Eutrophication effects on greenhouse gas fluxes from shallow-lake mesocosms override those of climate warming

Introduction

Fresh waters and their sediments are important locations of the biogeochemical processes involved in the cycling of the three main greenhouse gases (GHGs) – carbon dioxide (CO₂), methane (CH₄) and nitrous oxide (N₂O). Inland waters make a disproportionately large, relative to area, contribution to GHG emissions (Bastviken et al., 2011; Cole et al., 2007). Shallow lakes (<3 m average depth) are not only efflux hot spots (Bastviken et al., 2011; Tranvik et al., 2009) but also the most numerous type of water body globally, both by number and by area (Downing et al., 2006; Verpoorter et al., 2014). Furthermore, shallow lakes are very numerous at higher latitudes where temperature is rising fastest (Verpoorter et al., 2014).

At microbial levels, the processes controlling GHG production, that is metabolism, are temperature dependent as their rates are controlled by biochemical kinetics (Brown et al., 2004). Theoretical models have linked the metabolism of individual organisms to communities and ecosystems (Brown et al., 2004) to demonstrate how temperature drives the global carbon cycle (Allen et al., 2005; Yvon-Durocher et al., 2010) and methane fluxes from fresh waters (Yvon-Durocher et al., 2014). Empirical evidence for the ecosystem-scale temperature dependence of metabolism has, to a large extent, relied on the correlation between seasonal temperature change and seasonal variation in GHG fluxes (Allen et al., 2005; Enquist et al., 2003; Giardina & Ryan, 2000; Yvon-Durocher et al., 2014, 2010). Such studies may be adequate for demonstrating the temperature dependence of the biogeochemical processes, but the key question is whether they are useful for predicting the likely response of ecosystem-level GHG flux to climate change.

Seasonal temperature variation is associated with the dynamics of a number of abiotic and biotic factors that can affect GHG fluxes, including primary producer biomass (Michaletz et al., 2014) and switches in limiting resources – such as nutrients (Fernández-Martínez et al., 2014), food web structure (Schindler et al., 1997) and biotic interactions (Atwood et al., 2013). In addition, resource quantity, such as dissolved organic carbon, can control rates of methanogenesis regardless of temperature (Whalen, 2005). The controls of N₂O fluxes from lakes are less well established, but denitrification and nitrification are considered to be the main processes producing N₂O (Seitzinger et al., 2000) and, besides temperature and nitrogen availability, these two processes are greatly influenced by oxygen, pH and carbon availability.

Studies of net primary production in forest ecosystems have noted a dichotomy between the strong evidence for temperature dependence of ecosystem metabolism based on the correlation of seasonal temperature variation with seasonal variation in net primary production, contrasted with a lack of strong evidence for temperature dependence when comparing sites across space using annual temperature and flux data (Enquist et al., 2003). Furthermore, even where the seasonal data are used, the introduction of factors only indirectly linked to temperature, such as primary producer biomass and nutrient concentrations, revealed that, for forest ecosystems, biomass and nutrients are more important than temperature and precipitation in controlling net primary production (Fernández-Martínez et al., 2014; Michaletz et al., 2014). There may be a similar dichotomy in fresh waters. Analysis using the correlation between seasonal change in temperature and seasonal variation in GHG flux reveals a strong relationship between temperature and GHG fluxes (Yvon-Durocher et al., 2014, 2010), whereas when viewed as annual mean temperature and annual flux the relationship is much weaker, even for methanogenesis, the most temperature dependent of these processes (Bastviken et al., 2004; Yvon-Durocher et al., 2014). Thus, to make robust predictions of likely climate-change effects, it is important to test whether the relationship between temperature and gas flux, established from the correlation of GHG flux with seasonal temperature change, provides a reliable means to predict climate-change effects at the ecosystem level.

