Climate-forced changes in available energy and methane bubbling from subarctic lakes

1 Introduction

The search for a “master proxy” for methane production from wetlands and waters has a long history [Whiting and Chanton, 1993], but several recent papers [Gudasz et al., 2010; Yvon-Durocher et al., 2014] have pointed to a temperature dependency. This is not a new idea; the relationship between CH₄ production by Archaea and temperature has been known for decades [Boylen and Brock, 1973; Zeikus and Winfrey, 1976], and observations have long been made of system-wide temperature dependences as well [Atkinson and Hall, 1976; Seiler et al., 1984]. Yet the multitudes of competing effects which damp or enhance productivity have until recently prevented the establishment of direct links between lake energy content and CH₄ production. A recent long-term, high-frequency whole-lake CH₄ ebullition data set for three lakes in northern Sweden [Wik et al., 2013], coupled with high-resolution lake and sediment temperature data, has improved our ability to search for an appropriate proxy. One anticipated effect of future warming is a decrease in subarctic and Arctic lakes' ice-covered season length; such trends are already being observed in lake spring ice-out dates [Prowse et al., 2011; Weyhenmeyer et al., 2005].

It was previously hypothesized that a link between incoming shortwave (SW) solar radiation and seasonal CH₄ ebullition exists for shallow high-latitude lakes and ponds because the spring ice-out date marks the point at which a lake begins absorbing large amounts of SW solar energy [Wik et al., 2014]. Warming of the sediments after ice-out eventually leads to a large increase in CH₄ production rates, due to the temperature dependence of methanogens' CH₄ production rate. Once the sediment pore water is warmed and becomes saturated with CH₄, any additional CH₄ production can only go into the formation of bubbles [Wik et al., 2014]. Although there are numerous factors that impact short-term ebullition (instantaneous temperature, frontal system passages, lake internal mixing, etc.), on seasonal timescales, given sufficient organic substrate, energy availability to the sediment is the dominant production control; in many lakes this may be adequately expressed by incident SW [Wik et al., 2014]. Many high-latitude lakes lie in regions already noted to be experiencing the effects of climate change [Payette et al., 2004], and warming is widely expected to be greatest in the Arctic, potentially altering a major source of the powerful greenhouse gas CH₄ [Kirschke et al., 2013; Nisbet et al., 2014] to the atmosphere.

The strong dependency of seasonal CH₄ ebullition on seasonal energy proxies demonstrated previously (r² = 0.86 to 0.997) [Wik et al., 2014] suggest that other energy proxies may be useful in describing CH₄ ebullition as well (see also supporting information), if they capture the interseasonal variability. Intuitively, the ice-out date of Torneträsk should work as a proxy for seasonal CH₄ ebullition, as more ice-free days lead to more heating of the very low albedo water, as opposed to the ice-covered lakes. Unlike the ice-free period SW measurement, the time of ice-out has no impact on the energy absorbed by the lake after the ice is gone. Thus, a particularly cloudy, cold summer and a sunny, hot summer could have lower and higher than average total SW energy collection, respectively, with the same spring ice-out date. Figure 1 shows the relationship between Torneträsk ice-out date and the seasonal CH₄ ebullition observed in our study lakes. Developing a CH₄ ebullition proxy based on Torneträsk ice-out date allows for the possibility of a substantial backward projection of our lake ebullition records.

2 Methods

The sample collection and analysis methods are described in detail elsewhere [Wik et al., 2013]. Briefly, 40 half-meter-wide bubble traps were distributed across the different depth zones of the study lakes in Stordalen Mire (68°21′N, 19°02′E, 350 m above sea level). The traps are subsurface and include a restriction to collect only ebullitive CH₄, not diffusively released CH₄. Measurements began in two lakes, Inre Harrsjön and Mellersta Harrsjön in summer 2009; a third lake, Villasjön was added the following summer. During the ice-free season, the traps were sampled every 24–72 h. The collected gas was analyzed the same day for CH₄ concentration using a Shimadzu 2014 gas chromatograph equipped with a flame ionization detector. Concentrations were converted into daily mass fluxes of CH₄ [Wik et al., 2013].

