LETTER
doi:10.1038/nature14172
Decrease in CO2 efflux from northern hardwater lakes with increasing atmospheric warming ¨rn Wissel2, Benjamin M. Tutolo3, Gavin L. Simpson2 & Peter R. Leavitt1,2 Kerri Finlay1, Richard J. Vogt1{, Matthew J. Bogard1{, Bjo
Boreal lakes are biogeochemical hotspots that alter carbon fluxes by sequestering particulate organic carbon in sediments1,2 and by oxidizing terrestrial dissolved organic matter to carbon dioxide (CO2) or methane through microbial processes3,4. At present, such dilute lakes release 1.4 petagrams of carbon annually to the atmosphere3,4, and this carbon efflux may increase in the future in response to elevated temperatures5 and increased hydrological delivery of mineralizable dissolved organic matter to lakes6,7. Much less is known about the potential effects of climate changes on carbon fluxes from carbonaterich hardwater and saline lakes that account for about 20 per cent of inland water surface area4,8. Here we show that atmospheric warming may reduce CO2 emissions from hardwater lakes. We analyse decadal records of meteorological variability, CO2 fluxes and water chemistry to investigate the processes affecting variations in pH and carbon exchange9,10 in hydrologically diverse lakes of central North America. We find that the lakes have shifted progressively from being substantial CO2 sources in the mid-1990s to sequestering CO2 by 2010, with a steady increase in annual mean pH. We attribute the observed changes in pH and CO2 uptake to an atmospheric-warminginduced decline in ice cover in spring that decreases CO2 accumulation under ice, increases spring and summer pH, and enhances the chemical uptake of CO2 in hardwater lakes. Our study suggests that rising temperatures do not invariably increase CO2 emissions from aquatic ecosystems. Boreal lakes are important in global carbon (C) cycles because they receive ,2.9 Pg C per year from terrestrial sources3,4, permanently bury ,0.6 Pg per year as particulate C (refs 1, 2), and mineralize up to 50% of the remainder to CO2 and methane4 through bacterial activity in the water column11 and sediments12. In general, dilute unproductive lakes release more gaseous C than is fixed by aquatic photosynthesis11,13,14, whereas net CO2 uptake occurs in some productive basins when elevated nutrient influx intensifies primary production and labile organic C is incompletely mineralized by bacteria3,4,15. At present, the magnitude of C fluxes from boreal lakes is similar to those arising from global deforestation, oceanic CO2 sequestration and net terrestrial production4; however, future mineralization of organic matter is predicted to intensify under a warmer5 or wetter climate6,7. Less is known about how solute-rich hardwater lakes influence planetary C fluxes4,8, despite accounting for ,50% of inland waters by volume16 (23% by area)8, in part because pH regulates inter-annual variation in atmospheric CO2 exchange at these sites independently of microbial metabolism during summer8,9, and because controls of inter-annual variation in pH are poorly understood9,10. Typically, hardwater lakes are alkaline (8 , pH , 11), rich in dissolved inorganic C (DIC) derived from catchment sources of HCO32 and CO322, and evade (release) much more CO2 (up to 200 mmol C m22 d21) than do boreal lakes (up to 60 mmol m22 d21) when pH values are below 9.0 (refs 4, 8–10). At higher pH, CO2 is converted to HCO32 and CO322 (ref. 17), partial pressure of CO2 (pCO2 ) declines to below atmospheric values, and hardwater lakes sequester atmospheric CO2 (refs 8–10). Furthermore, DIC-rich
hardwater and saline lakes exhibit a high degree of spatial synchrony in mean summer pH18,19 and can rapidly vary the direction and magnitude of CO2 flux9. Thus, a better understanding of the mechanisms regulating inter-annual variation in pH and carbon processing of hardwater lakes is essential to quantify the contribution of these ecosystems to the global carbon cycle4,20. We analysed 16 years of meteorological and limnological data collected every two weeks during May to August from six lakes, a 28-year record of weekly chemical determinations at one lake, and surveys of water chemistry in an additional 20 (seasonal) to 70 (annual) lakes to identify factors regulating inter-annual