Effect of topographic slope on the export of nitrate in humid catchments

Excess export of nitrate to streams affects ecosystem structure and functions and has been an environmental issue attracting world-wide attention. The dynamics of catchment-scale solute export from diffuse nitrate sources can be explained by the activation and deactivation of dominant flow paths, as solute attenuation (including the degradation of nitrate) is linked to the age composition of outflow. Previous data driven studies suggested that 25 catchment topographic slope has strong impacts on the age composition of streamflow and consequently on in-stream solute concentrations. However, the impacts have not been systematically assessed in terms of solute concentration levels and variation, particularly in humid catchments with strong seasonality in meteorological forcing. To fill this gap, we modeled the groundwater flow and nitrate transport for a cross-section of a small agricultural catchment in Central Germany. We used the fully coupled surface and subsurface numerical simulator HydroGeoSphere to model 30 groundwater and overland flow as well as nitrate concentrations. We computed the water ages using numerical tracer experiments. To represent various topographic slopes, we additionally simulated ten synthetic cross-sections generated by modifying the mean slope from the real-world scenario while preserving the land surface micro-topography. Results suggest a three-class response of in-stream nitrate concentrations to topographic slope, from class 1 (slope > https://doi.org/10.5194/hess-2021-375 Preprint. Discussion started: 14 September 2021 c © Author(s) 2021. CC BY 4.0 License.

with smaller porosity due to high loam content [Yang et al., 2018]. Subsequently, six property zones were used ( Figure   1b), with zonal parameter values following the model in Yang et al., [2018] listed in Table 1.

Climates 150
The considered climate for the cross-sections was derived from the catchment 'Schäfertal' located in a region with temperate humid climate and pronounced seasonality. According to the meteorological data records from 1997 to 2007, the mean annual J and Q (per unit area) are 610 mm and 160 mm, respectively. Actual mean annual ET based on the ten-year water balance (J = ET + Q) is 450 mm. Mean annual potential ET is 630 mm [Yang et al., 2018]. The humid climate is representative for wet regions, quantified by an aridity index (J / potential ET ) of 155 1.0. The ET is the main driver of the hydrologic seasonality as the precipitation is more uniformly distributed across the year (Figure 1c). To acknowledge this fact, we selected the data records of the year 2005, and calculated the annual J and monthly-averaged potential ET. Using these averaged values in the study can accelerate the simulations and simplify the analysis while preserving the main characteristics of the meteorological forcing to the system.   [Yang et al., 2018]. The dashed lines represent the annual (J) and monthly (ET) averages. 165

Flow and nitrate transport
It is necessary to solve both groundwater and surface water flow because the spatially-explicit details in the model catchment including the specific flow paths and exchange fluxes are necessary to interpret the effect of varying 170 topographic slope on nitrate transport. We simulated the flow system using the fully coupled surface and subsurface numerical model HydroGeoSphere, which solves for variably saturated groundwater flow with the Richards' equation and for surface flow with the diffusion-wave approximation of the Saint-Venant equations [Therrien et al., 2010].
Additionally, the exchange flux between groundwater and surface water can be implicitly simulated. The nitrate transport is simulated in the groundwater flow, surface flow and exchanges fluxes by solving the advection-dispersion-175 diffusion equation describing the conservation of nitrate mass. The model has been successfully used to simulate https://doi.org/10.5194/hess-2021-375 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License. catchment hydrological processes and solute transport in many studies [e.g. Therrien et al., 2010;Yang et al., 2018], therefore governing equations and technical details are not explicitly repeated here.
The modeled subsurface of the cross-sections was discretized into 15 horizontal element-layers between land surface and aquifer base, with thinner layers in the upper part (~ 0.05 m) to better represent the unsaturated zone in more detail 180 and thicker layers in the lower part (~ 1 m). The cross-sections were 420 m long and uniformly discretized into 200 cells. Apart from that, each cross-section had a width (lateral direction perpendicular to the cross-section) of 100 m discretized uniformly into 10 cells. The reason for that was to avoid boundary influences that may have been caused by the lateral flow boundary condition (described later). In total, the discretization led to 30,000 block elements for the surface. The topmost 2,000 rectangles (200×10) were used to discretize the surface domain, where surface flow 185 was simulated.

