2.1 Study area and sampling design

The study was conducted in 50 stream sites in North Portugal. Surveyed watercourses belonged to the Ave, Cávado, Lima and Minho catchments. Climate is Atlantic with warm, dry summers (mean temperature: 11.5°C) with a mean annual precipitation around 1000 mm (Portal do Clima, http://portaldoclima.pt/en/; Trigo & DaCamara, 2000). Lithology is siliceous, dominated by granite rocks, but also including slates and schists.

Our sampling design, carried out during June–October 2020, aimed to represent realistic gradients of multiple abiotic factors that can potentially act as stressors for stream ecosystems (Brauns et al., 2022; Gutiérrez-Cánovas et al., 2022; Lourenço et al., 2023). We define a stressor as any abiotic variable that can be altered by human intervention, resulting in detectable changes in riverine biota or functioning (Birk et al., 2020; Sabater et al., 2019). To do this, our site selection covered crossed gradients of catchment size, catchment land-use cover and reach-scale riparian vegetation cover to capture wide ranges of nutrient enrichment, dissolved oxygen depletion, thermal stress, riparian canopy openness and discharge (Table S1). These crossed gradients represent a spectrum of stress intensities and combinations that simulate realistic global change scenarios for land-use intensification and climate change in the study area (Table S1), comparable to those used in manipulative experimental studies (e.g., Cross et al., 2022; Matthaei et al., 2010; Romero et al., 2018).

The study site selection encompassed a range of land-uses, from natural conifer and broadleaf forests to a variety of mixed land-use types, such as extensive agriculture (cereals, corn), pastures and diffuse urban areas. This produced a wide variation in nutrient concentrations (dissolved inorganic nitrogen ranged 0.03–4.69 mg L⁻¹) and dissolved oxygen concentration deficit (0.3–7.1 mg L⁻¹). The variability in local riparian vegetation structure and density at the reach level allowed us to capture a range of riparian canopy openness (33%–100%) (Pace et al., 2022). Additionally, variations in elevation (6–930 m a.s.l.) and local climate (July 2020's mean air temperature ranged 8.7–29.0°C) permitted us to capture a wide range of daily mean water temperatures (12.2–25.7°C). To explore whether the effect of stressors on stream metabolism and GHG production depends on stream hydrology and size, we covered wide ranges of flow velocity (0.007–0.323 m s⁻¹), stream width (1–35 m) and discharge (0.001–1.731 L s⁻¹).

2.2 Characterization of stressor and environmental gradients

At each sampling site, we characterized indicators of nutrient enrichment, oxygen depletion, thermal stress, riparian degradation and stream hydrology (Table S1). Dissolved inorganic nitrogen (DIN) was used as a proxy of nutrient enrichment, as it represents different nitrogen compounds (nitrite, nitrate and ammonium) and showed a positive correlation with dissolved phosphate (r_P = .56, p < .001). The concentration of these solutes was determined from water using a HACH DR/2000 spectrophotometer (Hach Company, Loveland, CO, US). Water samples were collected in plastic bottles, transported cooled and frozen until analysis. We calculated mean daily dissolved oxygen deficit in the water column (DO deficit) as an indicator of oxygen depletion and hypoxic conditions that can favour anaerobic respiration (Blaszczak et al., 2023; Gómez-Gener et al., 2020). For each river reach, DO deficit was calculated as the difference between the DO saturation concentration and mean daily DO concentration. DO was recorded every 10 min during 24 h using miniDOT loggers (PME, USA). One oxygen datalogger was placed and fixed to the riverbed of each sampling site at 10–30 cm depth using iron bars. We estimated mean water temperature as an integrative indicator of thermal stress. We calculated mean daily temperature for each site from continuous temperature recording of 32–46 days obtained using HOBO UA-002-64 dataloggers (Onset, USA). To represent the effects of riparian degradation, we estimated riparian canopy openness, which reflects increasing human and climatic impacts on riparian vegetation structure and an increase in light inputs (Feld et al., 2018; Munné et al., 2003; Pace et al., 2021). Based on the normalized difference vegetation index (NDVI) of the riparian vegetation (from pure riparian vegetation pixels), we calculated canopy openness for each 100-m reach as 1 − NDVI values (Pace et al., 2022). As such, the resultant values indicate greater canopy openness when they are close to 1. NDVI values were obtained from Sentinel-2 orthorectified surface reflectance images (S2A MSI L2A). To cover the whole study area, we selected two clear scenes (T29TNG and T29TNF tiles) from June 2020 (Copernicus Open Access Hub user interface, developed by the European Space Agency for Earth observation). To characterize hydrology and control for stream size, we used discharge—a crucial hydrological indicator that responds to variations in catchment size, climate, hydromorphology and land-use alterations (Bussi et al., 2018; Vicente-Serrano et al., 2019). Discharge was calculated as the product of mean width, mean depth and mean current velocity at the reach level. These variables were calculated across five transects over a 100-m reach. For each transect, we recorded at least five measures of depth and current velocity using a flow meter with a graduated stick (Electromagnetic Flow Meter-Model 801, Valeport, UK). We also characterized segment slope for each location from a digital elevation model. More details on stressor and environmental characterization are available in Lourenço et al. (2023).

