Temperature Sensitivity of Freshwater Denitrification and N2O Emission—A Meta‐Analysis

Freshwater denitrification removes a considerable amount of nitrogen from inland waters, which are under pressure from eutrophication and warming. However, incomplete denitrification can lead to the formation of N2O, a potent greenhouse gas, which can amplify climatic warming. Although temperature effects on denitrification are well studied in individual habitats and experiments, global patterns in temperature‐responses of denitrification and N2O emissions remain to be elucidated. Here, we investigated the temperature sensitivity (Q10) of denitrification and N2O emissions in freshwater ecosystems worldwide, using a meta‐analytic approach. To this end, Q10 values from in‐situ and temperature manipulation studies were related to environmental nutrient conditions, O2, pH, sediment organic matter (SOM), and geographic location. Temperature sensitivity of denitrification displayed a strong positive correlation with environmental nitrogen concentrations, pH and O2. Significant correlations with SOM and SOM:N ratios were observed as well, but the direction of the effect differed between in‐situ and temperature manipulation studies. Surprisingly, temperature sensitivity of N2O emissions did not correlate with pH, SOM, nutrient or O2 conditions. Temperature sensitivity of the ratio between N2O emission and NO3− concentration (adapted EF5 values) was 6.6 times higher in Australia and New Zealand compared to other geographic regions. As global temperatures and nitrogen deposition in freshwater ecosystems are expected to increase over the coming decades, our results suggest enhanced future denitrification, which may present a natural way to balance eutrophication. The observed temperature sensitivity of N2O emission factors, however, may indicate enhanced denitrification‐derived N2O emissions from freshwater ecosystems in a future warmer world.

Similar to other biological processes, denitrification is affected by temperature. Currently, the body of literature on the effects of global warming on denitrification in freshwater ecosystems is growing steadily (see for instance Palacin-Lizarbe et al. (2018) for a recent overview in lakes). While a straightforward increase in denitrification with warming is generally expected, examples of positive, negative and neutral responses can all be found in the literature (Brin et al., 2017;Tomaszek et al., 1997;Veraart et al., 2011). This strengthens the belief that the temperature sensitivity of denitrification is dependent on its surrounding environment. For example, deni trification takes place under hypoxic to anoxic conditions (Knowles, 1982), and temperature-regulated oxygen dynamics have been shown to amplify the effect of temperature on denitrification (de Klein et al., 2017;Veraart et al., 2011). Moreover, as canonical denitrification requires both nitrogen and organic matter (OM) as a substrate, the availability of both degradable OM and NO 3 − is expected to play a role in its response to elevated temperatures. Under low substrate availability, denitrification will be primarily constrained by substrate, minimizing temperature effects. At higher substrate availability, denitrification will be released from substrate limitation, possibly enabling a stronger response to increased temperatures.
Although the temperature effect on denitrification is well studied in individual habitats and experiments (see for instance (Klaus et al., 2018;Messer & Brezonik, 1984;Veraart et al., 2011)), less is known about the overarching patterns in temperature responses of denitrification and resulting N 2 O emissions. Global patterns in denitrification rates across freshwaters can be mirrored by variations in nitrogen loading to the respective water bodies (Seitzinger et al., 2006) and related socio-economic factors such as human population density and use of nitrogen-rich fertilizers (Seitzinger et al., 2000). Modelled aquatic N 2 O emissions are significantly higher in the northern hemisphere (Seitzinger et al., 2000). Projected temperature increases in temperate and polar regions by the end of the century are profoundly larger than those at lower latitudes (Pachauri et al., 2014). We therefore hypothesize that temperature sensitivity of denitrification in freshwater ecosystems displays spatial heterogeneity, following variation in nitrogen loading and temperatures.
To connect rate predictions to potential N 2 O emission under a range of global change scenarios, it is important to take into account the amount of N 2 O emitted relative to nitrogen load. This is expressed by the N 2 O emission factor EF5, which is used by the Intergovernmental Panel on Climate Change (IPCC) to calculate indirect N 2 O emissions from a suite of ecosystems (Well et al., 2005). In freshwater ecosystems, accurate estimates of nitrogen loading are not always available. Therefore, the N 2 O-N/NO 3 -N mass ratio can be used as an adaptation to calculate EF5 values (Hama-Aziz et al., 2017). This ratio reflects the efficiency of N 2 O production compared to the nitrogen substrate. Previous research indicates that EF5 values can differ substantially between sites (Clough et al., 2007) and can vary with environmental conditions such as temperature . However, to our knowledge there has been no global overview of this temperature sensitivity in freshwater ecosystems to date.
In this study, we summarize current knowledge on temperature sensitivity of global freshwater denitrification rates, N 2 O emissions, and adapted EF5 values, using a meta-analytic approach. First, we constructed a database composed of published data gathered from a systematic digital literature search. Using this database, we tested the hypothesis that temperature sensitivity (expressed as Q10) of denitrification rates and N 2 O emissions is dependent on substrate availability. Study-specific Q10 values were hypothesized to be positively correlated with nitrogen concentrations in the sediment and the water column and sediment organic matter. An additional positive correlation between temperature sensitivity and phosphorus (P) concentrations in the sediment and in the water column was anticipated, as P is indicative of eutrophic conditions that stimulate bacterial denitrification (Seitzinger & Nixon, 1985). We furthermore hypothesized a negative correlation between Q10 values and oxygen (O 2 ) concentrations.

