Explaining the Variability in High‐Frequency Nitrate Export Patterns Using Long‐Term Hydrological Event Classification

Runoff events play an important role in nitrate export from catchments, but the variability of export patterns between events and catchments is high and the dominant drivers remain difficult to disentangle. Here, we rigorously asses if detailed knowledge on runoff event characteristics can help to explain this variability. To this end, we conducted a long‐term (1955–2018) event classification using hydro‐meteorological data, including rainfall characteristics, soil moisture and snowmelt, in six neighboring mesoscale catchments with contrasting land use. We related these event characteristics to nitrate export patterns from high‐frequency nitrate concentration monitoring (2013–2017) using concentration‐discharge (CQ) relationships. Our results show that low‐magnitude rainfall‐induced events with dry antecedent conditions exported lowest nitrate concentrations and loads but exhibited highly variable CQ relationships. We explain this by a low fraction of active flow paths, revealing the spatial heterogeneity of nitrate sources within the catchments and by an increased impact of biogeochemical retention processes. In contrast, high‐magnitude rainfall or snowmelt‐induced events exported highest nitrate concentrations and loads and converged to similar chemostatic export patterns across all catchments, without exhibiting source limitation. We explain these homogeneous export patterns by high catchment wetness that activated a high number of flow paths and by higher nitrate availability during high‐flow seasons. Long‐term hydro‐meteorological data indicated an increased number of events with dry antecedent conditions in summer and a decreased number of snow‐influenced events. These trends will likely continue and cause increased nitrate concentration variability during low‐flow seasons and changes in the timing of nitrate export peaks during high‐flow seasons.

It is most likely that the spatiotemporal variability in the hydrological land-to-stream connectivity causes different responses in nutrient mobilization and transport as well (Stieglitz et al., 2003). With the advent of high-frequency measurements for nitrate and other nutrient concentrations (Kirchner et al., 2004;Rode, Wade, et al., 2016), we can now measure water quality at the same temporal resolution as water quantity to analyze in detail the connection of runoff event characteristics with nitrate export patterns. More specifically, we refer to runoff event characteristics as all related hydro-meteorological characteristics including antecedent conditions, the characteristics of the inducing precipitation event and the characteristics of the runoff event hydrograph (for example peak discharge). This also includes the proposed different runoff formation processes (Tarasova et al., 2020) that can potentially connect different sources of nitrate to the stream network. For example, from available hydroclimatic data, we can now distinguish events that are induced by rain-on-snow under wet antecedent conditions and low-magnitude rain-induced events with dry antecedent conditions (Tarasova et al., 2020). Note, that for the sake of consistency in the use of terms with previous studies (i.e., Musolff et al., 2015Musolff et al., , 2021Tarasova et al., 2020), we use the terms "discharge" and "runoff" synonymously, referring to the total volumetric water flow rate in the stream at a gauging point.
Several studies took advantage of high-frequency measurements and conducted a detailed analysis on nutrient mobilization and transport during runoff events and generally confirmed the importance of runoff events for nutrient export (e.g., Blaen et al., 2017;Burns et al., 2019;Fovet et al., 2018;Knapp et al., 2020;Rose et al., 2018). For example, Casson et al. (2010) and Pellerin et al. (2012) showed that high-magnitude rainon-snow events can account for a disproportional amount of annually exported nitrate loads. These studies also revealed large inter-event and inter-catchment variability of nutrient export dynamics. For example, Blaen et al. (2017) found a positive correlation between antecedent wetness and event nitrate concentrations in a catchment with mixed agricultural and forested land use. On the contrary, Knapp et al. (2020) found a negative correlation between antecedent wetness and event nitrate concentrations in a small mountainous catchment that is covered by forest and meadows. While both studies could explain parts of their findings by changes in the hydrological catchment connectivity, their differences might mainly be caused by different nitrate source availabilities, induced by different catchment characteristics such as land use. Knapp et al. (2020) summarized that the variability of event responses was driven by changes in source availability, hydrological connectivity, and biogeochemical reaction rates. The role of temperature-and soil-moisture-driven differences in biogeochemical reaction rates for nitrate export was also stressed by Lutz et al. (2020) and Rode, Angelstein et al. (2016). Both, hydrological connectivity and biogeochemical removal (or retention) are driven by environmental conditions and thus often have a similar seasonal timing in temperate climates. In consequence, connectivity and removal can be difficult to disentangle. For example, rain-on-snow events with a high hydrological connectivity typically occur in colder periods of lower ecosystem nitrate uptake and hence a higher nitrate availability. All these examples show that nitrate export 10.1029/2021WR030938 3 of 21 patterns can be potentially related to hydrological event characteristics, such as the contribution of meltwater or antecedent conditions as well as to biogeochemically controlled source availability. The response of nitrate export to runoff events is obviously highly variable between catchments of different configurations for example, with regard to land use and nitrate availability. Therefore, we see a need for larger-scale studies that analyze the connection of event characteristics and nitrate export patterns across the entire annual cycle and in contrasting catchments.
A common tool to reveal the relevant sources and flow paths for nitrate transport under changing hydrological conditions are concentration-discharge (CQ) relationships, which represent the directional relationship between concentrations and discharge (e.g., Bieroza et al., 2018;Bowes et al., 2015;Musolff et al., 2021;Vaughan et al., 2017). A negative slope of the CQ relationship can indicate high base flow concentrations that are diluted by water from newly activated flow paths (Bowes et al., 2015) or a depletion of nutrient sources (Vaughan et al., 2017). A positive CQ slope can indicate the additional activation of more shallow and younger (Musolff et al., 2015) or more distant nutrient source zones (Bowes et al., 2015). A chemostatic pattern is instead described by a CQ slope close to zero (Godsey et al., 2009;Musolff et al., 2015;Thompson et al., 2011) and is mainly attributed to ubiquitous and uniformly distributed N sources in agricultural catchments (Basu et al., 2010). The CQ slope is not necessarily consistent across time scales and can thus reveal complementary information on nutrient export during single runoff events compared to CQ relationships across seasons that integrate several events (e.g., Godsey et al., 2019;Knapp et al., 2020;Minaudo et al., 2019;Musolff et al., 2021).
Yet, a rigorous assessment of how much of the inter-event variability of nutrient export patterns can be explained by a more thorough understanding of runoff event characteristics and classifications of runoff formation processes is still missing and has not yet been applied across a range of hydro-climatic conditions and land use settings (Knapp et al., 2020;Tarasova et al., 2020). Studies that relate hydrological runoff events with nutrient transport are typically limited to single catchments and/or to relatively short time periods, which is often not more than one or two years, frequently excluding the cold seasons (e.g., Blaen et al., 2017;Carey et al., 2014;Knapp et al., 2020). Therefore, the full range of runoff event characteristics is not always covered and it remains unclear if analyzed event characteristics are representative across catchments and for the long-term catchment behavior. Moreover, it is largely unknown if runoff event characteristics are changing over longer time scales. In this study, we conduct an extensive assessment across catchments and time scales to explore to what extent runoff event characteristics and runoff formation processes govern nitrate export during and across runoff events. To this end, we related event characteristics such as the wetness state of a catchment, the nature of an inducing precipitation event and the temporal distribution of rainfall to nitrate concentrations and loads. For this, we used a 5-yr period of high-frequency water quality and hydro-meteorological data from six mesoscale Central European catchments with different land use settings. We classified runoff events according to their different hydro-meteorological conditions (Tarasova et al., 2020) and utilized CQ relationships to infer the relevant flow paths and source areas for nitrate transport and mobilization. We then combined the findings from such analysis with the changes in event characteristics and catchment state conditions over past decades obtained from long-term daily hydro-meteorological time-series to identify possible trends in the long-term runoff event characteristics that could impact nitrate export dynamics in the future.

