The research was undertaken at six river reaches in the Hampshire Avon upstream of Salisbury (Wiltshire, UK), representing sub-catchments of contrasting geology (clay, Greensand, and Chalk), and a gradient of BFI (Fig. 1; Table 1). The majority of the upper catchment of the Hampshire Avon (draining ca. 1390 km2 in total) is dominated by the Cretaceous Chalk geology, and the hydrogeological properties of these geological units are described in detail in Allen et al. (2014). Sites CW on the River Wylye and CE on the river Ebble are river reaches characterized by high baseflow indices (> 0.9) where Chalk provides the main source of groundwater (Allen et al., 2014). In the north and west of the Hampshire Avon catchment there are also significant groundwater contributions from geological formations of Upper Greensand which comprise fine-grained glauconitic sands and sandstones (Bristow et al., 1999). The sub-catchments of sites GN on the river Nadder in the west of the catchment, and GA in the north of the catchment, both comprised ca. 50 % Upper Greensand by area with BFI of 0.695 and 0.861, respectively. The two sites characterized by the lowest BFI, sites AS (0.372) and AP (0.234), are located in the sub-catchment of the river Sem underlain by impermeable Late Jurassic Kimmeridge Clay (usually a non-aquifer) and thin interbedded limestone from which limited groundwater flow may occur (Allen et al., 2014). Agricultural land use dominates the Hampshire Avon catchment with arable farming including horticulture comprising 42 % of land use, and improved grassland for dairy and beef production covering 23 % of the catchment. The distribution of arable and livestock farming varies with sub-catchment; improved grassland dominates in the clay catchment of the river Sem (AS and AP), where it supports intensive dairy production, whilst arable agriculture represents ca. 50 % of land use at the chalk sites (CW and CE), with sheep grazing and intensive pig production as minority land uses (Table 1).

Sites AS, GA, GN, and CE were instrumented for 2 years from June 2013 until June 2015. Stream stage was measured using pressure transducers (HOBO U20-001-01, Onset Corporation, USA, at AS, GA and GN; Levelogger Edge, Solinst, Canada, at CE) in a perforated stilling well, logging at 15-mins intervals. Regular (fortnightly when possible) manual measurements of discharge by the velocity–area method enabled construction of stage–discharge relationships for each site. Discharge values used in the analysis were scaled to mm day-1, using an assumed catchment area defined by the topographic divide for that point in the stream network. Rainfall was measured at 15 min intervals at AS, GA, and CE using a tipping bucket rain gauge (674, Teledyne Isco, USA) in order to calculate daily rainfall totals (mm day-1) for the study period. Details of exact locations of hydrological measurements can be found in Heppell et al. (2016a, b).

Temperature, pH, dissolved oxygen (optical), and electrical conductivity of river water were logged in situ at 30 min intervals using a water quality multiprobe (Manta 2, Eureka Water Probes, USA). An automatic water sampler (6712, Teledyne Isco, USA) collected water samples from the river every 48 h from June 2013 to June 2014 for analysis of water chemistry, and samples were collected fortnightly. Therefore, field and laboratory tests were undertaken to ensure that sample degradation over this time period was negligible. Furthermore, Milli-Q water was decanted into sample bottles in the field to create field blanks to ensure that no sample contamination occurred during transportation between the field and laboratory. Three riparian piezometers (screen depth installed in the soil C horizon, typically circa 2 m depth) with porewater sampling tubes at screen depth were installed in the banks at each site in summer 2013 to enable measurements of riparian hydraulic head and porewater samples to be collected for chemical analysis. Hydraulic head was measured using pressure transducers (HOBO U20-001-01, Onset Corporation, USA, at AS, GA and GN; Levelogger Edge, Solinst, Canada, at CE) validated with manual dips on a fortnightly basis. Porewater samples were collected from sampling tubes on the riparian piezometers every 2 months from February 2014 to June 2016 using a syringe and Tygon tubing. Samples were then filtered to 0.45 µm in the field.