Methods
To determine the global patterns in temperature-responses of denitrification and N 2 O emissions, we compiled a database from the published literature on denitrification and N 2 O emission rates in freshwater ecosystems under varying temperatures. We were specifically interested in two types of datasets: (a) in-situ measurements of temperature in freshwater ecosystems (e.g., lakes, reservoirs, wetlands, rivers and ditches) and (b) experimental temperature manipulations. This implied that the publication either included in-situ measurements of temperatures and denitrification rates or N 2 O emissions (henceforth referred to as in-situ temperature dataset), or the publication included freshwater experiments in which temperature was manipulated and denitrification and/or N 2 O was measured (temperature manipulation dataset). For this purpose, we carried out a systematic literature review using Web of Science (https://www.webofknowledge.com/) on the 6th of March 2020 using the following query on paper titles, abstracts and keywords: ((freshwater* OR lake* OR aquatic OR pond* OR reservoir* OR lagoon* OR river* OR stream* OR creek* OR brook* OR ditch* OR surface water* OR wetland* OR *marsh*) AND (temperature* OR warming OR heating OR Q10 OR "Q(10)") AND (denitrification OR N 2 O OR "nitrous oxide" OR "acetylene inhibition" OR denitrifying)). Initially, papers were screened for data on temperaturemeasured in the sediment or water -, denitrification rates and/or N 2 O emissions in the graphs and tables. Studies were included if the in-situ sites were permanently water-bearing (or at least for the duration of the study period) and contained at least two combined measurements of denitrification rates or N 2 O emissions and temperature. Whenever papers reported from several conducted experiments or measurements from distinct locations, these were considered as unique case-studies. To focus on freshwater studies, studies from salt marshes, estuaries or other tidal systems were not included. Studies on constructed wetlands and wastewater treatment systems were excluded as well. As our literature search rendered very few studies on the southern hemisphere, we supplemented the database with data on Uruguayan lakes from Veraart et al. (2012). Furthermore, the dataset was supplemented with a Finnish lake dataset (Rissanen et al., 2013). Data were extracted from publications using the Java program Plot Digitizer (Huwaldt & Stienhorst, 2013). When reported, data on sample size, nitrogen and phosphorus concentrations in the water column and/or sediment, sediment organic matter content (SOM), dissolved O 2 concentrations and pH were extracted as well. As denitrifiers may respond differently to short-term nutrient enrichment than long-term background nutrient concentrations, we recorded both responses separately. Total nutrient concentrations in the water column were calculated as the sum of nutrient enrichment and background concentrations. Not all studies reported on the same forms of nitrogen and phosphorus. Therefore, we improved comparability between studies by using the reported total nitrogen (TN) and total phosphorus (TP) concentrations as a proxy of lake N and P concentrations. In case-studies where TN and TP were not reported, TN concentrations were approximated by the sum of the individual N-species (NO 3 − , NO 2 − and NH 4 + ) reported, and TP concentrations were equated to Soluble Reactive Phosphorus or PO 4 3− concentrations. If needed, we converted units to ensure comparability between studies. Additionally, information on the measurement method, system type (lake, river, pond, stream etc.) and its geographic location (latitude, longitude, geographic region, climate zone and country) was included. This resulted in a database with a total of 140 papers that contained 215 unique case-studies for the in-situ temperature dataset and 144 for the temperature manipulation dataset ( Figure  S1 in Supporting Information S1).