Study Area
Event characteristics and nitrate export patterns were analyzed in six sub-catchments of the Bode River catchment (Figure 1), which is an intensively monitored catchment within the network of the TERestrial Environmental Observatories (TERENO, Wollschläger et al., 2017). Warme Bode (WB), Rappbode (RB) and Hassel (HS) are part of the Rappbode Reservoir Observatory (Rinke et al., 2013), whereas Silberhütte (SH), Meisdorf (MD) and Hausneindorf (HD) are three subsequent gauging stations of the nested Selke River catchment. All six catchments are located in the Harz Mountains and the Harz foreland in Saxony-Anhalt, Germany ( Figure 1). They have contrasting characteristics in regard to their size, land use, elevation, and mean annual precipitation (Table 1).

Long-Term Daily Data
We used long-term daily hydro-meteorological data (Figures S1-S6 in Supporting Information S1) to classify runoff events and to analyze potential trends in the characterization of runoff events. Daily discharge data was provided by the State Office of Flood Protection and Water Management of Saxony-Anhalt (LHW) and calculated to specific discharge [mm d −1 ]. In all catchments except HS and HD, discharge data is available from 1955until 2018. In HS and HD, discharge data records started in 1968and 1980, respectively, and lasted until 2018 Daily precipitation data over these time periods were provided by Germany's National Meteorological Service (Deutscher Wetterdienst, DWD) as interpolated station data at a spatial resolution of 1 km 2 (REGNIE; Rauthe et al., 2013). Daily average temperatures were interpolated to a 4 km grid from the DWD stations by External Drift Kriging using elevation as an explanatory variable (Zink et al., 2017). Daily soil moisture and snow water equivalent were calculated using the mesoscale Hydrological Model (mHM, Kumar et al., 2013;Samaniego et al., 2010;Zink et al., 2017).

High-Frequency Hourly Data
High-frequency hourly data were used to analyze exported nitrate loads and CQ relationships within and between runoff events from 2013 to 2017. Discharge data at a temporal resolution of 15 min were provided by the LHW, which we aggregated to hourly values [mm h −1 ]. Similar to nitrate concentration data (see below), a moving average was applied over a 5-hr window to reduce noise in the raw data and to stay consistent with the procedure applied in previous studies that used parts of the same data (e.g., Musolff et al., 2021;Rode, Angelstein, et al., 2016). Hourly precipitation data as reprocessed radar data were provided by the DWD with precipitation amounts adjusted to station observations and a spatial resolution of 1 km 2 (RADOLAN; Winterrath et al., 2017). Due to a lack of hourly temperature data, we reconstructed those from the daily data by using hourly weights based on month-specific sine functions obtained from long-term minimum and maximum temperatures to resemble the diurnal cycle of temperature. Hourly snow water equivalent and soil moisture data were simulated using mHM (Zink et al., 2017).
Nitrate concentration data were collected between 2013 and 2017 via TRIOS ProPS-UV sensors at 15 min intervals (Kong et al., 2019;Rode, Angelstein, et al., 2016), which we aggregated to hourly averages. Data from the WB catchment were previously published by Kong et al. (2019) and Musolff et al. (2021), data from the three Selke catchments (SH, MD, and HD) were previously published by Musolff et al. (2021), Rode, Angelstein, et al. (2016), Winter et al. (2021), and X. Yang, Jomaa, et al. (2018). For the processing of the nitrate concentration data, we refer to the references above and to our Supporting Information (Text S1 in Supporting Information S1). Briefly, raw data was restricted to a realistic range (0-100 mg N L −1 ), outliers were removed (Grubbs, 1950), a moving average was applied over a 5-hr window to smooth the data and concentrations were calibrated against grab samples analyzed in the lab (R 2 0.80-0.91, Figures S7 and S8 in Supporting Information S1). Note that the LHW gauging station at HS, where long-term and high-frequency discharge data were measured, is located upstream of the measurement point for concentration data, thus delineating a catchment size of around 29 km 2 for discharge data compared to 42 km 2 for measured concentration data ( Figure 1, Table 1). Nevertheless, area-specific discharge data [mm h −1 ] from upstream and downstream measurement points (available downstream between 2013 and 2014) showed a good agreement in their temporal dynamics with a R 2 of 0.88 and in absolute values with a small percentage bias of −3.0% ( Figure S9 in Supporting Information S1). We thus found area-specific discharge from the upstream station data to be suitable for further analysis at the downstream station.