variation in lake pH and CO2 flux within a 236,000 km2 region of the Northern Great Plains of North America (Extended Data Fig. 1). Our grassland study region represents more than 40% of all cultivated land in Canada and is composed mainly (75%) of agricultural fields and pastures, particularly within the 52,000 km2 Qu’Appelle River catchment (50u 009–51u 309 N, 101u 309–107u 109 W) of southern Saskatchewan21. Study lakes within this drainage basin vary tenfold in most morphometric, hydrological and limnological features (Extended Data Table 1), include both reservoirs (Wascana and Diefenbaker lakes) and sites with limited hydrological outflow (Last Mountain Lake), yet are all alkaline (mean summer pH ,8.8; 30–60 mg DIC l21) and well mixed (except occasionally stratified Katepwa Lake) and have common plankton composition and trophic relationships10,21. Analysis of 16 years of water chemistry and C flux data revealed that Qu’Appelle lakes have shifted progressively from being large CO2 sources in the mid-1990s to sequestering substantial amounts of CO2 at present (Fig. 1). The annual pH of these lakes has steadily increased from 8.3 6 0.1 in 1995 to 9.2 6 0.1 in 2010 (means 6 s.e.m.; n 5 6; Fig. 1a), whereas total inorganic carbon (TIC) (Fig. 1b), hydrological influx9 (not shown) and lake production9 (not shown) have remained essentially unchanged. The consequence of these shifts is that aquatic pCO2 has declined nearly tenfold in all lakes (Fig. 1c), atmospheric CO2 evasion has been decreased by nearly 100 g C m22 per summer (Fig. 1d), and lakes now sequester substantial quantities of CO2 (37.4 6 6.5 g C m22 per summer) (Fig. 1d). Despite the marked physical, hydrological and chemical differences between lakes (Extended Data Table 1), interannual variation in pH and CO2 parameters is highly coherent among ecosystems9 and shows spatial patterns of synchrony that are characteristic of ecosystem regulation by energy influx (air temperature, irradiance) rather than by mass influx (precipitation, runoff, solutes)19,22. Principal components analysis suggests that the mean summer pH of Qu’Appelle lakes increased as a function of both spring and annual air temperatures, was correlated inversely with the duration and date of ice melt, and was uncorrelated with other measured meteorological variables (Extended Data Fig. 2a). In particular, pH was elevated under warmer atmospheric conditions, including those associated with a negative Southern Oscillation Index and positive (warm) phase of the Pacific Decadal Oscillation, both of which represent mild winters and reduced ice cover23. In contrast, mean summer pH was not correlated strongly with any measured aspect of lake chemistry, other than ammonium
1
Limnology Laboratory, Department of Biology, University of Regina, Regina, Saskatchewan, Canada S4S 0A2. 2Institute of Environmental Change and Society, University of Regina, Regina, Saskatchewan, Canada S4S 0A2. 3Department of Earth Sciences, University of Minnesota, Minneapolis, Minnesota 55455, USA. {Present addresses: Department of Biology, Trent University, Peterborough, Ontario, Canada K9J 7B8 (R.J.V.); De´partement des Sciences Biologiques, Universite´ du Que´bec a` Montre´al, Montre´al, Que´bec, Canada H3C 3P8 (M.J.B.). 0 0 M O N T H 2 0 1 5 | VO L 0 0 0 | N AT U R E | 1
©2015 Macmillan Publishers Limited. All rights reserved
8.5
60
n.s.
r = 0.609, p = 0.016 3.0 2.5 2.0
d
150
Enhanced CO2 flux (g C m–2 per summer)
Log10 mean summer pCO 2 (μatm)
3.5
1.5
20 0
8.0
c
40
100
r = 0.608,
p = 0.016 100
50 0 –50
–100 1995 2000 2005 Year
2010
1995 2000 2005 Year
2010
Figure 1 | Temporal changes in summer pH and CO2 flux in six hardwater lakes of central Canada. a, Surface water pH; b, total inorganic carbon (TIC) concentration (mg C l21); c, log10 partial pressure of CO2 (matm); d, chemically enhanced flux of CO2 (g C m22 per summer). All time series are unweighted means and s.e.m. (n 5 6). Least-squares regression analysis revealed linear increases in pH and declines in pCO2 and CO2 flux when conducted using years with complete summer sampling. Least-squares regression analyses exclude 2000, a year lacking samples during late July to September. Mean pCO2 of the atmosphere (370 matm) is indicated in c with a horizontal dashed line.