Parameters and boundary conditions for flow
The key model parameters for simulating groundwater flow, surface flow and ET are listed in Table 1. Their values were taken from previous work [Yang et al., 2018], where a hydrological flow model was built and calibrated against 195 measured groundwater levels and Q for the entire catchment. For each cross-section, constant J and time-variant potential ET were applied to the aquifer top. HydroGeoSphere calculates actual ET from potential ET taking into account the modeled water content, leaf area index and root depth distributions. A free-drainage boundary condition was assigned to the topmost 1.5 m of the subsurface at the valley side (left side) boundary (Figure 1b), enabling subsurface discharge to the channel. A critical depth boundary [Therrien et al., 2010] was assigned to the left-side 200 edge of the land surface, allowing surface discharge to the channel. In the 3D catchment, surficial flowpaths can connect the surface water ponding in the depressions of the land surface. In our 2D model, it is unrealistic to force these surficial flowpaths to be parallel to the cross-section. Therefore, our model allowed for the lateral exit of surface water via the depressions of the land surface by assigning critical depth boundary conditions there. This lateral exit was also counted as surface discharge to the channel. Finally, the total discharge Q can be calculated by summarizing 205 the subsurface and surface discharge.

Parameters and boundary conditions for transport
The nitrogen pool in the soil is controlled by various complex processes. It is replenished by inputs from atmospheric deposition, biological fixation, animal manure from the pasture area, and fertilizer from the farmland on the hillslopes.
Nitrate-N that can be transported with water is formed from this (organic) nitrogen pool by a microbiological 210 immobile-mobile exchange process Van Meter et al., 2017]. In our study, it is not necessary to fully implement all the complexities of the different nitrogen pools and transformations into the model, because we focus on the in-stream concentration responses with regard to catchment topography, rather than on the full nitrogen cycle of the catchment. Therefore, we assumed that a nitrate source concentration Cj was associated with the precipitation. The Cj, which is time-variant, comprehensively defines the amount of dissolved nitrate that can enter 215 the storage along with precipitation. In this study, the Cj curve followed the time-variant nitrate leaching concentrations calculated by Nguyen et al., [2021] using a mesoscale nitrate export model ( Figure 2). Additionally, we also considered cases with constant Cj which is the average of the time-variant values. This constant Cj was used to quantify the source contribution to the variation of instream concentrations (described in section 3.3).
To simplify the transport processes, the nitrate transported with ET (representing plant-uptake) was assumed to not 220 alter the nitrate concentration of the water in storage, neglecting the evapoconcentration effect, because its potential effect to cause source variability was implicitly considered by forcing the source concentration to vary along the Cj curve. The denitrification in the system was described by the first order decay process with a degradation rate coefficient λ of 0.009 day -1 according to Nguyen et al., [2021] studying the area including the catchment Schä fertal.  In total, we simulated the flow and transport for 22 scenarios (11 topographic slopes × 2 for variable/constant nitrate sources). For each scenario, the simulations were run for 100 years with identical boundary conditions for each year.
The first 99 years were used as a spin-up phase to assure a dynamic equilibrium (i.e. to achieve simulated variables, such as heads and concentrations, being identical between years), and the last year was used for actual observation and analysis. The CPU time of each simulation was ~4 hours. 235

Water ages
The water stored in a catchment (storage), Q and ET is characterized by its age distribution, as it comprises water parcels of different age from precipitation events that occurred in the past. The age distributions need to be calculated for each aforementioned scenario to assess the responses of water ages on catchment topographic slope. Our model 240 setup (with virtual cross-sections and identical climate for each year) allowed us to perform long-term numerical tracer experiments and to extract the age distributions.
We assumed that inert tracers of uniform concentration existed in precipitation. The tracers were applied to the land surface as a third-type (Cauchy) boundary condition and were subjected to transport modeling. Tracer can exit the aquifer via the outfluxes Q and ET. We considered a period of 200 years for the tracer experiments, which was 245 sufficiently long to ensure convergence of the computed water ages. The 200 years period was partitioned into 2400 months ∆ . A different tracer was used for each of the periods resulting in a total of 2400 distinct tracers. The injection of tracer i started at the beginning of its associated period 0 and lasted throughout the period with the precipitation.