Calculations and Statistics
All calculations and statistics were carried out in R (R Core team 2013). Q10 values for denitrification rates and N 2 O emissions were calculated for each dataset using the function Q10 (package respirometry (Birk, 2017)). A cut-off of Q10 > 50 was used to remove extremely high Q10 values (n = 21; 5.9% of the analyzed dataset) from the in-situ dataset because these values might represent incorrect model fits. An overview of the studies incorporated in the two datasets can be found in Table S1 of Supporting Information S1. The overall system temperature coefficient θ s was derived from the Q10 values. A doubling of reaction rates over a 10° temperature range corresponds with a θ s of around 1.07 . Differences in Q10 values between the two datasets were tested with a one-way ANOVA (function aov). To improve ANOVA-assumptions of normality and homogeneity of variance, data were log-transformed in the analysis. As Q10 values significantly differed between these datasets, other categorical variables (geographic region, climate zone, measurement method) were tested using two-way ANOVA with log-transformed Q10 values (function aov). Correlations between denitrification and N 2 O emissions were determined by linear regression (function lm). Relationships between the Q10 values and nutrient conditions, pH, SOM, O 2 and latitude were investigated using generalized linear models (function glm) with a gamma distribution. When necessary, a log or inverse link was used in the glm models to improve normality of the residuals.

N 2 O Emission Factors (EF5)
Indirect N 2 O emission factors (EF5) were calculated according to Hama-Aziz et al. (2017). Ideally, N-N 2 O concentrations are related to N-NO 3 concentrations for this calculation. However, we adapted the calculation due to limited data availability. To be specific, N-N 2 O emissions (in μg m −2 h −1 ) were divided by the dissolved N-NO 3 concentration (in μg L −1 ). Our adapted EF5 values reflect the ratio of N 2 O emissions to NO 3 concentration and provide a relative measure for its conversion efficiency. Temperature sensitivity of these adapted EF5 values were calculated using the function Q10 (package respirometry (Birk, 2017)). A cut-off of Q10 > 50 was used to remove extremely high Q10 values (n = 24; 16% of the dataset). Relationships between the Q10 values and latitude were investigated using generalized linear models (function glm). The categorical variables "geographic region" and "climate zone" were tested using one-way ANOVA (function aov). Data was either logor square root-transformed to improved normality of the residuals. In the in-situ dataset, the geographic regions South America, Central America, Southern Europe and Southern Asia were excluded from analysis due to low sample size (n < 3).