In aquatic systems, nutrient levels and biotic interactions have the potential to affect the biogeochemical processes producing and consuming GHGs. For example, in a range of freshwater habitats, predation pressure was shown to influence CO₂ concentrations, largely by altering the biomass of the primary producers (Atwood et al., 2013). In lakes, nutrient enrichment has been shown to promote carbon sequestration by increasing biomass (Trolle et al., 2012), which results in increased carbon burial in sediments (Anderson et al., 2014). While it is well known how emergent vegetation acts as a shortcut for GHG emissions between sediments and atmosphere (Bastviken et al., 2004), few studies have considered how submerged plants affect biogeochemical cycles. Submerged plants affect almost all biological and chemical processes in shallow lakes, and shallow areas of deeper lakes, influencing the biomass of organisms across trophic levels from sediment microorganisms to fish (Jeppesen et al., 1998). Submerged plants may therefore influence GHG dynamics in shallow lakes, and recent work has highlighted their potential importance in carbon dynamics from small shallow lakes (Brothers et al., 2013; Natchimuthu et al., 2014; Walter Anthony et al., 2014).

Whether climate change will result in shallow lakes having positive or negative feedbacks to atmospheric GHG concentrations is not clear (Walter Anthony et al., 2014; Walter et al., 2006), in part due to the uncertainty surrounding how the temperature-dependent processes scale up within the framework of a complex ecosystem. Given their abundance and geographic location (Verpoorter et al., 2014), it is important to determine the relative importance of temperature in controlling GHGs fluxes from shallow lakes. Here, we use the longest running freshwater mesocosm experiment in existence (Liboriussen et al., 2005) to assess the drivers of GHG fluxes in shallow lakes. This experimental system offers a unique opportunity to assess the relative importance of nutrient enrichment, seasonal temperature variation and experimental warming in shaping GHG fluxes from a realistic facsimile of an ecosystem.

Materials and methods

Experimental set-up and data

The climate-change mesocosm experiment situated in Central Jutland, Denmark (56°14′N, 9°31′E), has been running continuously since August 2003 and is the longest running lake mesocosm experiment investigating the effects of climate change. It consists of twenty-four fully mixed, outdoor, flow-through mesocosms (diameter 1.9 m, water depth 1 m, retention time ~2.5 months). The experiment set-up is fully described in Liboriussen et al. (2005). In brief, it is a fully factorial-designed experiment with three temperature treatments: unheated ambient temperature (AMB), the A2 scenario as described by the IPCC (Houghton et al., 2001) (coded A2) and A2 scenario +50% (coded A2+). The mesocosms contain sediments taken from a number of shallow lakes and homogenized prior to the experiment starting; thus, the starting conditions for each individual mesocosm were identical.

The three temperature treatments are crossed with two nutrient treatments resulting in a total of six treatments each with four replicates each. The mesocosms are fed by ground water, and the low-nutrient mesocosms receive no additional nutrients. The loading of the high-nutrient mesocosms is 7 mg P m⁻² day⁻¹ and 27.1 mg N m⁻² day⁻¹, respectively. This resulted in mean values of 0.23 mg L⁻¹ TN and 12.8 μg L⁻¹ TP for the low-nutrient treatment and 3.53 mg L⁻¹ TN and 186 μg L⁻¹ TP for the high-nutrient treatment for the time period of the study presented here. The two nutrient treatments result in either a clear water shallow lake or a turbid water shallow lake at low and high levels, respectively (Scheffer et al., 1993).

Mesocosm water temperature reflects the seasonal variation in air temperature, with the experimental temperature treatments A2 and A2+ being 2–4 °C and 4–6 °C higher than in the reference mesocosm (AMB), respectively. This temperature difference among treatments is based on predictions of downscaled climate-change scenarios, using 1961–1990 as the reference period (Houghton et al., 2001) (Fig. 1).

Gas measurements and flux calculations

Aqueous CO₂, CH₄ and N₂O in the mesocosms were determined monthly between March and September 2011 and bimonthly from September 2011 to January 2012, eight years after the initiation of the experiment. Direct measurements of CO₂, CH₄ and N₂O were performed by headspace equilibration following the method described in Raymond et al. (1997). Briefly, a thermo-insulated 1.2-L bottle was carefully filled with water sampled at the water surface. Thereafter, the bottle was capped underwater with a specially designed lid equipped with rubber septa to enable the introduction of a 50 mL headspace of ambient air using a syringe. The bottle was then vigorously shaken for 120 s to allow for equilibration between gas and water phases. Samples of both the headspace and the ambient air were taken and brought to the laboratory for analysis. For each mesocosm, duplicate equilibrations were made. The temperature of the water was recorded at the time of sampling. The gas samples of CO₂, CH₄ and N₂O were stored at room temperature and in the dark before further processing, typically after 3–5 days, it is highly unlikely that there was a loss of sample integrity over this time period (Raymond et al., 1997).

