Two maars in Cameroon, i.e., lakes Monoun and Nyos, caused natural disasters in 1984 and 1986, respectively, when CO₂ was explosively released from the lake waters and resulted in casualties in the nearby communities (Tanyileke et al., 2019). Such explosive CO₂ releases are known as limnic eruptions; the term was first used by J. C. Sabroux at the conference regarding the Lake Nyos disaster, which was organized by UNESCO and the Cameroon government at Yaoundé, Cameroon, in March 1987 (Sigvaldason, 1989; Halbwachs et al., 2004).

A limnic eruption is an extremely rare geological phenomenon; other than the two cases at lakes Monoun and Nyos, no other limnic eruptions have been reported. In addition, it is not known whether limnic eruptions occurred in lakes Monoun and Nyos before the 1980s. In general, magmatic eruptions are driven by the degassing of H₂O dissolved in magma; conversely, limnic eruptions are driven by degassing of CO₂ dissolved in lake water. During limnic eruptions, the entire lake body corresponds to the magma chamber during a magmatic eruption. Magmatic eruptions can be initiated by the depressurization of the upper part of the magma chamber; conversely, the initiation of limnic eruptions is not clear. During the limnic eruptions, lake water was blown up from the lake surface by the explosive degassing of CO₂, and CO₂ dissolved in the lake water mixed with the ambient air and formed an air mass with low oxygen concentration. This air mass diffused from the lake to the surrounding area (Costa and Chiodini, 2015) and suffocated the people (Baxter and Kapila, 1989). Magmatic eruptions cannot be artificially suppressed; conversely, limnic eruptions can be suppressed by artificially removing dissolved CO₂ from the lake water (Halbwachs et al., 2020). Research on limnic eruptions has focused on the triggering mechanism and history of limnic eruptions in lakes Monoun and Nyos before the 1980s.

The Cameroon Volcanic Line (CVL) consists of alkaline volcanoes extending from Annobon Island in the Atlantic Ocean to the interior of the African continent (Fitton and Dunlop, 1985) and has branches within Cameroon. One of the branches extends to the east, reaching the Ngaoundéré Plateau; another branch goes to the north, reaching the Biu Plateau in Nigeria. Lake Monoun (5.579,784 °N, 10.587,654 °E) and Lake Nyos (6.438,545 °N, 10.298,798 °E) are located in the central part of the CVL.

Many maars are located on the CVL, of which 39 have been investigated by Kling (1988). Except for lakes Monoun and Nyos, no other lakes have high concentrations of dissolved CO₂ in their water, thereby causing limnic eruptions (Kling, 1988). Geologically, Lake Nyos is a young maar; it formed approximately 9 ka before present (BP) according to ²²⁶Ra/²³⁰Th of the lava (Aka and Yokoyama, 2013). CO₂ in lakes Monoun and Nyos is accompanied by He with a high ³He/⁴He ratio and MORB-type Ne (Nagao et al., 2010), suggesting the existence of mantle-derived degassing magmas beneath both lakes. Nagao et al. (2010) discovered also that magmatic He-laden water was discharged into Lake Nyos at −190 m depth, i.e., 20 m higher than the lake bottom; they suggested that the magmatic He-laden water traveled along the ring fault of a potential diatreme structure (Lockwood and Rubin, 1989). The consensus is that the driving force of the limnic eruptions was CO₂ dissolved in the lake water (Kling et al., 1987; Sigurdsson et al., 1987).

To prevent the recurrence of limnic eruptions in lakes Monoun and Nyos, degassing pipes were installed in both lakes; these pipes transport deep lake water safely to the lake surface (Halbwachs et al., 2004, 2020). CO₂ is separated from the lake water and diffused into ambient air. The amount of CO₂ dissolved in the water of Lake Nyos was 14.8 Gmol in 2001 (Kusakabe et al., 2008), when degassing started; by 2015, the amount had been reduced to 3.9 Gmol (Ohba et al., 2015). In Lake Monoun, the amount of dissolved CO₂ was 0.61 Gmol in 2003 (Kusakabe et al., 2008), when degassing started; by 2007, the amount had been reduced to 0.25 Gmol. The force transporting lake water through degassing pipes is the buoyancy of the CO₂ bubbles that are separated from the lake water. When the concentration of dissolved CO₂ decreases below a certain threshold, the transportation of lake water stops. Water flowing through the degassing pipe at Lake Monoun decreased significantly in 2010 (Yoshida et al., 2015), at which point the operation of the degassing pipe stopped; however, the water in Lake Monoun would inevitably become again saturated with CO₂, likely resulting in limnic eruptions. A mechanical device consisting of an electric rotary pump powered by a solar panel was installed to restore the water flow in the degassing pipe at Lake Monoun, thereby successfully resuming the degassing (Yoshida et al., 2015).