Runoff Event Identification and Classification
Runoff events were separated and classified using the recently developed approach by Tarasova et al. (2018Tarasova et al. ( , 2020, which allows for an automated separation and classification of runoff events. The approach is explained in detail in the cited studies and is, therefore, only briefly described here. As a first step, events from daily long-term and high-frequency data were identified using an automated event separation approach from Tarasova et al. (2018). Then, we adopted the event classification framework from Tarasova et al. (2020), developed for daily data resolution ( Figure 2). Each runoff event was classified by the characteristics of the inducing precipitation event ( Figure 2a, Layer 1) and the pre-event wetness state of the catchment (Figure 2a, Layer 2). The nature of precipitation events was identified by the ratio of meltwater volume (M vol ) and total precipitation volume (i.e., sum of rainfall and snowmelt, P vol ). Using a threshold of M vol /P vol = 0.05 (Figure 2a), events were classified as Rain or Snow-influenced events. The temporal distribution of precipitation was characterized by means of the temporal coefficient of variation of the precipitation rate (CV temp ) and by the ratio between the maximum precipitation rate during an event and precipitation volume (P max /P vol ). Events with a CV temp > 1 and P max /P vol > 0.5 were defined as intensity-dominated and all other events as volume-dominated ( Figure 2a). Third, the wetness state of a catchment was characterized by means of antecedent soil moisture (SM ant ). Using a threshold of maximum κ, with κ representing the catchment-specific curvature of the nonlinear relationship between event runoff coefficients and soil moisture (Tarasova et al., 2020), events were classified as Wet or Dry events ( Figure 2a). In total, this classification resulted in five event classes ( Figure 2b): (a) snow-influenced events (Snow), (b) rain-induced events that were volume-dominated and occurred under wet antecedent soil moisture conditions (Rain-Wet-Vol), (c) rain-induced events that occurred under wet antecedent conditions and were intensity dominated (Rain-Wet-Int), (d) rain-induced events that occurred under dry antecedent conditions and were volume-dominated (Rain-Dry-Vol) and (e) rain-induced events that occurred under dry antecedent conditions and were intensity dominated (Rain-Dry-Int). To assure comparability of event classes between two data sets of different resolutions, we classified events using the daily time series  and then assigned the respective classes to the corresponding events from the hourly time series (2013-2017).

Long-Term Trends in Event Characteristics
We used the non-parametric Mann-Kendall test (Kendall, 1998;Mann, 1945) to detect monotonic trends in the continuous event characteristics and event classes with a significance level of 5%. We considered the following continuous event characteristics: (a) antecedent soil moisture (SM ant ), (b) the ratio of meltwater volume and precipitation volume (M vol /P vol ), and (c) the ratio of maximum precipitation and precipitation volume (P max /P vol ), which is an indicator for intensity-or volume-dominated events, respectively ( Figure 2a). To reduce inter-event variability between those characteristics, we calculated seasonal averages for each year and analyzed these for seasonal long-term trends. Similarly, we analyzed seasonal trends in the annual contribution and total number of (a) Snow events (vs. Rain events), (b) Rain-Dry (vs. Rain-Wet) events, and (c) Intensity-dominated (vs. Volume-dominated) events.

Descriptors of Nitrate Export
To characterize nitrate transport from the high-frequency data, we chose four descriptors for the event scale: (a) median nitrate concentration (C med in mg N L −1 ), (b) average loads per event in kg N ha −1 yr −1 (this unit was chosen for a better comparison between catchments and events of different duration), (c) inter-event CQ slopes and (d) event-specific CQ slopes. Event-specific CQ slopes were assessed by fitting the parameter b from the following power-law relationship after Godsey et al. (2009) and Musolff et al. (2015) to the data of the individual events: where C(t) represents the time series of nitrate concentrations during a specific event in mg N L −1 , Q(t) represents the time series of discharge in mm h −1 , and a and b represent the intercept and linear slope of the CQ relationship in the log-log space. A parameter b < 0 describes a negative CQ slope, that is, decreasing concentrations with increasing discharge and therefore a dilution pattern. A parameter b > 0 describes a positive CQ slope, that is, increasing concentrations with increasing discharge and therefore an accretion pattern. Both scenarios are accounted for as chemodynamic patterns (Godsey et al., 2009;Musolff et al., 2015Musolff et al., , 2017. If parameter b is close to zero, there is no clear directional relationship. This pattern can be described as chemostatic under the assumption that the coefficient of variation of concentrations is much smaller than that of discharge (Godsey et al., 2009;Musolff et al., 2015Musolff et al., , 2017. Similar to the event-specific CQ slope, we analyzed the CQ relationship across all events within each catchment (i.e., the inter-event CQ slope) using the power law model from Equation 1 with C med and Q med of each event instead of C(t) and Q(t) within each specific event.

Statistical Analysis
All computations and statistical analyses were conducted in R (R Core Team, 2020). We used the non-parametric Kruskal-Wallis test (Kruskal & Wallis, 1952) to test for differences in loads, C med , Q med and the CQ slope between event classes and the Pairwise Wilcoxon Test (Wilcoxon, 1945) with Holms correction for multiple comparisons (Holm, 1979) to test for differences in-between the event classes, both at the significance level of 5%. In order to test the impact of event classes on the inter-event CQ slope, we tested the simple linear ln(C med )-ln(Q med ) regression against a linear regression model that includes event classes and their interactions with ln(Q med ). Both models were compared via the sample-size corrected Akaike Information Criterion (AICc; Akaike, 1973;Hurvich & Tsai, 1989;Sugiura, 1978). If accounting for event classes led to a substantial improvement (i.e., AICc decreased at least by 2, similar to Marinos et al., 2020) their impact was regarded as considerable. Otherwise, the added value from event classes compared to a simple CQ model was negligible for nitrate export estimations.

Long-Term and High-Frequency Runoff Event Characteristics
In total, we identified and classified 5,872 events over the long-term period (on average 14.5-19.0 events per catchment and year) and 388 events over the high-frequency time period (on average 9.6-16.2 events per catchment and year). Event classes generally differed more strongly between seasons than between catchments (Figure 3). In both long-term and high-frequency event classes, winter (December-February) was dominated by Snow and Rain-Wet-Vol events. Spring (March-May) showed the most even distribution of event classes and was the season with the highest percentage of Rain-Wet-Vol and Rain-Wet-Int events. Summer (June-August) and autumn (September-November) were dominated by Rain-Dry-Vol events and Rain-Dry-Int events. Differences between catchments reflect a decreasing percentage of Snow events and an increasing percentage of Rain-Dry-Vol and Rain-Dry-Int events from west to east (WB to HD catchment; Figure 3). Across all seasons of the long-term event classes, more than half of all events in the six catchments were classified as Rain-Dry events (from 51.2% in WB to 61.9% in HD). Around a fifth up to a quarter of all observed events were classified as Snow events (from 17.7% in MD to 25.0% in WB), and the proportion of Rain-Wet events ranged between 14.0% in HS and 23.9% in WB. Rain-induced events were more frequently volume-than intensity-dominated.
Runoff events had an average duration of 11.2 ± 9.1 days for long-term data and 11.2 ± 8.9 days during the period of high-frequency data, with considerable differences between event classes ( Figure S10 in Supporting Information S1). Intensity-dominated events (Rain-Dry-Int and Rain-Wet-Int) were shortest, lasting in average 5.5 ± 4.7 days and 8.6 ± 6.5 days for long-term data, respectively, followed by Rain-Dry-Vol events that lasted in average, 10.0 ± 8.1 days. Rain-Wet-Vol and Snow events were the longest events, lasting in average 14.4 ± 8.8 and 18.2 ± 10.2 days.