Sites AP and CW were a component of the Demonstration Test Catchment network (McGonigle et al., 2014; Outram et al., 2014). At AP, stream discharge was measured using a Mace Flow Pro to record paired stage height and velocity measurements at 15 min temporal resolution to which the velocity–area method was applied (Lloyd et al., 2016a, b). The Mace Flow Pro measurements were taken within a concrete section which meant that the cross-sectional area was stable. However, during high-flow events, the stage height overtops the concrete structure and out-of-bank flows occur. In these cases, a weir equation was implemented to account for the additional water flowing over the concrete section: Qi=CdbHi1:5, where Qi is the discharge at time point i (m3 s-1), Cd is the dimensionless coefficient of discharge, b is the weir crest breadth (m), and Hi is the stage height (m) above the bridge at time point i. Cd was set at 2.7 based on typical values from published literature (Brater and King, 1976). Discharge data for CW were obtained from the Environment Agency Gauging Station (gauge number 43 806), which provided 15 min resolution stage height data using a Thistle 24R incremental shaft encoder with a float and counterweight. For periods of modular flow, these data were used in conjunction with a stage–discharge curve to calculate discharge (ISO 1100-2, 2010). However, during non-modular flow periods, the stage heights are used alongside 15 min velocity measurements from a second ultrasonic gauge to calculate discharge using the velocity–area method (ISO 1088, 2007). At both sites daily river water samples were collected using automatic water samplers (Teledyne Isco 3700, USA) and collected weekly.

On return to the laboratory a sub-sample of river water from sites AS, GA, GN, and CE was filtered at 0.45 µm for analysis of nitrate and DOC. Nitrate concentrations were analysed using ion exchange chromatography (Dionex-ICS2500). The limits of detection (LOD) and precision were 8 µmol L-1 ± 7 %. These samples were then prepared for DOC analysis by acidification to pH < 2 with HCl and then analysis by thermal oxidation (Skalar) using the non-purgeable organic carbon (NPOC) method. The LOD of the DOC analysis was 42 µmol L-1 with precision of ±12 %. Accuracy was ensured by analysis of certified reference material (SPS-SW2 and TOIC4M14F1 for nitrate and DOC respectively) with each instrument run. Porewater samples from all sites were analysed using the same methods as for the surface water from AS, GA, GN, and CE.

River samples collected from sites AP and CW were filtered then analysed for nitrate using a Skalar San++ multi-channel continuous flow autoanalyser. This analysis was based on the hydrazine–copper reduction method producing an azo dye measured colorimetrically at 540 nm. DOC was analysed as non-purgeable organic carbon by coupled high-temperature catalytic oxidation using a Shimadzu TOC-L series analyser. For further details on sample collection and analysis at AP and CW sites see Yates et al. (2016).

BFI (BFI) for each site was calculated using the hydrograph separation procedure outlined in Gustard et al. (1992). Hydrographs with high BFI show relatively smooth characteristics and are indicative of major aquifers where water (and consequently solute) residence time in permeable bedrock will be of the order of decades, whereas a low BFI is characterized by a flashy hydrograph, with steep recession curves, and is indicative of a generally shorter residence time in the catchment before water reaches the stream channel, with quickflow comprising shallow, lateral preferential and overland pathways predominant during storm events. Soil moisture deficit (SMD) is defined as the amount of water (in mm) which would have to be added to the soil in order to bring it back to field capacity. SMD values were obtained from the UK Meteorological Office for MORECS square 169 (4000 east, 1400 north) for a medium textured soil type with predominantly grass cover.