The advective-dispersive multi-solutes transport was simulated using HydroGeoSphere. The first 199 years of the simulation period were used a spin-up phase to ensure a dynamic equilibrium of calculated ages, minimizing the 250 https://doi.org/10.5194/hess-2021-375 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License.
influence of the initial conditions. The last year was used for the actual observations and the computation of age distributions. Solving the transport of the 2400 tracers is computationally expensive. However, because the climate (flow boundary conditions) was identical for each year, the transport simulation was performed only for the first 12 tracers that covered the course of a year. Based on these results, the results for the other 2388 tracers were manually reproduced (e.g., by shifting the concentration breakthrough curves of the 12 tracers in time while maintaining the 255 shapes).
For each tracer, the breakthrough curves of the mass-fluxes of Q and ET, as well as the mass in storage were reported. For each scenario, the CPU time of the tracer experiment was ~8 hours. Based on the age distributions, we calculated the mean discharge age (t), which is equivalent to the mean discharge transit time (simply referred to as 'discharge age' in the following sections). We calculated the young water fraction in streamflow ( ), which is the fraction of streamflow with an age younger than three months (also referred to as 'young streamflow fraction' [Jasechko et al. 2016]). Similarly, the ET age (t) and the young water fraction in ET ( ) can be calculated as well (more 270 details are described in Text S1 of the supporting information). Their responses to a change in topographic slope were analyzed.

Assessment variables
The simulations of flow, nitrate transport and water age provided in-stream nitrate concentrations ( ), streamflow 275 ages (t) and young water fractions ( ) for each scenario. They all fluctuated seasonally over the course of a year.
The temporal means and standard deviations σ of these variables can be calculated. The temporal variation in can potentially be split and attributed to (i) the variability in the nitrate source concentration, referred to as source contribution, and (ii) the variability created by degradation associated with variable transit times, referred to as degradation contribution. To understand which of these processes has the dominant effect on CQ variability, we 280 quantified the source contribution by calculating the relative change of σ for when Cj switches from being timevariant to being constant between separate model scenarios (see section 3.1), as: The calculated source contribution ranges from 0 (degradation-dominated) to 100 % (source-dominated). Additionally, the Damköhler number ( = ̅̅̅ · λ, [Oldham et al., 2013]), which is a dimensionless ratio between the discharge 285 age and the reaction time, can be calculated to indicate the interplay between the rate of degradation and the timescale of transport. Da > 1 indicates a faster degradation time than transport time and vice versa.

Results and discussion
For all the scenarios, the simulated Q, in-instream nitrate concentrations CQ, the young water fractions YF, and the 290 water ages show seasonal fluctuations. and YFET. Young water in ET is low during the wet and high during the dry season, while young water in Q is low during the dry and high during the wet season. ET generally has larger young water fractions than Q as ET has a higher probability to remove young water from the shallow soil rather than the older water in the deeper aquifer. Especially during the dry season, most precipitation can be quickly removed by ET. Figure 3d shows that ET age ranges from 192 to 395 days, being older during the wet and younger during the dry season. Simulated discharge age ranges from 300 1259 to 1490 days, being younger during the wet and older during the dry season.