Carbon dioxide, CH₄ and N₂O concentrations were determined on a dual-inlet Agilent 7890 gas chromatograph (GC) system interfaced with a CTC CombiPal autosampler (Agilent, Nærum, Denmark) (Petersen et al., 2012). Aqueous concentrations in CO₂, CH₄ and N₂O were calculated from the headspace gas concentrations according to Henry's law and using Henry's constant corrected for temperature and salinity (Weiss, 1974; Weiss & Price, 1980; Wiesenburg & Guinasso, 1979).

The fluxes of CO₂, CH₄ and N₂O between the water and the overlying atmosphere were estimated as

(1)

where f_g (g m⁻² h⁻¹) is the flux of a specific gas g, k_g (m h⁻¹) is the piston velocity of the gas, β_g is the chemical enhancement factor, and C_wat,g - C_eq,g (g m⁻³) is the gradient of concentration between the concentration of gas dissolved in the water (C_wat,g) and the concentration of gas the water would have at equilibrium with the atmosphere (C_eq,g).

Gas transfer velocities for the different gases were calculated based on direct measurement of piston velocity in the mesocosms for O₂ ( = 0.0088 m h⁻¹ at 20 °C) (Jähne et al., 1987; Liboriussen et al., 2011):

(2)

where Sc_g is the Schmidt number calculated following the equation in Wanninkhof (1992) and including water temperature. We chose x = −2/3 as this factor is used for smooth liquid surface (Deacon, 1981). At high pH and low wind speeds, the transfer of CO₂ from the water to the atmosphere can be enhanced because of the chemical reaction between CO₂ and OH^-. This effect was taken into account, and the chemical enhancement β_CO2 was estimated as a function of pH following the model from (Bade & Cole, 2006). For N₂O and CH₄, β was set to 1 (i.e. no chemical enhancement).

Total greenhouse gas flux was calculated as the sum of CO₂, CH₄ and N₂O after conversion to CO₂ equivalents taking a global warming potential (GWP) of 25 for CH₄ and 298 for N₂O.

Gross primary production and ecosystem respiration calculation

Oxygen was measured every 30 min using oxygen probes (OxyGuard^®, two-wire probe, model 420), and a submersible pump (Sicce pumps, model: Micra) connected to the oxygen probe creates continuous horizontal water flow over the membrane and, thus, reduces the risk of instability due to fouling. All probes are calibrated weekly. Gross primary production (GPP) and ecosystem respiration (ER) were estimated using previously published formulas (Erlandsen & Thyssen, 1983; Mahlon et al., 1983)

(3)

where

T_W : water temperature °C

DO_sat: dissolved oxygen saturation concentration mg l⁻¹

DO: dissolved oxygen concentration mg l⁻¹

I_t: light intensity – photosynthetically active radiation (PAR) mol m⁻² 30 min⁻¹,

K₂ is the piston velocity which is 0.0088 h⁻¹ based on direct measurements (Liboriussen et al., 2011), and, ρ₂₀, π₂₀ and η are estimated parameters (using the secant nonlinear method, PROC NLIN, in the statistical SAS 9.3). Piston velocity varies with wind speed and convective mixing. However, at low wind speed (<3 m s⁻¹) there is very little effect on the piston velocity (Cole & Caraco, 1998), and this was the case in our study as the mesocosms were relatively well sheltered from the wind. The piston velocity for CH₄, CO₂ and N₂O included temperature correction. Given that conditions are identical across the mesocosms, the use of a single value still allows reliable comparison of the fluxes. We used this method rather than the simple Odum method (Odum, 1956) that has been commonly used in lake metabolism research as Eqn 3 has a term for the temperature dependence of respiration and temperature shows substantial diurnal variation in shallow lakes. Moreover, simultaneous calibration of the parameters ρ₂₀, π₂₀ and η (respiration and photosynthesis parameters) was demonstrated by stream researchers to be a robust method (Mahlon et al., 1983). Net ecosystem production (NEP) was determined as the difference between GPP and ER (mg O₂ l⁻¹ day⁻¹), and DO was represented as the dissolved oxygen percentage saturation (DO%), calculated as a function of T_W (Lin, 2001):