Results
Temperature sensitivity (Q10) values for denitrification tended to be higher in the temperature manipulation dataset compared to the in-situ dataset (3.6 ± 5.5 and 2.3 ± 4.7, respectively, mean ± SD), though this effect was not significant (F (1,206) = 3.15, P = 0.08). These Q10 values correspond to system temperature coefficient (θ s ) values of 1.14 and 1.09 in the temperature manipulation and in-situ dataset. The Q10 values for N 2 O emissions were significantly higher in the temperature manipulation dataset (4.5 ± 8.6; θ s = 1.16; n = 63) than those originating from the in-situ temperature dataset (2.2 ± 4.4, θ s = 1.08; n = 160; F (1,221) = 7.25, P < 0.01; Figure 2). Therefore, the in-situ temperature and temperature manipulation datasets are considered separately in further analyses. Q10 values calculated on denitrification rates were significantly higher than those calculated on N 2 O emissions (one-way ANOVA; F (1,429) = 11.35, P < 0.001).
As expected, denitrification rates correlated strongly with N 2 O emissions ( Figure 3, R 2 = 0.81, P < 0.001). There was no such correlation between the Q10 values of denitrification and N 2 O emissions observed (R 2 = −0.01, P = 0.39). Most studies in the datasets originated from Europe (n = 171), Asia (n = 112) and North America (n = 65) and there was a general lack of data from both the southern hemisphere (n = 42) and tropical regions in particular (n = 1) (Figure 1). Q10 values for temperature manipulation datasets were not significantly different between climate zones or measurement methods (Table S2 in Supporting Information S1). In-situ Q10 denitrification values, however, were significantly different between climate zones and measurement methods (Table S2 in Supporting Information S1). Tukey post-hoc tests indicated that these Q10 values were significantly lower in temperate compared to continental climatic regions (Table S3 in Supporting Information S1). Furthermore, in-situ Q10 denitrification values measured with the acetylene inhibition method were significantly higher than those  (Table S3 in Supporting Information S1, P = 0.03), while no significant differences between the other methodologies were observed.

Relationships Between Q10 denitrification , Q10 N2O and Environmental Variables
In the in-situ temperature dataset, Q10 denitrification values increased exponentially with N concentrations in the water column and sediment and with dissolved O 2 concentration, while Q10 denitrification decreased with SOM and P concentration in the water column (Table 1, Figure 4). Similarly, we observed a negative correlation between Q10 denitrification values with the SOM:N ratio in the sediment ( Figure 4, Table 1). The correlation between N concentrations and Q10 denitrification was not apparent for the individual N-species ( Figure S2, Table S4 in Supporting Information S1). These correlations were not observed in the temperature manipulation dataset. In this dataset, Q10 denitrification values correlated positively with SOM, pH values and sediment SOM:N ratio ( Figure 4). No correlations between Q10 denitrification values, N enrichment and sediment P concentrations were observed ( Figure  S3 in Supporting Information S1).   In contrast to Q10 denitrification , Q10 N2O values did not correlate with any of the selected environmental variables in either the temperature manipulation or in-situ temperature datasets (Table 1, Figure S4 in Supporting Information S1). In the temperature manipulation dataset, Q10 N2O values tended to increase with P concentration in the water column, but this correlation was non-significant (P = 0.06).

Temperature Sensitivity of N 2 O Emission Factors
The temperature sensitivity of adapted emission factors (EF5, see methods) was calculated for the studies that reported both N 2 O emissions and dissolved NO 3 − concentration. In the in-situ dataset, Q10 EF5 values were 1.82 ± 2.87 (mean ± SD, n = 87) and significantly differed between geographic regions (Table S5 in Supporting Information S1, Figure 5). To be specific, Q10 EF5 values were 6.6 times higher in the region Australia and New Zealand compared to regions on the northern hemisphere (Tukey-HSD test, P < 0.05, Table S6 in Supporting Information S1). Furthermore, no significant differences between climate zones were observed for the Q10 EF5 values calculated on the in-situ dataset. No correlations between Q10 EF5 values and distance from equator were observed in either in-situ and temperature manipulation dataset (Table 1, Figure S5 in Supporting Information S1).
In the temperature manipulation dataset, average Q10 EF5 values were 6.90 ± 10.98 (mean ± SD, n = 35). No significant differences between climate zones nor geographic regions were observed (Table S5 in Supporting Information S1).

