UV nitrate sensors have been used to monitor nitrate-nitrogen concentrations in surface water, springs, wastewater and drinking water supplies (Pellerin et al., 2012; Miller et al., 2016, 2017; Burkitt et al., 2017; Burns et al., 2019; Hansen and Singh, 2018; Shi et al., 2022). More recently they have been applied to in situ groundwater monitoring studies to elucidate temporal variability (Opsahl et al., 2017; Burbery et al., 2021; Haas et al., 2023; Husic et al., 2023). The use of UV nitrate sensors allows for reduced maintenance and calibration compared with methods that use reagents, while greatly increasing measurement frequency compared with discrete sampling programmes (Pellerin et al., 2013).

UV nitrate sensors measure the transmittance of light across a known path length (Pellerin et al., 2013; Huebsch et al., 2015). The amount of light transmitted is related logarithmically to the absorbance characteristics of the groundwater through the Beer–Lambert law (Huebsch et al., 2015). The Beer–Lambert law can be used to determine the concentration of nitrate nitrogen in groundwater based on the amount of light at a specific wavelength transmitted from the light source to the detector and processor (Pellerin et al., 2013). However, water turbidity and coloured dissolved organic matter (CDOM) may also absorb or scatter light at similar wavelengths to nitrate nitrogen and these concentrations also need to be accounted for as interference effects (Snazelle, 2016). Thus, UV nitrate sensors also measure the transmittance of light at wavelengths for common substances that cause interference and use algorithms to adjust a proportion of the absorbance rates to the nitrate-nitrogen concentration (Snazelle, 2016).

Using high-frequency UV nitrate sensors to monitor nitrate nitrogen in groundwater can consequently provide new insights into sources, processes and pathways that are temporally variable. As the mechanisms that control inorganic nitrogen movement are also highly variable across catchments, high-frequency monitoring can be applied to characterise a range of settings to gain new insights into inorganic nitrogen export through groundwater pathways (Liu et al., 2020; Burbery et al., 2021; Haas et al., 2023; Husic et al., 2023). Such temporally dense data provided by in situ UV nitrate sensors can then be used as the basis for time series analysis techniques traditionally used in surface water studies that have not previously been applied in groundwater quality analysis.

Chemographs, for example, plot concentration (y axis) over time (x axis) and are a useful tool for understanding pulsing and dilution behaviour, whereas concentration (y axis) over discharge (x axis) plots reveal the timing and phasing of mobilisation from local or distal sources (Evans and Davies, 1998; White, 2002; Dupas et al., 2016; Liu et al., 2021). These concentration–discharge plots often present as either concordant phasing between concentration and water discharge (i.e. a linear or non-linear trend), or as discordant phasing, whereby concentration varies with ascending or descending patterns in discharge. The latter may present as hysteretic curves that describe the relationship between concentration and discharge during discrete events (Evans and Davies, 1998).

Hysteresis can be described as source- or transport-limited behaviour. A clockwise loop describes where the source (or pathway length) of the solute is located nearby or is rapidly transported. Conversely, an anticlockwise loop indicates a distant source with a delay in solute transport to the monitoring site (Liu et al., 2021). In hydrological research, hysteresis analysis is traditionally applied to streams and rivers, where discharge measurements are more easily obtained (Lloyd et al., 2016; Vaughan et al., 2017; Baker and Showers, 2019). Where rainfall recharge is the driver of changes in groundwater levels and nitrate-nitrogen concentrations, hysteresis analysis can be used to identify the lag between these two variables following a recharge event. While several hysteresis studies have considered the effect of near stream groundwater on stream-derived hysteresis curves, high-frequency groundwater measurements have not been widely utilised (Aubert et al., 2013; Jacobs et al., 2018; Knapp et al., 2020; Winter et al., 2021; Gelmini et al., 2022).

Antecedent conditions play an important role in explaining the response of nitrate-nitrogen concentrations in groundwater after rainfall. The flushing of inorganic nitrogen after periods of drought can result in higher concentrations in groundwater from additional storage being mobilised in soil or the vadose zone (Van Metre et al., 2016; Opsahl et al., 2017; Leitner et al., 2020; Jutglar et al., 2021; Winter et al., 2023). Conversely, a more muted nitrate-nitrogen response may occur after consecutive recharge events temporarily exhaust the supply of nitrate nitrogen (Wang et al., 2020; Yue et al., 2023). As a result, each recharge event has a distinctive nitrate-nitrogen response due to antecedent conditions. Hysteresis analysis in groundwater (i.e. examining concentration versus groundwater level) may help to describe the temporal variability of these recharge events.