Results

Nutrient treatment dictated the ecological structure, that is dominance by submerged plants or phytoplankton, of the mesocosms. The low-nutrient mesocosms had clear water and submerged plants, and pelagic chlorophyll a biomass was almost always low (Fig. 1). The high-nutrient mesocosms generally had turbid water with abundant phytoplankton, reflected by high pelagic chlorophyll a (Fig. 1). On rare occasions, macrophytes or filamentous algae were relatively abundant, albeit at much lower abundance and for a relatively short period compared with the low-nutrient mesocosms (Fig. 1). Gross primary production and ecosystem respiration varied with the seasonal progression of temperature and were higher, particularly in mid-summer, in the high-nutrient treatment (Fig. 1).

Testing nutrient vs temperature effects on GHG fluxes

When considering the whole data set, including both high- and low-nutrient treatments, there were no detectable effects of temperature on the total annual GHG flux, which is in marked contrast to the quite strong nutrient effect (Table 1). Similarly, the monthly observations, analysed using mixed-effects models, demonstrated no significant temperature effects on GHG flux when high and low nutrient data were considered together, whereas nutrient treatment was highly significant (Table 2). The monthly observations showed that CO₂ flux (P = 0.0002, pseudo-R² = 0.055) and CH₄ flux (P = 0.0001, pseudo-R² = 0.079) were both lower at high nutrient levels, although the models had little explanatory power. High nutrient levels saw an increase in N₂O flux with a larger amount of variance explained by nutrient treatment (P = 0.0001, pseudo-R² = 0.320) (Fig. 2). The total annual flux data showed the same general pattern of lower CO₂ (P = 0.03, R² = 0.19) and CH₄ flux (P = 0.0001, R² = 0.49) but increased N₂O flux (P = 0.0006, R² = 0.52) at high nutrient levels (Fig. 3), with much larger proportions of the variance explained by nutrient treatment, particularly for CH₄ and N₂O.

Table 1. Testing treatment effects using anova (nutrients and temperature) on total annual flux data on the whole data set and then low- and high-nutrient treatments separately. The final model contains the factors selected as best explaining total annual flux data using likelihood ration tests. Where results are significant (in bold), the nature of the relationship is given (positive or negative to increased nutrients or temperature treatment) along with the P-value and the R² value

Data set	Test	CO2	CH4	N2O
Whole	Temperature treatment	ns	ns	ns
	Nutrient treatment	(negative) P  =   0.032, R 2  = 0.19	(negative) P  =   0.0001, R 2  = 0.49	(positive) P  =   0.00006, R 2  = 0.52
Low nutrient	Temperature treatment	ns	ns	ns
	Final model	TOTPVI (positive) P  =   0.015 R 2  = 0.46	ER (positive) MACPVI (negative) P  =   0.0089, R 2  = 0.75	MACPVI (negative) P  =   0.0399 R 2  = 0.36
High nutrient	Temperature treatment	ns	ns	ns
	Final model	TOTPVI (negative) Chla (negative) ES unstable (positive) P  =   0.0087, R 2  = 0.74	ns	ns

• NEP, net ecosystem production (mg O₂ l⁻¹ day⁻¹); ER, ecosystem respiration (mg O₂ l⁻¹ day⁻¹); TOTPVI, total submerged plant abundance, both macrophytes and filamentous algae (PVI); MACPVI, total macrophyte abundance excluding filamentous algae (PVI); ES, ecosystem stability; and Chla, chlorophyll a (μg l⁻¹).

