Nitrous Oxide Emissions From Drying Streams and Rivers

Streams and rivers are suffering more extreme and prolonged low flows than those naturally occurring without human intervention with potential relevant worldwide effects on the production of nitrous oxide (N2O), a greenhouse gas 300 times more potent than carbon dioxide. Here, we address the question of whether droughts and management‐induced low flows increase riverine N2O emissions causing positive feedback in response to climate change. Supported by field data, we model riverine N2O emissions for decreasing mean summer flows in a forested and agricultural watershed under chemostatic and chemodynamic dissolved inorganic reactive nitrogen (DIN) concentrations. Our results demonstrate that total N2O emissions decrease with increasing low flow severity in both watersheds regardless of DIN scenario; which imparts a form of climate resilience.

• Low flows lead to higher N 2 O emission per unit stream area in agricultural but not in forested reaches • Total riverine N 2 O emissions decrease with low flow severity regardless of land use • Increasing low flow severity does not induce positive feedback to climate change due to N 2 O emissions

Supporting Information:
Supporting Information may be found in the online version of this article. 2 of 10 on local land cover, water use, and changes in surface hydraulics (Marzadri et al., 2017). River morphology may change with discharge, for instance, dune size adjusts to local discharge because fine sediments are typically mobile even at low flows (Yalin, 1964). These morphological changes, in turn, impact hyporheic hydraulics and thus the time for biogeochemical transformation within streambeds. Moreover, nutrient concentrations may be discharge-dependent or behave as chemostatic (similar concentration regardless of discharge) depending on land uses (Basu et al., 2010;Levi et al., 2018). Denitrification reaction rate constant also changes depending on both solute concentration and surface hydraulics (see, Supporting Information S1 and Beaulieu et al., 2011;Davis & Minshall, 1999;Mulholland et al., 2008). These complex dependence of N 2 O emissions on local processes, from river morphology to biological reactivity, hinder predicting N 2 O emissions with decreasing discharge and prevent simple extrapolation of N 2 O trends observed under natural base flows to more extreme low flows conditions.
Here, we used the N 2 O emission model developed by Marzadri et al. (2017) and successively modified by Marzadri et al. (2021), which accounts for biological processing within the riverine hyporheic benthic and water column environments, to identify the hydrological and environmental conditions that cause riverine systems to increase or decrease total N 2 O emissions during low flows, at areal and whole-river network scales. We applied the model to two low-gradient watersheds with contrasting land-use land-cover: the Manistee River (Michigan, USA; ∼83% forested) and the Tippecanoe River (Indiana, USA; ∼82% agricultural). We collected two sets of 80 synoptic samples, providing water temperature and concentrations of NO 3 − , N 2 O, and ammonium (NH 4 + ), across both watersheds during a typical base flow condition during summer 2015. These model input data were supplemented by hydro-morphological measurements, including mean flow depth, velocity, width, substrate size, and bedform type at the same synoptic sites, thereby allowing the model to account for changes in stream hydraulics and morphology as discharge changes (Stewart et al., 2011) (see Supporting Information S1).

