High-frequency monitoring of catchment nutrient exports reveals highly variable storm event responses and dynamic source zone activation

Storm events can drive highly variable behavior in catchment nutrient and water ﬂ uxes, yet short-term event dynamics are frequently missed by low-resolution sampling regimes. In addition, nutrient source zone contributions can vary signi ﬁ cantly within and between storm events. Our inability to identify and characterize time-dynamic source zone contributions severely hampers the adequate design of land use management practices in order to control nutrient exports from agricultural landscapes. Here we utilize an 8 month high-frequency (hourly) time series of stream ﬂ ow, nitrate (NO 3 -N), dissolved organic carbon (DOC), and hydroclimatic variables for a headwater agricultural catchment. We identi ﬁ ed 29 distinct storm events across the monitoring period. These events represented 31% of the time series and contributed disproportionately to nutrient loads (42% of NO 3 -N and 43% of DOC) relative to their duration. Regression analysis identi ﬁ ed a small subset of hydroclimatological variables (notably precipitation intensity and antecedent conditions) as key drivers of nutrient dynamics during storm events. Hysteresis analysis of nutrient concentration-discharge relationships highlighted the dynamic activation of discrete NO 3 -N and DOC source zones, which varied on an event-speci ﬁ c basis. Our results highlight the bene ﬁ ts of high-frequency in situ monitoring for characterizing short-term nutrient ﬂ uxes and unraveling connections between hydroclimatological variability and river nutrient export and source zone activation under extreme ﬂ ow conditions. These new process-based insights, which we summarize in a export, with loads scaling with event duration; (ii) hydroclimatological variables would be signi ﬁ cant predictors of interevent variability in nutrient export, and (iii) differences in nutrient distribution within the catchment (i.e., spatially explicit zones of NO 3 -N in arable ﬁ elds versus more uniform distribution of DOC) would be re ﬂ ected in nutrient export patterns at the catchment outlet (stronger hysteretic behavior for NO 3 -N than DOC), providing an inverse approach to learn about source zone activation from observed nutrient export at the catchment outlet.