Generally, the seasonal fluctuation patterns of CQ, with both variable source and constant source, are highly correlated with the fluctuation pattern of the YFQ with Spearman rank-correlation coefficients of 0.81 and 0.93, respectively. The calculated Da for streamflow is 13, demonstrating that the degradation time-scale is significantly shorter than the https://doi.org/10.5194/hess-2021-375 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License. transport time-scale. This means young streamflow is the main contributor of nitrate mass as most of the nitrate in the 305 older water has been degraded before reaching the stream.  A similar three-class response can be observed for the wet-time CQ (blue line in Figure 4a), it is even more pronounced than the one for the mean CQ. The dry-time CQ decreases linearly from steeper to flatter landscapes, not exhibiting specific classes. The effect of topographic slope on CQ is hence dominated by the wet season response as most of the discharge was produced during the wet season. The response pattern of the YFQ is highly identical to the CQ response 330 patterns, also showing a three-class response (Figure 4b). This indicates that the topographic slope influences the CQ levels via changing the young water fraction. Figure 4c demonstrates that the discharge age TQ tends to be younger in steeper and older in flatter landscapes, especially during the dry season. This pattern did not show any correlation to the response of CQ, thus suggesting that discharge age TQ is not the most valuable predictor of CQ.  (1) For C1 (slope 1:20 -1:60), Figure 5a reveals that the hillslope part of the aquifer with a slope of 1:20 is largely unsaturated so that the flow paths in this area are characterized by vertical infiltration (Figure 5a). In contrast, the valley bottom is fully saturated. Overall, 34% of the subsurface domain (in volume) is characterized by vertical flow.
For this scenario two main discharge routes to the stream can be identified: (i) A fraction of the groundwater flows through the fully saturated zone and exits the aquifer to the stream, and (ii) another fraction exits the aquifer via 350 seepage near to where the groundwater table intersects the land surface, indicated by a large exchange flux (from subsurface to surface, positive). The seepage represents a preferential flow path allowing for rapid discharge via overland flow instead of slower discharge via the sub-surface with longer transit times. Reducing the topographic slope to 1:60 does not significantly change the flow pattern ( Figure 5b). However, the spatially averaged depth of the groundwater table is reduced from 1.5 m to 0.8 m (Figure 4d). This change leads to two main effects: (i) the infiltration 355 processes is weakened, indicated by the fact that the portion of subsurface domain characterized by vertical flow is reduced from 34% to 22%, and (ii) the shallow subsurface flow processes, such as seepage, are promoted, increasing the amount of water taking the short shallow flow paths in the system. This is proved by that the portion of streamflow generated by surface run-off increased from 3.2 m 3 /day to 7.1 m 3 /day (Figure 5a, b).
Subsequently, the contribution of young water to streamflow significantly increases when the slope decreases from 360 1:20 to 1:60, also supported by the computed TTDs (Figure 6a). Given that the groundwater storage significantly increases with decreasing slope, this effect is similar to the "inverse storage effect" that has been described in Harman, [2015] as the relative contribution of young water to stream flow increasing with increasing storage. Kim et al., [2016] also reported based on their lysimeter experiments that younger water was discharged in greater proportion under wetter conditions compared to drier conditions. However, the observed changes in groundwater table depth (thus 365 storage) in our study were caused by topography rather than by climate.
(2) For C2 (slope 1:60 -1:100), even though the groundwater table depth is still decreasing with decreasing slope, the flow pattern experiences a rapid change. The seepage flow vanishes because the groundwater table (fully or partially) disconnects from the land surface ( Figure 5c). The water that would have flown to the stream via seepage has to take slower flow paths in the subsurface to the valley bottom. The surface run-off dropped significantly from 370 7.1 m 3 /day to 0.5 m 3 /day (Figure 5c, d). Basically, decreasing the topographic slope reduces the horizontal component of the hydraulic head gradients, which is obvious as part of the precipitation falls at lower elevations instead of farther up the hillslope. The reduced head gradient generally slows down the groundwater flow velocity. Several hydrologic studies have described two different flow systems in aquifers: (i) a recharge-limited system where the thickness of the unsaturated zone is sufficient to accommodate any water-table rise and thus the elevation of the 375 groundwater table is limited by the recharge, and (ii) a topography-limited system where the groundwater table is close or connected to the land surface such that any fluctuation in groundwater table can result in considerable change in surface runoff [Werner and Simmons, 2009;Michael et al., 2013]. In our study, the aquifer of C1 is a partially topography-limited system (e.g. Figure 5a, b) (the hillslope is recharge-limited while the valley bottom is topographylimited). In C2 the aquifer is transformed into a fully recharge-limited system (from Figure 5b  (3) For C3 (slope 1:100 -1:1000), the aquifer is fully recharge-limited without any preferential flow via land surface.