The objectives of this study are to quantify event-based nitrate-nitrogen dynamics from a forestry-dominated land use as the principal source of inorganic nitrogen in groundwater. The study uses 4 years of high-frequency UV nitrate sensor data at a case study location in Waitaha / Canterbury, Aotearoa / New Zealand. We investigate how high-frequency measurements of nitrate nitrogen and groundwater levels, along with the analysis techniques that these measurements enable, can help elucidate nitrate-nitrogen dynamics from source to discharge point. The specific analysis techniques that this study utilises include hysteresis analysis of high-frequency nitrate-nitrogen concentrations and groundwater levels to understand how nitrate-nitrogen stores and pathways change during rainfall events, seasonally and in response to broader variations in climate conditions. These data will also be used to refine estimates of nitrate-nitrogen loads and leaching by improving the understanding of temporal dynamics of groundwater nitrate-nitrogen discharges to riverine systems to inform freshwater management strategies.

The Culverden Basin of Waitaha / Canterbury, Aotearoa / New Zealand, is located in a transitional area between the continental collision (transform fault) of the Australian and Pacific plates (Alpine Fault) to the west and the Hikurangi subduction zone to the east (Armstrong, 2000; Barrell and Townsend, 2013; Brough, 2019). The rapid uplift from the compression of the Australian and Pacific plate collision (Early Miocene) resulted in large volumes of sediment eroding and subsequently infilling the developing basins (Armstrong, 2000). These Quaternary gravel deposits have created aquifers that are the main source of groundwater in the Culverden Basin and surrounding area (Poulsen, 2012).

Groundwater occurs in Quaternary age glacial outwash gravels (Burnham formation) from the Hurunui and Waiau Uwha Rivers (Poulsen, 2012) that are generally less than 50 m thick (Armstrong, 2000). The alpine rivers in the Culverden Basin (Hurunui and Waiau Uwha) are incising the previously deposited alluvial fans as a result of continued uplift and lower rates of sediment deposition in post-glacial riverine environments (Armstrong, 2000). Due to the incising alpine rivers, groundwater in the surrounding alluvial terraces is often at a higher elevation than the height of these rivers. As a result, groundwater and land use are highly connected to the rivers that flow through the Culverden Basin, with groundwater vulnerable to nitrate-nitrogen accumulation in areas of limited recharge (Abraham and Hanson, 2006; Knottenbelt, 2023).

Mean annual precipitation in the Culverden Basin is 576 mmyr-1 (1981–2010), with nearly half (46 %) of rainfall over that period falling in winter months (Macara, 2016). On average, the Culverden Basin experiences 99 d of rainfall per year (more than 0.1 mm) and 82 d with more than 1 mm of rainfall falling (Macara, 2016). During summer months, potential evapotranspiration can be high, with an annual average (Penman PET) of 845 mm from 2018–2024 (NIWA, 2024). This indicates there can be a soil moisture deficit during summer months.

A 24 m deep well was installed downgradient and adjacent to a 25 km2 area of Pinus radiata forest in the Culverden Basin, approximately 1.5 km from the true left bank of the Hurunui River (172.792, -42.844, WGS 84). The drill core showed that the local stratigraphy is composed of a very thin layer of silty pallic soil (∼0.30m) with distinct alluvial deposits visible at the surface. Pallic soils in this region are known for their low water holding capacity, tendency to crack during summer and low carbon content relative to other soil types in Aotearoa / New Zealand (Hewitt et al., 2021). As such, these soils are recognised as being physiographically poor attenuators of inorganic nitrogen and tend to promote the leaching of excess nutrients (Hewitt et al., 2021). The thin stony soil and highly permeable gravel aquifer at the site results in rapid changes to groundwater conditions after rainfall. Immediately upgradient is an area of active pine (Pinus radiata) forestry (Fig. 1). During the study period (2020 to the end of 2023), sections of pine forest were logged and cleared. Through satellite imagery, it was observed that in areas where logging had occurred there was no significant plant regrowth. Downgradient of the site is an area that has been converted from forestry to irrigated beef. The land use of the broader region is dominated by pastoral agriculture, particularly beef and dairy grazing.