2.3 Stream metabolism

We used two approaches to characterize stream metabolism, including estimates of daily ecosystem GPP and respiration (ER) and estimates of algal production covering a longer time frame (~5 weeks).

Net ecosystem production was calculated as the difference between GPP and ER. We transformed oxygen-based rates to carbon metabolic rates (mmol C m⁻² day⁻¹) using a CO₂:O₂ ratio of 138:106 (Torgersen & Branco, 2007). To evaluate the sensitivity of the metabolic estimates to the quantification of the gas exchange coefficient, we explored the relationship between k₆₀₀ and ER (Figure S1) and found no significant relationship (r_P = .236, p = .098). To assess the performance of metabolic estimations, we inspected plots comparing observed and modelled diel DO values (Figures S3 and S4) and calculated mean absolute error (MAE) of the modelled DO values (Table S2; Figure S5). As a result, we discarded the metabolism rates of four sample sites due to non-plausible values (GPP < 0) or because they show very high MAE (i.e., reflecting inaccurate estimation of metabolic rates). We retained 92% of our initial data set (n = 46). See Table S2 for more details.

To calculate algal production (mg Chl a m⁻² day⁻¹), we measured the accumulation of chlorophyll-a on grazing-protected colonization tiles (10 × 10 cm unglazed tiles) over a period of ~35 days (range: 32–46 days). We used grazing-protected tiles to avoid noise caused by grazing. At each sampling site, we deployed three concrete bricks including one attached tile, which was surrounded by petroleum jelly to prevent grazing by crawling grazers. During the recovery day, we brushed tiles and preserved the algal material at −20°C under dark conditions. We then determined chlorophyll-a concentrations by spectrophotometry, following extraction in 90% acetone (Jeffrey & Humphrey, 1975). Algal production rates were calculated as the chlorophyll-a concentration per unit of surface divided by the number of colonization days.