Further reducing the topographic slope to 1:1000 mainly changes the spatial distribution of the unsaturated zone (comparing Figure 5d with 5c). Because the groundwater table depth (thus the storage) more or less remains 385 unchanged (Figure 4d), interestingly, here the "inverse storage effect" does not apply anymore and cannot explain the increase of the YFQ when the topography becomes flatter.
However, on flatter landscapes, local flow cells are more likely to form, where water infiltrates to the aquifer and eventually exits the aquifer via ET rather than via flow to the stream ( Figure 5d, the local flow cells are more pronounced in the dry season, see Figure S1d in the supporting information). Simply put, precipitation falling farther 390 from the stream has a lower chance to reach the stream and a higher change to be intercepted by ET on its way to the stream, because the flow velocity is much lower due to the smaller horizontal component of hydraulic head gradient.
While precipitation water close to the stream has a higher chance to contribute to streamflow. We hypothesize that the increase of the YFQ, as indicated by the computed TTDs (Figure 6b), is due to this reduction of the longer flow paths and the persistence of shorter flow paths. 395 To further verify our hypothesis, we mapped the land area contributing to the streamflow (streamflow generation zone) using a particle tracking algorithm in HydroGeoSphere [Yang et al., 2018]. Figure 7 demonstrates the streamflow generation zone in February for the slope 1:100 and 1:1000, respectively. For the aquifer with a slope of 1:100, the zone extends further into the hillslope, with relatively younger streamflow generated close to the stream and old streamflow (i.e. age > 5 years) generated further up the hillslope. When the slope is reduced to 1:1000, the streamflow 400 generation zone is much closer to the stream. The hillslope that used to generate old streamflow does not contribute to streamflow anymore. This means that in flatter landscapes, the evolution of local flow cells reduces the connectivity between the stream and the more distant hillslopes by intercepting the longer flow paths at the land surface before they can reach the stream (Figure 7b), leading to an increase in the YFQ. We refer to this as the "local flow cells effect".   In summary, we identified three classes for the response of in-stream concentrations to topographic slope under a humid climate. When the landscape becomes flatter, the hydraulic head gradient as the main driving force, changes the aquifer from a partially topography-limited system with preferential overland flow (C1) to a recharge-limited 425 system that is more likely to form local flow cells (C3). For the aquifer of C2, which is a transitional class between C1 and C3, YFQ and nitrate concentrations experience a sharp drop once the preferential overland flow paths cannot be maintained. For the aquifer of C1 (or C3), decreasing slopes tend to generate a higher fraction of young streamflow and export nitrate at higher concentrations. However, the former is dominated by the "inverse storage effect" while https://doi.org/10.5194/hess-2021-375 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License. the latter is dominated by the "local flow cells effect". In this sense, the response of in-stream concentrations to 430 topographic slope is threshold-like rather than monotonous.

In-stream nitrate seasonal variations
Simulated results demonstrate significant seasonal variations of CQ for all the scenarios ( Figure S2 in the supporting information). Basically, this variation in CQ is caused either by the fluctuation of the nitrate source input, or by 435 fluctuations in degradation time associated with the varying transit times. The calculated source contribution for our simulated scenarios indicates that only 2 to 33 % of the variation of CQ can be attributed to the fluctuation of source concentrations (Figure 8a). This means that the nitrate concentration fluctuations in all simulated cross-sections are dominated by the variability in degradation time (transit time). In other words, significant seasonal variation of the nitrate concentration in streamflow can be expected under the considered humid climate even when nitrate is applied 440 to the aquifer in a constant manner without any variation. These seasonal fluctuations of transit time and CQ were frequently explained using the "inverse storage effect" [Harman, 2015;Yang et al. 2018]: during the wet season Q has a strong preference for young water associated with higher concentrations, which would not occur during dry periods due to the deactivation of the shallow fast flow processes. This effect was revealed in the computed TTDs for Q indicated by the shift between wet and dry seasons ( Figure S3 in the supporting information). 445 The response of the source contributions to topographic slope is threshold-like (Figure 8a): the source contributions in C1 were significantly higher than the ones in C3. Especially for the landscapes of C3, the fluctuation of CQ was hardly impacted by source variability. Mechanically, the seasonal source fluctuation is more likely to be damped by relatively longer transit times in C3 landscapes, which are relatively flat.