A TriOS NICO UV nitrate sensor was installed to monitor nitrate-nitrogen concentrations in groundwater at a depth of 20 m, which is below the lowest groundwater level. Results were reported as nitrate nitrogen in milligrams per litre. The path length was set at 5 mm to account for the wide range of nitrate-nitrogen concentrations observed at the site. However, the path length and associated accuracy (approximately ±5%+0.5mgL-1) of the UV nitrate sensor cannot be optimised for all observed nitrate-nitrogen concentrations and, as a result, measurements of nitrate nitrogen can be overestimated. The high-frequency measurements were recorded at 15 min intervals with a 212 nm wavelength from a xenon light source. Interference absorption was also measured at 254 nm for organic compounds and at 360 nm for turbidity. The TriOS NICO also derives and outputs the sensor quality index (SQI), a rating of the measurement quality used for quality assurance. Due to the diameter of the well, the TriOS NICO UV nitrate sensor was vertically suspended. A Zebratech wiper was fitted to the TriOS NICO to reduce interference from turbidity settling on the lens or biofouling.

Groundwater levels were measured at 15 min intervals through a float and weight system (Unidata 6541). Rainfall volume was also measured at 15 min intervals by a tipping bucket rain gauge (Davis AeroCone Rain Collector). Additionally, a 12 V submersible pump (Proactive Super Twister) was permanently installed in the well to improve access to discrete groundwater quality sampling.

Nitrate-nitrogen standards were periodically (3-monthly) used to check the precision of the TriOS NICO at different concentrations. The accuracy of the UV nitrate sensor measurements was also compared with laboratory tested validation measurements of discrete nitrate-nitrogen samples (Fig. 2). Validation samples also included parameters that could cause interference, such as dissolved organic carbon and turbidity. Regular quarterly groundwater monitoring also occurs at the site, and involves a full suite of parameters including metals, nutrients and bacteria.

The UV nitrate sensor measurements were compared with discrete validation samples to check the accuracy of the UV nitrate sensor (Fig. 3). The coefficient of determination (R2) value of 0.93 indicated a strong regression line of best fit and that both the discrete and high-frequency measurements are similar (n=34). The best fit between UV nitrate sensor measurements and validation samples occurs when nitrate-nitrogen concentrations are less than 5 mgL-1. Figure 3 indicates that the UV nitrate sensor can overestimate (y=1.13x±0.058) high concentrations of nitrate nitrogen (i.e. the difference between discrete and high-frequency data).

The post processing of data from the UV nitrate sensor involved outlier and step change detection as well as gap-filling when data were not recorded. Outliers were identified by comparing high-frequency and discrete samples. We adjusted the UV nitrate sensor data where the difference between high-frequency data and discrete sample readings was greater than the accuracy of the UV nitrate sensor. Step changes due to interference effects and no measurements due to loss of signal or an error were also identified. Linear interpolation was used to fill gaps in data from outliers, step changes and no measurements.

A 1-D land surface recharge model was calibrated using nearby lysimeter data to determine when rainfall recharge occurred. The parameters used in the land surface recharge model included precipitation and Penman PET data sourced from a nearby weather station (Fig. 1). Soil water capacity (86.0 mm), an evaporation reduction function (10.0) and a drainage threshold (50.0) were derived from the nearby lysimeter (Fig. 1) (Bidwell and Burbery, 2011), while the crop factor (0.77) was sourced from Allen et al. (1998). In this model, rainfall recharge occurs when rainfall is above the soil water storage capacity. Evaporation, plant uptake and runoff are factors that reduce soil water storage. River recharge was not considered to be a strong control on groundwater levels in the model because of the site's mean groundwater level (173 metres above sea level (m a.s.l.)) compared with the elevation of the Hurunui River (162 m a.s.l.). The abstraction of groundwater was also not considered when interpreting groundwater levels as there are no nearby wells that extract groundwater.