Discussion
This meta-analysis compiles and synthesizes current knowledge on temperature sensitivity of freshwater denitrification and N 2 O emissions. A total number of 359 Q10 values from 144 temperature manipulation case-studies and 215 in-situ case-studies of temperature and denitrification/N 2 O could be incorporated into the analysis. Our results showed a positive correlation between temperature sensitivity of denitrification and environmental nitrogen concentrations, pH and O 2 (see Figure 6 for a conceptual overview of our meta-analytic results). Significant correlations with SOM and SOM:N ratios were observed as well, but the direction of the effect differed between in-situ and temperature manipulation datasets. Surprisingly, temperature sensitivity of N 2 O emissions did not correlate with any environmental variables included in the analysis. Possibly, this could be due to the multitude of physical processes and microbial guilds producing N 2 O, which can all directly or indirectly be affected by various environmental conditions. Additionally, various inhibiting factors for N 2 O consumption, such as oxygen (Ji et al., 2018) and pH (Blum et al., 2018), could also be a factor in this discrepancy. Below, we discuss the observed relationships between the temperature sensitivity of freshwater denitrification, N 2 O emissions, N 2 O emission factors and environmental conditions. Note. SOM, sediment organic matter. Significant factors and values are indicated in boldface, with ***P < 0.001, **P < 0.01, *P < 0.05, ˙ P < 0.10.

Substrate Availability Drives Temperature Sensitivity of Denitrification
Worldwide, lakes are estimated to cover about 5*10 6 km 2 (Verpoorter et al., 2014) and rivers and streams 0.8*10 6 km 2 (Allen & Pavelsky, 2018). Understanding the impact of global warming on the ecological processes in freshwater ecosystems is important to predict the behaviour of these ecosystems in the future. The results of our study demonstrate that temperature sensitivity of freshwater denitrification relates positively to the surrounding nitrogen availability in both sediment and water. Specifically, we observed a positive correlation between Q10 denitrification and nitrogen concentrations in both the water column and the sediment in the in-situ temperature dataset. This synergistic interaction between nitrogen availability and temperature responses is commonly observed in a myriad of ecosystems, ranging from lakes (Palacin-Lizarbe et al., 2018) and estuaries (Nowicki, 1994) to alpine meadows (Chen et al., 2017).  Table S5 of Supporting Information S1. has shown that denitrification activation energy can inversely correlate with nitrate concentrations, further supporting our meta-analytic results. The commonality between our results and other studies in freshwater and terrestrial ecosystems highlights the synergistic link between nitrogen as a driver of instantaneous denitrification rates and temperature as a regulator of the enzymatic process (Wallenstein et al., 2006).
In the in-situ denitrification dataset, this positive correlation with N conditions co-occurred with a negative correlation with sediment organic matter (SOM). SOM encompasses all organic material in the sediment and consists of a combination of recalcitrant and labile organic carbon, of variable nitrogen content (Schnitzer & Khan, 1975). We anticipated a positive correlation between SOM and temperature sensitivity as organic carbon is an electron donor in canonical denitrification and can thus limit denitrification rates. This positive correlation was observed in the temperature manipulation experiments, but we observed a negative correlation between SOM and Q10 denitrification in the in-situ dataset. This discrepancy highlights potential differences between conclusions drawn on laboratory experiments and in-situ measurements and show the difficulty in interpreting and upscaling experimental biogeochemical data to field scenarios. Possibly, competition between denitrification and other microbial processes might be more prominent under in-situ conditions. For instance, dissimilatory nitrate reduction to ammonium (DNRA) competes for organic carbon and nitrogen sources (Burgin & Hamilton, 2007). Following denitrification and DNRA stoichiometry, the ratio between SOM and N availability in the sediment can be a strong factor in determining the outcome of this competition, favouring DNRA at higher SOM:N ratios (Burgin & Hamilton, 2007;van de Leemput et al., 2011). In our work, we indeed observed that in-situ temperature sensitivity of denitrification was negatively correlated with sediment SOM:N ratios. Moreover, a study by  conducted in urban wetlands showed DNRA benefited more from the addition of organic carbon than denitrification did. In a follow-up study, these authors also showed that DNRA is more sensitive to temperature changes than denitrification . Therefore, while DNRA and denitrifiers can coexist under a variety of C:N ratios (van den Berg et al., 2016), the temperature response of denitrification may be limited by competition with DNRA at high SOM:N ratios.
In contrast to our hypotheses, we observed a positive correlation between temperature sensitivity of denitrification and O 2 concentration in this meta-analytic dataset. We expected that this temperature sensitivity would decrease with higher oxygen availability, as denitrification rates increase with temperature and hypoxia (de Klein et al., 2017;Veraart et al., 2011) and riverine work of Li and colleagues showed a reverse correlation of N 2 O concentration with oxygen (Li et al., 2021). However, effects of O 2 on N 2 O emission are inherently more difficult to predict, due to the contribution of both aerobic nitrification and anaerobic denitrification. For example, work of Liikanen et al. (2002), included in our dataset, showed higher N 2 O production in oxic than anoxic conditions, hinting at a larger role for nitrification than denitrification in N 2 O production in their eutrophic systems. At the same time, the enzyme responsible for the reduction of N 2 O to N 2 (NosZ) is known to be sensitive to O 2 inhibition (Dalsgaard et al., 2014), which could indicate reduced transformation rates of N 2 O to N 2 under high O 2 concentrations. The contrasting findings in studies included in our meta-analysis may be explained by site-specific characteristics affecting the balance between underlying processes driving both denitrification and N 2 O emission. In addition, reported correlations between O 2 and denitrification in-situ also depend on where and how O 2 is monitored, and denitrification is measured. Reporting changes in O 2 profiles when measuring temperature effects on denitrification will provide better insight in their interactive dynamics.