Methods
The Marzadri et al. (2021) model does not require calibration to predict the dimensionless N 2 O flux (F*N 2 O) and it is fully characterized by reach-scale, hydro-morphological characteristics, such as roughness, streambed morphology and discharge, and how these metrics change as a function of flow regime, combined with stream water temperature, and stream dissolved inorganic nitrogen (DIN) concentrations (see Supporting Information S1). F*N 2 O represents a type of N 2 O emission factor, at the reach scale resolution , and is defined as the ratio between the flux of N 2 O (FN 2 O) per unit stream area and the in-stream flux of DIN species per unit stream cross-section area (NH 4 +  ]), with V as the mean reach flow velocity. F*N 2 O value depends on the dimensionless Damköhler number, Da D , which accounts for reachscale hydro-morphological and biochemical characteristics. The Da D is defined as the ratio between the residence time τ in the compartment where denitrification chiefly occurs and the characteristic time of denitrification τ D : Da D = τ/ τ D . Two relevant stream compartments are considered: (a) the hyporheic zone, including the overlying benthic zone, with a characteristic residence time defined as the median of the travel time within the hyporheic zone τ 50 , and (b) the water column, with a characteristic time identifiable as the turbulent vertical mixing time scale t m = D/(0.067[g·D·s 0 ] 0.5 ), where g is the gravitational acceleration, D is the mean flow depth and s 0 is the friction slope (Rutherford, 1994). In small systems (i.e., streams), denitrification occurs chiefly in the benthic-hyporheic zone, but as systems widen (i.e., in rivers) and their slope declines, water depth increases and denitrification occurs mostly in the water column. These two situations are represented by using two different Da D numbers: Da DHZ = τ 50 /τ D for headwater streams, and Da DS = t m /τ D for rivers. River morphology (dunes but not pool-riffle, whose size depends on effective discharge [e.g., Tubino, 1991]) and local hydraulics are dynamically linked to discharge such that the model can account for direct impact of discharge to flow regime, and indirectly on stream bedform size (see Supporting Information S1). For both Da D numbers, the biological processes are represented via the denitrification time scale, τ D , which is the inverse of the denitrification reaction rate and is quantified as a function of the stream denitrification uptake rate, v fden (see Supporting Information S1). The v fden is a mass transfer coefficient, which is operationally defined as "demand relative to concentration" and its value decreases as NO 3 − concentrations increase (Davis & Minshall, 1999) typically in the form of v fden = a·[NO 3 − ] b with a and b being regression coefficients (Beaulieu et al., 2011;Mulholland et al., 2008).
The model consists of three equations, whose application depends on stream size, expressed as channel width, W, because the main source of N 2 O emissions shifts from the hyporheic zone in small streams (HZ Model), chiefly benthic zone for intermediate streams (BZ Model) and then to the water column for rivers (WC Model):  (Beaulieu et al., 2008(Beaulieu et al., , 2009) and 16 reaches from the second iteration of the Lotic Intersite Nitrogen eXperiment (LINXII) (Beaulieu et al., 2011;Hall et al., 2009;Mulholland et al., 2008). The obtained equations were validated against measured emissions from 400 reaches of streams and rivers worldwide during summer base flows with riverine water temperatures ranging between 11 and 36°C. The third equation of the model (F*N 2 O WC ) was obtained analyzing measured emissions from streams and rivers with width larger than 30 m in terms of the Damköhler number that considers the major role played by the water column in controlling emissions (Da DS ); this is expressed differently from the first two equations and is a regression-based model that we validated with independent data ( Figure S1 in Supporting Information S1).
Stream N 2 O measurements along with hydro-morphological data collected during drought conditions are scarce.
To our knowledge, we collected and used all those measurements available in the literature: nine headwater streams from a small Swedish agricultural catchment (C6, Kyllmar et al., 2006; Figure S2 in Supporting Information S1) and eight reaches of the San Joaquin River (Hinshaw & Dahlgren, 2012). The performance of HZ model in predicting N 2 O emissions during low flow conditions was tested with the data from the Swedish streams (r 2 = 0.53, Figure S3 in Supporting Information S1), and that of the BZ model with the data from San Joaquin River (r 2 = 0.43, Figure S3 in Supporting Information S1). In the two cases, only reaches within the HZ (W ≤ 10 m) and BZ bounds (W > 30 m) were included in the analysis.
We quantified the role of low flows in N 2 O emissions by applying the above framework to the Manistee River (MI, USA, forested) and the Tippecanoe River (IN, USA, agricultural). The Manistee River is forested over 83% of its 3,616 km 2 watershed, stream flow discharges range between 22 and 14,000 L/s and summer low-flow NO 3 − concentrations average 120 μg N/L. The Tippecanoe River has 82% of its 4,496 km 2 basin in row-crop agriculture, discharges range between 0.9 L/s and 22,500 L/s and summer low-flow NO 3 − concentrations average 1,860 μg N/L. Both watersheds have reaches with dune-like or pool-riffles morphologies with sediments composed by gravel and sand. Information from the grain size distribution allowed estimation of the hydraulic conductivity of the streambed sediment (Salarashayeri & Siosemarde, 2012).