Introduction
Riverine nutrient loading (N, P, and C) is increasing in many catchments worldwide due to changes in land management, farming practices, and increasing urbanization [Gruber and Galloway, 2008;Burt et al., 2010;Thomas et al., 2016]. The resulting eutrophication of waterbodies often leads to changes in pH, turbidity, and dissolved oxygen availability, and thus, can be detrimental to aquatic ecosystem structure and functioning [Smith and Schindler, 2009;Friberg et al., 2010]. In addition to ecological impacts, high riverine nutrient concentrations can cause significant socio-economic implications by impairing freshwater ecosystem services including drinking water supply, recreational opportunities, and aesthetic qualities such as taste or odor [Millennium Ecosystem Assessment, 2005;Bennett et al., 2009]. Moreover, drinking water supplies with high concentrations of nitrate and disinfectant by-products (associated with the removal of aromatic dissolved organic matter) have been linked to adverse public health impacts including cancer, diabetes, and mutagenic diseases [Ward et al., 2005;Carpenter et al., 2013;Ritson et al., 2014]. Given the global relevance of increasing river nutrient concentrations, there is a critical need to develop a thorough mechanistic understanding of the variability and controls on nutrient mobilization and export from river catchments. In particular, identification of the dominant landscape source zones that contribute to nutrient export is often challenging [Bishop et al., 1994;Pacific et al., 2010;Grabs et al., 2012], despite the importance of this for their effective management.
Riverine nutrient concentrations can exhibit highly dynamic and nonlinear behavior [Krause et al., 2015] that has been observed over a wide range of temporal scales [Bowes et al., 2009;Wade et al., 2012;Halliday et al., 2015;Bieroza et al., 2014]. Traditional approaches to nutrient monitoring, constrained by laboratory and BLAEN ET AL.
HIGH-FREQUENCY STORM EVENT MONITORING 1 personnel costs, were restricted to relatively coarse temporal sampling resolutions (i.e., days to weeks), although a growing body of evidence now suggests that short-term (event-based) variability in nutrient concentrations is not captured by infrequent sampling regimes [Bowes et al., 2009]. Recent developments in optical sensor technology are now enabling in situ, continuous measurements of riverine nutrient dynamics at subhourly temporal resolutions [Blaen et al., 2016]. Such high-resolution data have the potential to adequately monitor the real dynamic behavior of catchment nutrient exports and, thus, improve estimates of nutrient loads and facilitate detailed new insights into the patterns, drivers, and organizational principles of catchment nutrient fluxes.
Studies of riverine nutrient dynamics have shown that catchment exports are linked strongly to changes in hydrological and climatological conditions. For example, precipitation inputs can play an important role in determining nutrient loads by flushing solutes from near-stream sources [Huang and Chen, 2009]. Moreover, antecedent soil moisture, temperature, and groundwater conditions can alter the potential for transformation (e.g., mineralization) and accumulation of nutrients in shallow subsurface flow paths [Agehara and Warncke, 2005]. In particular, extreme flow conditions caused by episodic storm events and seasonal snowmelt have been shown to exert major influences on nutrient export patterns and dynamics [Pellerin et al., 2012;Saraceno et al., 2009;Basu et al., 2010;Khamis et al., 2017]. During these events, changes in surface and subsurface flow paths can modify riparian connectivity to the river catchment and lead to the activation of distant solute and particulate source zones that would not usually contribute to catchment nutrient export under base flow conditions. Storm events therefore may trigger hydrological and hydrochemical "hot moments" that contribute specifically to nutrient dispersal and exports from catchments. Therefore, understanding how nutrients are mobilized and transported from different source zones during these periods is critical to produce accurate estimates of the timing and magnitude of catchment nutrient fluxes in support of adequate and efficient integrated catchment water quality management [Wilson et al., 2013;Carey et al., 2014]. Such information is also needed to generate predictions of how river nutrient loads and catchment exports are likely to change under future climate regimes. This latter point is particularly important given that most climate change scenarios suggest an increase in the magnitude and frequency of episodic precipitation events and soil drying through drought across many areas of the world [Kendon et al., 2014;Mann et al., 2017], particularly where local temperatures are <25°C [Wang et al., 2017], which in turn are likely to drive changes in other hydrological variables that influence nutrient flux and in-channel processing through river catchments [Garner et al., 2015].
Previous studies characterizing variability in responses of catchment nutrient exports to storm events have focused mainly on single parameters, such as nitrate [Chen et al., 2012;Carey et al., 2014], fluorescent dissolved organic matter [Saraceno et al., 2009], or phosphate [Bowes et al., 2005], although in recent years a growing number have considered varying responses between nutrient types [Drewry et al., 2009;Pellerin et al., 2012;Outram et al., 2014]. Moreover, very few studies have investigated how changes in hydroclimatology, both during and preceding storm events, control river nutrient export and source zone activation under extreme flow conditions, despite a clear requirement for this information for the effective management of catchment water resources both now and under future climate regimes. Headwater rural catchments dominated by agricultural land use are often important sources of riverine nutrient pollution, particularly nitrogen [Mellander et al., 2012], and therefore, understanding controls on nutrient fluxes from these areas is an important component of developing strategies to prevent and mitigate excess nutrient loading of downstream ecosystems.
To address this research gap and provide detailed mechanistic system-level understanding of variable storm event controls on nutrient source zone activation and catchment export, we analyzed high-resolution records of streamflow, nitrate (NO 3 -N), dissolved organic carbon (DOC), and a suite of hydroclimatological variables from a headwater agricultural stream. We hypothesized that (i) storm events would contribute disproportionately to nutrient export, with loads scaling with event duration; (ii) hydroclimatological variables would be significant predictors of interevent variability in nutrient export, and (iii) differences in nutrient distribution within the catchment (i.e., spatially explicit zones of NO 3 -N in arable fields versus more uniform distribution of DOC) would be reflected in nutrient export patterns at the catchment outlet (stronger hysteretic behavior for NO 3 -N than DOC), providing an inverse approach to learn about source zone activation from observed nutrient export at the catchment outlet.

Site Description
Catchment nutrient exports were analyzed for the Wood Brook at the Birmingham Institute of Forest Research (www.birmingham.ac.uk/bifor) field site in the UK between March and November 2016 ( Figure 1). This second-order stream drains a 3.1 km 2 catchment ranging elevation from 90 to 150 m above mean sea level and is situated in a nitrate vulnerable zone (a conservation designation for land draining into nitratepolluted waters; Directive 91/676/EEC). Land use was dominated by arable farming of potatoes and winter wheat and a mixture of young and mature deciduous woodland, primarily English oak (Quercus robur), hazel (Corylus avellana), hawthorn (Crataegus spp.), and sycamore (Acer pseudoplatanus). Young woodland areas were planted with saplings in 2014, prior to which the land was used for arable farming for more than 20 years. Tile drains are present in some arable and young woodland areas of the catchment. In spring 2016, arable farmland areas were dosed repeatedly with soil fertilizer (ammomium nitrate and ammonium sulfate) with application rates of 45-80 kg N/ha, which are typical for these crop types in the UK [e.g., Outram et al., 2014]. Mean annual temperature at the site is 9°C, and mean annual precipitation is 690 mm [Norby et al., 2016]. Catchment geology is composed of red Permo-Triassic sandstone overlain by superficial deposits of glacial till up to 10 m thick, with organic-rich, sandy clay topsoils between 0.15 and 0.6 m thick.