By 1987, when the conference on the Lake Nyos disaster took place, there was a consensus that limnic eruptions are caused by the CO₂ gas dissolved in the lake water; however, the mechanism of their early eruptive stages remained unknown, and several hypotheses were proposed. Sigurdsson et al. (1987) suggested that the limnic eruption in Lake Monoun was caused by sediment that fell to the bottom of the lake because of a landslide on the scarp of the lake and disturbed the lake water saturated with CO₂. Giggenbach (1990) suggested that the limnic eruption in Lake Nyos was caused by the inflow of cold rainwater into the deeper layers of the lake, thereby lifting the CO₂-saturated lake water. Kusakabe et al. (2008) proposed that limnic eruptions at Lake Monoun occurred spontaneously without any triggering events.

Woods and Phillips (1999) conducted laboratory experiments and numerical analyses regarding Lake Nyos, when its lake water contained high concentrations of CO₂; they found that the upwelling of a small amount of CO₂ bubbles at the bottom of the lake could be magnified 10⁴–10⁵ times at the lake surface, thereby resulting in a limnic eruption. Kozono et al. (2016) estimated numerically that the upwelling of small CO₂ bubbles in the middle depths of Lake Monoun could be amplified, thereby leading to a limnic eruption. Elucidating the mechanism of limnic eruptions is important for forecasting potentially catastrophic future limnic eruptions. In this study, we attempted to elucidate the limnic eruption process at Lake Monoun based on the detailed lake-bottom bathymetry obtained by Alain et al. (2019) and the original observations of CO₂-enriched lake water.

The most important material in this study was a detailed bathymetric map of the bottom of Lake Monoun. The following information is necessary for interpreting the lake-bottom bathymetry. According to Sigurdsson et al. (1987), lava flowing from the Mfomben Crater dammed the Panke River and formed Lake Monoun (Figure 1); after the formation of Lake Monoun, the Makwet scoria cone adjacent to Lake Monoun was deposited. Although the radioactive formation age of Lake Monoun has not been obtained, the age of the last eruptive activity in the area surrounding Lake Monoun may have been a few centuries ago, considering the young volcanic topography of the Makwet scoria cone (Sigurdsson et al., 1987). Lake Monoun is a part of the Panke River; the river water enters the lake through the northeast inlet, and the lake water exits the lake through the west outlet. According to Sigurdsson et al. (1987), the 1984 limnic eruption caused waves of 5 m higher than the surface, which crashed onto the eastern shore of Lake Monoun (W in Figure 1). These large waves destroyed the vegetation within 100 m from the shore.

The bathymetric map of the lake bottom shown in Figure 2 was obtained through a multibeam sonar survey conducted in November 2014 (Alain et al., 2019). An image of the basin with resolution higher than that of Figure 2 is provided as the supplementary figure. The high-resolution image was analyzed for extracting features that seemed to be related to the 1984 limnic eruption.

We obtained vertical profiles of temperature, electrical conductivity, and pH using a CTD probe (Ocean Seven Model 316, IDRONAUT) at several points in the lake in 2015. After submerging the probe in the lake water, temperature, electrical conductivity, and pH were measured every second. In situ lake-water sampling revealed the total concentration of carbonate species. Owing to the high concentration of dissolved CO₂, effervescence from deep water reaching the surface would result in loss of most of dissolved CO₂; to address this problem, a plastic syringe containing 10 ml of 5 M KOH solution was used (Kusakabe et al., 2000). This method is known as the “MK method.” The syringe was attached to a mechanical device and would be sent to the desired depth, where approximately 30 ml lake water would be obtained by the syringe. The amount of sampled lake water would be determined more accurately based on the weight difference of the syringe before and after sampling. Each sample was analyzed in the laboratory for the total carbonate content (i.e., CO_2aq + HCO₃‾ + CO₃ ²‾) using the microdiffusion–titration method (Conway, 1950). This total concentration of carbonate species is denoted as Ct. In this study, CO₂ dissolved in lake water is expressed as “CO_2aq” and was the driving force behind the limnic eruption; however, the CO_2aq concentration cannot be measured directly and was theoretically estimated from the Ct and pH values of lake water. The advantage of the MK method is its high accuracy in determining Ct values. The analytical error of the method is approximately ±2.5 mmol/L. The data from the CTD probe were combined with the analytical results of the MK method to estimate the CO_2aq concentration profile.

Based on the high-resolution images obtained from the multibeam sonar survey (Alain et al., 2019), three basins were recognized within Lake Monoun (Figure 2). The maximum depths in the eastern basin (EB), central basin (CB), and western basin (WB) were −100, −56, and −44 m, respectively (Figure 2). To define the water exchange between basins, the depth profile, as shown in Figure 3A, was obtained along line L1 (Figure 2), which passed through the saddles on the ridges separating WB, CB, and EB. The depth of the ridge separating WB and CB was −42 m, i.e., 2 m higher than the bottom of WB. Below −22 m depth, the waters in WB and CB mixed laterally (Figure 3A). The depth of the ridge separating CB and EB was −22 m, i.e., 34 m higher than the bottom of CB. Below −22 m depth, the waters in CB and EB could not be laterally mixed (Figure 3A).