We collected all the required hydro-morphological and water quality information during a synoptic sampling effort in summer 2015, at 80 locations in each watershed, collected within a 24-hr period, at baseflow conditions. To apply the model to each reach of the riverine network, we spatially distributed these original measurements of temperature, NH 4 + and NO 3 − concentrations along the two riverine networks using the Spatial Stream Network (SSN; R) (Ver Hoef et al., 2014) model and Spatial Tools for the Analysis of River Systems (STARS; ArcGIS) , which uses a one-dimensional geostatistical interpolation along the streams. Mean flow velocity, depth and width were spatially distributed with a set of regime equations developed from the 80 locations at each watershed (Equations S9 and S10 in Supporting Information S1). Bedform type and size were derived by combining Montgomery and Buffington stream classification (Montgomery & Buffington, 1998) with hydro-morphological relationships for dune and pool-riffle. Bedform prediction was verified with visual inspection at the sites. Stream discharge was predicted at each reach of the river networks with the GIS tools ArcHydro and Hec-GeoHMS, which provided the input conditions for the Land Surface Model (Piccolroaz et al., 2016) that predicted the lateral inflows to the river network. Flow generation was accomplished with a continuous soil-moisture accounting model based on the SCS-CN methodology (Avesani et al., 2021). Stream flows were then routed with a validated (averaged Nash-Sutcliffe index larger than 0.7) Muskingum flow routing scheme (Cunge, 1969) at the reach scale.
The resolution of the model and input data allows us to explore the response of N 2 O emissions longitudinally, from headwaters to mainstem reaches, along the forested and agriculture watershed, and at the watershed scale under different scenarios of decreasing low flows (i.e., increasing drought severity). The mean daily discharge (Q sum ) in summer is a common statistic used for water resource planning with important ecological and water use implications (WMO, 2008). Thus, we used Q sum as our reference discharge for current flows and selected three alternative mean summer flows, which may replace the current Q sum in future scenarios resulting from climate-induced increase in drought severity or low flows induced by water uses. These data represent current daily base flow conditions, and their 25th and 5th percentiles, which we quantify from daily discharges monitored at six US Geological Survey (USGS) gauging stations located within the two watersheds with the R package lfstat (Koffler & Laaha, 2012). We define these three alternative summer mean flows as moderate low Q sum , very low Q sum , and extremely low Q sum . Current information on DIN concentration and temperature distributions were then used as reference values for simulating N 2 O fluxes. We considered the two most likely end member scenarios for FDIN: constant DIN concentrations (i.e., chemostatic scenario) and constant FDIN (i.e., conservation of mass flux, chemodynamic scenario) (Basu et al., 2010;Levi et al., 2018). In chemostatic scenario, a flow reduction gives rise to a contemporaneous reduction in catchment inputs (Mosley, 2015) and thereby a reduction of FDIN (see example, Figure S4a in Supporting Information S1). However, in chemodynamic scenario, an increase in DIN concentrations may offset the lower flows resulting in a constant FDIN, rather than constant DIN concentrations, with decreasing discharge (Hinshaw & Dahlgren, 2012) (see example, Figure S4b in Supporting Information S1).
Total N 2 O emissions (TE) at the watershed scale were quantified at daily time steps with: where N r is the number of the reaches that compose the stream network, W r is the width of the r-th reach, L r is its length, F*N 2 O r is its dimensionless N 2 O emission estimated with the power law scaling model (Equation 1), and FDIN r is its flux of DIN.