(NH41) concentration (Extended Data Fig. 2b). Together these patterns are consistent with previous observations from other hardwater lakes and suggest that prolonged ice cover arising from cold winters favours increased CO2 accumulation under ice and declines in underice pH after CO2 hydration and carbonic acid formation24,25. Least-squares regression analysis of the decadal time series for Qu’Appelle lakes also showed that warmer winter temperatures were correlated negatively with both the duration of ice cover (Fig. 2a) and the date of ice melt (Fig. 2b), as has been noted elsewhere23,26. During years of prolonged cover, the date of ice melt is delayed by up to 20 days and the pH of Qu’Appelle lakes during spring (see below) and summer can be depressed by up to 1 pH unit (Fig. 2c). As shown in diverse lake districts, prolonged ice cover allows the accumulation of CO2 from mineralized organic matter, which in turn hydrates to lower pH through formation of carbonic acid27,28. Furthermore, as pH is depressed, chemical equilibria dictate that a higher proportion of DIC is present as free CO2, which can evade to the atmosphere17. In Qu’Appelle lakes, summer pH values below 9.0 were associated with substantial CO2 evasion, whereas these lakes captured up to 50 g C m22 during summers with a mean pH of .9.0 (Fig. 2d). Detailed study of Buffalo Pound Lake, Saskatchewan, Canada, within the Qu’Appelle catchment illustrates the linkage between the duration of ice cover, the metabolic production of CO2 and the depression of pH in spring and summer waters (Fig. 3). This lake has been monitored continuously at weekly intervals since 1979, with comprehensive chemical analysis from 1985 to 2003. Here we found a strong negative relationship between the duration of ice cover and the mean lake water pH during March, the month immediately preceding ice melt (Fig. 3a). In addition, spring pH was correlated positively with coeval determinations of oxygen content (Fig. 3b), suggesting that variations in mineralization of organic matter by microbes underlie both patterns27,28. Finally, we found a strong linear relationship between pH during March and values recorded in subsequent months (Fig. 3c), suggesting that variation in under-ice conditions can alter pH during the following summer.
Statistical and geochemical modelling of winter water chemistry from 1985 to 2003 reveals that metabolic production of CO2 was the main control of inter-annual variation in the spring pH of Buffalo Pound Lake. For example, elastic net analysis explained 81% of observed deviance in winter pH and showed that microbial metabolism under ice was the principle predictor of the pH at spring ice melt, with a nearly fourfold greater standardized coefficient (0.14) than either HCO32 or Ca21 (0.04), the only other significant and substantial model predictors (Extended Data Fig. 3). Similarly, geochemical modelling demonstrated that underice CO2 evolution (O2 decline 3 respiratory quotient of 1.2 5 CO2 production) was sufficient to depress the pH from values observed at ice formation (8.32 6 0.06; mean 6 s.e.m.) to those (7.83 6 0.07) similar to values observed at the winter pH minimum (7.78 6 0.08) or date of ice melt (8.06 6 0.08). Finally, geochemical modelling revealed that spring ice melt resulted in a short-lived release of CO2 but that the resultant increase in water-column pH was too small to uncouple the linear relationship between spring and summer pH (Extended Data Fig. 4). Together, these patterns suggest that variation in ice cover regulates the magnitude and direction of atmospheric CO2 exchange by controlling spring and summer pH, altering the duration of the ice-free period and changing the proportion of time in which water-column pH is above or below the threshold of 9.0 (Fig. 2d). Monitoring of other regional lakes since 2002 (refs 18, 29) has revealed that pronounced inter-annual variation in mean and seasonal pH is common and synchronous within the grassland region of central Canada (Extended Data Fig. 5). For example, the mean summer pH of Qu’Appelle lakes during 2002–2009 was highly correlated with that of ,20 DIC-rich closed-basin lakes within an independent 100,000 km2 region (Extended Data Fig. 5a), whereas the rate of seasonal increase in pH was not significantly different between the two groups of lakes (Extended Data Fig. 5b). Furthermore, the chemical and hydrological properties of these closed-basin sites are representative of an additional ,50 DIC-rich hardwater and saline lakes surveyed during the past decade18,29. As shown elsewhere, inter-annual variation in pH within these closed-basin lakes a 180