Stream Water Sampling
A stream water quality monitoring station was deployed at the catchment outflow to provide continuous in situ measurements of water and nutrient fluxes ( Figure 1). The station was equipped with a pressure transducer (Adcon, Austria) for stage measurements, an OPUS UV spectral sensor (TriOS GmbH, Germany) for optical measurements of NO 3 -N and DOC concentrations, and a Manta 2 multiprobe (Eureka, TX, USA). The Manta 2 was equipped with sensors for ancillary measurements of water temperature, electrical conductivity, and turbidity. A stage-discharge relationship (R 2 = 0.89; Figure S1 in the supporting information) was derived from salt dilution gauging measurements [Hudson and Fraser, 2005] and was consistent throughout the study period.
Channel geometry and restrictions in water depth of this second-order stream meant that most sensors (except the pressure transducer) were not submerged in the stream but housed in an insulated kiosk 1 m from the stream bank. An ISCO (Lincoln, NE, USA) 3710 peristaltic pump was used to pass 1 L of water every hour from an intake point in the thalweg of the stream through two flow cells containing the OPUS UV sensor and Manta 2, respectively. The intake was protected with a coarse (1 mm) nylon mesh to prevent damage to the pump tubing from large particulates. Instruments were programmed to acquire sample readings 3 min after the completion of each pumping cycle. Data were uploaded every 3 h via a telemetry system to an internet server for storage and quality control.
All sensors were cleaned every 2 to 4 weeks using acetone on optical windows and mild detergent on other components. Tubing was flushed with 10% HCl on each cleaning occasion to inhibit the development of biofilms. In addition, the Manta 2 had an automatic wiper that cleaned all sensors every 10 min throughout the monitoring period. Grab samples were collected to validate and adjust NO 3 -N and DOC readings by in situ sensors, with efforts focusing on storm events when concentration range was greatest. Therefore, an ISCO 3700 autosampler was used to collect hourly samples during storm events (n = 83), while additional manual grab samples were collected throughout the monitoring period (n = 13). All samples were filtered through 0.45 μm nylon filters (Thames Restek, UK) into sterile HDPE bottles, kept cool, and frozen within 6 h for later analysis using a Skalar (Breda, Netherlands) SAN++ continuous flow analyzer for NO 3 -N and a Shimadzu (Kyoto, Japan) TOC-L analyzer for DOC.

Hydroclimatological Sampling
Hydroclimatological variables were measured continuously throughout the study period to provide insights into the environmental processes that drive variability in water and nutrient fluxes at the catchment outflow. The aim was not to characterize the full range of spatial variability within the catchment, but rather the relative temporal changes in local hydroclimatological conditions that occurred between storm events. Therefore, precipitation was measured every 15 min using an ARG100 tipping bucket rain gauge (EML, UK) positioned in the center of the catchment (elevation 110 m above sea level (asl)) and away from trees or other structures which could influence collection at a distance >1.5 times the height of the nearest object.

10.1002/2017JG003904
Volumetric soil moisture content at 10 cm depth was measured at 15 min intervals using a 5TM probe (Decagon Devices, WA, USA) positioned in a young woodland clearing in the south of the catchment (elevation 100 m asl). Air temperature was measured every 15 min using a Vaisala HMP155 probe positioned on a meteorological monitoring tower at 1.2 m above the ground in the south of the catchment. Groundwater levels were recorded hourly using a Mini-Diver (Van Essen Instruments B.V., Netherlands) located in a 10 m deep borehole in the south of the catchment. Comparison of precipitation, soil moisture, air temperature, and groundwater measurements showed good agreement (typically r > 0.9, p < 0.01) with other probes deployed for shorter timescales with regard to temporal variability (data not shown), indicating that these measurements were representative of conditions across the catchment.

Data Analysis 2.4.1. Data Validation and Calculation of Nutrient Loads
Unless otherwise specified, all data analysis was conducted using the statistical software R version 3.1.1 [R Core Team, 2016]. Servicing, instrument malfunction, and data communication failure led to periodic data losses causing gaps in the measurement time series. Missing data for short gaps <3 h (n < 20) were filled by linear interpolation. Longer gaps were not filled and accounted for 7% of the possible 5924 hourly measurements in the monitoring period. NO 3 -N and DOC data were smoothed by a running median filter (window width = 8) using the R package robfilter to reduce the influence of noise in the time series. In addition to hourly concentration measurements, load time series for NO 3 -N and DOC were calculated as where P L is the load for the parameter of interest (mg/hr), P C is the concentration of the parameter of interest (mg/L), and Q is stream discharge at the time of sampling (L/s). Note that for a pure dilution effect of discharge on concentration, the relative changes in P c and Q compensate to keep P L constant.