The bathymetric features of EB were extracted (Figure 4) by visual inspection of the supplementary high-resolution images. Three major depressions (i.e., D1, D2, and D3) were found in the area near the basin scarp. The depths at points p1 and p1o were −98.3 and −97.1 m, respectively. The depth ranging from −98.3 to −97.1 m corresponds to the interior of D1. The depth of D1 relative to the surrounding lake bottom was 1.2 m. The depths at points p2 and p2o were −102.1 and −99.9 m, respectively. The depth ranging from −102.1 to −99.9 m corresponds to the interior of D2. The depth of D2 relative to the surrounding lake bottom was 2.2 m. The depth of D3 relative to the surrounding lake bottom was not precisely determined but appeared to be approximately 1 m or less. Besides the aforementioned major depressions, small depressions were also evident (blue spots in Figure 4). Four major mounds (i.e., M1, M2, M3, and M4) were found near the scarp of the basin. Besides the aforementioned major mounds, smaller mounds were also evident (red spots in Figure 4). Many grooves (green curves in Figure 4) were evident on the scarp of the basin. Most of the grooves were located in the northern half of the scarp. An area free of grooves was found on the northern wall of the basin (F in Figure 4). This area is adjacent to a small ridge (R in Figure 4).

Figure 3B shows the depth profile along line L2 in Figure 4. The EB was gently inclined from north to south. The cross sections along lines L3, L4, and L5 in Figure 4 are shown in Figures 3C–E, respectively. Lines L3 and L5 straddle mounds M1 and M2, respectively. The depth profiles along lines L3 and L5 revealed a raised section in the corresponding mound section (Figures 3C,E).

The temperature, electrical conductivity (C₂₅), and pH of the lake water were measured along the depth of each basin using a CTD probe, in which C₂₅ is the electrical conductivity normalized at 25°C using the following equation: C 25 = C { 1 + 0.02 ( t c − 25 ) } − 1 , (1) where C is the raw electrical conductivity measured at Tc (°C). On 28 February 2015, CTD observations were conducted at points p1, p1o, p2, p2o, and eb (Figure 4). On 1 March 2015, CTD observations were conducted at the points of the closed circles in WB and CB (Figure 2). The closed circle for EB in Figure 2 is identical to point eb in Figure 4. The measured temperature, C₂₅, and pH values are shown in Figures 5A–C, respectively.

The temperature profile of Lake Monoun has features that are not found in ordinary lakes (Figure 5A). In EB, the temperature at the surface was high, while it reached a minimum at approximately −30 m depth. Further below, the temperature increased as the depth increased, and a thermocline appeared at approximately −80 m depth. Under the thermocline, a mixed layer of approximately 10 m thick was developed. Under the mixed layer, the water temperature increased further toward the bottom of the lake. In WB and CB, the thermocline appeared at approximately −33 m depth. Below the thermocline, the temperature increased slightly toward the bottom of WB and CB. Toward the bottom of WB, the temperature profile in CB was almost identical to that in WB. Over the entire depth range, excluding the surface, the temperature in EB significantly decreased relative to the temperature in 2003 (Kusakabe et al., 2008).

C₂₅ in EB was low and approximately constant from the surface to approximately −30 m depth (Figure 5B); further below, it increased with depth, and a chemocline appeared at approximately −80 m (the term “chemocline” used here refers to the boundary in which the chemical properties of lake water change rapidly). Below the chemocline, a mixed layer of approximately 10 m thick was developed. Under the mixed layer, C₂₅ increased toward the bottom of EB. The C₂₅ profiles in WB and CB revealed chemoclines at approximately −33 m depth (Figure 5B). C₂₅ in CB increased in the layer below the chemocline toward the bottom. The highest C₂₅ value in the CB profile was close to the C₂₅ value observed in 2003 at EB. Over the entire depth range, except for the surface and near the bottom, C₂₅ in EB decreased significantly relative to C₂₅ in 2003.

The pH of the lake water in WB, CB, and EB was weakly alkaline near the surface (Figure 5C). In EB, the lake water was close to neutral, and the pH was almost constant from −10 to −50 m depth. Further below, the pH fell with increasing depth, and a chemocline appeared at approximately −80 m depth. Below the chemocline, a mixed layer of approximately 10 m thick was developed. Below the mixed layer, the pH rose toward the lake bottom. The pH profiles in WB and CB were different from those in EB, exhibiting left-pointing “projections” locally falling pH at the chemocline at around −33 m depth (right arrow in Figure 5C). Over the entire depth range, except for the surface and near the bottom, the pH in EB significantly rose relative to the pH in 2003.