We investigated nitrate-nitrogen concentrations and flow data in the Hurunui River to understand the impact of the nitrate-nitrogen load from groundwater. The flow of the Hurunui River was measured at the State Highway One bridge site (Fig. 1) using a rating curve to relate gauged flow and water level stage height. The stage height was measured at 5 min intervals, from which the mean was calculated and reported as a daily flow average. The high-frequency nitrate-nitrogen concentration measurements in the Hurunui River were obtained using a TriOS NICO, at the State Highway One bridge, approximately 30 km downstream of the groundwater monitoring site (Fig. 1). The depth of the UV nitrate sensor in the Hurunui River was adjusted to account for summer and winter season flows. The UV nitrate sensor measured concentration of nitrate nitrogen at 15 min intervals, and was calibrated using monthly validation samples. Due to this UV nitrate sensor being in a braided river, the sensor experienced additional interference effects and required more maintenance. There were periods where no data were available because the sensor was out of the water for calibration or repairs or the data were discarded because the SQI and discrete samples indicated the data were not accurate.

Regional groundwater levels across Waitaha / Canterbury were analysed as an indicator of climate variation (rainfall recharge) over the study period. The purpose of this analysis is to show the climate variation that controlled rainfall recharge over the study period. Monitoring wells with at least 10 years of groundwater levels in the Waitaha / Canterbury region were used to show monthly groundwater level percentiles over time. Groundwater levels were categorised as very low (bottom 10 %), low (10 %–25 %), average (25 % to 75 %), high (75 %–90 %) or very high (top 10 %) and then aggregated to determine the number of Waitaha / Canterbury wells in these percentiles for each month.

Increases in nitrate nitrogen and the associated groundwater level were graphed for each event between 2020 and the end of 2023. To determine when hysteresis analysis should occur, we considered the accuracy of the UV nitrate sensor and groundwater levels as well as the variation in baseline conditions. For these reasons we only analysed events with an increase in nitrate nitrogen over 1 mgL-1 (25 events). To visually show the rate of change in nitrate nitrogen and groundwater levels, the hysteresis curves were split into five equal sections, each represented by a different colour on the graph. The hysteresis curves were categorised as having hysteresis (or not) and by the direction of hysteresis. As multiple recharge events often occurred consecutively, the nitrate-nitrogen concentration or groundwater level did not always recover to pre-event conditions. Therefore, the next event began at the lowest nitrate-nitrogen concentration after the previous event.

We used the Hysteresis, Area, Residual and Peak (HARP) analysis method in R (R Core Team, 2021; Roberts et al., 2023) to compare metrics on hysteresis curve area, residual (difference in concentration between the rising and falling limbs) and the proportion of time into the event that peak groundwater level and nitrate-nitrogen concentration were observed in 2020 (15 events) and 2021 (10 events). The analysis in R included truncating data to form a closed loop and the normalisation of groundwater levels and nitrate-nitrogen concentrations. The hysteresis area was represented as a decimal fraction between 0 and 1, with larger values indicating more hysteresis. The times to peak nitrate nitrogen or groundwater level are percentages that indicate the times along the hysteresis curve (or event) where the maximum values occur (Roberts et al., 2023). The residual analysis was used to determine if the environment shifted to new baseline concentrations after recharge events. The results of the residual analysis were not analysed further due to nitrate-nitrogen concentrations returning to pre-event conditions. Recharge events were not analysed using HARP in 2022 (6 events) and 2023 (2 events) as they did not present as hysteresis curves. The R script used to process the hysteresis data is available on Zenodo (Roberts, 2023).

The hysteresis type analysis identifies the storage, mobilisation and flow pathways of nitrate nitrogen in the catchment that contribute to the observed concentrations at the monitoring well (Evans and Davies, 1998; Knapp et al., 2020 Liu et al., 2021). Clockwise hysteresis curves indicate that the source of nitrate nitrogen is nearby (short pathway), rapidly mobilised and progressively depleted along the hysteresis curve (Husic et al., 2023). In contrast, anticlockwise hysteresis curves represent longer flow pathways and less immediately mobile stores of nitrate nitrogen, where the groundwater level change is initially greater (Roberts et al., 2023). Complex or “figure of eight” hysteresis curves indicate that the source or flow path of nitrate-nitrogen changes during the event (Baker and Showers, 2019; Liu et al., 2021). The area of the hysteresis curve is a metric that can be used to compare events. A larger area shows a greater difference in the rate of change between nitrate nitrogen and groundwater levels (Roberts et al., 2023).