Global Heterogeneity in Temperature Sensitivity of N 2 O Emission Factors
Ecosystems near the poles, especially in the Arctic, are warming at least twice as fast as ecosystems at lower latitudes (polar amplification; (Pachauri et al., 2014;Smith et al., 2019)). Globally, nitrogen inputs in aquatic ecosystems are increasing simultaneously (Ackerman et al., 2019;Reay et al., 2008), and indications of nitrogen effects on the biogeochemistry of lakes have been documented as far north as 70° (Wolfe et al., 2006). In our meta-analysis, no latitudinal patterns in any of the temperature sensitivity values were observed. Nonetheless, the temperature sensitivity of N 2 O emission factors (EF5) exhibited spatial heterogeneity. Specifically, in-situ Q10 EF5 values were significantly higher in the region of Australia and New Zealand compared to other geographic regions. Though sample sizes were small, this may indicate that N 2 O emission factors are sensitive to changes in surrounding temperatures and that this sensitivity differs spatially. IPCC methodology of EF5 values is aimed to determine an average N 2 O emission factor per ecosystem. In our analysis, we showed that EF5 values can increase with a factor 1.8-6.9 with a 10° temperature increase. As the IPCC has reduced their EF5 estimates for rivers and lakes between their 1997 and 2006 reports (IPCC, 1997(IPCC, , 2006Tian et al., 2019), we would advise to take the temperature-dependence into account in future decision making as the world is continuing to warm.
We should note that most of the studies included in our meta-analysis were conducted in Europe, Northern America and Asia. A knowledge gap therefore exists for tropical ecosystems and ecosystems south of the equator in general, as only one study from South America (Veraart et al., 2012), one from Africa (Roland et al., 2017) and few from Australia and New Zealand were identified (see for instance Crawshaw et al. (2019), an overview is presented in Table S1 of Supporting Information S1). While additional datasets may have been published (see for instance Borges et al. (2015) for an overview on N 2 O emissions from African rivers), these were not picked up by our systematic literature search. Studies on temperature sensitivity in some ecosystems may have been published in other languages than English. Because our literature search focused on English literature, these potential studies are currently not considered in our analysis. Taking non-English studies into account can alter meta-analytic outcomes significantly (see e.g., Konno et al. (2020)). This knowledge gap should be considered when interpreting our meta-analytic results.