Storm Event Delineation
Storm events were delineated following a rule-based system using the R package hydromad. First, the base flow component of the hydrograph was separated using a three pass recursive digital filter with a constant of 0.96 [Nathan and McMahon, 1990]. Subsequently, storm events were defined as periods when total stream discharge exceeded base flow by 20% for more than 24 h. No minimum time period between events was stipulated. Only events that had NO 3 -N and DOC measurements available for >90% of the duration of the event were considered for analysis. Following the automated selection process, storm event data were examined individually and a small number of outliers were excluded manually based on local site knowledge (e.g., crop irrigation immediately adjacent to the stream resulted in short-term hydropeaking during a dry weather window).

Hydroclimatological Drivers of NO 3 -N and DOC Export During Storm Events
Variability in NO 3 -N and DOC dynamics between storm events was characterized by calculating the following response variables for each event: maximum event concentration (C max ), percent increase and decrease in concentration since the start of the event (P inc and P dec , respectively), and the total load during the event (L tot ).
To understand the hydroclimatological processes that drive nutrient dynamics during storm events, specific hydroclimatological variable properties of individual storm events were analyzed (Table 1) with the hypothesis (H 2 ) that they would provide explanation and predictive capacity for nutrient availability and transport in the catchment. For example, temperature and soil moisture have been previously found to influence the release of nitrogen from organic sources [Agehara and Warncke, 2005], while the volume of rainfall before and during an event may be important for creating hydrological connections between catchment nutrient source zones and the stream network. Hydroclimatological variables were examined using a principal component analysis (PCA) with all data standardized and centered prior to analysis using the ade4 R package.
To identify the key drivers of nutrient dynamics, we used multiple linear regression and an informationtheoretic model selection approach after Johnson and Omland [2004]. All data were initially screened for collinearity, and variables with correlation coefficients >0.7 or variance inflation factors >3 were removed from the analysis [Zuur et al., 2010]. Ten predictor variables (R max , R dur , t Qmax ,Rain 1 , Rain 7 , T mean14 , Soil max7 , Δ t-1 , Q maxt-1 , and day of year (DOY); see Table 1) were then taken forward to fit global models, using ordinary least squares regression, for the five response variables (C max , P inc , P dec , L tot , and hysteresis index-see below) of both NO 3 -N and DOC. Predictor variables were transformed where necessary and standardized to improve the interpretability of regression coefficients, i.e., effect sizes [Schielzeth, 2010], and the robustness of the statistics. We then fitted all possible subset models of the global model and ranked them using AIC values  [Lange et al., 2014] with regression coefficients calculated as weighted averages [Burnham and Anderson, 2003].

Interstorm Variability in NO 3 -N and DOC Source Zone Contributions
Analysis of hysteretic behavior between solutes and discharge can provide insights into the sources and transport mechanisms of nutrients within a river catchment [Bowes et al., 2005;Chen et al., 2012;Outram et al., 2014;Lloyd et al., 2016b] To better understand the processes controlling nutrient-specific export patterns observed at the catchment outlet (see hypothesis 3), NO 3 -N and DOC were characterized as hysteresis indices (HI) for each storm event using methods developed by Lloyd et al. [2016a]. The HI for each storm quantifies the "fatness" and direction of the hysteresis loop, whereby clockwise hysteresis behavior is represented by positive HI values and anti clockwise behavior by negative values. Unlike previous methodologies, the HI is calculated for storm data which are first normalized by flow and chemical concentration to enable valid comparisons between storm events with varying concentrations and flow rates. The index is then calculated at multiple points throughout the storm using: where HI Qi is the index at percentile i of discharge (Q), C RL_Qi is the chemical value on the rising limb at percentile i of Q, and C FL_Qi is the chemical value at the equivalent point in discharge on the falling limb. The percentiles of discharge (Q i ) are defined by where Q max is the peak discharge, Q min is the discharge at the start of the event, and k is the point along the loop where the calculation is being made; in this case the index was calculated at every 5% of discharge. Finally, the overall HI was calculated using the mean of the 19 values obtained across the storm loop. HI metrics were regressed against PCA factor scores and NO 3 -N and DOC response variables to explore further how changes in hysteresis patterns were linked to hydroclimatological drivers and nutrient dynamics.