Given that the seasonal CQ variation can be attributed more to the variation in transit times (thus to the variation in the 450 YFQ), it was expected that the standard deviations of CQ and YFQ (Figure 8b, c) had similar responses to the topographic slope. Both of the responses exhibit a threshold-like pattern, similar to the response of the mean CQ (Figure 4a). This is because CQ during the dry season is generally low, regardless of whether the landscape is steeper or flatter. The overall response of σ(CQ) to topographic slope is determined by the response of the relatively high CQ during the wet season, and can be interpreted in the same three-class pattern: σ(CQ) increases with the decrease of slope within C1 455 (or C3), suggesting that flatter landscapes tend to export nitrate with more seasonal fluctuations in CQ for C1 (or C3).
However, for C2, a significant drop in this fluctuation can be expected when the landscape transforms from C1 to C3.
As a result, the maximum seasonal variation was reached at the slope of 1:60. For the mechanistic interpretation please refer to section 4.1.    Jasechko et al., [2016] reported that (the logarithm of) catchment topographic slope was significantly negatively correlated with young streamflow fractions with a spearman rank correlation of -0.36. This conclusion was made statistically based on their observed 254 sites. Our numerical study based on the eleven cross-sectional aquifers with different slopes but identical climate conditions resulted in more physically-based information that goes beyond such 470 statistical correlations. Our results show that young streamflow fraction and slope possess a threshold-like three-class https://doi.org/10.5194/hess-2021-375 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License. slope, therefore the magnitude of seasonal variations depended on how high the CQ rises during the wet seasons. This indicates that, for similar catchments in temperate humid climates, a high mean in-stream concentration level also means a high seasonal variation. The source contribution to seasonal variations is higher for C1 landscapes (> 0.2) 510 than for C3 landscapes (almost zero). This implies that changes in the nitrate source input due to, e.g., changing crop type, land use or fertilizer application amount, are more likely to cause a detectable short-term (e.g. seasonal) response of the in-stream concentration for mountainous catchments. For flat landscapes, this response would be weaker.

Limitations and outlook 515
The cross-comparison between cross-sectional aquifers with differing topographic slopes provides physically-based insights into the effects of slope on nitrate export responses in terms of mean concentration level and seasonal variations. However, this study is limited in scope in that it neglects other factors that may also have important impacts on the young streamflow and nitrate export processes: First, the modeled cross-sectional aquifers were unconfined with an impermeable base and prescribed heterogeneity. 520 Our model conclusions may be limited to the regional scale. Other catchment characteristics such as landscape aspect, catchment area, aquifer permeability or drainage ability, aquifer depth, stream bed elevation and fractured bedrock permeability can potentially change the flow patterns and age composition in streamflow [McGlynn et al., 2003;Broxton et al., 2009;Sayama and McDonnell, 2009;Stewart et al., 2010;Jasechko et al., 2016;Heidbüchel et al., 2013Heidbüchel et al., , 2020. For example, aquifers with high permeability or highly fractured bed rock are more likely to use deep 525 rather than shallow flow paths and preferential discharge routes that lead to rapid drainage. Apart from that, it was reported that hydrological features such as precipitation variability, ET, antecedent soil moisture are also significantly linked to transit times [Sprenger et al., 2016;Wilusz et al. 2017;Evaristo et al., 2019;Heidbüchel et al., 2013Heidbüchel et al., , 2020.
For example, compared to uniform precipitation, event-scale precipitation is more likely to trigger rapid surface runoff and intermediate flow, such that the contribution of young water from storage to streamflow can be increased. 530 Therefore, further research should consider a more complex model structure involving various heterogeneity and climate types.