Figure 6 shows the profiles in Figure 5 enhanced near the bottom of EB. At all points, temperature increased toward the lake bottom (Figure 6A). Except for the profile at point p1, the thermal gradient (i.e., −dT/dz; T: temperature; z: depth) was common and almost constant, suggesting an established stable thermal gradient in the lake water near the bottom. Inside D1, the thermal gradient was high, suggesting discharged hot water at the bottom of D1. However, the thermal gradient inside D2 was similar to that at points other than p1. At point p1, C₂₅ increased toward the bottom, where the highest C₂₅ value was recorded (Figure 6B), suggesting discharged high C₂₅ water at the bottom of D1. At points eb and p1o, C₂₅ sharply increased at the lake bottom. At point p2o, C₂₅ peaked before reaching the lake bottom. The interior of D2 was filled with high C₂₅ water, with values close to the maximum value at point p2o. At point p2, C₂₅ decreased toward the bottom, suggesting no discharge of water with high C₂₅ within D2. At point p1, the pH started to fall at the upper limit of D1 and then turned to rise toward the bottom (Figure 6C). At points p1, p1o, and eb, pH rose at the lake bottom. At point p2o, the pH peaked before the bottom. The interior of D2 was filled with high-pH water; the value was close to the maximum value at point p2o. At point p2, the pH fell at the bottom, suggesting no discharge of water with high pH at the bottom of D2.

On 2 March 2015, lake water was collected at point p1 using the MK method. Supplementary Table S1 lists the Ct values determined by the MK method. The Ct values are plotted in Figure 5D, Figure 6D. Supplementary Table S1 also presents the theoretically estimated [CO_2aq] and [HCO₃‾] based on the chemical equilibrium among CO_2aq, HCO₃‾, and CO₃ ²‾ (in this study, brackets, i.e., [ ] indicate the concentration of chemical species in mmol/L). The theoretical relationship between [HCO₃‾] and Ct is given by the following equation (Stumm and Morgan, 1996): [ H C O 3 − ] = C t ( a H   γ H C O 3 K a 1   + 1 + K a 2   γ H C O 3 a H   γ C O 3 ) − 1 . (2)

Here, a_H is the activity of H⁺ ion given by, a H = 10 − p H . (3) K_a1 and K_a2 in Equation 2 denote the acidity constants of CO_2aq and HCO₃‾, respectively. K_a1 and K_a2 are functions of temperature, and the concrete formula is given by Ohba et al. (2015). γ_HCO3 and γ_CO3 in Equation 2 are the activity coefficients of HCO₃‾ and CO₃ ²‾, respectively, given by the Davies equation (Butler, 1991): log ⁡ γ = − 0.5 z 2 ( I 0.5 1 + I 0.5 − 0.2 I ) ( 298 T ) 2 3 , (4) where z, I, and T denote the charge of the species, ionic strength, and temperature in Kelvin, respectively. Based on the cationic composition of the water at −96.5 m depth in Lake Monoun in 2006 (Kusakabe et al., 2008), I can be estimated as follows: I = 1.463   [ H C O 3 − ] / 1000 . (5)

[HCO₃‾] was calculated from Ct using the following procedure:

Step 1

γ_HCO3 and γ_CO3 were initially set equal to 1.

Step 2

K_a1 and K_a2 were calculated from the temperature of the lake water.

Step 3

a_H was calculated from the pH of the lake water.

Step 4

Substituting the above values into Equation 2, the initial [HCO₃‾] was calculated from Ct obtained using the MK method.

Step 5

Substituting the initial [HCO₃‾] into Equation 5, I was obtained.

Step 6

Substituting the ionic strength, ion charge, and lake water temperature into Equation 4, γ_HCO3 and γ_CO3 were obtained.

Step 7

Finally, steps 2, 3, and 4 were repeated.

In general, the electrical conductivity (C₂₅) of lake water is proportional to the concentration of the dissolved ions. According to Kusakabe et al. (2008), the main anion contained in the lake water of Lake Monoun was HCO₃‾. The concentrations of anions other than HCO₃‾, such as Cl‾ and SO₄ ²‾, were negligibly lower than [HCO₃‾]. Therefore, a positive correlation is expected between C₂₅ and [HCO₃‾]. Figure 7 depicts C₂₅ versus [HCO₃‾] (Supplementary Table S1), in which the correlation is approximated by the following quadratic equation in terms of C₂₅: y = 2.204 × 10 − 6 x 2 + 5.670 × 10 − 3 x + 4.109 , (6) where x and y denote C₂₅ and [HCO₃‾], respectively. Based on Equation 6, the C₂₅ profiles were converted to continuous profiles of [HCO₃‾], as shown in Figures 5E, 6E. The Ct profiles can be calculated from the [HCO₃‾] profile using Equations 2–5. The Ct profiles are shown in Figures 5D, 6D and are consistent with the discrete Ct values obtained using the MK method. [CO_2aq] is obtained from [HCO₃‾] using the following equation: [ C O 2 a q ] = γ H C O 3 [ H C O 3 − ] 10 − p H K a 1 − 1 . (7)

Since [CO₃ ²‾] is less than one-thousandth of [HCO₃‾] at a pH of 7.3 or lower, [CO₃ ²‾] can be safely neglected in the case of Lake Monoun, while [CO_2aq] can be simply expressed as the difference between Ct and [HCO₃‾] as follows: [ C O 2   a q ] = C t − [ H C O 3 − ] . (8)

The obtained [CO_2aq] profiles are shown in Figures 5F, 6F.