In order to quantify the relationships between nitrate, DOC, and DOC : nitrate molar ratio with BFI, and to understand how any relationship varied intra-annually, a linear mixed-effects modelling approach was used. Linear mixed-effects models account for missing data, which is a common issue associated with long-term field datasets, and the inclusion of repeated measures in the analysis (Blackwell et al., 2006). The “lmer” function in R (R Core Team, 2016) package lme4 (Bates et al., 2015) was used to perform a linear mixed-effects analysis of the relationship between BFI as the independent measure, and either nitrate concentration, DOC concentration, or DOC : nitrate molar ratios as the dependent variable. The nitrate and DOC concentration of river water recorded at each site over the same time period (i.e. from samples collected at simultaneous 48 h time intervals from June 2013 until June 2014) was used in the analysis. BFI was entered as a fixed effect. We accounted for the influence of repeated measures by including time (Julian day) as a random intercept and slope in the model. The “lme” function in R package “nlme” (Pinheiro et al., 2016) was used to fit a linear mixed effects model to porewater data to investigate differences in nitrate and DOC concentrations between CE and CW (the Chalk sites) and all the other sites (AS, AP, GA, and GN).

For the purposes of considering the relationship between BFI and nitrate concentrations in the wider Hampshire Avon catchment, nitrate concentrations in river water samples collected between June 2013 and June 2014 were obtained from the Environment Agency Harmonised Monitoring Scheme (HMS) records. Average annual nitrate concentration was calculated for each site, but those with less than 12 samples in the 12-month period were removed from the analysis (number of samples ranged from 12 to 56 depending on the site). BFI for each Environment Agency site was estimated using the Flood Estimation Handbook, which uses the Hydrology of Soil Types (Boorman et al., 1995) methodology because there is not a gauging station at every location. Baseflow indices derived in this manner are referred to as BFIHOST to distinguish them from BFI values derived using our own discharge data. Pearson correlation analysis was used to explore relationships between solutes (nitrate and DOC) and BFI.

Annual loads of nitrate and DOC for sites AS, GA, and CE were calculated as kilograms per hectare by integrating paired concentration and discharge data collected on a 48 h basis from June 2013 to June 2014. Any missing solute data (maximum gap of 10 days due to equipment failure) were infilled using seasonal concentration–discharge relationships derived for each site. Seasonal loads are expressed as a percentage of total annual load for each site.

The first year of study (June 2013–2014), on which these results are focused, was characterized by pronounced cycles of soil wetting and drying due to alternating periods of unusually wet and dry weather (Fig. 2). Due to a combination of lower-than-average rainfall (ca. 50 % of 1910–2015 long-term average for the region) and high temperatures (> 28 ∘C for a 10–12 day period in July) over the summer of 2013, SMD reached a maximum of 140 mm for a 4-week period in August and September 2013. A period of unsettled weather in October and November 2013 (224 mm rainfall in total) reduced the SMD to 0 mm. After a brief return to dry, settled conditions, a series of deep Atlantic low-pressure systems brought a prolonged period of heavy rain to the entire Hampshire Avon catchment. A total of 161 mm rain fell in December 2013 (190 % of the 1961–1990 long-term average), with a maximum daily rainfall total of 58 mm on 23 December, followed by a further monthly total of 205 and 148 mm in January and February 2014, 261 and 259 % of the long-term averages, respectively. January 2014, in particular, was the equal wettest on record since 1910. SMD and rainfall patterns in 2014 were not as extreme as those in 2013, returning to monthly values that were much closer to the long-term averages. SMD reached peak values of 129 mm by the end of the summer in early October 2014, and autumn rainfall during October and November caused wetting-up of the soil to reduce SMD to 0 mm by mid-November 2014. By March 2015, warmer weather, combined with lower-than-average rainfall (< 50 % of long-term average), caused SMD to steadily increase until the end of the study period in June 2015.