Water and surface sediment temperatures were measured year-round during the study period using HOBO Water Temp Pro v2 temperature loggers. Temperature loggers were arranged at 0.1, 0.3, 0.5, and 1 m depths in the water; additional deeper sensors were included in Inre Harrsjön and Mellersta Harrsjön. In all lakes, the deepest sensors were embedded in the surface sediment. Incoming shortwave (SW) radiation was measured using pyranometers over the surface of Inre Harrsjön and at Abisko Scientific Research Station (ANS), in Abisko. The ANS SW record is continuous since mid-1984. We calculated total SW by summing daily ground level SW averages. Seasonal daily average SW was determined by dividing the total SW from lake ice-out date until 30 September. Because bubbling essentially stops or decreases rapidly by that point [Wik et al., 2013] due to lake cooling, 30 September was used as an end date. The daily average SW climatology for Abisko for the years 1985–2013 is shown as Figure S2 in the supporting information and is described by equation 1:

(1)

where SW(d) is the shortwave climatology for Abisko for 1985–2013, and d is day of the year. On average, 2.92 GJ m⁻² of SW are available to the lakes each year.

Approximately 750 m north of Inre Harrsjön is the southern shore of Torneträsk, a 168 m deep glacial lake. Torneträsk is the sixth largest lake by area in Sweden (third largest by volume and second deepest), with an area of 33,200 Ha and length of 70 km. Spring ice-out date observations began for Torneträsk in 1916. They have continued every year since except 1927. The Torneträsk ice-covered season length has been noted to be decreasing [Christensen et al., 2004] with a trend of −0.31 d yr⁻¹ for the 1961–1990 period [Weyhenmeyer et al., 2005].

Between 2009 and 2013, Torneträsk ice-out on average occurred 23 ± 7 days after ice-out on the three studied lakes within Stordalen. There is reason to believe that this 23 day lead time is typical, because there has been only a slight change in ice-covered season length, and air temperature, for much of the past century [Callaghan et al., 2010]. The use of ice-out dates as proxies for spring temperatures and climate has been applied previously [Livingstone, 1997]. An error in estimating the ice-out date of our study lakes of ±1 week does not appreciably change the reliability of the correlations reported here or previously [Wik et al., 2014]. This is largely due to the fact that ice-out date shifts of a few short spring days do not greatly increase in SW energy input to the lakes. Also, each earlier ice-free spring day is shorter, and its potential SW contribution is less.

The observed Torneträsk ice-out date trend and the offset between the small Stordalen lakes and Torneträsk can be combined with the SW climatology in equation 1 to predict ice-free season SW (SSW) as a function of year:

(2)

where predicted seasonal SW (SSW) (W m^− 2 y^− 1) is a function of year (y) from 1961 and predicted ice-out dates, “nint” is the “nearest integer” function, and d is day of year, based on 1985–2013 Abisko SW climatology. This function assumes a constant annual ice-out date of day 142 before 1970 (based on no discernible trend in the 1916–1969 time period) and a constant ice-out date of 110 after 2050 (due to limited SW early in the year, earlier ice-out dates become progressively much more difficult to achieve). Similarly, equation 3 predicts seasonal CH₄ ebullition (mg m^− 2 y^− 1), from the equation 2 output, based on 2009–2013 observations:

(3)

The values generated by equations 2 and 3 reflect only the trend in the Torneträsk ice-out data and so are a generalized model of the Stordalen lakes. In Figure 2, we use the observed Torneträsk ice-out dates to more realistically represent the interannual variability. Using the relationship shown in Figure 1 and the 98 year record of Torneträsk ice-out dates, we calculated the expected annual CH₄ ebullition from 1916 to 2008 for our study lakes in Stordalen (Figure 2). The range of interannual variability in 2009–2013 was approximately 1000 mg m² yr⁻¹. Nonetheless, there are clear long-term trends in the data. Up until about 1970, there is little discernable trend in ice-out date and calculated CH₄ ebullition. For the 1970–2008 period, the trend is about +14 mg m² yr⁻¹. If the entire 1916–2008 period is taken together, the trend is +3.9 mg m² yr⁻¹. This is in agreement with other work which has shown that climate-driven changes in the Stordalen area became more pronounced in the latter third of the twentieth century [Payette et al., 2004].