Results and Discussion
The effect of the DIN scenario considered on F*N 2 O was large because DIN concentration influenced the calculation of Da D via v fden . For the chemodynamic scenario, F*N 2 O values remained almost constant ( Figure S5 in Supporting Information S1), while values increased for the chemostatic scenario. In the latter case, the model predicted a statistically significant increase in F*N 2 O with lower flows in the agricultural headwater (HW, stream orders 1 to 3) and main stem (MS, stream order >3) reaches, but not for the forested headwater and mainstem reaches (Figure 1; Kruskal-Wallis non-parametric test results in Table S4 in Supporting Information S1). The impact of low flows on F*N 2 O in agricultural reaches reflects its land use footprint characterized by narrower streams (including ditches) compared to the forested watershed ( Figure S5 in Supporting Information S1). As such, the hydro-morphological characteristics especially in HW are more sensitive to changes in discharge in the agricultural than in the forested watershed, as indicated by the 30% decline in mean W simulated for extreme low Q sum (Figures S6 in Supporting Information S1). In contrast, forested HW streams have a less pronounced reduction in width (half of the agricultural, 15%) and smaller changes in Damkhöler numbers (1.4% increase) with low-flow severity than the agricultural HW, which shows a 26% increase in Damkhöler numbers (Figures S6 and S7 in Supporting Information S1). In contrast to HW, simulated low flow conditions have subtler impacts on main stem hydromorphology (e.g., mean W reductions of 7% and 13% from Q sum to extreme low Q sum for the forested and agricultural MS, respectively), which drives the non-significant changes in F*N 2 O in river forested MS, but still significant in the agricultural.
However, an increase in F*N 2 O does not necessarily lead to an increase of FN 2 O from riverine systems, because changes in FN 2 O depend also on DIN availability (FDIN), which is influenced by land use, land cover, and discharge (Basu et al., 2010;Marinos et al., 2020;Mosley, 2015). Both FDIN scenarios did not significantly change FN 2 O in either HW or MS reaches in the forested watershed. Conversely, the chemodynamic scenario caused significant increase of FN 2 O with decreasing low flows in the agricultural HW reaches, whereas the chemostatic scenario caused a significant reduction in FN 2 O with low flow severity in the agricultural MS reaches (Figure 1 and Kruskal-Wallis test results in Table S4 in Supporting Information S1).
We found that the distribution of DIN concentrations measured during the summer synoptic sampling showed a clustering of high and low concentrations within sub-basins, rather than a longitudinal pattern from headwater to mainstem reaches (Figures 2a and 2b). As such, the change in N 2 O emissions per unit streambed area (ΔFN 2 O) induced by the extreme low flow scenario showed a systematic decrease from HW to MS reaches (except for order 1), suggesting that low flows accentuate increases in FN 2 O from headwater reaches, especially from stream orders 2 and 3 (Figures 2g and 2h). The same results were found for total emissions quantified from headwater or mainstream reaches, which further underscores the importance of headwater streams in controlling N 2 O emissions at the watershed scale (Table S5 in Supporting Information S1). We found that this pattern was accentuated in agricultural reaches compared to forested ones (Figures 2c, 2h and Figure S8 in Supporting Information S1), likely because the former had a stronger reduction in width, which contributed to decrease FN 2 O ( Figure S6 in Supporting Information S1). This size change is critical because the hyporheic zone is the dominant source of N 2 O yield in HW streams, but its role decreases rapidly as size increases, from HW to MS. Brackets report the p value significance (*p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001) of the statistically significant changes in F*N 2 O and FN 2 O with low flow severity with the non-parametric Wilcoxon test. If not reported the trend is not statistically significant. Only the agricultural watershed has some scenarios with statistically significant changes: F*N 2 O from headwaters and mean stems for the chemostatic scenario and FN 2 O from headwaters for the chemodynamic scenario significantly increase with increasing low flow severity, but FN 2 O values decrease for the main stems with the chemostatic scenario (also see Kruskal-Wallis test in Table S4 in Supporting Information S1). Mean FN 2 O increased for the chemodynamic scenario (i.e., constant DIN flux) regardless of watershed type, but FN 2 O decreased for the chemostatic case for the forested watershed ( Figure 3a). Overall, the FN 2 O patterns for the forested watershed showed less dependence on the DIN scenario considered compared to the agricultural watershed, suggesting that forested watersheds are more resilient to increasing the occurrence of low flows.