b 125
r = –0.705, p = 0.015
170
Ice melt (day of year)
9.0
80
160 150 140
r = –0.831, p = 0.002
120 115 110
105 130 –8 –7 –6 –5 –4 –3 –2 –1 –8 –7 –6 –5 –4 –3 –2 –1 Mean February-April temperature (ºC)
c 10.0
r = –0.684, p = 0.020
9.5 9.0 8.5 8.0 105
110 115 120 Ice melt (day of year)
d
150
Enhanced CO2 flux (g C m–2 per summer)
9.5
b
Ice duration (days)
r = 0.618, p = 0.014
Mean summer pH
Mean summer pH
a 10.0
Mean summer TIC (mg C l–1)
RESEARCH LETTER
100
125
r = –0.711, p = 0.014
50 0 -50
–100 8.0
8.5 9.0 9.5 Mean summer pH
10.0
Figure 2 | Effects of winter temperature on ice cover, lake chemistry and CO2 flux in lakes of central Canada. a, b, Least-squares regression analysis of meteorological and lake variables showing correlation of warmer winter temperatures (mean daily uC during February to April) with decreased duration of ice cover (a) and earlier date of ice melting the following spring (b). c, Correlation of date of ice melt with summer pH. d, Correlation of summer pH with CO2 flux from hardwater lakes during summer. Regression analyses were as in Fig. 1, using 11 years with complete data.
2 | N AT U R E | VO L 0 0 0 | 0 0 M O N T H 2 0 1 5
©2015 Macmillan Publishers Limited. All rights reserved
LETTER RESEARCH 190 r = –0.575, p = 0.001
8.0
170
7.5
150
7.0
1990 Year
15
March bottom-water O2 (mg O2 l–1)
b
1980
2000
Ice cover (days)
March bottom-water pH
a 8.5
130 2010
r = 0.832, p
doi:10.1038/nature14172
Decrease in CO2 efflux from northern hardwater lakes with increasing atmospheric warming ¨rn Wissel2, Benjamin M. Tutolo3, Gavin L. Simpson2 & Peter R. Leavitt1,2 Kerri Finlay1, Richard J. Vogt1{, Matthew J. Bogard1{, Bjo
Boreal lakes are biogeochemical hotspots that alter carbon fluxes by sequestering particulate organic carbon in sediments1,2 and by oxidizing terrestrial dissolved organic matter to carbon dioxide (CO2) or methane through microbial processes3,4. At present, such dilute lakes release 1.4 petagrams of carbon annually to the atmosphere3,4, and this carbon efflux may increase in the future in response to elevated temperatures5 and increased hydrological delivery of mineralizable dissolved organic matter to lakes6,7. Much less is known about the potential effects of climate changes on carbon fluxes from carbonaterich hardwater and saline lakes that account for about 20 per cent of inland water surface area4,8. Here we show that atmospheric warming may reduce CO2 emissions from hardwater lakes. We analyse decadal records of meteorological variability, CO2 fluxes and water chemistry to investigate the processes affecting variations in pH and carbon exchange9,10 in hydrologically diverse lakes of central North America. We find that the lakes have shifted progressively from being substantial CO2 sources in the mid-1990s to sequestering CO2 by 2010, with a steady increase in annual mean pH. We attribute the observed changes in pH and CO2 uptake to an atmospheric-warminginduced decline in ice cover in spring that decreases CO2 accumulation under ice, increases spring and summer pH, and enhances the chemical uptake of CO2 in hardwater lakes. Our study suggests that rising temperatures do not invariably increase CO2 emissions from aquatic ecosystems. Boreal lakes are important in global carbon (C) cycles because they receive ,2.9 Pg C per year from terrestrial sources3,4, permanently bury ,0.6 Pg per year as particulate C (refs 1, 2), and mineralize up to 50% of the remainder to CO2 and methane4 through bacterial activity in the water column11 and sediments12. In general, dilute unproductive lakes release more gaseous C than is fixed by aquatic photosynthesis11,13,14, whereas net CO2 uptake occurs in some productive basins when elevated nutrient influx intensifies primary production and labile organic C is incompletely mineralized by bacteria3,4,15. At present, the magnitude of C fluxes from boreal lakes is similar to those arising from global deforestation, oceanic CO2 sequestration and net terrestrial production4; however, future mineralization of organic matter is predicted to intensify under a warmer5 or wetter climate6,7. Less is known about how solute-rich hardwater lakes influence planetary C fluxes4,8, despite accounting for ,50% of inland waters by volume16 (23% by area)8, in part because pH regulates inter-annual variation in atmospheric CO2 exchange at these sites independently of microbial metabolism during summer8,9, and because controls of inter-annual variation in pH are poorly understood9,10. Typically, hardwater lakes are alkaline (8 , pH , 11), rich in dissolved inorganic C (DIC) derived from catchment sources of HCO32 and CO322, and evade (release) much more CO2 (up to 200 mmol C m22 d21) than do boreal lakes (up to 60 mmol m22 d21) when pH values are below 9.0 (refs 4, 8–10). At higher pH, CO2 is converted to HCO32 and CO322 (ref. 17), partial pressure of CO2 (pCO2 ) declines to below atmospheric values, and hardwater lakes sequester atmospheric CO2 (refs 8–10). Furthermore, DIC-rich