2.4 GHG concentrations and emissions

We measured the partial pressure of CO₂ and CH₄ in water with the headspace method. Using 60-mL polypropylene syringes, we collected three samples of 30 mL of stream water 10 cm below surface, plus another 30 mL of ambient air. This procedure ensured an equal ratio of ambient air and water. Next, each syringe was vigorously shaken for 1 min to ensure an adequate gas equilibration, followed by 10 min of immersion in stream water to maintain constant equilibrium temperature. The water temperature was recorded using a multiparameter probe (Multi 3630 IDS Set KS2, WTW, Germany). From each equilibrated sample, we injected 20 mL of air into a 12-mL glass vials equipped with crimped rubber stoppers (Exetainers, Labco, UK) until analysis. The concentrations of CO₂ and CH₄ in water and air samples were analysed on an Agilent 7820A gas chromatograph equipped with a flame ionization detector. Aqueous CO₂ and CH₄ concentrations at field conditions were determined from measured headspace gas volume fractions and concentrations, based on the barometric pressure at the sampling site, field water temperature, laboratory equilibration temperature and the appropriate Henry's law constants (Sander, 2015). We collected GHG samples and measured DO data for metabolism in coincident days. We also calculated the molar ratio of CH₄ and CO₂ (CH₄:CO₂; Gómez-Gener et al., 2020; Stanley et al., 2016) and the proportion of CO₂-equivalent pCH₄ relative to the total sum of CO₂-equivalent pCH₄ and pCO₂. In this latter estimation, we assumed that CH₄ had 28 times more warming potential than CO₂ over a 100-years horizon (IPCC, 2014). We used CO₂-equivalent units to better represent the contribution of CH₄ to the overall global warming potential of dissolved gases. However, it is important to note that a fraction of these gas concentrations will not evade stream water because of potential CH₄ oxidation (Stanley et al., 2016) and photosynthetic re-assimilation of CO₂ (Solano et al., 2023).

2.5 Data analysis

Initially, we used Pearson's correlations to understand associations between primary production (GPP and algal production), ER and heterotrophy (NEP), and between concentrations and fluxes of GHG forms. We also determined correlations between DO deficit, heterotrophy and GHGs. To explore multiple stressor effects on stream metabolism, GHG concentrations and the molar ratio CH₄:CO₂, we used linear regression models and a multi-model inference approach (Burnham & Anderson, 2002). All models included DIN, water temperature, riparian canopy openness and discharge as predictors. Models predicting responses of pCH₄ and the molar ratio CH₄:CO₂ also included DO deficit as a predictor because high DO deficit values are associated with hypoxic conditions, anaerobic metabolism and methanogenesis (Blaszczak et al., 2023; Gómez-Gener et al., 2020; Stanley et al., 2016). We conducted multi-model inference through a two-step process. First, using Akaike information criterion for small sample sizes (AICc), we evaluated whether global models for each response variable should include a pairwise interactive term besides the additive stressor terms. We retained a model with an interaction if it showed the minimum AICc value and it differed in more than two AICc units respect to the additive model. We thus fitted four models, one containing only additive terms and the other three also including a pairwise stressor interaction each. A recent synthesis revealed that nutrient enrichment is an overarching stressor for riverine ecosystems, which can interact with light, thermal and hydrological stressors across spatial scales (Birk et al., 2020). Thus, we tested three related interactive terms where nutrient enrichment effects can be modulated by riparian degradation (DIN × canopy openness), thermal stress (DIN × water temperature) or hydrology (DIN × discharge). Second, based on the selected global model for each response variable, we quantified stressor coefficients, statistical support and importance, using the function dredge() from the MuMIn R package (Bartoń, 2016). This function produces the models for all potential combinations of predictors included in the global model. Based on their AICc values, we ranked the alternative models for each response variable and retained those with an AICc difference ≤7 with respect to the highest-ranking model. We also derived the explained variance (R²) and Akaike weights to determine the explanatory power and the relative likelihood of each model (statistical support), respectively. For each response variable, based on model's Akaike weights, we obtained the mean-weighted partitioned variance for each predictor (Hoffman & Schadt, 2016). To visualize the overall response of these models and using model's Akaike weights, we calculated a weighted-average of their standardized regression coefficients and predictions across the retained models (ΔAICc ≤7).

Using the fitted values from the selected models for CO₂ and CH₄, we explored how CO₂-equivalent concentration could change within the observed ranges of nutrient enrichment (DIN: 0.25–5 mg L⁻¹) and oxygen depletion (DO deficit: 0.25–7. mg L⁻¹). To do so, we considered that CH₄ had 28 times the warming potential of CO₂ over a 100-year horizon (IPCC, 2014). We use CO₂-equivalent units to better represent the warming potential of CH₄ relative to CO₂ for different stressor levels, given that these GHGs could require different mitigation strategies. We focused on nutrient enrichment and oxygen depletion because they were the two most important predictors of CO₂ and CH₄ in our linear regression models, and because of their importance in management policies and restoration actions in a multiple stressor context (Birk et al., 2020; Feld et al., 2018; Spears et al., 2021).