The annual nitrate-nitrogen load from groundwater to the Hurunui River between 2020 and 2023 at the site was calculated using trapezoidal integration of the high-frequency UV nitrate sensor data (Richards, 1998). The timing of the nitrate-nitrogen load was analysed and compared with flow as well as nitrate-nitrogen concentrations in the Hurunui River. The groundwater discharge component was calculated using Darcy's law: 1Q=KiA, where Q is groundwater flow (m315min-1), K is hydraulic conductivity (m15min-1), i is hydraulic gradient and A is cross sectional area of groundwater flow (m2). We modified the hydraulic conductivity units to be metres per 15 min to align with the recorded nitrate-nitrogen data interval. The saturated thickness of the aquifer was calculated using the high-frequency depth to groundwater (highest elevation) and the elevation of the Hurunui River (base). An upward gradient was also observed in nearby deeper wells, further supporting the sub-horizontal flow of nutrients to the Hurunui River. The hydraulic gradient was estimated using data from previous studies (Neil Thomas and Bas Veendrick, unpublished) in the area. The hydraulic gradient was varied in the time series for increasing groundwater levels (highest), decreasing groundwater levels and stable groundwater levels (lowest).

To calculate the nitrate load from the area upgradient of the monitoring borehole, the scipy.integrate Python subpackage (Virtanen et al., 2020) was used to calculate the area under the curve of the high-frequency UV nitrate sensor data. Trapezoidal integration was used to approximate the area because of the UV nitrate sensor data characteristics, including a high number of linearly spaced intervals, lack of symmetry and data smoothness. The area under the curve for the high-frequency nitrate-nitrogen data was calculated to give 15 min intervals of nitrate nitrogen in milligrams per litre (mgL-115min-1).

To calculate the volume of water moving in a horizontal direction from the aquifer to the Hurunui River, a time-variable Darcy's law equation was used at 15 min intervals. The hydraulic gradient and saturated thickness were varied based on the high-frequency groundwater levels. Due to the upward gradient in the adjacent deep well (screened at 98–101 m b.g.l.), we assumed that a high proportion of the shallow groundwater moves horizontally, discharging to the Hurunui River. Therefore, the application of Darcy's law was deemed to be appropriate for this scenario. As the hydraulic conductivity of the aquifer is heterogeneous and the gradient also varies, we used a range of potential hydraulic conductivity (Freeze and Cherry, 1979) and hydraulic gradient values to understand variation in the load calculation. This resulted in a 25 % increase or decrease in exported nitrate nitrogen (Table 2).

The calculated volume of groundwater discharged from the aquifer at 15 min intervals was multiplied by the concentration of nitrate nitrogen per 15 min interval and summed to give the total annual load of nitrate nitrogen: 2Load=K∑i=1nciqi12ti+1-ti-1, where K is a unit conversion constant, i is the sample in the ith position, ci is concentration (mgL-115min-1), qi is discharge (m315min-1) and t is the time period for the ith sample. Statistics on the timing of the nitrate-nitrogen flux were calculated by integrating specific time periods and comparing values with the total exported load.

The observed pulses of nitrate nitrogen during the study period reflect complex interactions between forestry harvesting and hydrological conditions. To understand the drivers of nitrate leaching and storage, we examined the timing of forestry harvesting and patterns of rainfall recharge. There was a strong relationship between prior forestry harvesting and rainfall recharge driving nitrate-nitrogen mobilisation. In contrast, low nitrate-nitrogen concentrations were consistently observed during dry periods, aligning with phases of nitrate-nitrogen storage and low groundwater levels. These findings suggest that both land use changes and hydrological variability play a critical role in understanding pulses of nitrate nitrogen, which we explore in detail next.

The highest concentrations of nitrate nitrogen in groundwater were observed during winter months (June–August), coinciding with the majority of rainfall recharge. Nitrate-nitrogen concentrations showed a strong flushing response to rainfall recharge (Fig. 6) that resulted in elevated concentrations for short periods of time. The average duration of elevated nitrate-nitrogen concentrations was 8 d (n=33), before returning to baseline conditions. When rainfall recharge or increases in groundwater levels did not occur, nitrate-nitrogen concentrations in groundwater were below 1 mgL-1. Apart from winter months, these conditions were present for most of the study period.

