrated at high accuracy, direct gas pressure measurements offer the fastest perspective to detect changes in gas load of Lake Kivu.