Long-Term Trends and Changes in Event Characteristics
In agreement with increasing temperature due to climate change (IPCC, 2013), Mann Kendall trends analysis indicated a decrease in the number and proportion of Rain-Wet events in summer, which was significant in half of the catchments (WB, SH, and MD; Table S1 in Supporting Information S1). This decrease goes along with a significant decrease in antecedent soil moisture in spring and/or summer in WB, RB, and MD catchments. Moreover, the number and/or proportion of Snow events decreased significantly in spring in WB, RB, SH, and MD catchments (Table S1 in Supporting Information S1). These trends go along with a significant decrease in the proportion of meltwater volume per event (M vol /P vol ) in winter and spring in all catchments except HS. Only one catchment (WB) showed a significant decrease in the number and proportion of intensity-vs. volume-dominated events, which occurred during summer. In contrast, the RB, SH, and MD catchments showed a significant increase in P max /P vol during summer, but no significant trend in the total number or proportion of intensity-or volume-dominated events (Table S1 in Supporting Information S1).

Nitrate Export During Runoff Events
Runoff event and nitrate export characteristics differed between catchments. Event runoff decreased roughly from west to east, along the precipitation gradient (Table 1) with highest average Q med in the WB catchment (1.1 mm hr −1 ) and lowest average Q med in the HD catchment (0.3 mm hr −1 ). Nitrate export during runoff events varied across catchments in line with their land use patterns (Table 1). Catchments with the highest percentage of agricultural land use exported in average highest C med (HS and HD with 2.5 mg N L −1 and 2.3 mg N L −1 ), followed by mixed land use catchments (MD and SH with 1.6 mg N L −1 and 1.5 mg N L −1 ) and lowest average C med was observed in the dominantly forested catchments RB and WB (0.6 mg N L −1 and 0.7 mg N L −1 ).

Nitrate Loads
Runoff events had a prominent role for annual nitrate export. The cumulative duration of all identified events from the high-frequency data was on average 39.6% (30.8%-48.1% depending on the catchment) of the analyzed time period (2013-2017), while they accounted on average for 51.2% (44.8%-63.3%) of all nitrate loads (Text S2 in Supporting Information S1). In relation to catchment area (Table 1), the HS catchment transported highest median nitrate loads across all event classes (5.5 kg N ha −1 yr −1 ) followed by HD (1.8 kg N ha −1 yr −1 ), MD (1.7 kg N ha −1 yr −1 ), WB (1.6 kg N ha −1 yr −1 ), SH (1.4 kg N ha −1 yr −1 ) and lowest median loads were exported from RB catchment (1.0 kg N ha −1 yr −1 ). Between event classes, lowest loads were transported during Rain-Dry-Int and Rain-Dry-Vol events, which were responsible for around 25.6% (14.8% in MD up to 41.6% in HD) of all event-driven loads. Highest loads were transported during Rain-Wet and Snow events, which were responsible for around 74.4% (58.4%-85.2%) of all event-driven loads (Figure 4). Kruskal Wallis test showed significant differences in exported nitrate loads between the event classes in all catchments. Results of the pairwise Wilcoxon Test indicated that these differences are mainly driven by the differences between Rain-Dry-Int or Rain-Dry-Vol events and Rain-Wet-Vol or Snow events, whereas no significant difference between Rain-Dry-Vol and Rain-Dry-Int events, nor between Rain-Wet-Vol and Snow events were detected. Rain-Wet-Int events were generally too low in their frequency (n = 1-7) to be compared reliably (Figure 4).

Inter-Event Concentration-Discharge Relationships
The inter-event CQ relationship is characterized by the slope between ln(C med ) and ln(Q med ) of all events within one catchment. It shows consistently positive CQ relationships in the log-log space, indicating that C med increased with Q med but with a different slope depending on the catchment ( Figure 5). In line with transported loads (Figure 4), Rain-Dry-Vol and Rain-Dry-Int events are mainly located in the lower part of the CQ relationship, representing low-magnitude events (low Q med ) with low concentrations (low C med ) that occur mainly during summer and autumn (Figure 3). Rain-Wet-Vol and Snow events that occurred mainly in winter and spring (Figure 3) are located on the upper part of the CQ relationship, showing the highest Q med and C med ( Figure 5). Additionally, some Rain-Dry-Vol events are located at the upper end of the CQ relationship. These events occurred mainly during autumn and often extended into the winter period with higher Q med and C med . Rain-Wet-Int events occurred only occasionally and represent mainly events of a lower magnitude in winter and spring with medium C med and Q med , plotting in between the other event classes.
The inter-event CQ relationship could account for most of the variance in C med with an R 2 varying between 0.51 and 0.91 ( Figure 5). Except for the SH catchment, adding information on event classes to the model did not improve its performance in terms of AICc compared to a simple CQ model. This indicates that Q med was the most powerful predictor of C med and no or only a small part of additional variance was explained by the event classes themselves. In the SH catchment, event classes clearly improved the linear regression model (AICc decreased by 13.6 units). While no clear differences between Rain-Dry-Vol and Rain-Dry-Int events nor between Rain-Wet-Vol and Snow events are visible ( Figure S11 in Supporting Information S1), the main event class differences in SH was a higher intercept of Rain-Wet-Vol and Snow events compared to Rain-Dry-Int and Rain-Dry-Vol events.