The Ct profiles at all points exhibited almost constant values from the surface to approximately −33 m depth (Figure 5D). At points in WB and CB, a chemocline appeared at −33 m depth; at points outside WB and CB, a chemocline appeared at around −80 m depth, while a mixed layer of approximately 10 m thick was developed below. The characteristics of the Ct profile (Figure 5D) are similar to those of the C₂₅ profile (Figure 5B); nevertheless, they exhibit the following differences: the C₂₅ value at the bottom of CB was close to the C₂₅ value at the bottom of points eb, p2, and p2o (Figure 5D), whereas the Ct value at the bottom of CB was only one-tenth of the Ct value at the bottom of points eb, p2, and p2o (Figure 5D). At point p1, Ct continued to rise toward the lake bottom within D1 (Figure 6D), suggesting the discharge of high-Ct water at the bottom. At point p1o, adjacent to point p1, Ct peaked near the bottom (Figure 6D). Point p2, in the interior of D2, was filled with water; the value of Ct at point p2 was slightly higher than that of the ambient lake water represented by the water at point eb, except for near the bottom. The Ct value at point p2 decreased at the lake bottom, suggesting that no high-Ct water was discharged at the bottom of D2. The Ct values at points eb and p2o peaked near the lake bottom (Figure 6D). The [HCO₃‾] profiles (Figure 6E) were similar to the Ct profiles (Figure 6D), except for points p1o and eb, in which [HCO₃‾] did not decrease at the lake bottom. The [CO_2aq] profiles (Figure 6F) were similar to the Ct profiles (Figure 6D). Allowing equilibrium between CO_2aq and HCO₃‾, the [HCO₃‾]/[CO_2aq] ratio was 0.19 and 0.43, respectively, when the pH was 5.65 and 6.00, which was the pH range of lake water below −95 m depth (Figure 6C). The similarity between the [CO_2aq] and Ct profiles is reasonable, considering the dominance of CO_2aq over HCO₃‾. At point p1, [CO_2aq] increased toward the lake bottom, suggesting the discharge of high-[CO_2aq] water at the bottom of D1. Inside D1, [CO_2aq] exceeded the value observed at point eb in 2003 (Figure 6F), suggesting that the high-[CO_2aq] water at the bottom of D1 could be the source of the high-[CO_2aq]-mixed layer developed below the chemocline at approximately −81 m depth. The [CO_2aq] profile at point p1 did not reach the saturation concentration of CO_2aq estimated using Henryʼs constant of CO₂ (Fernandez-Prini et al., 2003).

Three large depressions, i.e., D1, D2, and D3, were found at the bottom of EB, and hot water rich in CO_2aq and HCO₃‾ was discharged at the bottom of D1. However, within D2, [CO_2aq] and [HCO₃‾] decreased toward the bottom, indicating no discharge of water rich in CO_2aq and HCO₃‾. At points p1o, eb, and p2o, [CO_2aq] peaked above the lake bottom (Figure 6F). The [CO_2aq] peak height was the highest at p1o, followed by eb and p2o, while it decreased with distance from D1 (Figure 4). The above observations suggest the wide horizontal distribution of high-[CO_2aq] hot water along the bottom of EB.

At points p1o and eb, pH rose sharply at the lake bottom (Figure 6C). The pH behavior indicated that the [HCO₃‾]/[CO_2aq] ratio increased near the lake bottom. The [HCO₃‾]/[CO_2aq] ratio and pH are related through the following formula: p H = l o g ( γ H C O 3 [ H C O 3 − ] [ C O 2 a q ] ) − l o g K a 1 . (9) FeCO₃ siderites were present in the bottom sediments of Lake Monoun, and the lake water near the bottom was supersaturated with respect to FeCO₃ (Sigurdsson et al., 1987). Fine FeCO₃ particles could be suspended in the lake water near the bottom. The pH increase, i.e., the [HCO₃‾]/[CO_2aq] ratio increase, near the bottom of the lake is indicative of the reaction between CO_2aq and FeCO₃ particles as follows: F e C O 3 + C O 2 a q + H 2 O = F e 2 + + 2 H C O 3 − . (10)

In the above reaction, 1 mol CO_2aq is consumed, and 2 mol HCO₃‾ is generated; when it occurs, the [HCO₃‾]/[CO_2aq] ratio as well as the pH rises. Figure 8A shows the relationship between water temperature and [HCO₃‾] below −95 m depth. As shown, LW is an end-member representing the lake water. The water temperature, [CO_2aq], and [HCO₃‾] of LW were assumed to be 22.3°C, 80 mmol/L, and 18 mmol/L, respectively. End-member E2 represents lake water after reaction with FeCO₃. The temperature, [CO_2aq], and [HCO₃‾] of E2 were assumed to be 23.9°C, 74 mmol/L, and 30 mmol/L, respectively. Through the reaction, the [CO_2aq] decrease of LW was 6 mmol/L, and the [HCO₃‾] increase was 12 mmol/L. The data points distributed on the line connecting LW and E2 (Figure 8A) indicate a linear relationship between temperature and [HCO₃‾] near the lake bottom, especially in the range of low [HCO₃‾]. The hot water at the bottom of D1 is assumed to be end-member E1. As shown in Figure 8A, the lake water located in the area to the right of the line connecting LW and E2 can be a mixture of LW and E1. Figure 8B shows the relationship between [CO_2aq] and [HCO₃‾] below −95 m depth. The bottom-water compositions at points p2 and p2o were close to E2, suggesting that the bottom water at p2 and p2o was lake water reacting with FeCO₃. The other points cannot be easily explained by a simple mixture of LW and E1. As shown by the red dashed line and arrow, if mixing between LW and E1 is followed by the reaction with FeCO₃, the observed relationship between [CO_2aq] and [HCO₃‾] can be explained because [HCO₃‾] increases and [CO_2aq] decreases, as indicated by the red arrow.