The overall model fit was assessed using Fisher's C-test, in which small and non-significant values (p-value > .05) indicate a good fit of the model (Shipley, 2013). Following these criteria, we removed stressor interactions and stressors links with non-significant p-values and low predictive capacity to increase model performance (i.e., reducing C-test value and increasing p-value). We focussed on the observations for which complete data were available (n = 46). For each stressor, we quantified the contribution of the effects of local and distal processes on GHGs by summing their standardized effect sizes across the SEM structure without consideration of the direction of the relationship (absolute standardized effect size). For indirect stressor–GHG relationships, we multiplied stressor effect sizes by the effect size of the variable mediating their effects on each GHG (i.e., NEP and/or DO deficit).

For all models, some variables were log-, logit- or squared-root- transformed to reduce distribution skewness and all variables were z-standardized (mean = 0, SD = 1) to allow the comparison of model coefficients. In all models, predictor combinations showed variance inflation factors <3, suggesting an acceptable degree of collinearity (Zuur et al., 2009). Model residuals were visually assessed to verify linear model assumptions (Zuur et al., 2009). All statistical analyses were performed using the R statistical software version 3.4.1 (R Development Core Team, 2011). The data and R code required to reproduce our results are available at the GitHub (https://github.com/tanogc/multiple_stressors_ghg/) and Zenodo repositories (Gutiérrez-Cánovas et al., 2024).

3.1 Spatial variation of stream metabolism and GHG concentrations and emissions

The stream reaches in our study showed a median GPP of 22.2 mmol C m⁻² day⁻¹ and a median algal production of 0.072 mg Chl a m⁻² day⁻¹ (Figure S6). However, there was considerable variation in these values, ranging from 0.1 to 226.7 mmol C m⁻² day⁻¹ for GPP and from 0.002 to 0.790 mg Chl a m⁻² day⁻¹ for algal production. In contrast, the streams exhibited generally high rates of respiration (ER), with a median of 76.6 mmol C m⁻² day⁻¹, ranging from 6.5 to 613.9 mmol C m⁻² day⁻¹. Consequently, all the streams were heterotrophic, with a median NEP of −56.1 mmol C m⁻² day⁻¹ and a range of −474.9 to −6.4 mmol C m⁻² day⁻¹. ER was positively correlated with both GPP (r_P = .85, p < .001; Figure 2a) and algal production (r_P = .53, p < .001). Likewise, increasing primary production was associated with greater heterotrophy, as indicated by the negative relationship between NEP and GPP (r_P = −.68, p < .001) and between NEP and algal production (r_P = −.51, p < .001). Increasing heterotrophy was also linked to a greater DO deficit (r_P = −.41, p < .01; Figure 2b).

Surface waters were generally oversaturated with respect to the atmosphere for both pCO₂ and pCH₄, resulting in net emissions of CO₂ and CH₄ (Figure 2c,d; Figure S6). Specifically, the median pCO₂ and pCH₄ concentrations were 1064 and 4.3 μatm, respectively, and the median CO₂ and CH₄ emissions were 708 mmol CO₂ m⁻² day⁻¹ and 1.8 mmol CH₄ m⁻² day⁻¹, respectively. Only two sites were found to be CO₂ sinks (yellow dots in Figure 2c). The mean of the molar ratio CH₄:CO₂ was 0.014, with a range of 0.002–0.041. Similarly, the contribution of CH₄ to the total CO₂-equivalent concentration was highly variable across the study area, with a median of 12.9% and a range of 2.2%–29.6%. Streams with high GHG concentrations tended to have greater emissions, as indicated by the positive correlation between concentrations and emissions for both CO₂ (r_P = .67, p < .001; Figure 2c) and CH₄ (r_P = .76, p < .001; Figure 2d). Higher concentrations of CO₂ (r_P = .74, p < .001) and CH₄ (r_P = .60, p < .001) were associated with a greater DO deficit. The positive correlation between pCO₂ and pCH₄ (r_P = .50, p < .001) suggests that they shared similar spatial pattern and controls. The median of the estimated k₆₀₀ was 16.1 m day⁻¹, but ranged from 3.2 to 56.0 m day⁻¹ (Figure S7).