3 Results

This SW climatology, combined with existing observations of CH₄ ebullition onset timing [Wik et al., 2013], allows us to describe, using average present-day Stordalen lake ice-outs (based on 2009–2013 data), that ≈28% of the annual SW reaches the surface prior to ice-out (when most is reflected by the high albedo of the ice and snow cover) and hence has a limited effect on total lake energy (Figure 3). Ebullition in the shallow zone of the lakes (<2 m) becomes much greater in late June [Wik et al., 2013], by which time ≈55% of the annual SW has been received. Ebullition from the deep zones (>4 m depth) of the lakes (late July) requires in total ≈75% of the annual SW. About 69% of the annual SW (1.99 GJ m⁻²) reaches the lakes between ice-out and our ebullition monitoring cut-off date of 30 September. Only ≈3% of the annual SW reaches the lakes after 30 September. By the end of September, the lakes are losing heat much faster than it is replenished by solar heating, and the lakes, along with the methanogens in the sediment, are cooling. These observations of energy requirements suggest constraints on seasonal CH₄ ebullition based on lake morphologies and local climate. As an alternative to the extrapolations shown in Figure 2, the annual seasonal CH₄ ebullition from the study lakes from 1961 can be approximately described by equation 3. This equation does not include a slight increase in SW that has been observed at Abisko since 1985 of 1700 J m⁻² d⁻¹, whereas the actual 2009–2013 data implicitly include this increase.

The 2009–2013 CH₄ ebullition data (Figures 2 and 3) show an interannual variability of similar scale to that for the calculated CH₄ ebullition from 1916 to 2008 from Torneträsk ice-out dates and the SW climatology. Further, the measured interannual variability of 2009–2013 of 1091–1850 mg m⁻² yr⁻¹ lies totally within the calculated interannual variability of 519–2316 mg m⁻² yr⁻¹ from 1916 to 2008, or, omitting the outlier years of 1951 and 2002, 641–1948 mg m⁻² yr⁻¹. Figure 2 suggests that years of low CH₄ ebullition have become less frequent; for instance, the average for 1916–1936 was 1140 mg m⁻² yr⁻¹, while the average for 1988–2008 was 1487 mg m⁻² yr⁻¹, an increase of 30%.

Published simulations of lake-ice thermal structure driven by ERA-40 reanalysis data set modified for 2040–2079 forcings from the Special Report on Emissions Scenarios A2 [Nakicenovic et al., 2000], suggested that the 2040–2079 period, the region around Abisko may have ice-out dates 20–30 days earlier than in the 1960–1999 period [Dibike et al., 2011]. The average ice-out on Torneträsk for 1960–1999 was day 164. The ice-out date range suggested from the model results [Dibike et al., 2011] is between days 134 and 144. If the 23 day offset between Torneträsk and Stordalen lake ice-outs holds, the Stordalen lakes' average ice-out date would be between days 111 to 121 when only 17–22% of the annual SW would have been received by the lakes (present average ice-out is day 130). Ice-out dates of 111–121 suggest that the average seasonal ebullitive CH₄ flux from the Stordalen mire lakes in the years 2040–2079 would be ≈2000–2400 mg CH₄ m⁻² yr⁻¹, due to the ice-out date shift of at least 44 days compared to the 1916–1936 average of 166.

Such a hypothetical future shift from the present average ice-out day of 130 to day 111 represents the addition of 0.35 GJ m⁻² to the Stordalen lakes during their ice-free season. Integrating over the average ice-free season corresponds to adding 8100, 3900, and 60000 GJ of energy to Inre Harrsjön, Mellersta Harrsjön, and Villasjön, respectively. Compared to 1916, the estimate for 2060 represents the addition of 23000, 11000, and 170000 GJ to Inre Harrsjön, Mellersta Harrsjön, and Villasjön, respectively.