In the numerical simulation of limnic eruptions by Kozono et al. (2016), the following initial conditions were adopted: the lake water was saturated with CO₂ at approximately −50 m depth. The [CO_2aq] of the lake water below −50 m depth was equal to [CO_2aq] at −50 m depth. Below −50 m depth, the lake water was unsaturated with CO₂. The conditions described previously were based on the CO_2aq profile (eb-2003 in Figure 5F) observed by Kusakabe et al. (2008). Sigurdsson et al. (1987) supposed that the lake water was supersaturated with respect to CO_2aq before the 1984 limnic eruption, which was triggered by the disturbance of the lake water by a landslide. However, in the initial condition by numerical simulation, the lake water below −50 m depth is unsaturated with CO₂. Actually, in 1986, when 2 years later of limnic eruption, the [CO_2aq] of lake water deeper than −65 m was unsaturated with respect to CO_2aq (Figure 5F). The falling sediment displaced lake water from near the chemocline to the deep layer. Because the difference between the CO₂ partial pressure of the displaced lake water and the surrounding pressure increases, degassing of CO₂ from the displaced lake water is not likely to occur. Therefore, it is unlikely that any falling sediment triggered the limnic eruption.

In this study, [CO_2aq] at the bottom of D1 was estimated to be 275 mmol/L. From Henryʼs constant of CO₂ (Fernandez-Prini et al., 2003), the equilibrium gas pressure of CO₂ was estimated to be 8.2 × 10⁵ Pa, which was lower than the hydrostatic pressure at the lake bottom (i.e., 1.0 × 10⁶ Pa). At the bottom of D1, CO_2aq was unsaturated. According to Issa et al. (2013), CO₂ dissolved in the water of Lake Monoun was accompanied by CH₄; they found that the contributions of CO₂ and CH₄ to the total pressure of the gas phase equilibrated with water near the lake bottom were 63 and 37%, respectively. This suggests that the total gas pressure in equilibrium with the lake water could be 1.3 × 10⁶ Pa, thereby exceeding the hydrostatic pressure, and the bubbles consisting of CO₂ and CH₄ were always present at the bottom of D1.

Sigurdsson et al. (1987) discovered a trace of a landslide in W, as shown in Figure 4. Mound M1 is located close to W, and it has likely formed by the sediment slumped from W. All mounds at the bottom of the EB were located along the scarp of the basin. The volume of M1 can be estimated by comparing the cross section along L3, which straddles M1, and the cross section along L4 (Figures 3C,D). The shaded area in Figure 3C corresponds to the body of M1 by comparing the cross sections and assuming that the terrain below M1 is equal to the cross section along L4. The cross-sectional area, length, and volume of M1 were 73 m², 83 m, and 6 × 10³ m³, respectively. The horizontal cross-sectional area of ​the EB at −50 m depth was 1.02 × 10⁵ m². If the entire mass of M1 slumped from W, then the mass pushed the chemocline at −50 m depth upward by approximately 0.06 m.

According to the limnic eruption scenario of Sigurdsson et al. (1987), a landslide occurred at W, and the collapsed sediment agitated the lake water near the bottom, thereby triggering the limnic eruption. What role does D1 play in the limnic eruption scenario described by Sigurdsson et al. (1987)? D1 was an outlet for hot water supplying CO_2aq, which was the driving force of limnic eruption. In the limnic eruption scenario of Sigurdsson et al. (1987), D1 can play its role regardless of its location at the bottom of the lake. Because M1 is located directly under W, M1 is considered to be sediment deposited by a landslide. D1 does not necessarily need to be adjacent to M1. In the limnic eruption scenario of Sigurdsson et al. (1987), the coupling of M1 and D1 was considered to be a coincidence; however, in our limnic eruption model, the coupling of M1 and D1 is not a coincidence but an inevitable consequence.

As shown in Figure 5F, [CO_2aq] near the bottom of the lake in 1986, 2003, and 2015 was similar and approximately equal to 160 mmol/L, suggesting that the hot water with high [CO_2aq] discharging at D1 was diluted by the lake water with low [CO_2aq]. The mixing ratio between the two waters was probably constant throughout the period between 1986–2015. It is likely that the sustained discharge of hot water with high [CO_2aq] at D1 occurred even before the 1984 limnic eruption.