Nitrate concentration in surface water of our sub-catchments is significantly positively correlated with BFI (r=0.749, p<0.001), whereas DOC concentration in our surface water samples exhibits a significant negative correlation with BFI (r=-0.881, p<0.001) (Fig. 3a, b, Table 2). The linear mixed effects model analysis indicates that BFI has a significant effect on nitrate (χ2(1)=19, p<0.0001) and DOC (χ2(1)=497, p<0.0001) concentrations, with an increase in BFI of 0.5 leading to a difference in average increase in surface water nitrate concentrations of 260 µmol L-1 and a reduction in DOC concentrations of 840 µmol L-1 between the clay and Chalk sites. Inclusion of time as a random effect (both slope and intercept) improved the model fit for both nitrate and DOC, indicating that temporal dynamics associated with these determinands are important to consider. The sites of lower BFI exhibit marked variations in nitrate concentration in autumn and winter, which change the slope (although not the overall direction) of the nitrate and BFI relationship, and highlight the importance of seasonality. Overall, the respective increase in nitrate, and decrease in DOC concentration with BFI, broadly reflects the patterns in concentrations of DOC and nitrate in the riparian zones associated with each geology. Nitrate concentrations in riparian porewaters were significantly higher in the Chalk sites compared to the clay and Greensand sites (F(1,146)=105, p<0.0001), whereas DOC concentrations were significantly lower in the Chalk sites compared to the others (F(1,146)=38, p<0.0001). The relationship between DOC : nitrate molar ratio and BFI is non-linear and can be best described by a power function ((DOC : nitrate) = 0.453 × BFI-2.575, r2=0.638, p<0.001, Fig. 3c).

The relationship between nitrate and BFIHOST was tested for 17 additional sites within the Hampshire Avon catchment using Environment Agency Harmonised Monitoring Scheme data collected between June 2013 and June 2014. Figure 4a shows that across the Hampshire Avon, there is a significant, positive, linear relationship between nitrate and BFIHOST (r=0.951) with a regression model indicating that BFIHOST accounts for 90.4 % of the variation in nitrate concentration. There is also a significant, positive correlation between nitrate concentration and % arable land use (r=0.839, p<0.001). Although % arable and BFIHOST are positively correlated (r=0.881), a tolerance value (a test for collinearity) of 0.224 indicates that multiple linear regression can be used in this instance (Field, 2000). Multiple regression shows, however, that BFIHOST alone produces the best model, with the forced inclusion of % arable resulting in no significant improvement to the model fit (Table 2).

From this point forward, data from three sites only are presented as illustrative of the hydrochemical signatures from a range of BFIs across our three geologies; Chalk (site CE, high BFI), Greensand (site GA, intermediate BFI), and clay (site AS, low BFI). There is a marked difference in the response of electrical conductivity to discharge across the three sites (Fig. 5a–c). At the Chalk site, CE, a maximum electrical conductivity of 0.570 mS cm-1 is maintained across the full range of recorded discharge. At the Greensand site, GA, electrical conductivity is maintained at ca. 0.650 mS cm-1 until discharge exceeds 1 mm d-1 and then a decline in electrical conductivity with increasing discharge is observed. An examination of electrical conductivity by season indicates that geogenic solute concentration was lowest at the Greensand site during winter 2014, and concentrations were comparable in spring, summer, and autumn (Fig. 5b). At the clay site, AS, there are two different relationships between electrical conductivity and discharge; a constant electrical conductivity of ca.  0.520 mS cm-1 is maintained at lower discharges of 0.001–0.3 mm d-1, whilst a log-linear decrease in electrical conductivity is observed between 0.2 and 3.5 mm d-1, and there is some overlap between the two patterns of behaviour. Box plots of electrical conductivity by season indicate highest concentrations of geogenic solutes in summer, intermediate concentrations in autumn and spring, and lowest concentrations in winter (Fig. 5c).

Inter-site comparisons of the response of nitrate, DOC, and DOC : nitrate molar ratio to variations in discharge are illustrated in Fig. 6. There is a significant, positive correlation between log-nitrate and log-discharge for all sites, with the slope of the regression relationship increasing with BFI (CE < GA < AS; Table 3). Visual examination of the relationship between nitrate and discharge for AS and GA suggests more than one trend is apparent and this is investigated in detail by considering seasonality below. There is also a significant, positive correlation between log-DOC and log-discharge, although in this case the slope of the regression relationship increases in the following order: AS < CE < GA (Table 3). However, again there is marked scatter in the relationship and this is investigated further below. There is a similar significant, proportional increase in DOC : nitrate molar ratio with increasing discharge at both CE and GA (slopes of 0.199 and 0.196, respectively, on a log–log basis, Table 3), whilst AS has a much weaker relationship, exhibiting far greater scatter.