Table 2. Tests for significant treatment effects using monthly observations of the different GHG fluxes applying mixed-effects models. Temperature treatment effects are not given for the whole data set as they were all nonsignificant. Temperature treatment effect details: T1 is Reference vs A2 scenario (+2–4 °C); and T2 is Reference vs A2+ (+4–6 °C). Significant relationships in bold

Data set	Treatment	Gas	Slope	df	Pseudo-R2	P
Whole	Nutrient	CO 2	−0.049	206	0.055	0.0002
	Nutrient	CH 4	−0.364	206	0.079	0.0001
	Nutrient	N 2 O	0.392	206	0.320	0.0001
Low nutrient	Temperature	CO 2	T1 −0.062 ± 0.025 T2 −0.049 ± 0.025	97	0.058	0.036
	Temperature	CH4	T1 T2	97	0.0003	0.959
	Temperature	N2O	T1 T2	97	0.036	0.144
High nutrient	Temperature	CO2	T1 T2	97	0.005	0.6865
	Temperature	CH 4	T1 0.461 ± 0.181 T2 0.644 ± 0.181	97	0.043	0.0016
	Temperature	N 2 O	T1 −0.337 ± 0.15 T2 0.120 ± 0.15	97	0.068	0.01

What best explains GHG fluxes at different nutrient levels?

Total annual flux

At low nutrients, total annual CO₂ flux was strongly negatively related to macrophyte abundance (P = 0.015, R² = 0.46). At high nutrient levels, total annual CO₂ was strongly negatively related to chlorophyll a. In the latter case, ecosystem stability and macrophyte abundance were also important factors. Ecosystem instability, characterized by a large drop in primary producer abundance, resulted in large effluxes of CO₂, where macrophytes were abundant and CO₂ flux was reduced (Table 1).

At low nutrients, the best model explaining total annual CH₄ flux contained ecosystem respiration and macrophyte abundance, excluding filamentous algae (P = 0.0089, R² = 0.75). Respiration was positively related to CH₄ flux, whereas macrophyte abundance was negatively related to CH₄ flux. At low nutrients, the model best explaining total N₂O flux also contained only macrophyte abundance, excluding filamentous algae, as submerged plants correlated with reduced N₂O flux (P = 0.0399 R² = 0.36). At high-nutrient treatment, there were no significant predictors of total annual flux for CH₄ or N₂O.

Monthly observations

The results of the mixed-effects models suggest that primary producer biomass, either submerged plants or phytoplankton at low and high nutrient levels, respectively, is important for shaping the CO₂ flux (Fig. 4a, b). At low nutrient levels, the selected model for CO₂ flux included macrophyte abundance and NEP and an interaction term (P < 0.0001, pseudo-R² = 0.62) (Table 3). At high nutrients, chlorophyll a biomass, macrophyte abundance and ecosystem stability (ES) were the key factors explaining a large proportion of the variance in CO₂ flux (P < 0.0001, pseudo-R² = 0.56) (Fig. 4b). At high nutrients, this direct relationship between primary producer biomass (chlorophyll a) and CO₂ flux was complicated by the occasional occurrence of submerged plants or filamentous algae, which use CO₂ to build biomass, and thus reduced CO₂ flux, below what would be expected for a given chlorophyll a value. Another factor shaping the CO₂ flux was ecosystem stability; unstable ecosystems that experienced a crash in the abundance of chlorophyll a had a much larger efflux of CO₂ than expected for a given chlorophyll a value (Table 3, Fig. 4b).

Table 3. Final models chosen by the model selection process using monthly observations and mixed-effects models explaining each of the GHG fluxes from the shallow-lake mesocosms: pseudo-R² values are high, indicating good explanatory power

Nutrient level	Gas	Slope	Pseudo-R2	P
Low	CO2	NEP −0.061 ± 0.008	0.62	<0.0001
		TOTPVI −0.036 ± 0.009
		NEP:TOTPVI 0.018 ± 0.008
Low	CH4	TOTPVI −0.17 ± 0.048	0.37	0.0027
		ER 0.27 ± 0.084
Low	N2O	NEP 0.07 ± 0.02	0.20	<0.0001
		ES1 −0.51 ± 0.10
		TOTPVI 0.54 ± 0.12
		ES1:TOTPVI −0.60 ± 0.12
High	CO2	ES1 −0.80 ± 0.06	0.62	<0.0001
		TOTPVI −0.86 ± 0.11
		Chla −0.12 ± 0.02
		ES:TOTPVI 0.75 ± 0.12
High	CH4	ER 2.87 ± 1.21	0.43	0.0028
High	N2O	Chla −0.59 ± 0.10	0.13	<0.0001
		ES1 −0.48 ± 0.23

• NEP, net ecosystem production (mg O₂ l⁻¹ day⁻¹); TOTPVI, total macrophyte abundance (PVI); Chla, chlorophyll a (μg l⁻¹); ES, ecosystem stability; ER, ecosystem respiration (mg O₂ l⁻¹ day⁻¹).