Woods and Phillips (1999) reproduced the limnic eruption in Lake Nyos at −210 m depth through laboratory experiments and numerical analysis. They found that a small amount of CO₂ bubble flow at the lake bottom entrained CO_2aq-saturated lake water, and the rising flow was strengthened by CO₂ degassing. The CO₂ flux at the lake surface could reach 10⁵ times that at the bottom of the lake. Under the initial conditions in the numerical simulation by Kozono et al. (2016) for Lake Monoun, the lake water is only saturated with CO_2aq at −50 m depth, and a slow upward flow of lake water (i.e., >0.01 m/s) near −50 m depth is necessary for initiating the first degassing of CO₂ (Figure 9A). Kozono et al. (2016) suggested a perturbation of the layer boundary caused by double-diffusive convection and a seiche induced by external forcings, such as strong wind causing the slow updrift of lake water. In this study, we propose that bubbles rising from D1 cause a slow upward flow of lake water. The first degassing of CO₂ generates buoyancy in the lake water, and a parcel of lake water containing CO₂ bubbles starts to rise. The rising parcel of lake water entrains lake water enriched in CO_2aq below −50 m depth (Figure 9B). The degassing of the entrained lake water increases the rising speed. The increased velocity of lake water results in explosive degassing (i.e., limnic eruption). Considering the chemocline of [CO_2aq] in 1986 (Figure 5F), the layer above −65 m depth suffered degassing. The slow upward flow of lake water around −50 m depth was the “seed” of the limnic eruption. The hot water discharged from D1 was accompanied by bubbles. The bubbles rising from D1 and reaching −50 m depth probably produced a slow upward flow of lake water (Figure 9A). As shown in Figure 9B, the 1984 limnic eruption was likely initiated above D1. The lake-water wave generated by the limnic eruption impacted the lake shoreline at W, thus causing a landslide and resulting in the M1 deposits. In this model, the coupling between D1 and M1 was not a coincidence but an inevitable consequence.

The coupling of D2 and M2 may be the trace of another limnic eruption before 1984, strongly implying the probability of future recurrence of limnic eruptions at Lake Monoun if the artificial CO₂ degassing stops. Shanklin (1992) collected oral testimonies by several traditional ethnic groups in the area near Lake Monoun from 1981 to 1986. Several oral testimonies showed that the lake exploded, and subsequently the ethnic groups left the lake area. The oral testimonies suggest that limnic eruptions have occurred in the past at Lake Monoun or another lake, which is consistent with the idea that the coupling of D2 and M2 is likely the trace of a limnic eruption that occurred before 1984.

Many grooves were observed on the scarps of EB (Figure 4). It is likely that the waves generated by the limnic eruption collided with the scarp, thereby destabilizing the surface layer while part of the surface layer collapsed to form grooves. The distribution of grooves was limited to the slopes of the northern half, suggesting that the waves arriving at the southern shore were weak. These probabilities are consistent with the estimation that the starting point of the limnic eruption was directly above D1 and far from the southern shore. Near the inlet of the Panke River, there was an area free of grooves (F in Figure 4). As shown in Figure 3B, EB gently inclined from north to south. It is likely that the gentle inclination was caused by the sedimentation of suspended materials carried by the Panke River, which flowed into EB from the north side. The grooves in area F may have been filled with a suspended material carried by river water, thereby erasing the grooves.

A diatreme structure generally develops beneath a maar (Lorenz, 1973). Nagao et al. (2010) suggested that CO₂-rich fluid rose along the ring fault of the potential diatreme structure developed beneath Lake Nyos. In Lake Monoun, D1, D2, D3, and other small depressions are distributed on the bottom of EB along its scarp. This distribution suggests that a diatreme structure also exists beneath Lake Monoun (Figure 2), while the CO₂-containing fluid rose along the ring fault of the diatreme and was discharged on D1. However, D2 seems to be a trace of the outlet, in which CO₂-enriched water was discharged before 1984.

The [HCO₃‾] profile near the bottom of CB in 2015 was similar to that of EB in 2003 (Figure 5E). For example, in 2003, the [HCO₃‾] value near the bottom of CB was close to the [HCO₃‾] value at the same depth in EB. Furthermore, the [HCO₃‾] profile in CB had a bending point near −49 m depth (right arrow in Figure 5E). The [HCO₃‾] profile in EB in 2003 also had a bending point at −50 m depth (left arrow in Figure 5E). This suggests that the estimated profiles in WB and CB in 2003 were identical to those in EB in 2003, and the [HCO₃‾] profiles in WB and CB in 2015 were remnants of the profile established in 2003. As shown in Figure 3A, the depth of the EB chemocline in 2003 was close to the depth of the saddle between CB and EB. By 2003, the dense lake water that had accumulated in the hypolimnion of EB may have flowed into CB and WB beyond the saddle. Here, the hypolimnion is the region beneath the chemocline where mass transfer is restricted to and from the shallow layer of lake water. Since 2003, when artificial degassing began, the EB chemocline has deepened rapidly; however, the dense lake water that flowed into CB and WB could not return to the EB beyond the saddle and was left behind as of 2015.