Event-Specific Concentration-Discharge Relationships
Across all catchments, most events (72.4%) were characterized by a positive event-specific CQ slope, indicating an increase of nitrate concentrations with increasing discharge (Figure 6). We found that event-specific CQ slopes in all catchments showed a large variability between low-magnitude events (low Q med ), whereas CQ slopes for high-magnitude events (higher Q med ) collapse to a slightly positive CQ slope that is roughly between 0.1 and 0.3, close to a chemostatic pattern (Figure 6a). Some catchment-specific differences can be observed between CQ slopes during low-magnitude events (Figure 6b). The more forested and pristine catchments dominantly showed positive CQ slopes, whereas the agriculturally dominated catchments HS and HD tended toward close-to-zero or negative CQ slopes.
Rain-Dry-Int and Rain-Dry-Vol events cause most of the variability between event-specific CQ slopes ( Figure 6). These two event classes are distinguished by the temporal distribution of the inducing rainfall, being either intensity-or volume-dominated. To assess whether the difference in the temporal distribution of rainfall explains any additional variability in event-specific CQ slopes of low-magnitude events, we compared both classes using the Kruskal-Wallis test (Kruskal & Wallis, 1952). We found significantly higher event-specific CQ slopes for Rain-Dry-Int events compared to Rain-Dry-Vol events in half of the catchments (Figure 7; WB, SH, and MD), all of them showing >60% forest cover (Table 1). Median event-specific CQ slopes for Rain-Dry-Vol events were 0.18, 0.32, and 0.05 in WB, SH, and MD, respectively, and 0.39, 0.51, and 0.35 for Rain-Dry-Int events. In contrast, no Figure 6. Event-specific CQ slopes (slope of nitrate concentrations and discharge in the log-log space) against the specific median discharge (Q med ) of each event shown for all catchments in one plot (a) and with logarithmic x-axis and separated by catchments to visualize differences in events with low Q med between catchments (b). Colors of dots indicate the five event classes, dot sizes indicate the event load. Gray-shaded areas indicate event-specific CQ slopes close to zero (between −0.1 and 0.1).
significant difference in C med and Q med between Rain-Dry-Int and Rain-Dry-Vol events could be detected ( Figure  S12 in Supporting Information S1), except for the SH catchment, where C med and Q med were significantly higher during Rain-Dry-Vol events.

Impact of Runoff Event Characteristics on Nitrate Export
The aim of this study was to thoroughly assess how much of the inter-event variability in nitrate export patterns can be explained by event characteristics and runoff formation classes across a range of hydro-meteorological conditions and different types of catchments. We argue that the extensive data set used in combination with a systematic runoff event classification and a nitrate export characterization using CQ relationships leads to a broader transferability of the identified, dominant drivers of event-scale nitrate export patterns beyond the catchments analyzed in this study.
Our results show that the average level of exported nitrate concentrations and loads during an event are catchment specific, depending on the land use and related N input. Higher export was observed in catchments with more agricultural land use, which have a higher N input through fertilizer application, whereas lower export was found in the more forested catchments, were N input, stemming from atmospheric deposition and biological fixation, is typically lower (Ebeling et al., 2021). However, we could also show very similar patterns in event-scale nitrate export across all catchments that were strongly dependent on the event magnitude (in regard to runoff) and season ( Figure 8). Low-magnitude events during summer and autumn exported lowest concentrations and loads and high-magnitude events in winter and spring exported highest nitrate concentrations and loads (Figure 8a). The variability of event-specific CQ slopes decreased with increasing event magnitude, indicating an increasing homogenization of the dominant drivers for nutrient export for increasing event magnitudes (Figure 8b).
In the following, we discuss the impact of runoff event characteristics on nitrate export during those different conditions in more detail. Furthermore, we discuss the transferability of results to other areas and how well the analyzed runoff event characteristics from the high-frequency period (2013-2017) represent the long-term runoff event characteristics. Moreover, we discuss the observed trends in long-term runoff event characteristics and their implications for future nitrate export.