4 Discussion

Using an ebullition estimate of 2200 mg CH₄ m⁻² yr⁻¹, in the predicted range for the 2040–2079 period, this represents a near-doubling of the 1916–1936 average of 1140 mg CH₄ m⁻² yr⁻¹. About 24% of this increase in ebullition is already apparent in the present day. The projections in Table 1 do not account for the depletion of organic material available for methanogenesis in the lakes. Numerous other studies expect increased availability [Duc et al., 2010] in a warmer climate. Due to high interannual variability, several years have already reached 75% of the maximum seasonal methane ebullition output predicted for 2060.

The three studied postglacial lakes in Stordalen are broadly typical of pan-Arctic lakes, although the studied lakes are, unlike pan-Arctic thaw lakes, long-term landscape features. Although the ebullitive CH₄ flux ranges discussed here cover the range of observed fluxes from most other high-latitude lake types, including thermokarst lakes and thaw features, the total ebullitive CH₄ flux from the Stordalen lakes is somewhat lower than such fluxes reported elsewhere [Bastviken et al., 2011] for other high-latitude lakes. This means that ebullition projections for wider areas made for the Stordalen lakes would likely be smaller than projections made from other lakes.

Pan-Arctic projections are of great interest and are dependent on lake area remaining substantially unchanged in the future. Future changes in lake area remain an active research area, with studies suggesting both fewer [Smith et al., 2005] and in certain situations more lakes [Jorgenson and Osterkamp, 2005] in a warming Arctic. Although most studies predict general landscape drying based on thawing permafrost, in the period before 2080 discussed here, it has been suggested that a transient increase in lake area may occur [Avis et al., 2011]. Such feedbacks are not accounted for in the predictions made here; any increase in lake area would almost certainly lead to higher total CH₄ ebullition from lakes.

5 Conclusions

In future scenarios, if the range of interannual variability remains similar to today, with annual quasi-stochastic forcing by available SW, previous winter's length and average temperature, a range may be predicted of 2000 to 2400 ± 500 mg CH₄ m⁻² yr⁻¹. Recent years have already approached this range (Figure 2). It is difficult to imagine the ice-out date for the small Stordalen lakes shifting much earlier than about day 110, as the available daily SW decreases rapidly earlier in the year, and may not be able to melt the previous winter's ice cover. However, we do not have adequate data for early ice-out years to evaluate how strong the effect is nor have we evaluated lake ice thickness for such scenarios. Warmer autumns may lengthen the CH₄ ebullition period, though relative lack of SW after 30 September is a strong limit on this effect. Nonetheless, warmer autumns could make our future projections slight underestimates. It is also worth considering if the supply of organic substrate to the lakes may be inadequate to supply the higher levels of CH₄ productivity we have projected; the highest observed production to date is 1850 mg CH₄ m⁻² yr⁻¹ (year 2013 average).

With the predicted temperature increases in the coming century in the subarctic region, multiple feedbacks may be triggered, leading to a large increase of organic carbon availability in the lakes due to horizontal mobilization of labile organic material during permafrost thaw [van Huissteden et al., 2011]. The effects of such changes remain to be understood. Unlike other published predictions, ours are not based on air temperature changes, but rather SW flux changes during the ice-free season that may or may not be concomitant with future warming, as the interannual observed patterns of cloudiness may change. However, it is unlikely that increases in cloudiness could overwhelm the effect of shifting the ice-out date earlier. The results of this study suggest the possibility of a robust annual statistical model of ecosystem responses to climate forcing. Such a statistical model may be used as an important tool for future responses of lake ecosystems to climate change. Further, our results emphasize the importance of continued monitoring of methane emissions and the need for detailed monitoring at selected representative sites in the Arctic region. We suggest that CH₄ ebullition for high-latitude lakes is seasonally modulated by available SW; SW itself is modulated by lakes' ice-out date.

Competing Financial Interests

The authors declare no competing financial interests.