3.2 Multiple stressor effects on river metabolism and GHGs

Our results suggest that stressor effects on river metabolism tend to be additive. GPP was positively associated with discharge and temperature (Figure 3a). In addition, DIN and discharge had a positive effect on algal production and respiration (Figure 3b,c), but a negative effect on NEP (Figure 3d). Discharge was the most important predictor of GPP (r² = 45.3%; Figure 3a) followed by temperature (r² = 16.2%; Figure 3a), whereas both DIN and discharge were the most important predictors of ER (r² = 45.1%; r² = 15.7, respectively; Figure 3b), NEP (r² = 29.1%; r² = 22.2; Figure 3c) and algal production (r² = 26.9%; r² = 18.9; Figure 3d). Riparian canopy openness had a lower explanatory power and exhibited individual effects that tended to overlap with zero.

CO₂ concentrations showed a positive relationship with DIN but declined with discharge (Figure 4a). Both DIN and discharge were the most important predictors of pCO₂ (DIN r² = 29.0%; discharge r² = 21.4%). CH₄ concentrations increased with DO deficit, water temperature and discharge (Figure 4b). DO deficit was the best pCH₄ predictor (r² = 30.7%), followed by water temperature (r² = 19.8%) and discharge (r² = 2.8%). The effects of DIN on pCH₄ were partly modulated by water temperature (r² = 7.3%), showing an opposing-type interaction.

Our pCO₂ and pCH₄ models showed that nutrient enrichment and oxygen depletion can jointly increase the overall amount of stream CO₂-equivalent concentrations, with a nonlinear increase of CH₄ concentration as oxygen depletion grows (Figure 5). We found that at oligotrophic (DIN = 0.25 mg L⁻¹) and oxic (DO deficit = 0.25 mg L⁻¹) conditions, the baseline CO₂-equivalent concentration was 959 μatm. However, when nutrient enrichment and oxygen depletion reach high values (DIN = 5 mg L⁻¹, DO deficit = 7 mg L⁻¹), the CO₂-equivalent concentration can increase up to 2225 μatm (132% increase). The proportion of CH₄ in the total amount of CO₂-equivalent concentration increased with oxygen depletion from a mean of 4.4% at oxic conditions to a mean of 20.5% at high oxygen depletion. Remarkably, even at oxic conditions, high nutrient enrichment can double CO₂-equivalent concentration (1977 μatm, 106% increase) compared with baseline conditions.

Our models also showed that the molar ratio CH₄:CO₂ increased with discharge and water temperature (Figure S8), which were the most important predictors (discharge r² = 22.5; water temperature r² = 19.7%). Water temperature effects were modulated by DIN and followed an opposing-type interaction, which explained part of the variance (r² = 12.2%).

3.3 The role of local and distal processes in mediating multiple stressor impacts on GHGs

Structural equation modelling results showed that only DIN (standardized effect size, SES = 0.475, p = .006) and discharge (SES = −0.615, p < .001) had significant direct effects on pCO₂ (Figure 6a). The indirect links between stressors and pCO₂ via NEP were not statistically supported (SES = −0.224, p = .156). Effects of DIN and discharge indicated a greater contribution of distal processes compared with local processes on pCO₂ (Figure 6b).