Low-Magnitude Events
Low-magnitude and often relatively short rain-induced events occurred mainly during summer and autumn and coincided with dry antecedent soil moisture conditions, classified as Rain-Dry-Int or Rain-Dry-Vol. These events . Event-specific CQ slopes (slope between nitrate concentrations and discharge in the log-log space) for all six study catchments and the two event classes Rain-Dry-Int and Rain-Dry-Vol, representing rain-induced runoff events that occurred under dry antecedent soil moisture conditions with either intensity-or volumedominated rainfall. Orange asterisks in the header indicate significant (p < 0.05) differences between the event classes for a particular catchment. Gray-shaded areas indicate event-specific CQ slopes close to zero (between −0.1 and 0.1).
exported lowest nitrate concentrations and loads and showed highly variable event-specific CQ slopes. We can explain these relatively low nitrate loadings by a decreased hydrological connectivity (i.e., a low fraction of activated flow paths) with lower antecedent soil moisture and by a lower nitrate availability due to higher biogeochemical removal and biological uptake in summer and autumn, compared to winter and spring. As a result from decreased hydrological connectivity, only nitrate sources in close proximity to the stream network and from sources connected via deeper groundwater flow paths are connected to the stream network (Musolff et al., 2015;Stieglitz et al., 2003;J. Yang, Heidbüchel, et al., 2018). These flow paths during the dry period are generally characterized by longer transit times and thus enable more nitrate uptake and removal via denitrification (Ebeling et al., 2021;Ehrhardt et al., 2019;Kumar et al., 2020;Nguyen et al., 2021). As a result from increased biological activity with higher temperatures, nitrate uptake and removal increases, especially in streams and in the riparian zones (Baird et al., 1995;Lutz et al., 2020;Rode, Angelstein et al., 2016), which can lead to reduced nitrate availability compared to colder seasons. While certainly both nitrate availability and hydrological connectivity Figure 8. Framework of the relationship between runoff event classification and nitrate export characteristics. (a) The inter-event CQ relationship with median event-scale nitrate concentrations and discharge (C med and Q med ). Gray lines show the inter-event CQ slope for the individual catchments in this study. Catchments with higher N input (i.e., more agricultural land use) show a higher intercept than catchments with lower N input (i.e., more forest). (b) The event-specific CQ slopes for all catchments and their decreasing variability with higher event magnitude (Q med ). Event-specific CQ slopes for low-magnitude events tended to be more positive for catchments with low N input and more negative for catchments with high N input. (c) Concept of potential underlying mechanisms in regard to catchment wetness enhancing hydrological connectivity and biogeochemically controlled nitrate source availability in catchments of overall low vs. high N input (i.e., forest vs. agricultural land use).
play a role for event-scale nitrate export, it is beyond the scope of this study to fully disentangle their individual contributions.
Seasonal differences in discharge and concentrations shaped the inter-event CQ relationship that was positive across catchments, reflecting higher concentrations during high flow in winter and spring and lower concentrations during low-flow conditions in summer and autumn. At the event level, CQ slopes can reveal complementary information on the underlying processes of nitrate mobilization and transport within seasons (Godsey et al., 2019;Musolff et al., 2021;Winter et al., 2021). We found large variability in event-specific CQ slopes during low-magnitude events, which we explain by an increased relevance of different environmental factors such as (a) catchment characteristics and the spatial distribution and connectivity of N sources within a catchment, (b) riparian and in-stream biogeochemical processes and (c) the spatial and temporal distribution of the inducing precipitation event. In regard to catchment characteristics, we could show that the more forest-dominated catchments (WB, RB, SH, MD, Figure 1) showed mainly positive event-specific CQ slopes during low-magnitude events, whereas the more agriculturally dominated catchments (HS and HD) tended towards negative event-specific CQ slopes (Figure 6b). The dilution patterns in the agricultural catchments can be explained by relatively high base-flow concentrations (reflecting high N input from fertilization) and the spatial distribution of N sources within these catchments. For example, Musolff et al. (2021) argued that due to large buffer strips (100 m) in the HS catchment, there are no or only few nitrate sources in the riparian zones. Hence, events of low magnitude that activate only proximate flow paths from this area could cause the observed dilution pattern. In the HD catchment, Winter et al. (2021) found that a disproportionally large part of event runoff is generated in the upstream area that is mainly covered by forests and thus exports lower nitrate concentrations. Runoff from this area can thus dilute higher concentrations in base flow, which are largely generated by groundwater from the downstream agricultural areas. Hence, the preferential mobilization from certain areas of lower N availability, here riparian zones or upstream areas, can cause a dilution pattern in catchments with an overall high N input. In contrast, the dominant accretion pattern in the more forested catchments might be explained by a flushing of proximate shallow nitrate sources, likely from the upper soil layers of the riparian zones as also suggested by Musolff et al. (2021) for the WB catchment and a sub-catchment of RB. In regard to in-stream and near-stream processing, several studies argued that biogeochemical processes such as nitrate uptake and denitrification in-stream or in the riparian zones have a stronger relative impact on nitrate export during low-magnitude events (e.g., Marinos et al., 2020;Moatar et al., 2017). Hence, variability in these processes through, for example, varying instream temperature (Rode, Angelstein, et al., 2016) or in the riparian zone partly due to stream water infiltration (Lutz et al., 2020;Nogueira et al., 2021) might be responsible for the observed higher variability between event-specific CQ slopes. This is supported by a study from Heathwaite and Bieroza (2021), who found that nutrient export dynamics during low-magnitude events can be considerably influenced by diurnal cycling.
By separating runoff event classes into intensity-and volume-dominated precipitation, we could show that the impact of the temporal distribution of precipitation can explain another part of the variability in mobilization patterns during low-magnitude events. Intensity-dominated events (Rain-Dry-Int) showed higher event-specific CQ slopes compared to volume-dominated events (Rain-Dry-Vol) in half of the catchments. Those catchments comprise forested or mixed land use and showed overall positive event-specific CQ slopes for both Rain-Dry-Int and Rain-Dry-Vol events (Figures 1 and 7). Both event classes are rain-induced with dry antecedent conditions. During Rain-Dry-Int events however, runoff is generated by a shorter and rather intense rainfall, whereas during Rain-Dry-Vol events, the duration of rainfall is typically longer with a lower ratio of the maximum precipitation rate compared to the total precipitation volume (Tarasova et al., 2020). As argued further above, nitrate mobilized during low-magnitude events in those forested catchments may mainly stem from shallow and proximate N sources . One possible explanation for the difference in event-specific CQ slopes might be that relatively short but intensive runoff events preferentially activate proximate and shallow flow paths and mobilize those shallow N sources. This mobilization then causes an increase in nitrate concentrations that is reflected by the positive event-specific CQ slope. Longer volume-dominated events might create a higher, yet delayed hydrological connectivity with more distant sources than those near-stream N source zones, which is reflected in a decreasing event-specific CQ slope. As such, CQ slopes during volume-dominated events approximate more chemostatic patterns and show a higher similarity with higher-magnitude runoff events under wet antecedent conditions (see Section 4.1.2).