Nitrate concentrations at the Chalk site, CE, show little variation with season or discharge, whereas DOC concentrations appear to follow two trends: (i) a slight increase in DOC concentration with discharge in spring and winter and (ii) elevated concentrations of DOC which are unrelated to discharge in spring (Fig. 7a). Consequently, DOC : nitrate molar ratios remain low (< 1) throughout the year (Table 4).

At the Greensand site, GA, both nitrate and DOC concentrations increase with discharge (irrespective of season) until a breakpoint is observed at 1.5 mm d-1. At this point, during the winter storms of 2013–2014, nitrate concentrations start to decline with increasing discharge, whereas DOC concentrations drop to < 500 µmol L-1 and a new, positive trend in increasing DOC with increased discharge is observed with a gentler slope (Fig. 7b). As a consequence, the positive relationship between DOC : nitrate ratios and discharge also show a similar breakpoint, but the DOC : nitrate ratio remains below 3 : 1 throughout the year (Table 4).

At the clay site, AS, there are two trends in the concentration–discharge relationship for nitrate (Fig. 7c). Concentrations are highest (200–400 µmol L-1) during the autumn storms of intermediate discharge that followed the summer drought of 2013. The winter storms of 2014 are associated with highest discharge, but lower nitrate concentrations (ca. 100 µmol L-1). This contrasts with DOC which shows a plateau in concentration (ca. 1000 µmol L-1) with increasing discharge, irrespective of season. Nitrate and DOC concentrations were plotted against electrical conductivity to test whether nitrate and DOC arose from a linear combination of old (long residence time) and new (short residence time) water, but this was not the case (data not shown) suggesting that variations in supply and/or in-stream processing of these solutes occur through the seasons. At AS, there are two observable trends in DOC : nitrate molar ratio: (i) highest and the greatest variability in DOC : nitrate ratios is observed during summer low-flow conditions, and (ii) there is an increase in DOC : nitrate ratios with discharge irrespective of season (Fig. 7d). Consequently, during autumn, values of DOC : nitrate ratios were generally equal to or less than 5, whilst values significantly greater than the threshold of 4 observed by Taylor and Townsend (2010) predominated during spring, summer, and winter (Table 4).

Over 50 % of the annual DOC load was exported from our sub-catchments during winter months, irrespective of geology. In the spring, 22–28 % of the annual DOC load was transported, with summer and autumn months together responsible for < 20 % of the total weight of DOC leaving each sub-catchment (Table 4). Winter was also an important season for nitrate export with between 45 and 66 % of the total annual nitrate load being exported. Spring export of nitrate was important in both Chalk and clay sub-catchments (ca. 30 % of annual load) and in the clay, autumn export of nitrate was also of comparable magnitude to spring (Table 4).

Our six sites exhibit a range of BFI (0.207–0.905) indicating a gradient from river water with 80–90 % groundwater contribution to total flow in the chalk geology, 70–80 % groundwater contribution in the Greensand, and only 20–55 % groundwater characteristic at the sites underlain by clay geology. Our calculation of BFI for the six sites, based on our 2-year discharge dataset, compared favourably with the BFI estimated from HOST (Gustard et al., 1992).