Due to low prior rainfall, groundwater levels were the lowest during the study period in 2020 and continued to be low until June 2021, when a large storm event (100 mm of rainfall over 2 d) resulted in significant rainfall recharge (Fig. 6c). After June 2021, groundwater levels continued to increase in response to successive rainfall recharge events. Outside of winter months, rainfall recharge also occurred in November 2020 and February 2022. For November 2020, the highest nitrate-nitrogen concentration was 5.9 mgL-1 because of rainfall recharge. In February 2022, however, the highest observed nitrate-nitrogen concentration was 12 mgL-1; this coincided with prior forestry harvesting and intense rainfall.

Nitrate-nitrogen concentrations in the Hurunui River showed concordant increases with flow and rainfall. Delayed peaks of nitrate-nitrogen concentrations in the Hurunui River were also observed after rainfall events. These delayed peaks of elevated nitrate-nitrogen concentrations occurred during base flow conditions. Data were missing in the Hurunui River nitrate-nitrogen time series due to maintenance and interference effects.

We found that the relationship between nitrate-nitrogen and groundwater levels varied between years based on the number and magnitude of prior recharge events and interannual groundwater level trends (Figs. 7 and 8). Large rainfall recharge events resulted in clockwise (source-limited) hysteresis curves and higher concentrations of observed nitrate nitrogen. In 2020, there were periods of receding groundwater levels that resulted in increasing nitrate nitrogen with declining groundwater levels after recharge (Fig. 7). For the last event at the end of winter in 2021, there was a positive linear relationship between increases and decreases in groundwater level and nitrate nitrogen (Fig. 8). In 2022 and 2023 all recharge events produced clockwise loops that were not complete due to an increasing trend in groundwater levels. All graphs showing the relationship between groundwater levels and nitrate-nitrogen concentrations are available in Fig. S1 in the supplement.

HARP analysis indicated that the average hysteresis curve area of first flush events was larger, at 0.65, compared with an average of 0.35 for subsequent events. The average proportion of time to reach peak groundwater level and nitrate-nitrogen concentration during events was lower in 2021, at 50 % and 55 % of events, respectively, compared with 2020, at 77 % and 70 % (Fig. 9). We compared the correlation (Pearson's r) of rainfall volume 10 d prior to an event and time to peak nitrate-nitrogen concentrations and groundwater levels between 2020 and 2021. In 2021, there was a moderately strong negative relationship between 10 d prior rainfall volume and peak nitrate nitrogen (-0.65) and peak groundwater level (-0.61), indicating that peaks occurred earlier in the event for larger rainfall events. There was no significant relationship between nitrate-nitrogen concentrations and prior rainfall volumes in 2020.

Pulses of high nitrate-nitrogen concentrations were observed in groundwater after rainfall recharge in winter months. The highest concentrations of nitrate nitrogen (up to 26 mgL-1) in groundwater occurred in winter months (June–August), with the winter median concentration over 4 years generally between 2.4 mgL-1 and 5.7 mgL-1 (Fig. 6a). In late June the daily median concentrations were between 11.3 and 18 mgL-1. During summer, autumn and spring months, daily median nitrate-nitrogen concentrations were generally below 2.4 mgL-1.

The nitrate-nitrogen load from groundwater was found to be highest during 2020, with an estimated 7.25 tyr-1 exported to the Hurunui River (Table 2). There were higher-intensity rainfall recharge events in 2021 and there was more rainfall recharge (265 mm) but the nitrate nitrogen exported was slightly less than 2020, at 6.91 tyr-1. There was less nitrate nitrogen exported from groundwater to the Hurunui River in 2022 and 2023, at 5.97 and 6.88 tyr-1, respectively. We also calculated the exported load using the annual and quarterly measurements, which over- (or under-) estimated the load, depending on whether they measured the winter pulses of nitrate nitrogen (Fig. 2 and Table 3).

The majority of nitrate-nitrogen leaching occurred immediately after rainfall recharge for short time periods. In 2020, 50 % of the total nitrate nitrogen that was leached occurred during 13 % of the year as a result of the observed nitrate-nitrogen spikes after rainfall. In 2021, the rapid leaching of nitrate nitrogen was also observed as 80 % of the total nitrate nitrogen was exported during the heavy rainfall events in winter months, which accounted for 25 % of 2021. Short pulses of leaching were also observed in 2023. Due to the short duration of high nitrate-nitrogen concentrations in 2022, only 10 % of the total nitrate nitrogen was leached during each observed pulse. As a result, there was more nitrate nitrogen exported outside of the observed pulses in 2022.