hardwater and saline lakes exhibit a high degree of spatial synchrony in mean summer pH18,19 and can rapidly vary the direction and magnitude of CO2 flux9. Thus, a better understanding of the mechanisms regulating inter-annual variation in pH and carbon processing of hardwater lakes is essential to quantify the contribution of these ecosystems to the global carbon cycle4,20. We analysed 16 years of meteorological and limnological data collected every two weeks during May to August from six lakes, a 28-year record of weekly chemical determinations at one lake, and surveys of water chemistry in an additional 20 (seasonal) to 70 (annual) lakes to identify factors regulating inter-annual variation in lake pH and CO2 flux within a 236,000 km2 region of the Northern Great Plains of North America (Extended Data Fig. 1). Our grassland study region represents more than 40% of all cultivated land in Canada and is composed mainly (75%) of agricultural fields and pastures, particularly within the 52,000 km2 Qu’Appelle River catchment (50u 009–51u 309 N, 101u 309–107u 109 W) of southern Saskatchewan21. Study lakes within this drainage basin vary tenfold in most morphometric, hydrological and limnological features (Extended Data Table 1), include both reservoirs (Wascana and Diefenbaker lakes) and sites with limited hydrological outflow (Last Mountain Lake), yet are all alkaline (mean summer pH ,8.8; 30–60 mg DIC l21) and well mixed (except occasionally stratified Katepwa Lake) and have common plankton composition and trophic relationships10,21. Analysis of 16 years of water chemistry and C flux data revealed that Qu’Appelle lakes have shifted progressively from being large CO2 sources in the mid-1990s to sequestering substantial amounts of CO2 at present (Fig. 1). The annual pH of these lakes has steadily increased from 8.3 6 0.1 in 1995 to 9.2 6 0.1 in 2010 (means 6 s.e.m.; n 5 6; Fig. 1a), whereas total inorganic carbon (TIC) (Fig. 1b), hydrological influx9 (not shown) and lake production9 (not shown) have remained essentially unchanged. The consequence of these shifts is that aquatic pCO2 has declined nearly tenfold in all lakes (Fig. 1c), atmospheric CO2 evasion has been decreased by nearly 100 g C m22 per summer (Fig. 1d), and lakes now sequester substantial quantities of CO2 (37.4 6 6.5 g C m22 per summer) (Fig. 1d). Despite the marked physical, hydrological and chemical differences between lakes (Extended Data Table 1), interannual variation in pH and CO2 parameters is highly coherent among ecosystems9 and shows spatial patterns of synchrony that are characteristic of ecosystem regulation by energy influx (air temperature, irradiance) rather than by mass influx (precipitation, runoff, solutes)19,22. Principal components analysis suggests that the mean summer pH of Qu’Appelle lakes increased as a function of both spring and annual air temperatures, was correlated inversely with the duration and date of ice melt, and was uncorrelated with other measured meteorological variables (Extended Data Fig. 2a). In particular, pH was elevated under warmer atmospheric conditions, including those associated with a negative Southern Oscillation Index and positive (warm) phase of the Pacific Decadal Oscillation, both of which represent mild winters and reduced ice cover23. In contrast, mean summer pH was not correlated strongly with any measured aspect of lake chemistry, other than ammonium
1
Limnology Laboratory, Department of Biology, University of Regina, Regina, Saskatchewan, Canada S4S 0A2. 2Institute of Environmental Change and Society, University of Regina, Regina, Saskatchewan, Canada S4S 0A2. 3Department of Earth Sciences, University of Minnesota, Minneapolis, Minnesota 55455, USA. {Present addresses: Department of Biology, Trent University, Peterborough, Ontario, Canada K9J 7B8 (R.J.V.); De´partement des Sciences Biologiques, Universite´ du Que´bec a` Montre´al, Montre´al, Que´bec, Canada H3C 3P8 (M.J.B.). 0 0 M O N T H 2 0 1 5 | VO L 0 0 0 | N AT U R E | 1
©2015 Macmillan Publishers Limited. All rights reserved
8.5
60
n.s.