Seasonal-and Event-Based Variability in Streamflow, NO 3 -N and DOC
The mean stream discharge during the monitoring period was 12.7 L/s. Stream base flow declined gradually from March to August and then rose slightly from mid-August onward (Figure 2a). A total of 568 mm precipitation was recorded. The mean daily precipitation was 3.9 mm and showed no evidence of seasonal variability. Total runoff was 508 mm over the study period with a runoff coefficient of 0.82 (a flow duration curve is presented in Figure S2). Streamflow generally responded rapidly to precipitation inputs; discharge typically increased by >100%, and in some cases up to 400%, within a few h of each rain event, and then returned to base flow conditions over the following 2-3 days. A large increase in discharge occurred over a period of approximately 6 h at the end of the monitoring period following sustained rainfall. Continued monitoring of discharge into December showed that flow conditions returned to base flow after approximately 1 week (data not shown). Air temperature and groundwater level both exhibited strong seasonal patterns and were not influenced to a large extent by short-term storm events. In contrast, soil moisture conditions were highly responsive to precipitation inputs (Figure 2b). In situ measurements of NO 3 -N and DOC displayed linear relationships with laboratory measurements (NO 3 -N R 2 = 0.97; DOC R 2 = 0.79; Figures S3 and Table S1 in the supporting information). While error associated with the DOC relationship (root-mean-square error (RMSE) = 0.77 mg/L) was greater when compared to NO 3 -N (RMSE = 0.37 mg/L), bias was lower (percent bias; 0% versus 3.2%). Mean stream NO 3 -N and DOC concentrations were 6.01 mg/L and 12.98 mg/L, respectively, for the entire monitoring period. No long-term trends were observed in NO 3 -N concentrations (Figure 2c), but DOC concentrations rose throughout the beginning of the study period, peaked in late August, and then declined through autumn (Figure 2d). DOC exhibited diurnal fluctuations during base flow conditions with peak concentrations in early morning and the lowest concentrations in late afternoon, but this pattern was less evident during storm events (Figure 3). In contrast, diurnal fluctuations in NO 3 -N were less apparent.
A total of 36 storm events were identified by the event delineation process, of which 29 were selected for detailed analysis (Figures 2 and S4), covering a range of different hydroclimatological conditions throughout the study period (Table 1). The mean time between events was slightly less than 5 days. Selected storm Journal of Geophysical Research: Biogeosciences 10.1002/2017JG003904 events represented 31% of the 248 day monitoring period and accounted for 42% of total NO 3 -N and 43% of total DOC exported from the stream catchment during that time. Note, however, that the total number of storms throughout the monitoring period was higher than the 29 selected here for detailed analysis. NO 3 -N and DOC concentrations varied between storm events and exhibited both increases and decreases in concentration in response to increased stream discharge (Table 2 and Figure S5). NO 3 -N concentrations exhibited an overall increase for 12 of the 29 storm events, while DOC concentrations exhibited an overall increase for 23 events. Mean nutrient concentrations were higher during events than at base flow throughout the monitoring period (Figure 3), albeit with considerable variability around the mean and with limited observations for storms in autumn relative to other seasons. NO 3 -N concentrations varied more intensively between events than DOC: in particular, the mean decrease in NO 3 -N concentration throughout all the storm events was 16.2%, compared with 3.2% for DOC (Table 2). In contrast to concentration dynamics, nutrient loads mirrored trends in stream discharge, with both NO 3 -N and DOC loads increasing with stream discharge.
PCA analysis of hydroclimatological variables for each storm event produced four components with eigenvalues >1. The first component explained 34% of the variance (Figure 4), with positive loadings for soil moisture and antecedent rainfall and negative loadings for event rainfall duration and time to peak discharge. The second component explained 19% of the variance with positive loadings for total rainfall and discharge range and negative loadings for air temperature and groundwater level. In total, the four components explained 77% of the variance in the data set. Summer storm events generally had negative scores for both axes 1 and 2, while no clear patterns were found for spring and autumn events (Figure 4).

Hydroclimatological Controls on NO 3 -N and DOC Stream Fluxes
NO 3 -N and DOC export dynamics were significantly predicted by hydroclimatological variables (Table 3). DOC load was strongly associated with rainfall duration and mean air temperature (effect sizes: 0.57 and À0.53) and moderately associated with soil moisture content (0.37). NO 3 -N load was strongly associated with mean air temperature (À0.69) and moderately associated with rain duration and 7 day antecedence (0.44, 0.37). For both DOC and NO 3 -N, maximum concentrations were strongly associated with maximum rainfall intensity; however, 7 day antecedence and time since last event were also found to be important for explaining observed NO 3 -N dynamics. Concentration increases were associated with maximum rainfall intensity for both nutrients. However, DOC was also influenced by air temperature and rainfall duration, while NO 3 -N was influenced by time since last event.

Variability in NO 3 -N and DOC Source Zones Under Different Hydroclimatological Conditions
Clockwise and anticlockwise hysteresis patterns were observed in NO 3 -N and DOC concentrations during storm events (Table 2). Examples are presented in Figure 5. Overall, the HI value for NO 3 -N exhibited more negative (i.e., anticlockwise) behavior (range À0.8 to 0.69) than the HI for DOC (range À0.66 to 0.94) and

10.1002/2017JG003904
was lower on an event-specific basis (mean difference 0.1). The HI for NO 3 -N was significantly negatively correlated with PC2 (r = À0.33, p < 0.05) and PC4 (r = À0.48, p < 0.01) of the PCA analysis of hydroclimatological variables (as described above and in Figure 4). In contrast, the HI for DOC was not significantly correlated with any PC (Figure 6). No significant correlation was found between the HI for NO 3 -N and the HI for DOC (r = 0.06, p > 0.05). Regression analysis revealed that rainfall antecedence was the strongest control on the HI index for DOC (i.e., greater 7 day rainfall led to increased clockwise hysteresis), although this relationship was still relatively weak when compared to controls on NO 3 -N. For NO 3 -N mean air temperature and soil water were moderately associated with an increase in clockwise hysteresis, while maximum rainfall and 7 day antecedence were weakly associated with anticlockwise behavior.