In contrast, pCH₄ SEM revealed that the effects of DIN and discharge on DO deficit and pCH₄ were significantly mediated by NEP (Figure 6c). Specifically, NEP was negatively related to DO deficit (SES = −0.540, p < .001), whereas DO deficit had in turn a positive effect on pCH₄ (SES = 0.663, p < .001). In addition, water temperature had a direct positive effect on pCH₄ (SES = 0.399, p = .006), but their indirect effects via NEP and DO deficit were not statistically supported. DIN effects reflected a greater contribution of local processes to pCH₄, whereas temperature and discharge effects indicated a greater influence of distal processes (Figure 6d).

4.1 Multiple stressor effects on GHGs

Our models revealed that nutrient enrichment and discharge, for pCO₂, and oxygen depletion and temperature, for pCH₄, were the most important drivers of GHG concentrations. These findings are consistent with recent research indicating that intensified land use and the resulting stream pollution can amplify stream metabolism (Ardón et al., 2021; Brauns et al., 2022) and GHG concentrations and fluxes (DelVecchia et al., 2023; Yu et al., 2017; Zhang et al., 2021). However, our study provides novel insights into how combined stressor effects alter both stream metabolism and GHGs. Specifically, we found that nutrient enrichment was more strongly associated with pCO₂ than with pCH₄. This pattern is probably reflecting the stronger positive effects of nutrients on aerobic respiration and heterotrophy (Cross et al., 2022). In contrast, pCH₄ was primary driven by oxygen depletion and temperature, indicating its strong dependence on conditions favourable for methanogenesis (Stanley et al., 2016). As such, our results show that temperature had a positive effect on CH₄, and the molar ratio CH₄:CO₂ (Yvon-Durocher et al., 2011, 2014), but that the effects of temperature can be modulated by an opposing interaction with DIN. Manipulative studies have found that nutrient enrichment and warming can jointly increase stream heterotrophy (Cross et al., 2022), but our models show a low temperature dependence of respiration. Nonetheless, we found a positive relationship between primary production and heterotrophy that might partly reflect increases in nutrient enrichment.

Consistent with the previous research (Liu & Raymond, 2018), we found that discharge was negatively associated with pCO₂, likely reflecting increased evasion rates due to more turbulent flow under higher discharge. In contrast, our results suggest that discharge may enhance pCH₄, possibly by promoting primary production, biomass accumulation and subsequent oxygen depletion in the most productive areas (Stevenson, 1996). In addition, the positive relationship between discharge and CH₄ might reflect direct hydrological import of this gas from distal sources. These findings differ from previous studies reporting increased CH₄ concentrations during low flows as a result of prolonged residence times and consequent oxygen depletion (Gómez-Gener et al., 2020; Hosen et al., 2019). Such contrasting results evidence the complex interplay between hydrology and GHGs, suggesting potential nonlinear effects of discharge on primary production and pCH₄. Furthermore, we found that riparian degradation, represented by riparian canopy openness, had a weak effect on metabolism and GHG concentrations, perhaps because the sampling time (late spring and summer) ensured sufficient light inputs even in streams with dense riparian cover. To better understand its role in explaining GHG concentrations in streams, future studies would be required to capture how seasonal variations in light and leaf inputs can influence multiple stressor effects on ecosystem functioning.