r = 0.609, p = 0.016 3.0 2.5 2.0
d
150
Enhanced CO2 flux (g C m–2 per summer)
Log10 mean summer pCO 2 (μatm)
3.5
1.5
20 0
8.0
c
40
100
r = 0.608,
p = 0.016 100
50 0 –50
–100 1995 2000 2005 Year
2010
1995 2000 2005 Year
2010
Figure 1 | Temporal changes in summer pH and CO2 flux in six hardwater lakes of central Canada. a, Surface water pH; b, total inorganic carbon (TIC) concentration (mg C l21); c, log10 partial pressure of CO2 (matm); d, chemically enhanced flux of CO2 (g C m22 per summer). All time series are unweighted means and s.e.m. (n 5 6). Least-squares regression analysis revealed linear increases in pH and declines in pCO2 and CO2 flux when conducted using years with complete summer sampling. Least-squares regression analyses exclude 2000, a year lacking samples during late July to September. Mean pCO2 of the atmosphere (370 matm) is indicated in c with a horizontal dashed line.
(NH41) concentration (Extended Data Fig. 2b). Together these patterns are consistent with previous observations from other hardwater lakes and suggest that prolonged ice cover arising from cold winters favours increased CO2 accumulation under ice and declines in underice pH after CO2 hydration and carbonic acid formation24,25. Least-squares regression analysis of the decadal time series for Qu’Appelle lakes also showed that warmer winter temperatures were correlated negatively with both the duration of ice cover (Fig. 2a) and the date of ice melt (Fig. 2b), as has been noted elsewhere23,26. During years of prolonged cover, the date of ice melt is delayed by up to 20 days and the pH of Qu’Appelle lakes during spring (see below) and summer can be depressed by up to 1 pH unit (Fig. 2c). As shown in diverse lake districts, prolonged ice cover allows the accumulation of CO2 from mineralized organic matter, which in turn hydrates to lower pH through formation of carbonic acid27,28. Furthermore, as pH is depressed, chemical equilibria dictate that a higher proportion of DIC is present as free CO2, which can evade to the atmosphere17. In Qu’Appelle lakes, summer pH values below 9.0 were associated with substantial CO2 evasion, whereas these lakes captured up to 50 g C m22 during summers with a mean pH of .9.0 (Fig. 2d). Detailed study of Buffalo Pound Lake, Saskatchewan, Canada, within the Qu’Appelle catchment illustrates the linkage between the duration of ice cover, the metabolic production of CO2 and the depression of pH in spring and summer waters (Fig. 3). This lake has been monitored continuously at weekly intervals since 1979, with comprehensive chemical analysis from 1985 to 2003. Here we found a strong negative relationship between the duration of ice cover and the mean lake water pH during March, the month immediately preceding ice melt (Fig. 3a). In addition, spring pH was correlated positively with coeval determinations of oxygen content (Fig. 3b), suggesting that variations in mineralization of organic matter by microbes underlie both patterns27,28. Finally, we found a strong linear relationship between pH during March and values recorded in subsequent months (Fig. 3c), suggesting that variation in under-ice conditions can alter pH during the following summer.