At both low and high nutrient levels, ecosystem respiration was the factor that explained most variance in the CH₄ flux (Fig. 4c, d and Table 3). At high nutrient levels, there were some differences between the warmed and the ambient temperature mesocosms (Fig. 4d); however, this was not enough for the model selection process to include temperature treatment as a significant factor. At low nutrient levels, in addition to the effects of ecosystem respiration, submerged plant abundance was associated with a significant drop in CH₄ flux (ER and macrophyte abundance P < 0.0001, pseudo-R² = 0.51), abundant plants apparently reducing CH₄ fluxes below what was predicted for a given level of ecosystem respiration (Fig. 4c).

The patterns of N₂O flux were less clear; however, there was no clear temperature effect, but at high nutrients higher chlorophyll a led to lower N₂O (Table 3). Macrophyte abundance and ecosystem stability were the best predictors of N₂O flux from the low-nutrient mesocosms, with more plants and an unstable system leading to more N₂O (Table 3).

Discussion

To understand and predict how climate change will impact GHG fluxes from shallow lakes, it is important to determine how the processes controlling those different GHG fluxes will react to climate change from within the complex interconnected framework of the ecosystem. Mesocosm experiments can provide just that framework, balancing the inevitable abstraction of reality against the potential to isolate the experimental effects not only on ecosystem structure but also on ecosystem processes (Benton et al., 2007). What emerges from the different levels of analyses presented here is that temperature increase, within the expected range of climate change, has either no effect or very little significant effect on GHG fluxes from shallow lakes. Furthermore, where a positive temperature effect was detected, it had the potential to be confounded by nutrient levels. The total annual flux data illustrate how large the impact of nutrient levels is compared with that of temperature treatment (Fig. 3).

The separate analyses of data from low- and high-nutrient treatments provided a more sensitive test of the existence and nature of temperature effects on the different GHG fluxes. Given the positive cellular-level temperature dependence of the processes producing some GHGs, the existence of only one, rather weak, positive temperature treatment effect is noteworthy. This positive relationship, for CH₄ flux at high nutrients, is in agreement with studies that have used the framework of metabolic theory to determine the temperature dependence of CH₄ fluxes from cells to ecosystems (Yvon-Durocher et al., 2014). However, while both the warmer treatments (A2 and A2+) had higher CH₄ flux compared with the ambient temperature, the flux from the warmest treatment (A2+) was not significantly higher than that from the A2 treatment, despite a 2 °C difference. This suggests that even for methanogenesis, the most temperature dependent of the processes generating GHGs, other factors can have a significant influence on the ecosystem-level CH₄ flux. Metabolic theory also predicts that warmer temperatures will stimulate respiration to a greater degree than primary production, resulting in a net increase in CO₂ efflux (Yvon-Durocher et al., 2010). The negative association between temperature treatment and CO₂ efflux at low nutrients is therefore the opposite pattern to that predicted by metabolic theory. Our findings are in agreement with a recent study which found increased temperature was associated with reduced CO₂ efflux from boreal lakes (Finlay et al., 2015). The final significant temperature treatment effect on N₂O at high nutrients had a complex response, as the A2 scenario had higher emissions than the ambient but the A2+ treatment lower than ambient, suggesting an indirect or nonlinear influence of temperature. The low explanatory power (Table 2), or lack of significance (Table 1) of temperature treatment, suggests that at the ecosystem scale, other factors can be more important than temperature in shaping GHG fluxes from shallow lakes.