Second, several main simplifications were used in the formulation of nitrate transport processes. (i) Transport modelling employed a constant degradation rate coefficient assuming that transit time was the only factor to determine degradation. This assumption neglected other factors that can spatially and temporally affect denitrification rates, such 535 as temperature, redox boundaries (e.g., high oxygen concentration in shallow flow paths), amount of other nutrients (e.g. carbon), which also contribute to the seasonality in nitrate concentrations [Böhlke et al., 2007]. Apart from that, we did not account for the long-term (decades [Van Meter et al., 2017]) nitrate legacy effect as the dissolved nitrate in groundwater reservoirs degraded continuously in our model, which would not occur in older reservoirs where the denitrification is very slow or deactivated (e.g. due to the lack of carbon source). (ii) In our simulations, the 540 complexities of the nitrogen pool were simplified by integrally defining a source concentration curve. The variability of the source input was implicitly considered by forcing the source concentration to vary along that curve over the course of a year. The accurate simulation of CQ would depend on a realistic estimation of the input source curve. However, it is not that important in our study as we were focused on understanding how the response changes with regard to topographic slope rather than on accurately reproducing CQ. (iii) The nitrate source was uniformly applied 545 across the land surface in our modelling. However, strong source heterogeneity may exist in catchments. For example, the source concentrations vary between land uses or along the soil profile [Zhi et al., 2019]. This spatial source heterogeneity could affect the seasonal variations of CQ Zhi et al., 2019] and should be considered in further research.
Despite these limitations, the numerical experiments in this study could clearly identify a three-class response of young 550 streamflow and nitrate export to topographic slope under a humid seasonal climate, and show that hydraulic gradient is an important factor causing the differences between the classes. This was achieved by using the advantages of a physically-based flow simulation that allows for a more mechanistic evaluation of flow processes, which would be impossible with a purely data driven analysis based on, e.g., isotopic tracers only.

Conclusions
Previous data driven studies suggested that catchment topographic slope impacts age composition of streamflow and consequently the in-stream concentrations [Jasechko et al., 2016]. We attempted to find more mechanistic explanations for these effects. We chose a cross-section from the small agricultural catchment 'Schäfertal' in Central Germany and generated eleven synthetic cross-sections of varying topographic slope. The groundwater and overland 560 flow, and the nitrate transport in these cross-sections were simulated using a coupled surface-subsurface model. Water age compositions for Q and ET were determined using numerical tracer experiments. Based on the calculated flow patterns, in-stream nitrate concentration CQ and young water fractions in streamflow YFQ, we systematically assessed the effects of varying catchment topographic slopes on the nitrate export dynamics in terms of the concentration level (annual mean) and its seasonal variability. The main conclusions of this study are: 565 • Under the considered humid climate, CQ is related to topographic slope by a three-class response. When the landscape becomes flatter, the hydraulic head gradient is the main driving force, changing the aquifer from a partially topography-limited system with preferential overland flow (C1) to a recharge-limited system that is more likely to form local flow cells (C3). For landscapes falling into the classes C1 or C3, flatter landscapes tend to generate more young streamflow and export nitrate of higher CQ. However, for the former this is due 570 to the "inverse storage effect" and for the latter this is due to the "local flow cells effect". For the transitional class C2, YFQ and nitrate concentration decrease sharply once the flatter landscapes are no longer able to maintain the fast preferential overland flow paths.
• For catchments in temperate humid climates with considerable seasonality in wetness conditions, the seasonal variation of CQ is dominated by the variability in transit times and in turn degradation, rather than by the 575 variability in the nitrate source. Especially for the aquifer of the C3 class, significant seasonal variation of CQ can be generated even without any variability in the nitrate source. https://doi.org/10.5194/hess-2021-375 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License.
• The response of the seasonal variation of CQ to topographic slope is similar to the one of the mean CQ. For the landscapes of the C1 or C3 classes, seasonal variation tends to be more pronounced for flatter landscapes.
However, for the C2 class, a significant decrease in this variation can be expected when fast preferential 580 overland flow paths are switched off on flatter landscapes.
Overall, this study provides a mechanistic perspective on how catchment topographic slope affects nitrate export patterns. The use of a fully-coupled flow and transport model extends the approach to investigate the effects of catchment characteristics beyond the frequently used tracer data-driven analysis. It can be used for similar studies of other catchment characteristics and for other solutes. The results of this study reveal potential implications for the 585 management of stream water quality and agricultural activity, in particular for catchments in temperate humid climates with pronounced seasonality. Given the limitations of this study, future work should be devoted to improve the degradation formulation, to investigate further catchment characteristics, as well as to consider various climate types.