Statistical and geochemical modelling of winter water chemistry from 1985 to 2003 reveals that metabolic production of CO2 was the main control of inter-annual variation in the spring pH of Buffalo Pound Lake. For example, elastic net analysis explained 81% of observed deviance in winter pH and showed that microbial metabolism under ice was the principle predictor of the pH at spring ice melt, with a nearly fourfold greater standardized coefficient (0.14) than either HCO32 or Ca21 (0.04), the only other significant and substantial model predictors (Extended Data Fig. 3). Similarly, geochemical modelling demonstrated that underice CO2 evolution (O2 decline 3 respiratory quotient of 1.2 5 CO2 production) was sufficient to depress the pH from values observed at ice formation (8.32 6 0.06; mean 6 s.e.m.) to those (7.83 6 0.07) similar to values observed at the winter pH minimum (7.78 6 0.08) or date of ice melt (8.06 6 0.08). Finally, geochemical modelling revealed that spring ice melt resulted in a short-lived release of CO2 but that the resultant increase in water-column pH was too small to uncouple the linear relationship between spring and summer pH (Extended Data Fig. 4). Together, these patterns suggest that variation in ice cover regulates the magnitude and direction of atmospheric CO2 exchange by controlling spring and summer pH, altering the duration of the ice-free period and changing the proportion of time in which water-column pH is above or below the threshold of 9.0 (Fig. 2d). Monitoring of other regional lakes since 2002 (refs 18, 29) has revealed that pronounced inter-annual variation in mean and seasonal pH is common and synchronous within the grassland region of central Canada (Extended Data Fig. 5). For example, the mean summer pH of Qu’Appelle lakes during 2002–2009 was highly correlated with that of ,20 DIC-rich closed-basin lakes within an independent 100,000 km2 region (Extended Data Fig. 5a), whereas the rate of seasonal increase in pH was not significantly different between the two groups of lakes (Extended Data Fig. 5b). Furthermore, the chemical and hydrological properties of these closed-basin sites are representative of an additional ,50 DIC-rich hardwater and saline lakes surveyed during the past decade18,29. As shown elsewhere, inter-annual variation in pH within these closed-basin lakes a 180
b 125
r = –0.705, p = 0.015
170
Ice melt (day of year)
9.0
80
160 150 140
r = –0.831, p = 0.002
120 115 110
105 130 –8 –7 –6 –5 –4 –3 –2 –1 –8 –7 –6 –5 –4 –3 –2 –1 Mean February-April temperature (ºC)
c 10.0
r = –0.684, p = 0.020
9.5 9.0 8.5 8.0 105
110 115 120 Ice melt (day of year)
d
150
Enhanced CO2 flux (g C m–2 per summer)
9.5
b
Ice duration (days)
r = 0.618, p = 0.014
Mean summer pH
Mean summer pH
a 10.0
Mean summer TIC (mg C l–1)
RESEARCH LETTER
100
125
r = –0.711, p = 0.014
50 0 -50
–100 8.0
8.5 9.0 9.5 Mean summer pH
10.0
Figure 2 | Effects of winter temperature on ice cover, lake chemistry and CO2 flux in lakes of central Canada. a, b, Least-squares regression analysis of meteorological and lake variables showing correlation of warmer winter temperatures (mean daily uC during February to April) with decreased duration of ice cover (a) and earlier date of ice melting the following spring (b). c, Correlation of date of ice melt with summer pH. d, Correlation of summer pH with CO2 flux from hardwater lakes during summer. Regression analyses were as in Fig. 1, using 11 years with complete data.
2 | N AT U R E | VO L 0 0 0 | 0 0 M O N T H 2 0 1 5
©2015 Macmillan Publishers Limited. All rights reserved
LETTER RESEARCH 190 r = –0.575, p = 0.001
8.0
170
7.5
150
7.0
1990 Year
15
March bottom-water O2 (mg O2 l–1)
b
1980
2000
Ice cover (days)
March bottom-water pH
a 8.5
130 2010
r = 0.832, p