Discussion
In this study we used high-frequency in situ monitoring to characterize complex nutrient dynamics and unravel connections between hydroclimatological variability and river nutrient export under stormflow conditions. Our results provide new insights into how storm events affect catchment hydrological connectivity and lead to the short-term activation of nutrient-specific source areas that would not otherwise contribute to catchment nutrient fluxes under base flow conditions.

Temporal Dynamics of Streamflow, NO 3 -N, and DOC
We observed temporal variability in streamflow, NO 3 -N, and DOC dynamics across a range of temporal scales (i.e., seasonal, diurnal, and event). At the seasonal scale, streamflow decreased through summer in response to relatively warm, yet not unusually dry, conditions with high evapotranspiration potential, and then  [Jaffé et al., 2008]. Such DOC concentrations are akin to those reported for a similar sized agricultural catchment in northern France [Morel et al., 2009] but relatively high when compared to values reported from many mixed-use catchments in both Europe and North America [Raymond and Oh, 2007;Monteith et al., 2007]. Base flow NO 3 -N concentrations of around 6.5 mg/L were similar to those reported previously for agricultural catchments [Lloyd et al., 2016b;Thomas et al., 2016] and were likely driven by current fertilizer application and legacy fertilizer  At diurnal scales, streamflow exhibited minor fluctuations during base flow conditions, most likely driven by contrasts in evapotranspiration demand between day and night but also potentially due to temperaturedriven changes in water viscosity and associated changes in bed sediment hydraulic conductivity [Schwab et al., 2016;Constantz et al., 1994]. Observed daily cycles in NO 3 -N and DOC concentrations during base flow-dominated periods are likely to reflect a combination of minor dilution effects (in relation to diurnal changes in streamflow) and in-stream assimilatory uptake by microbial communities, which Rode et al.
[2016a] showed can be high in agricultural catchments compared to those dominated by other land uses. In contrast, photochemical degradation of DOC [e.g., Spencer et al., 2007;Lu et al., 2013] is unlikely to have had a substantial impact on diel DOC dynamics because the stream channel was incised and shaded by riparian vegetation for much of its length.
Compared to diurnal patterns, storm events induced considerably greater variability in streamflow dynamics and NO 3 -N and DOC concentrations, with no evidence for chemostatic behavior during any events [Moatar et al., 2017]. The rapid response of streamflow to precipitation inputs can be attributed in part to the numerous ditches and tile drains in the upper catchment. These landscape features have the capacity to export significant quantities of water, and with that nutrients, from agricultural catchments over relatively short time periods [Li et al., 2010;De Schepper et al., 2015], particularly from areas dominated by clay soils, resulting in flashy hydrological regimes that transfer solutes quickly to fluvial networks [Bowes et al., 2005;Cassidy and Jordan, 2011]. Responses of NO 3 -N and DOC concentration dynamics varied between nutrients and between storm events (as discussed in more detail below). NO 3 -N concentrations were typically diluted on the rising limbs of storm hydrographs, most likely due to the rapid delivery of relatively low-concentration water transferred to the stream channel from near-surface soil flow paths in the early stages of each event [Outram et al., 2014;Dupas et al., 2016], whereas patterns in DOC concentrations generally exhibited flushing behavior through storm events. In contrast to the relatively high variability between NO 3 -N and DOC concentrations, stream solute loads were tightly coupled to discharge for both parameters. This is because the relative change in streamflow during storm events (i.e., orders of magnitude) was substantially higher than the associated change in solute concentration (i.e., multiples), meaning that increases in discharge outweighed dilution effects on concentration (see equation (1)). Therefore, storm events represented important components of the annual hydrograph for mobilization and export of NO 3 -N and DOC from the catchment; thus, supporting our first hypothesis that storm events would contribute disproportionately to nutrient export relative to their duration [Basu et al., 2010;Raymond et al., 2012;Mellander et al., 2012]. The insights provided by these high-resolution data sets underscore the value of automated in situ sensors to capture both interevent and intraevent variabilities in storm event dynamics relative to conventional sampling methods using manual grab samples or autosamplers of limited capacity [Blaen et al., 2016;Bowes et al., 2009].