BFI and logEC–log⁡Q plots are useful complementary approaches to interpreting hydrological and hydrochemical pathways operating in the sub-catchment associated with each site. Electrical conductivity is an aggregated measure of geogenic solute response in the sub-catchment, and provides an indication of relative contributions of old groundwater (long residence time) and new (short residence time) water arising from routes such as shallow throughflow, preferential pathways and overland flow to the river. The study allowed the full range of flows at the sites to be sampled because two extreme conditions in the UK were captured: the summer drought of 2013 and the extremely wet winter of 2013–2014. In the Chalk, the logEC–log⁡Q plots show groundwater (old water) dominance during the period of flooding, because electrical conductivity is maintained through the entire range of flows, including at the highest discharge approaching 10 mm d-1. At the Greensand site, the sharp decline in electrical conductivity at discharges > 1.5 mm d-1 provides evidence of dilution of total dissolved solutes by new water, which occurs only during the wet winter of 2014. At the clay site, EC–Q relationships demonstrate that quickflow pathways, most likely involving preferential delivery enabled by field drainage (both agricultural and army camp drains from World War II) installed due to the risk of seasonal waterlogging on the slowly permeable local clay soils (Denchworth and Wickham soil series), are operational throughout autumn, winter, and spring months. Under summer baseflow conditions, the field drains are inactive and any river flow (almost negligible during the summer drought of 2013) is provided by springs draining the aquifers of the Upper Greensand and Wardour Formation (Allen et al., 2014), or direct discharges from septic tanks, and drains connecting farm yards to the stream.

Average annual nitrate concentrations in surface waters of the Hampshire Avon catchment increase with increasing BFI. In a UK-wide study, Davies and Neal (2007) used linear regression to consider how catchment characteristics control mean nitrate concentrations in UK rivers. Nitrate concentrations were explained by land use (% arable and % urban), topography (expressed as % upland), effective rainfall (mm) and BFI. Therefore, on the basis of these prior national analyses, it would be predicted that % arable and BFI would be the most important explanatory factors. For the Hampshire Avon, stepwise regression analysis showed limited co-linearity between BFI and % arable, and forced entry regression indicated that BFI was the better explanatory variable for mean nitrate concentrations. In the UK, historical fertilizer applications have led to elevated concentrations of nitrate in both Chalk and Upper Greensand aquifers, currently in the range 500–645 µmol L-1 (Department for Environment Food & Rural Affairs, 2002; Burt et al., 2011; Howden et al., 2011; Wang et al., 2016). Although the Chalk aquifer of the Hampshire Avon has been designated as a groundwater nitrate vulnerable zone (NVZ) under the EU Nitrate Directive (Directive 2000/60/EC), the time taken for water to move from the soil surface, through the unsaturated zone to the aquifer can result in a decadal scale time-lag between implementation of management practice and any observed response in groundwater or river nitrate concentrations (Allen et al., 2014; Wang et al., 2012). We observe an increase in nitrate load in baseflow with increasing BFI (Chalk > Greensand > clay) in line with previous research by Tesoriero et al. (2013), and our riparian porewater samples indicate significantly higher nitrate concentrations in the soil C horizon of the Chalk sites in comparison to Greensand and clay sites. However, it is an over-simplification to suggest that the gradient of annual average nitrate concentrations with BFI can be explained solely by different ratios of nitrate-rich groundwater to relatively nitrate-poor quickflow components of the hydrograph over an annual cycle. If this were the case, then nitrate concentrations would be highly correlated with electrical conductivity, and they are not. Instead, our analysis suggests that additional N transformation processes, and exchange with other N species forms instream, driven by seasonality and varying land use and management contribute to the observed patterns that we see, and this is discussed below.

Our six sites provide evidence that average annual DOC concentrations decline with increasing BFI in the Hampshire Avon catchment. Unfortunately, the Environment Agency does not collect DOC data in the rivers of the Hampshire Avon region so we cannot investigate the wider applicability of the DOC trend. Wetland area is often cited as an important control on DOC concentrations in a catchment (Morel et al., 2009), but our sub-catchments all comprise < 0.6 % wetlands by area. Data from the Environment Agency indicate that groundwater concentrations of DOC in the catchment are generally < 83 µmol L-1. Porewater samples from the grassland riparian zone at each site show elevated DOC concentrations in comparison to regional groundwater, and the Chalk sites (high BFI) have significantly lower DOC concentrations in soil C horizons compared to the Greensand and clays, suggesting that soil type and underlying geology could influence the concentration at which DOC is delivered to the stream in these sub-catchments. Once again, DOC concentrations in the surface water cannot be explained by a mix of old and new water alone, and seasonality plays an important role in controlling the flux of DOC through river water.