These other factors affecting the different GHG fluxes are suggested by the final models for both annual (Table 1) and monthly observations (Table 3). Determining which variable, or combination of variables, controls variation in GHG fluxes from data collected over a seasonal cycle is problematic as many of the potential drivers covary, increasing as summer progresses and then falling in the autumn. For example, temperature, as opposed to temperature treatment, has proved to be excellent at explaining patterns in ER (Yvon-Durocher et al., 2012) and CH₄ fluxes (Yvon-Durocher et al., 2014), where correlations between seasonal temperature variation and seasonal change in metabolic processes or CH₄ flux have been used. A pertinent question, however, is whether seasonal temperature measurements can be used, not only to explain the observed patterns but also to predict the effect of increased annual temperature on the different GHG fluxes. The experimental approach used here is uniquely placed to address this question, and the absence of strong temperature treatment effects suggests that while seasonal temperature variation strongly correlates with GHG flux, it does not represent a relationship capable of predicting ecosystem-scale response to climate change.

There are large uncertainties associated with teasing out important relationships shaping GHG fluxes using seasonally collected, covarying, explanatory data. Thus, the variables chosen by the models as optimal at explaining the GHG fluxes from the monthly seasonally varying observations (Table 3, Fig. 4) should be treated with a degree of caution. The total annual flux data are, however, free from such problems. The variables selected by analyses of both monthly observation and total annual flux were similar, engendering a higher degree of confidence in the veracity of the results. Both analyses identified primary producer abundance as strongly linked to CO₂ flux, independent of temperature treatment (Table 3, Fig. 4a, b). At low nutrients, this was macrophyte abundance, and at high nutrients, phytoplankton biomass was the most important factor, a relationship demonstrated in lakes with time series data (Trolle et al., 2012). Ecosystem respiration was included in the final model explaining the diffusive flux of CH₄ for the mesocosms both with and without added nutrients. Increased respiration results in lower oxygen and perhaps more prevalent anoxia in the sediment, a condition crucial to the carbon decomposition processes that ends with CH₄ formation (Whalen, 2005). Perhaps a more surprising variable selected at low nutrient concentrations, from both monthly and total annual flux data, was macrophyte abundance, which was associated with a reduction in the diffusive flux of CH₄ (Fig. 4c). Phytoplankton biomass did not have the same effect on CH₄ diffusive flux (Fig. 4d) at high nutrient concentration, indicating that the reduction in CH₄ flux is not linked to the levels of primary producer abundance per se, but to some other characteristics of submerged plants. Submerged plants are the single most dominant influence on the trophic structure and the function of shallow-lake ecosystems, influencing a range of biological and chemical processes (Jeppesen et al., 1998), providing habitat for periphyton, invertebrates and fish and changing sediment structure and chemistry (Barko et al., 1991). Furthermore, submerged plants link the water column to the sediments, changing sedimentary oxygen dynamics (Sand-Jensen et al., 1982) which can reduce methanogenesis (Jespersen et al., 1998). Methane consumption, by methane-oxidizing bacteria, can reduce emissions from lakes to a small proportion of production (Bastviken et al., 2008); however, the short distance between source (the sediment) and the atmosphere has been thought to limit the capacity for CH₄ consumption in shallow lakes (Cole et al., 2007). Recent research, however, suggests that submerged plants can provide excellent habitat for methane-oxidizing bacteria, resulting in the consumption of large quantities of CH₄ in plant beds (Yoshida et al., 2014). Furthermore, in streams, large quantities of CH₄ can be consumed over very short distances (Shelley et al., 2014b) with the potential to match CH₄ production (Shelley et al., 2014a). Thus, biotic interactions between submerged plants and methanogenic and/or methanotrophic bacteria resulting in increased consumption and/or decreased production are two plausible mechanisms explaining the lower diffusive flux of CH₄ where submerged plants were abundant (Fig 4c).