Hydroclimatic Drivers of Nutrient Export During Storm Events
In this study, stream nutrient dynamics could be accurately predicted from a relatively small number of variables characterizing the hydroclimatic variability of storm events, incorporating both present and antecedent hydroclimatic conditions, thereby supporting our second hypothesis that hydroclimatological variables would be important predictors of variability in nutrient export. Mean daily precipitation totals were very similar to values reported by Leeuw et al. [2015] for precipitation observations across England and Wales, suggesting that our studied storm events are broadly representative of those that occur at a national scale. Precipitation intensity during storm events was particularly important for predicting maximum concentrations and relative increases in concentrations for both NO 3 -N and DOC. This is likely due to rapid runoff through the near-surface soil horizon during events of high precipitation intensity, thus mobilizing and routing the product of recently decomposed and nitrified organic matter to the river channel [Bernal et al., 2002]. Previous studies have found evidence of nutrient source exhaustion following consecutive storm events [e.g., Outram et al., 2016]. We found no evidence of this for DOC; however, the time interval between two events was an additional control on NO 3 -N concentrations, with higher maximum and relative increases positively associated with longer time intervals. This suggests that pore water with high concentrations of NO 3 -N is flushed from the soil during storm events and that extended dry periods between events are required for mineralization and nitrification of organic N to occur [Inamdar and Mitchell, 2006;Darwiche-Criado et al., 2015]. Moreover, close linkages between variability in nutrient loads and bulk stream discharge during storm events indicate that aquatic export of both NO 3 -N and DOC is primarily transport-limited in this historically agricultural catchment. For NO 3 -N, this is most likely to be attributable to the ready surplus of nitrogen derived from current and previous fertilizer use in the upper catchment, which may remain not only in soil horizons but also in groundwater aquifers that are connected hydrologically to the stream [Krause et al., 2009]. Transport limitations on NO 3 -N export from anthropogenically modified catchments, particularly those impacted by crop management and fertilizer application, have been identified by previous studies [e.g., Carey et al., 2014]. Moreover, Basu et al. [2010] suggest that such patterns in nutrient export are likely to continue into the future until accumulated legacy stores are depleted. Similarly, DOC export has also been linked to climatic drivers at annual timescales. For example, Raymond and Saiers [2010] showed an increase in DOC flux with discharge following a power relationship in forested watersheds in the U.S. However, in contrast, we found a negative relationship between air temperature and DOC load. This may be due to differences in the timing of tree senescence between North America and Europe (i.e., the longer time taken for leaf fall in Europe results in peak DOC leaching later in the season) or alternatively may be caused by larger storm events occurring in spring and autumn months when air temperatures are generally lower at the study site.