4.2 The role of local and distal processes in GHG concentrations

Our main novelty was in demonstrating that multiple stressor effects alter GHG concentrations through biogeochemical processes occurring at different spatial scales, matching our main hypothesis. Our predictions were generally supported given that CO₂ only responded to distal processes, but the controls of CH₄ were more complex than expected. These results suggest that multiple stressors can produce impacts throughout the whole fluvial continuum, giving rise to hotspots of GHG production and emission at heavily modified catchments (DelVecchia et al., 2023; Yu et al., 2017; Zhang et al., 2021). Coupled with its higher water solubility, the increase of pCO₂ in response to human impacts stems from a variety of upstream and lateral catchment sources. Despite that internal metabolism typically produces a limited fraction of stream CO₂ (10%–19% in Hotchkiss et al., 2015; a median of 29% in Solano et al., 2023), increased metabolism following nutrient enrichment (Ardón et al., 2021; Cross et al., 2022) can give rise to an intense hydrological export of CO₂ to downstream areas. As such, land-use intensification impacts can also contribute to increase aquatic GHG concentrations across the fluvial continuum through the export of organic carbon (Drake et al., 2018). Furthermore, oxygen depletion seems to respond to both distal and local nutrient inputs and to discharge, which can partly explain the frequently observed positive associations between pCO₂ and pCH₄ (DelVecchia et al., 2023). This indirect stressor effect was supported by our SEMs and revealed the importance of distal processes in controlling local pCH₄ via increased nutrient enrichment and oxygen depletion. In addition, our results support the idea that oxygen depletion and warming caused by distal processes can favour in-stream methanogenesis (Stanley et al., 2016; Yvon-Durocher et al., 2017). This reflects the need to consider the complex interplay of factors that regulate GHG production in streams in response to global change.

Our study offers novel insights into the impact of stressors on the mechanisms regulating GHGs in streams. Previous studies did not separate the contributions of distal and local processes to local GHG concentrations (DelVecchia et al., 2023; Yu et al., 2017; Zhang et al., 2021). However, our SEMs allowed us to better understand the different pathways and spatial scales through which stressors impact GHGs and the role of local metabolism in mediating the impacts of combined stressors. Although we found that direct stressor effects were relevant for both CO₂ and CH₄, previous studies overlooking the role of local metabolism in driving multiple stressor impacts on GHGs might have been missing important biogeochemical mechanisms. In addition, we built model structures that limit spurious relationships between stressors and GHGs. For example, while the previous literature commonly reports strong negative associations between dissolved oxygen and pCO₂ (DelVecchia et al., 2023; Yu et al., 2017; Zhang et al., 2021), these associations are likely a result of enhanced aerobic respiration rather than direct control on pCO₂ (Rocher-Ros et al., 2020). Similarly, relationships between reduced solutes and pCH₄ shown in previous studies are more likely coincidental than explanatory (e.g., Yu et al., 2017; Zhang et al., 2021), whereas their main driver is oxygen depletion.

4.3 Management implications

Our study highlights the potential co-benefits of policies aimed at reducing the number and intensity of stressors in order to protect river health and mitigate climate change. Our results suggest that avoiding specific stressor levels and combinations is crucial to maintain near-natural GHG concentrations. Specifically, our findings highlight that addressing nutrient pollution should be prioritized to reduce anthropogenically induced fluxes of GHG from streams. This is because nutrient enrichment was found to play a key role in influencing metabolic rates and pCO₂, and because most of the observed warming potential in streams, including our study sites, can be attributed to this gas (Campeau & Del Giorgio, 2014; Stanley et al., 2016). In addition, the abatement of nutrient pollution can further help to reduce GHGs concentrations by limiting the risk of eutrophication and oxygen depletion. Avoiding eutrophic and hypoxic conditions is also important for preventing CH₄ emissions in the context of increasing water temperatures (Kaushal et al., 2010; Kelleher et al., 2021), as these conditions will jointly exacerbate GHG emissions from polluted streams (Rosentreter et al., 2021; Yvon-Durocher et al., 2017).

The strong influence of distal processes on CO₂ production reinforces the need for adopting catchment-scale management perspectives to preserve stream health and avoid excess GHG production. Both diffuse and point-source pollution can increase stream metabolism and pCO₂ throughout the fluvial continuum. Given the global length of stream networks (Allen & Pavelsky, 2018) and their increasing degradation by multiple stressors (Schinegger et al., 2016), there is an urgent need to consider anthropogenically induced GHG concentrations in freshwaters as an important threat to climatic stability. Potential solutions include improving waste water treatment (Pereda et al., 2020), promoting soil nutrient retention (Wagg et al., 2014) and using mitigation actions that remove nutrient excess, such as riparian restoration (Feld et al., 2018). Future research should address the most effective measures and scales at which these actions should take place to limit GHG fluxes from streams.