The [CO_2aq] profiles in WB and CB in 2015 were almost vertical toward the lake bottom (Figure 5F), suggesting no discharge of high-[CO_2aq] water at the bottoms of WB and CB. In 2015, the highest [CO_2aq] value in WB and CB was 10.8 mmol/L at −39 m depth, which was much lower than the [CO_2aq] value of 58.7 mmol/L in EB in 2003 at the same depth. Considering the above difference in [CO_2aq] and the similar [HCO₃‾] profiles, CO_2aq in the hypolimnion of WB and CB was preferentially lost. The difference between the properties of CO_2aq and HCO₃‾ is that the former has no electrical charge, whereas the latter has an electrical charge. Therefore, CO_2aq escapes as it can diffuse into the atmosphere as gas through the lake surface while HCO₃‾ does not.

The flux of CO₂ gas released from the surface of a crater lake can be interpreted as a sign of volcanic activity (e.g., Mazot and Bernard, 2015). It would be interesting to see how much CO_2aq stored in the hypolimnion of Lake Monoun contributed to the CO₂ gas released from the lake surface. According to Issa et al. (2014), CO₂ gas was released from the entire lake surface of Lake Monoun to the ambient atmosphere at a rate of 21.8 ton/day in 2013. The volume of WB and CB below −22 m depth is 1.5 × 10⁶ m³. The total amount of CO_2aq was estimated at 8.4 Mmol by integrating the [CO_2aq] value in 2015. Assuming that the [CO_2aq] profiles in WB and CB are the same as those in EB in 2003, the total amount of CO_2aq stored in lake water below −22 m depth was 74 Mmol. Therefore, 74 Mmol of CO_2aq dissolved in the lake water in WB and CB in 2003 was estimated to have decreased to 8.4 Mmol in 2015. Assuming a constant rate of the CO_2aq decrease, the rate is estimated at 14.7 kmol/day (or 0.647 ton/day). This amount of flux is only approximately 3% of the flux observed by Issa et al. (2014), which was 21.8 ton/day. Among the CO₂ gases emitted from the entire lake surface of Lake Monoun, CO₂ originating from the hypolimnion of WB and CB is expected to be negligible.

Near the lake bottom in EB of Lake Monoun, a stratified structure developed in which the temperature and [HCO₃‾] increased toward the bottom while maintaining a linear relationship between temperature and [HCO₃‾]. The [HCO₃‾] increase could be due to the reaction between CO_2aq in the lake water and possible FeCO₃ particles suspended in lake water. The reacted CO_2aq is converted to HCO₃‾. In addition to the stratified structure, hot water enriched in CO_2aq and HCO₃‾ was diffused horizontally along the lake bottom. Hot water was discharged from D1. At the bottom of D1, the temperature, pH, [CO_2aq], and [HCO₃‾] were 24.4°C, 5.67, 275 mmol/L, and 73.6 mmol/L, respectively.

In WB and CB of Lake Monoun, a hypolimnion was sustained below −30 m depth. In 2015, the hypolimnion seemed to keep CO_2aq and HCO₃‾ dissolved in the lake water in 2003. The total amount of CO_2aq stored in the hypolimnion in 2015 and 2003 was estimated to be 8.4 and 74 Mmol, respectively. If CO_2aq in the hypolimnion diffused out through the lake surface, a 14-kmol/day flux would be expected, corresponding to 3% of the CO₂ flux through the entire lake surface in 2013. The CO₂ flux through the lake surface in 2003 seemed to be dominated by the CO₂ originating from EB, and the contributions of CO_2aq stored in the hypolimnion of WB and CB were limited.

The hot water discharged from the bottom of D1 could have been accompanied by bubbles affected by the partial pressure of CH₄ dissolved in the lake water. The bubbles rising from D1 reached the chemocline at −50 m depth and formed a weak upward flow of lake water. The upward flow of lake water at −50 m depth was the “seed” of the limnic eruption (Kozono et al., 2016). The upward flow of lake water induced the initial degassing of CO₂. The initial degassing of CO₂ was amplified by the entrainment of CO₂-enriched lake water below −50 m depth, resulting in the 1984 limnic eruption. The wave generated by the limnic eruption impacted the east shore, causing a landslide and forming mound M1 beside D1. The coupling of D1 and M1 was regarded as a trace of the limnic eruption in 1984. Another coupling, i.e., that of D2 and M2, is likely the trace of another limnic eruption that occurred earlier than 1984. Lake Monoun may have experienced several limnic eruptions in the past. If the artificial degassing of CO₂ is not continued in Lake Manoun, the water containing high concentrations of CO_2aq released from D1 will increase the CO_2aq concentration in the lake water, and the bubbles rising from D1 will cause a limnic eruption. In the future, the flux of CO_2aq supplied from D1 may increase and exceed the flux of CO_2aq removed by the artificial degassing, potentially increasing the amount of CO_2aq accumulated in the lake water. The regular monitoring of the CO_2aq amount in lake water should also be continued.