Identification of Nutrient Source Zone Activation Under Contrasting Storm Events
Substantial variation in hysteretic behavior was observed in concentration-discharge relationships between storm events for both NO 3 -N and DOC. This variability was quantified through the application of the HI, which enables robust comparison of data between storm events of different magnitudes [Lloyd et al., 2016a]. Our modeling results suggested that observed hysteresis patterns were driven by a combination of the hydroclimatological conditions of each particular storm event, coupled with the antecedent hydroclimatological conditions in the period leading up to that event [Darwiche-Criado et al., 2015]. For NO 3 -N, increased anticlockwise hysteresis patterns were associated with hydroclimatological drivers such as higher rainfall intensity during the event and also with high rainfall and lower air temperature and soil moisture in the preceding Journal of Geophysical Research: Biogeosciences 10.1002/2017JG003904 week. For DOC, model results were less significant, but interevent variation in DOC hysteresis behavior appeared to be driven largely by antecedent rainfall rather than by the hydroclimatological conditions experienced throughout each event. However, given the moderate errors associated with the relationship between in situ and paired laboratory measurements of DOC, we suggest that further work is needed to explore carbon export from agricultural catchments. It is likely that different DOM pools of varying quality and composition contribute to the DOC flux causing time-variable relationships between absorbance and DOC quantity. Hence, coupled in situ monitoring of humic-and protein-like fluorescence could improve understanding of DOC behavior under stormflow conditions [Khamis et al., 2017].
Given that neither NO 3 -N nor DOC dynamics appeared to be supply-limited in this catchment, as inferred from the tight coupling between streamflow and nutrient loads, it may be assumed that interstorm variability in hysteresis behavior indicates predominantly changes in the activation of different nutrient source zones within the catchment [Chen et al., 2012]. The lack of relationship between the HI for NO 3 -N and the HI for DOC reflects differences in the timing of nutrient delivery measured at the catchment outflow, thereby providing strong support for our third hypothesis that nutrient export dynamics would vary between NO 3 -N and DOC. This may be associated with both differences in the spatial distribution of nutrient source zones within the catchment and also differences in nutrient-specific transport mechanisms through surface and subsurface flow pathways as summarized in Figure 7.
The most obvious sources of NO 3 -N are the current and former arable farmland areas (but also potentially also contaminated groundwater aquifers) in the upper catchment [Rozemeijer and Broers, 2007;Krause et al., 2009;Raymond et al., 2012]. As such, NO 3 -N hysteresis patterns are hypothesized to reflect hydroclimatological impacts on catchment flow pathways during storm events, resulting in hydrological connectivity and activation of nutrient source zones to the channel network that would not contribute to aquatic NO 3 -N export during base flow [Bowes et al., 2009]. This is supported by the observed link between HI and precipitation intensity and antecedent rainfall: thus, certain NO 3 -N sources (e.g., those in the surface or uppermost soil layers) are mobilized when particularly extreme streamflow conditions coincide with high levels of soil saturation. These conditions result in a greater proportion of precipitation converted to overland runoff or preferential flow through subsurface tile drainage networks [Rozemeijer and Broers, 2007;Darwiche-Criado et al., 2015;van der Grift et al., 2016] with corresponding reductions in the influence of catchment subsurface storage on runoff and NO 3 -N export dynamics [Teuling et al., 2010]. In contrast, during drier periods, lower soil moisture conditions lead to greater infiltration potential, and thus, water and nutrient transport are dominated by subsurface flow pathways (Figure 7).
In contrast to NO 3 -N, interpretation of the physical processes underpinning observed hysteresis behavior in DOC was less clear, possibly caused by a more homogenous spatial distribution of DOC across the catchment (e.g., shallow soils, riparian zones, and instream production) relative to that of NO 3 -N ( Figure 7). However, the tendency of DOC toward more positive (clockwise) hysteresis is suggestive of source areas that are hydrologically well connected to the stream channel with the potential to deliver DOC rapidly at the start of storm events. The low percent bias in the relationship between in situ and laboratory measurements provides confidence in the direction of individual hysteresis patterns. One potential explanation for this behavior is that leaf litter inputs from forested areas of the stream network, particularly in the lower catchment, are entrained and decomposed by microbial communities in hyporheic streambed sediments and then remobilised under high flow conditions (Figure 7), thereby releasing DOC rapidly into the water column [Bernal et al., 2002]. This hypothesis is supported by samples acquired from sediment pore waters at 10-20 cm depth in April 2016, when seasonal leaf litter inputs would be expected to be minimal, which showed mean DOC concentrations (25.4 ± 22.3 mg/L) in the streambed to be more than double those in the water column, in some cases up to five times higher (maximum measured DOC concentration 82.6 mg/L). In contrast, N concentrations in the streambed were considerably lower than in the water column, suggesting that streambed sediments were not a major source of NO 3 -N to the water column during storm events.

Conclusions and Implications
In the context of growing demands on agricultural production and the challenges posed by climate change [Garnett et al., 2013], understanding the hydroclimatological drivers of catchment nutrient dynamics, and in particular exports, is critical to developing accurate predictions of water quality in river ecosystems. Our study contributes to this important field of research by developing new mechanistic process understanding from examining interactions between hydroclimatological drivers, streamflow, and nitrogen, and carbon concentrations at high temporal resolution. The use of high-frequency in situ sensors to capture short-term streamflow and solute concentration dynamics through storm events facilitated insights into the processes controlling highly dynamic catchment exports that would be impossible to achieve using discrete sampling methods [Rode et al., 2016b;Blaen et al., 2016]. Our modeling results highlighted the importance of key hydroclimatological variables, notably rainfall intensity and antecedent conditions, which drive the mobilization and transport of nitrogen and carbon through stream catchments. Given that precipitation regimes in many parts of the world are expected to shift to higher frequencies, and with more extreme events, in the near future [Kendon et al., 2014;Mann et al., 2017], this finding is important for developing predictive models to assess the potential implications for catchment water quality parameters. As such, our results suggest that management interventions to improve downstream water quality should focus not only on reducing contemporary catchment nutrient inputs (i.e., fertilizer application) but also on developing measures aimed at mitigating the impact of legacy sources that are activated under certain hydroclimatological conditions until their eventual depletion in future. Furthermore, our analysis of differences in nutrient source zone areas indicates that nutrient-specific mitigation measures are required to target particular landscape areas that contribute disproportionately to aquatic nutrient export. These findings highlight the value of high-resolution temporal nutrient time series provided by in situ sensors, as well as the need for more spatially distributed water quality monitoring across river networks to unpick natural system variability. If combined with additional analyses of water age distributions [e.g., Morgenstern et al., 2015], particularly at fine temporal resolutions using field analysers [von Freyberg et al., 2017], such work will further our ability to assess the relative importance of different nutrient source zones and flow pathways within catchments. Recent advances in sensing technology, coupled with decreasing costs [Blaen et al., 2016], mean that high-frequency nutrient measurements from in situ sensors are likely to play an increasingly important role in developing techniques for the effective management of catchment water resources in future.