Technical Recommendations-Upscaling, FAIR Data and New Scientific Discoveries
It is extremely difficult, or even impossible, to mimic the complexity of a real ecosystem in a laboratory setting. Experimental set-ups allow us to reduce environmental complexity while simultaneously increasing control. In laboratory settings, ecological interactions are greatly reduced relative to natural ecosystems, which can alter observed responses (Englund & Cooper, 2003;Goldenberg et al., 2018). This may explain the difference in temperature sensitivity between the datasets, as Q10 values were higher in temperature-manipulation experiments than in in-situ datasets. While experimental studies are extremely valuable and necessary to gain mechanistic understanding of the influence of environmental pressures, considerable care should be taken when extrapolating these findings to natural conditions. Another platform for the potential upscaling of denitrification studies is modeling studies. Nitrogen processing models such as Global NEWS (Mayorga et al., 2010) and nutrient loading models such as PCLake+ (Janssen et al., 2019) and VEMALA (Huttunen et al., 2016), often include denitrification as a nitrogen conversion term. Moreover, other ecosystem models such as the MyLake-Sediment model (Markelov et al., 2019) and the terrestrial NEMIS model (Hénault & Germon, 2000), even use (derivatives of) Q10-values to describe the temperature sensitivity of denitrification. Our meta-analytic results provide an overview of in-situ Q10-values for a wide range of ecosystems (Table S1 in Supporting Information S1), which can potentially aid the modeling of the temperature sensitivity of denitrification for specific freshwater ecosystems.
It is quite common in scientific publications to include only the measured variables that are valuable for the scientific question of the paper. In meta-analyses, however, auxiliary data, combined with meta-data, which are not crucial for the specific paper of interest become valuable to address new research questions. The publication of auxiliary data, either in Supporting Information S1 or in an online repository, has become more and more common in ecological research (Culina et al., 2020). In our opinion, denitrification research, which has been an active field of investigation for over a century, could rapidly move forward by FAIR (Findable, Accessible, Interoperable Reproducible) data practises (Alves et al., 2018;Wilkinson et al., 2016).
The field of environmental denitrification research is continuously advancing. For instance, scientific discoveries in the nitrogen cycle such as anammox (Strous et al., 1999), comammox (Daims et al., 2015;van Kessel et al., 2015) and non-denitrifying N 2 O reduction (Jones et al., 2013;Sanford et al., 2012) can shed new light on previous scientific publications. A limitation of meta-analytic studies is the description of processes in the original paper, which may be partly outdated after these new scientific discoveries. In this study, we used a broad definition of denitrification (being the reduction of nitrate and nitrite to N 2 O and/or N 2 ) and a range of methodologies. Overall, no strong differences in Q10-values were observed between methods (Table S2 in Supporting Information S1). However, in the in-situ denitrification dataset, acetylene inhibition studies reported higher temperature sensitivity than studies that used a 15 N-tracer method. The 15 N-tracer method cannot always distinguish between denitrification and coupled DNRA-anammox pathways and the balance of these pathways is a current hot topic in research. The contribution of anammox to N 2 production in lake sediments can be substantial (Crowe et al., 2017) and depends on environmental conditions such as oxygen and light (Valiente et al., 2022). Recent work by Tan et al. (2020) suggests that sediment denitrifiers are more thermotolerant than anammox bacteria, which may explain our observed lower temperature sensitivity in 15 N-tracer studies. Overall, we cannot guarantee that the 15 N-tracer studies included in our meta-analysis did not include the DNRA-anammox pathway and think this is a valuable and interesting research avenue for further exploration.

Conclusion
Using a meta-analytic approach, we show that temperature sensitivity of denitrification is positively correlated with nitrogen concentrations and that this increase in temperature sensitivity is non-linear. In general, one degree temperature increase is expected to amplify denitrification and N 2 O emissions between 8 and 14%. Both global temperatures and nitrogen deposition in freshwater ecosystems are expected to increase over the coming years (Brunner et al., 2020;Pachauri et al., 2014;Vuuren et al., 2011). Our analysis therefore predicts overall increased freshwater microbial denitrification rates in the near future, which may present a way to naturally counteract the harmful effects of eutrophication. However, temperature sensitivity of N 2 O emission factors was observed as well, potentially indicating that warming and eutrophication may enhance future denitrification-derived N 2 O emissions from freshwater ecosystems.

Data Availability Statement
The data and R scripts that support the findings of this study are publicly available after the 6th of August, 2022 at the DANS EASY archive (https://easy.dans.knaw.nl) via https://doi.org/10.17026/dans-xy8-mycd (dataset 246823) under the creative commons zero waver (CC0) license.