The presented results show that nitrate-nitrogen concentrations in groundwater varied in accordance with seasonal, climatic and disturbance-based changes in nutrient availability. We observed large pulses of nitrate nitrogen in groundwater during every winter period (June to August) of the study, with a maximum concentration of 26 mgL-1. In winter biological activity decreases, reducing nutrient uptake; this in turn increases the labile pool of excess nitrate nitrogen (Gundersen et al., 2006; Davis, 2014). We infer that the reduced nutrient demand, increased rainfall and saturated soil conditions induced higher nutrient losses from the forest soils during winter (Bechtold et al., 2003; Sebestyen et al., 2014).

The decrease in forestry cover over the study period was a driver of high nitrate-nitrogen concentrations (Rosén and Lundmark-Thelin, 1987; Sebestyen et al., 2014). Notable harvesting occurred in October 2021 and in January 2022 (Fig. 4). After forestry harvesting, we observed a larger pulse of nitrate nitrogen outside of winter months. In February 2022 (directly after the forestry harvesting), the maximum concentration was 12 mgL-1, compared with 5.9 mgL-1 in November 2020, when prior forestry harvesting did not occur. The larger volume of rainfall (75 mm) during February 2022, compared with November 2020 (35 mm), was also a factor in the higher observed nitrate-nitrogen concentration. This indicates that there was more nitrate nitrogen available after harvesting practices (Bauhus and Bartsch, 1995). There was also more nitrate nitrogen available during first flush events (higher nitrate-nitrogen concentrations) at the start of winter, which were exacerbated in 2021 by the low-recharge conditions observed in 2020 and early 2021 (Fig. 5). Nitrate-nitrogen leaching continued after harvesting because of windrowed debris and the limited regrowth of plants, after harvesting, that could uptake nutrients (Likens et al., 1970; Slesak et al., 2009; Devine et al., 2012).

We identified that the relative importance of biogeochemical processes, such as the mineralisation and immobilisation of nitrogen, shifted in favour of the hydrological mobilisation of nitrogen after the intense rainfall event in May 2021. We infer that the catchment's ability to uptake nitrogen was reduced during this dry period (Winter et al., 2023) and was then continually disrupted due to forestry harvesting and the suppressed regrowth of understorey plants (Green et al., 2023). These conditions led to higher-than-expected nitrate-nitrogen concentrations during rainfall recharge (Green et al., 2023). The results observed in this study highlight the temporal variation in biogeochemical processing rates in response to forestry harvesting and subsequent nutrient losses driven by hydrological conditions.

The recent application of UV nitrate sensors to in situ groundwater poses challenges when comparing results with other studies with lower-frequency measurements. At the study site, there is unlikely to be significant denitrification due to the thin low-carbon pallic soils and gravel aquifers (Wilson et al., 2020). Therefore, we expect nitrate-nitrogen concentrations to be higher than for other studies reporting stream concentrations that have longer hydrological flow paths leading to increased denitrification potential and stream uptake (Table 1). Other studies measuring inorganic nitrogen concentrations in drainage water generally report higher concentrations (Table 1), but it is unclear whether peak concentrations were captured using lower temporal resolution techniques. This study highlights the benefits of high-frequency measurements in settings where hydrological events drive rapid changes in nitrate-nitrogen concentrations and groundwater levels.

We used a novel approach of comparing high-frequency groundwater levels and nitrate-nitrogen concentrations to understand event-scale responses to recharge. Hysteresis events in 2021 (Fig. 8) generally showed initial rapid increases in concentration (clockwise loop), indicating local leaching sources that were rapidly mobilised, while in 2020 groundwater was less hydrologically connected to recharge, resulting in more complex hysteresis loops as a result of slower drainage pathways (Fig. 7). Across both years, the initial winter recharge event had a higher average area (0.65) compared with subsequent events (0.35). This finding indicates that stores of nitrate nitrogen were initially rapidly mobilised but subsequent events resulted in diluted or depleted stores of nitrate nitrogen.