High-Magnitude Runoff Events
In contrast to low-magnitude events in summer and autumn, runoff events in winter and spring were mainly snow-influenced (Snow) or rain-induced (<5% snowmelt) and generated by volume-dominated precipitation under wet antecedent conditions (Rain-Wet-Vol, Figures 2 and 3). These two event classes were found to be the largest runoff events in regard to median discharge (Q med ) and caused the highest nitrate concentrations and loads (Figures 4 and 5). Approximately three quarters of all event-driven loads were exported during Snow and Rain-Wet-Vol events, which is in agreement with other studies that reported exceptionally high nitrate export during large rain-on-snow events (Crossman et al., 2016;Koenig et al., 2017;Sebestyen et al., 2009;Seybold et al., 2019). These results underline the important role of Snow and Rain-Wet-Vol event classes for nitrate export and show that missing information on the winter period, which is often the case (e.g., Blaen et al., 2017;Carey et al., 2014;Knapp et al., 2020;Wollheim et al., 2017), can lead to a lack of information about the most relevant events for the export of nitrate loads.
In temperate climates, rain-on-snow events often form the largest runoff events of the year due to the cumulative effect of rainfall and additional input from snowmelt (Casson et al., 2014;Pellerin et al., 2012). Nevertheless, we identified Rain-Wet-Vol events (not influenced by snowmelt) that caused comparable or even higher Q med , especially in the Selke catchment (Figures 5d-5f). Those events transported comparably high nitrate loads (Figure 4), fell on the same or a very similar inter-event CQ slope ( Figure 5) and showed similar event-specific CQ slopes ( Figure 6) as the Snow events. This indicates that both event classes, Snow and Rain-Wet-Vol, activate the same or very similar N sources within a catchment, despite their differences in the meltwater fraction.
Similar to Stieglitz et al. (2003), we argue that high-magnitude events during winter and spring can activate all relevant nitrate sources within a catchment, including distant sources (Bowes et al., 2015) and shallow and younger N sources (Fovet et al., 2018;Musolff et al., 2017Musolff et al., , 2015J. Yang, Heidbüchel, et al., 2018). During winter and spring, discharge and antecedent soil moisture are generally higher, which leads to more active flow paths compared to summer and autumn (Stieglitz et al., 2003;J. Yang, Heidbüchel, et al., 2018). At the same time, lower temperatures cause a reduced N demand of ecosystems that can result in a higher nitrate source availability (Baird et al., 1995;Rode, Angelstein, et al., 2016). Together, the flow path activation in a highly saturated catchment and a relatively high nitrate availability can explain the high nitrate concentrations and loads observed in the studied catchments (Figures 4 and 5). Moreover, they can explain the low variability in event specific-CQ slopes (Figure 6), because if all flow paths are activated and sufficient nitrate sources are available, no changes in nitrate mobilization through bypassing, activation of additional N sources or source depletion can occur.
Remarkably, the event-specific CQ slopes during high-magnitude events did not show any signs of dilution (Figure 6). Other studies have reported such dilution pattern during precipitation events across the whole year including high-magnitude events, which might indicate source depletion (Kincaid et al., 2020;Vaughan et al., 2017) or high base flow concentrations from deeper groundwater that are diluted by water with lower concentration from newly activated zones (Fovet et al., 2018;Rose et al., 2018). Here, we consistently reported slightly positive CQ slopes (roughly 0.1-0.3) that reflect a milder increase of concentrations compared to that of discharge, indicating increasingly chemostatic export patterns with increasing event runoff. This is further supported by the fact that the event-specific coefficient of variation of concentrations is much smaller than that of discharge (Musolff et al., 2015) with a median ratio of 0.28 for high-magnitude events (Q med > 1 mm hr −1 ). These patterns provide strong evidence for a transport rather than a source limitation of nitrate in all six catchments (Basu et al., 2010), even in the forest dominated catchments, which is alarming in terms of water quality. In the agricultural catchments (HS and HD), fertilization is likely the main nitrate source. In the mixed land use catchments (SH and MD) and the more forested catchments (WB and RB) smaller patches of agriculture and the overall high atmospheric N deposition in the Harz Mountains (Kuhr et al., 2014;Winter et al., 2021) are likely to be the dominant nitrate sources. In addition, long transit times and chemostatic export patterns in the agricultural lowland catchment HD indicate substantial N legacies belowground, which might keep nitrate concentrations at a high level independent from the event size (Winter et al., 2021). On the contrary, shorter transit times in the upper Harz Mountain catchments likely prevent such accumulation of long-term legacies (Ehrhardt et al., 2019;Nguyen et al., 2021;Winter et al., 2021;J. Yang, Heidbüchel, et al., 2018). Still, our results suggest that even without such long-term legacies, younger nitrate sources that get connected to the stream network during high-magnitude events provide sufficient supply to maintain chemostatic export patterns. Additionally, Ohte et al. (2004) and Sebestyen et al. (2009) showed that atmospheric N stored in the snowpack can considerably contribute to nitrate export during snow-influenced events in a forested catchment. However, the strikingly similar nitrate export patterns during Snow and Rain-Wet-Vol events with comparable event size hint at similar nitrate sources for both event classes and thus not at the melting snowpack as a key source.

Long-Term Trends of Event Characteristics and Their Implications for Nutrient Export Patterns
We analyzed nitrate export patterns for a 5-yr period of high-frequency nitrate concentration data (2013-2017), which is not sufficient to estimate any long-term trends. However, in this study, we bring together a short-term high-frequency analysis with long-term runoff event characterization and classification from daily data, which allowed us to detect long-term trends in runoff event characteristics and to discuss their possible impact on nitrate export patterns.
We found a decrease of soil moisture in summer, which aligned along a decrease of wet compared to dry events. This is in agreement with increasing summer temperatures over Europe (Briffa et al., 2009;IPCC, 2013) and other studies that report a decreasing contribution of summer precipitation (Szwed, 2019) and an increased risk for summer droughts in large parts of Europe (Hari et al., 2020;Pal et al., 2004). Here, we found that runoff events generated during dry catchment conditions are associated with a lower event magnitude (i.e., lower event runoff) proportionally lower nitrate concentrations and loads and a higher variability in event-specific CQ slopes, compared to wet conditions. Therefore, possibly drier antecedent conditions resulting from increasing future temperatures (IPCC, 2018;Pal et al., 2004) might lead to a decrease in nitrate export in summer periods but also to a higher variability in concentrations, due to more variable and partly higher event-specific CQ slopes. However, also nitrate availability is likely to be affected by changing climatic conditions with nitrate uptake and removal rates, but also mineralization, either increasing with increasing temperatures or decreasing because of soils drying out (Hartmann et al., 2013;Mosley, 2015). N that is not exported nor taken up during dry seasons is accumulated in the catchment and can be mineralized and flushed with rewetting in autumn (Mosley, 2015). First runoff events after especially dry summer periods were often reported to cause disproportionally high nitrate export peaks, which can cause severe water quality deteriorations and further increase the inter-annual variability of nitrate concentrations (Jarvie et al., 2003;Morecroft et al., 2000;Mosley, 2015;Oborne et al., 1980). In summary, we see evidence for an increased variability of nitrate concentrations and export dynamics with increasingly dry conditions in summer and autumn.
In addition to the increasingly dry summer conditions, we found a decrease in the contribution and number of snow-influenced events (Snow) as well as a decrease in the proportion of meltwater during winter and spring. These events exported highest nitrate loads; hence from the perspective of hydrological transport, a decrease of high nitrate export peaks could be expected, which was also reported by Sebestyen et al. (2009) for a mountainous forested catchment. However, winter precipitation is predicted to substantially increase in most of Europe (Stahl et al., 2010). The resulting larger rain-induced events could potentially counterbalance the decreased number of Snow events and trigger similarly high event runoff and nitrate export, as observed in the SH, MD, and HD catchments (Figures 5d-5f). As such, the timing of nitrate export peaks would not be restricted to the melting period but to the entire high flow season in winter and early spring, given sufficient nitrate supply. Additionally, several studies predict that an earlier start of snowmelt due to increasing temperatures causes a time shift of discharge and nitrate export peaks towards earlier in the year (Clow, 2010;IPCC, 2014;Sebestyen et al., 2009). In summary, we do not see clear evidence for a change in nitrate loading during high-magnitude winter and spring events but we do see evidence for a change in the timing of nitrate export peaks.