The Chalk site (CE) is near-chemostatic with respect to total dissolved solutes and nitrate. This means that the absolute concentration of geogenic solutes and nitrate is maintained at higher discharge, so that discharge drives solute load and hence the export of solutes to the coast. Here we use the definition of near-chemostatic expressed in Godsey et al. (2009) as a slope of close to zero on a log⁡(C)–log⁡(Q) plot, where C is concentration and Q is discharge. It has been suggested that chemostatic behaviour for nutrients arises if sources accumulate in the landscape, e.g. as legacy of nitrate management (Basu et al., 2010). Here nitrate has accumulated in groundwater (Wang et al., 2016) and it is the dominance of this old water under high discharge that gives rise to the near-chemostatic effect and transport-limited system. DOC is also transport rather than supply limited at this site, showing a slight increase in concentration with increasing discharge, and a more pronounced increase in spring which is not associated with a rise in discharge. In fact all three sites – on Chalk, Greensand, and clay – have elevated DOC concentrations in spring, which could arise from production, leaching and export of DOC from catchment soils as soil temperatures rise (Aubert et al., 2013), and/or in-stream production.

At the Greensand site, there appears to be a threshold of discharge of ca. 1.5 mm d-1 in winter above which there is evidence of different hydrological flowpath(s) or sources of water to the river with lower electrical conductivity compared to other seasons. Riparian head is closely correlated with discharge and shows two distinct regions of linearity which converge at a discharge of between 1 and 1.5 mm d-1. At this threshold, riparian head is at 60–80 cm below the ground surface suggesting that the water table is at the base of the soil C horizon. As the water rises up through the soil horizons during the winter, the electrical conductivity in the river water drops indicating a supply of new water from soil in the riparian zone and potentially from the surrounding fields. Conceptualizations of solute transport from other researchers include differing contributions from near stream riparian areas with rising and falling groundwater, arising from a combination of soil solute concentration and near-stream lateral water flux (Prior and Johnes, 2002; Seibert et al., 2009), and/or increased connectivity and fraction of active catchment contributing water, with emphasis on the lateral dimension (Basu et al., 2010). Above the threshold of 1.5 mm d-1 the DOC and nitrate concentrations in the river reflect a combination of groundwater contribution and the depth-integrated mass flux of each solute from the soil A, B, and C horizons. The reason for a decline in nitrate concentrations in river water above the threshold, whilst DOC concentrations increase, can be ascribed to the different depth-distributions of nitrate and DOC pools in the soil. The extent of the lateral connectivity between surrounding fields, the riparian zone, and the river channel in these low-gradient, intermediate BFI systems is not well characterized, and should be an area of further study.

Our two clay sub-catchments are dominated by artificially drained soils of the Kimmeridge Clay Series, and the field under-drainage will be a major control on the hydrological and hydrochemical response of the river. This is evident in the rapid fall in electrical conductivity in response to rainfall events (Fig. S1 in the Supplement) and in the variation in electrical conductivity with season which arises from the mix of rapid (via drainflow) and slow pathways of water during storm events, and suggests that the drains operate through much of the year (spring, autumn and winter). Concentrations of DOC in the surface waters of the two clay sites (167–2000 µmol L-1) are comparable to the range reported in drainage waters from permanent grassland in South West England (Sandford et al., 2013). Increases in DOC concentrations in drainage water during rainfall events have previously been explained as being due to increased lateral flows through the upper soil horizons (Neff and Asner, 2001), which are generally relatively carbon enriched compared to lower soil horizons. Here, flushing of DOC from soil aggregates and subsurface micropores contributes to rising concentrations during storm events (Jardine et al., 1990; Chittleborough et al., 1992). Sandford et al. (2013) reported molar DOC : nitrate ratios of 18–25 at times of highest DOC export in drainage water (which is at the upper end of our observations for surface water of our clay catchment), and they also found that the molar DOC : nitrate ratio increased with discharge. The comparability of results suggests that our findings may have wider applicability to other catchments of mineral soils dominated by drained grassland.