The spatial distribution of changes in FN 2 O demonstrates how biogeochemistry, hydrology, and land use practices interact to influence N 2 O emissions from different reaches within a stream network (Figures 2c-2h). However, it is essential to understand the response of entire watersheds to increasing low flow conditions by transitioning N 2 O emissions from a reach-to a whole fluvial network-scale perspective. We calculated the total N 2 O emissions (TE) over the entire watershed at daily time scale by accounting for each reach surface area (See Equation 2 in Section 2). Whereas average FN 2 O may slightly decrease or increase up to 38% depending on the DIN concentration scenario and watershed considered (Figure 3a), channel width, which controls the riverine surface areas (reach length is fixed), is expected to decrease or at the most remain constant with increasing low flow severity (Figure 3a, star symbols). For the three low flow scenarios, the model predicts a decrease in total daily N 2 O emissions for both watersheds regardless of the DIN scenario considered (i.e., chemodynamic or chemostatic), primarily driven by the reduction in channel width (Figure 3b).
All scenarios result in a decrease of TE emission between 10% and 20%. The reduction in stream width (grey star symbol Figure 3a) compensate the almost 40% increase in FN 2 O (Figure 3a pink square open symbols). This effect is stronger in headwater than main stem reaches (Table S5 in Supporting Information S1). Our analysis indicates that the compensatory role of changes in channel morphology is critical if we are to understand the feedback between climate change and its effect on the biogeochemistry of streams and rivers. Whereas the results obtained at the reach scale may suggest that lower flows, and eventually higher DIN concentrations in headwater reaches, could lead to overall higher N 2 O emissions (Figures 2g and 2h and Figure 3a), our watershed-scale analysis points toward the opposite direction: total emissions of N 2 O at the watershed scale would likely decrease with increasing drought conditions in the future because reduction in flow and stream width will hold down N 2 O emissions even in agricultural watersheds where increases in DIN concentration are expected (Figure 3b).

Conclusions
Our results highlight that the impact of drought on N 2 O emission factor expressed as F*N 2 O and emissions per unit area of stream (FN 2 O) was maximized in headwater reaches but decreased with increasing stream size in the river mainstems. This finding underscores the importance of managing (e.g., DIN runoff and channelization) small headwater streams, especially in agricultural lands. Drought conditions significantly decrease FN 2 O in those reaches where no changes in DIN concentration are expected (i.e., chemostatic scenario). Yet, in most stream reaches, the increase in low flow condition (i.e., drought severity) lead to increases in FN 2 O for the scenario of future increases in DIN concentrations and constant DIN fluxes, which is the most critical situation in streams draining agricultural lands with long-lasting legacy effects. Conversely, changes in FN 2 O are subtle, and not statistically significant for stream reaches at the forested watershed. However, our conclusions on the implications of increasing drought conditions for N 2 O emissions drastically change when results are integrated at the whole fluvial network scale because of the interplay between changes in FN 2 O and channel width. While FN 2 O at the reach scale may increase or decrease with drought conditions, channel width will mostly decrease. The combined effect of these two variables results in an amplified reduction of N 2 O emissions from forested and agricultural networks regardless of the DIN scenario. These results stress the intertwined link between in-stream biological processing and hydrological conditions; hydrology does not only control the mean residence time of water, and with that, the Damköhler number and the characteristic time of denitrification, but also drives the expansion and contraction of the fluvial network, which is a fundamental feature for understanding how changes in biogeochemical fluxes per unit stream area resonate at the whole fluvial network scale. Overall, our results show no evidence that climate change-induced drought would cause positive feedback on N 2 O emissions, but rather suggests that low flow conditions would contribute to a decrease watershed-scale N 2 O emissions in both agricultural and forested landscapes.