How Representative Are the Obtained Results for These and Other Catchments?
The classification of runoff events from long-term time series in this study allowed for a consistent characterization of typical hydro-meteorological and catchment-state conditions, their seasonality, and temporal changes (i.e., trends) in their configuration beyond the limited time period of available high-frequency nitrate measurements. To our knowledge, this placement of short-term nitrate export dynamics into a larger context of long-term runoff event characteristics has never been conducted before. Runoff event classes from the shorter and more recent high-frequency period (with available nitrate concentration data) deviated from the long-term average runoff event classes mainly in their proportion of Rain-Dry events (which mainly increased in summer) and in their proportion of snow-influenced events (which mainly decreased in spring). These deviations can help us understand possible trajectories of runoff event characteristics and their impact on nitrate export in the future. Additionally, these long-term runoff event characteristics allow us to embed the observed catchments into a larger group of catchments with very similar runoff event characteristics, classified by Tarasova et al. (2020). The six studied catchments match well with the clusters that characterize runoff events in the Central Uplands of Germany (including the Harz Mountains where this study is located) and in the Alpine Foreland (Tarasova et al., 2020). Over a time period from 1979 to 2002, the majority of runoff events in these clusters were Rain-Dry events, while approximately 15%-25% were snow-influenced events (Snow) and the number of events characterized by volume-dominated rainfall prevailed over intensity-dominated rainfall (Tarasova et al., 2020). This is well in line with our results that include more recent years (until 2018) and show >50% Rain-Dry events, 18%-25% Snow events, and more events characterized by volume-dominated rainfall than intensity-dominated rainfall. Based on this, we argue that our observed runoff event classes are representative for many upland areas and forelands of higher mountain ranges in a temperate climate.
To get a representative picture of nitrate export during those runoff events, one needs to consider that export also depends on additional factors, such as the amount and distribution of N sources within a catchment, which are strongly driven by land use patterns Musolff et al., 2017) as well as biogeochemical processing. By analyzing the impact of these representative runoff event characteristics on nitrate export across different hydro-climatic conditions and in six catchments that span a significant range of different land use types and other characteristics (Table 1), we are confident that the presented results are generally transferable to other upland areas and mountain forelands in a temperate climate. However, the analysis of long-term runoff event characteristics in combination with concentration data of five years only, does not allow inference on long-term trajectories of nitrate transport. We cannot easily assume a biogeochemical stationarity at decadal or longer time scales. However, there is evidence of this stationarity in catchments with high N loadings and thus ubiquitous and strong N sources that reflect in chemostatic export patterns (Basu et al., 2010). This may be the case in our catchments but may not be transferable to catchments with limited N availability or a different N input history.
Nonetheless, by including an extended set of hydro-meteorological variables that goes beyond the limited set of event characteristics used in previous studies, we could disentangle a large part of the variability in nitrate export patterns and create results that are better transferable to other catchments and time periods, assuming that catchment functioning for nitrate cycling and retention remains similar. A hydrological classification can thus be seen as one prerequisite for creating transferrable results to better compare the partly contradicting results between different studies (e.g., Knapp et al., 2020;Koenig et al., 2017;Rose et al., 2018;Vaughan et al., 2017;Winter et al., 2021) and to create a more coherent picture of the hydrological processes that shape nitrate export dynamics.

Conclusions
In this study, we conducted a rigorous assessment of the impact of runoff event characteristics and their classes on high-frequency nitrate export across multiple years, hydro-climatic conditions, and across six contrasting mesoscale catchments. We used long-term runoff event characteristics to embed the relationship between event runoff and nitrate export into a larger hydrological description of events and catchments. This new framework allowed us to identify potential long-term trends in nitrate export and their implications under a changing climate. We found that nitrate export differed substantially between runoff events with different characteristics, and strong drivers being event magnitude and a pronounced seasonality. With our findings, we argue that the variability and timing of nitrate export is likely to change with a changing frequency of event types that is driven by future global warming that is, projected changes in temperatures and other hydro-meteorological conditions.
Lowest nitrate concentrations and loads were transported during low-magnitude rain-induced events with dry antecedent soil moisture (Rain-Dry-Int and Rain-Dry-Vol), which occurred mainly during summer and autumn. These lower nitrate loadings, compared to high-flow seasons, can be explained by a small fraction of active flow paths, longer transit times, and a lower nitrate availability through higher uptake and denitrification rates during the vegetation period. Additionally, we found an increasing variability of event-specific CQ slopes with decreasing event size. We explain this high variability by an increased relevance of different environmental factors for nitrate export dynamics, such as the spatial distribution of nitrate sources and their connectivity to the streams, as well as the spatial and temporal distribution of precipitation (i.e., volume-or intensity-dominated) and biogeochemical processes in-stream and in the riparian zone. Consequently, more frequent dry spells will likely lead to more variable and less predictable water quality in rivers and streams.
In contrast, highest nitrate concentrations and loads were exported during high-magnitude snowmelt-induced (Snow) or volume-dominated rain-induced events under wet antecedent conditions (Rain-Wet-Vol), which occurred mainly during winter and spring. Nitrate mobilization, represented by event-specific CQ slopes, was surprisingly homogeneous among high-magnitude events across all catchments and land-use types, showing a relatively small increase of nitrate concentrations compared to discharge (approximately chemostatic conditions). We explain this by the activation of all relevant flow paths within a catchment that facilitate the land-to-stream connection of all relevant N sources and by higher, not limiting, nitrate availability during the dormant seasons.
As classes for high-magnitude events, that is, Snow and Rain-Wet-Vol, showed a very similar nitrate export behavior, we suggest that not the meltwater fraction, but instead other common characteristics such as event size, catchment saturation, and nitrate availability are the main drivers of nitrate export during high-magnitude runoff events. No dilution patterns (negative event-specific CQ slope) were observed for those events; hence, even forest-dominated catchments showed no sign of N source depletion, which could be a warning sign for future water quality trends. Increasing temperatures might cause a change in the timing of large nitrate export peaks within the high flow season, but we could not find evidence for a change in the amount of nitrate export in regard to hydrological transport, because declining snowfall (and consequently snow-influenced events) could potentially be compensated for by increasing winter rainfall.
Runoff event characteristics in this study are generic and hence comparable between catchments. Therefore, we argue that they are also representative for other upland or foreland areas in temperate climates. Covering a range of different catchment characteristics, for example, dominantly forested vs. mainly agricultural land cover, allowed us to analyze various catchment configurations and the respective event-driven nitrate export patterns and thus to represent a range of possible generic relationships between runoff event types and nitrate export. The potential of a hydrological event classification to create transferable results should be further exploited by analyzing event-driven nitrate export across even wider ranges of catchment characteristics and climatic conditions and by applying this approach to event-driven export of other solutes and particulates. Establishing robust relationships between runoff event characteristics and water quality dynamics, and relating them to long-term trends in runoff event characteristics, as introduced here, would be an informative tool for understanding possible directions of future changes in water quality.