The elevated concentrations of nitrate observed in the river Sem in Autumn 2013 (Fig. S2) provide some additional evidence to support results from dynamic modelling using INCA-N which show that drought conditions followed by wetting-up of soil (as predicted in future climate change scenarios) can give rise to high nitrate loads in rivers (Whitehead et al., 2006). However, we observed this flushing effect most markedly in the clay sub-catchment of the Hampshire Avon where the majority of nitrate is likely to be delivered rapidly to the stream through shallow subsurface pathways connected to topsoil, as opposed to the Chalk sub-catchments where groundwater contributions of nitrate dominate.

Here, we have shown that for our six tributaries of the Hampshire Avon, DOC : nitrate ratios are negatively correlated with BFI, but the relationship is non-linear. As far as we are aware, we are the first to demonstrate such a relationship, which, if more widely applicable to other lowland, agricultural catchments, might provide a useful means of predicting annual-averaged riverine nitrate and DOC concentrations.

The molar DOC : nitrate ratios fall in the lowest range recorded across multiple land use types in the US LINXII study (Mulholland et al., 2015), but vary over 2 orders of magnitude, suggesting order of magnitude variations in whole-stream nitrate uptake velocity in river reaches across our contrasting geologies (0.05–0.4 mm min-1; see Fig. 7 in Rodríguez-Cardona et al., 2016). Nitrate uptake velocity is the vertical movement of nitrate to the riverbed measured using the whole-stream “Tracer Additions as Spiraling Curve Characterization” method. The metric represents nitrate uptake efficiency, and can be interpreted as whole-stream nitrate removal through, for example, denitrification and/or assimilatory processes, although the method does not allow for discrimination of these processes. On the basis of the relationship between DOC : nitrate and BFI demonstrated in this study, we can hypothesize that the clay sub-catchments are associated with higher whole-stream nitrate removal than our Greensand and Chalk systems. Although we have no direct measurements of whole-stream nitrate removal for these sites, we have measured in situ rates of nitrate removal in the riverbed at these six sites using a modified push–pull technique (Jin et al., 2016), and the highest rates of nitrate removal were found at the two clay sites (see Table 4 in Jin et al., 2016). Whether DOC : nitrate ratios control nitrate removal may also depend on the net heterotrophic or autotrophic nature of our sub-catchments. In a net autotrophic reach, nitrate removal might correlate with physical factors such as light and temperature, which control photosynthetic activity, and hence the in-stream production of labile carbon which, in turn, is then tightly coupled to nitrate reduction. In contrast, in a net heterotrophic reach in our lowland, arable landscape, nitrate removal may depend on DOC : nitrate ratios driven by hydrological pathways delivering labile dissolved organic and inorganic carbon (Rodriguez-Cardona et al., 2016).

This study, amongst others, has revealed significant differences in the relationship between DOC : nitrate and discharge dependent on both geology and seasonal effects (Tiemeyer and Kahle, 2014; Thomas et al., 2016). The Chalk site exhibited little variation in DOC : nitrate with discharge due to the dominance of groundwater contribution at both high and low flows. At the Greensand site, there is a linear increase in DOC : nitrate with discharge irrespective of season. However, during the elevated flows in the winter, when riparian and rain water contributes increasingly to the discharge, causing a drop in electrical conductivity, a sharp change in nitrate and DOC concentration is observed resulting in an overall drop in DOC : nitrate during a time when > 66 % of the total nitrate export occurs. In contrast at the clay site, lowest DOC : nitrate values and highest nitrate concentrations are associated with autumn storms of intermediate discharge, which export 26 % of total annual nitrate load. These trends highlight contrasting seasons of risk associated with high nitrate export in combination with low DOC : nitrate ratios at the Greensand and clay sites. Our research gives added impetus to the need to control autumn run-off from drained, grassland catchments supporting intensive livestock farming. Our study also suggests that during winter, periods of lateral flow and over-bank flooding in areas of intermediate BFI, such as Greensand, may export a significant proportion of the annual nitrate load with little opportunity for in-stream nitrate processing or removal.