We observed nitrate-nitrogen concentration increases shifting from surface-based recharge to groundwater levels, mobilising vadose zone storage. This was evident in event f (Fig. 8) at the end of winter in 2021 as there was a linear relationship between increasing groundwater levels and nitrate-nitrogen concentrations. We think this occurred due to successive antecedent rainfall recharge events over a short period of time flushing and exhausting soil storage. Mobilisation then occurred in the vadose zone as groundwater levels rose, intercepting slower drainage or stored pore water with higher nitrate-nitrogen concentrations from previous events (Ascott et al., 2017; Burbery et al., 2021). Several other hysteresis curves after the June 2021 recharge event (near the end of winter) also exhibited limited hysteresis. This indicates that the dual components of soil storage and vadose zone storage were mobilised during these events.

In concentration–discharge studies, the input of near-stream groundwater is recognised (Knapp et al., 2020; Winter et al., 2021; Gelmini et al., 2022) but the timing and variation of groundwater solute discharges is typically unknown. The novel hysteresis analysis shows that there was significant inter- and intra-annual variation between recharge events. These findings can be a valuable addition to improve hysteresis analysis in riverine environments, where groundwater is a significant contributor to baseflows. Ideally, future riverine hysteresis studies would be conceptualised to consider groundwater hysteresis, so that high-frequency groundwater concentrations and water levels are available.

Groundwater concentration–groundwater level analysis can provide future riverine studies with more information on proximal (clockwise) or distal (anticlockwise) groundwater sources of nitrate nitrogen. As well as the source of nitrate nitrogen, the timing of greatest surface water–groundwater connection (highest groundwater level) compared with highest nitrate-nitrogen concentration could be investigated. This study highlights that the nitrate-nitrogen concentration in groundwater can be just as varied as stream concentrations at the event scale and is a valuable addition in understanding riverine hysteresis processes.

High-frequency monitoring indicates that there are distinct pulses of nitrate nitrogen to the Hurunui River from forestry land use (Fig. 6). The distinct surface runoff and groundwater discharge pathways result in different timings of nitrate nitrogen in the Hurunui River. The surface runoff component appears to occur simultaneously with high flow volumes in the Hurunui River, while the lag associated with the groundwater travel time indicates that the groundwater discharge generally occurs when there are base flow conditions in the Hurunui River (Fig. 6).

The hysteresis analysis and exported load estimates suggest that nitrate-nitrogen stores were small and quickly depleted during recharge events. The largest nitrate-nitrogen export observed in 2020 resulted from several factors: the shortest time period since the last forestry disturbance in 2018, preceding dry conditions (Winter et al., 2023) and repeated wetting and recharge events (Anaya et al., 2006). In contrast, subsequent years saw larger but less frequent rainfall events, leading to different export characteristics (Table 2). Annual load estimates were comparable to other studies estimating the nutrient loads from forestry land use, despite the high concentrations we observed (Di and Cameron, 2002; Gundersen et al., 2006; Davis, 2014). However, when analysing nutrient concentrations in the Hurunui River, the effects of forestry-related nutrient losses cannot be isolated from those of surrounding land uses.

The high-frequency UV nitrate sensor data and integration method were better at constraining the nitrate-nitrogen concentration and therefore the load. Using discrete annual and quarterly values did not accurately estimate the nitrate-nitrogen load, depending on the timing of the sample (Table 3). The results from this study highlight the importance of choosing an appropriate monitoring frequency and statistical description that can describe pulses of nitrate nitrogen. There are still some weaknesses in this method of estimating the nitrate-nitrogen load as the volume of groundwater that was discharging to surface water was uncertain. Additionally, the variation in nitrate-nitrogen concentration across the aquifer thickness was also not measured. Multiple UV nitrate sensors at different depths or multiple pumps with a flow cell would improve this information.

The data from the high-frequency UV nitrate sensor provide important information on the timing of leaching and improvements in constraining nitrate-nitrogen loads. For regulators that aim to improve water quality, these findings provide more understanding of the effects of changing land use and allow more targeted approaches to policy as there is greater information available on the timing and causes of nutrient losses. The strong hydrological controls on the leaching of nitrate nitrogen can be applied to future climate change projections. With increased periods of droughts and intensity of storm events, we can expect to see more pulses of nitrate nitrogen in the future, which may exacerbate existing water quality stressors on water resources (Srinivasan et al., 2021).