The absence of evidence for a direct effect of temperature on GHG fluxes, described above, contrasts markedly with the evidence of indirect effects from the analysis of temperature treatment effects on ER, macrophyte abundance, chlorophyll a and NEP (at low nutrient concentrations) (Table 4). There is no suggestion here that temperature increased ER to a greater extent than GPP, as predicted from metabolic theory (Yvon-Durocher et al., 2010), the slope of the models not being different (Table 4). In fact, at low nutrients, the positive effect of temperature treatment on NEP reflects that higher temperature promoted GPP over ER, which links to the greater abundance of macrophytes at higher temperature treatment. The data suggest that it is a combination of primary producer abundance and temperature that determines ecosystem-level metabolic rates, reflected by the fact that a much larger proportion of the variance in ER, GPP and NEP was explained by macrophyte or chlorophyll a abundance at low and high nutrients, respectively (Table 4). Furthermore, it suggests that the presence of plants rather than phytoplankton promotes GPP over ER. This has important implications for how the ecological condition of a lake might interact with climate change in shaping GHG fluxes from shallow lakes.

The results here demonstrate that eutrophication, ecological condition, that is the dominance of macrophytes vs. phytoplankton, and temperature change will combine to shape future GHG fluxes from shallow lakes. The data indicate that despite its lower warming potential, CO₂ fluxes dominate the total GHG flux from shallow lakes (Fig. 3); however, in the absence of data on the ebullitive flux of CH₄, the contribution of CH₄ to the total GHG flux is underestimated (see Appendix S1). The results further suggest that the three GHGs will respond differently to the indirect effects of warming dependent upon the following: a) nutrient level, b) identity of the dominant primary producer and c) the abundance of that dominant primary producer. CO₂ flux appears to be controlled more by primary producer abundance than by temperature. Whether this reduction in CO₂ flux will be offset by increases in CH₄ flux is indirectly linked to temperature, via ecosystem respiration rates and the identity of the dominant primary producer. Where macrophytes are abundant, there may be a reduction in the CH₄ diffusive flux, whereas at higher nutrient levels, where phytoplankton is the dominant primary producer, evidence suggests that the CH₄ flux will increase with rising temperatures. The controls of N₂O flux are more cryptic and appear more linked to the supply of nitrogen, and the potential climate-change effects remain unclear.

These results suggest that for many shallow lakes, in particular those dominated by macrophytes, there will be no positive feedbacks to atmospheric GHG concentrations as a result of climate change. The indirect nature of the effects of warming on GHG fluxes means that they can be confounded by changes in primary producer abundance or biomass. In particular, the significant positive effect of temperature treatment on macrophyte abundance (Table 4) has implications for the effects of climate change at higher latitudes where low-nutrient shallow lakes abound. The data show that warming may lengthen the growing season for submerged plants and increase abundance and biomass, increasing carbon influx through primary production and also reducing CH₄ flux via biotic interactions with methanogenic and methanotrophic bacteria. These findings fit well with recent work investigating long-term patterns of carbon flux in lakes in the thermokarst landscape, suggesting that increases in submerged plant abundance played a key role in the shift of these lakes from sources to sinks of carbon (Walter Anthony et al., 2014).

In regions more directly affected by human activity, particularly by agriculture, eutrophication is likely the crucial factor in determining the role of climate in the GHG flux. For these regions and in warmer climates, it is more difficult to predict how climate change will impact the relative abundance of macrophytes and phytoplankton in shallow lakes. The results here showed that rising temperatures increased macrophyte abundance for the whole data set and at low nutrients (Table 4). This is at odds with other work that has suggested that increased temperatures reduce the likelihood of a shallow lake being macrophyte-dominated (Kosten et al., 2009). The mechanisms behind this increased likelihood that plants are less abundant were obscure, and predicting the effect of climate change on submerged plants at higher nutrient concentrations is not straightforward (Kosten et al., 2011). Furthermore, the presence and abundance of submerged macrophytes in lakes is controlled not only by nutrient loading, but also by biotic interactions such as top-down cascades from fish to periphyton (Jones & Sayer, 2003) which further complicates prediction of climate-change effects on plant abundance. Future research should perhaps investigate the role of ecosystem instability, and a rapid reduction in the biomass of phytoplankton or macrophytes was associated large effluxes of GHGs, mostly CO₂ but also CH₄. Such events have been observed in natural lake systems and are associated with elevated nutrient levels (Sayer et al., 2010). However, how temperature increase might interact with eutrophication to impact ecosystem stability is currently very uncertain.