The study area centres around the small town of Tinwald (population 3000) situated on the south bank of the Hakatere/Ashburton River and located on the large coalescing alluvial plain known as the “Canterbury Plains” (Fig. 1a). The Canterbury Plains were built up by rivers fed by glaciers over several million years. Deposition in the Tinwald area (Fig. 1b) was mainly by the South Branch Hakatere/Ashburton River and its ancestors (Barrell et al., 1996). The alluvial deposits are poorly stratified greywacke gravel dominated with silts and sands which become finer towards the coast. Oil well exploration drilling and seismic surveys of the Ashburton–Hinds areas indicate thicknesses of over 1000 m of alluvial gravels overlying marine sediments (Jongens et al., 2012).

Existing wells in the Tinwald study area are almost all less than 100 m, and over half are less than 40 m deep (Aitchison-Earl, 2019). Wells are generally screened within post-glacial (Holocene) or last-glacial (late Quaternary) deposits. Shallow wells and springs are common close to the river, within the Holocene age deposits. There is little geological impedance for movement of groundwater between shallower and deeper screened wells.

The regional groundwater flow direction is parallel to the Hakatere/Ashburton River. State Highway 1 runs through the study area (Fig. 1b) and was originally built to take advantage of drier conditions at the inland point of the old “Hinds swamp”. The swamp has been largely drained but influences soil types, with deeper, poorly drained organic soils with less leaching and greater denitrification potential located coastwards of the highway. Soils are lighter and more freely draining with greater nitrate leaching risk inland of the highway and adjacent to the Hakatere/Ashburton River (Landcare Research, 2015).

In the Tinwald study area, groundwater in two wells with depths less than 50 m had mean residence times of 12 and 63 years based on CFC (chlorofluorocarbon) and tritium measurements (Stewart et al., 2002; van der Raaij, 2013). Groundwater residence times generally increase with depth in other wells in the greater Ashburton area. A trial site for managed aquifer recharge (MAR) has been operating since 2016 just outside of the study area (Fig. 1b).

The closest long-term rainfall site is part of a climate station at Ashburton Council (Fig. 1b). Annual average rainfall at Ashburton Council is around 730 mm (measured between 1909 and 2017), ranging from 382 to 1147 mm. There is little seasonality in rainfall, which averages 61 mm a month. Groundwater recharge was reported by Thorpe and Scott (1999) based on lysimeter measurements of soil drainage at Winchmore (10 km north of Ashburton; Fig. 2). In the 10-year period (1961–1971), average recharge was 293.5 mm yr-1 with average rainfall of 730 mm yr-1 and potential evapotranspiration (PET) of 765 mm yr-1. Average monthly recharge was much higher in winter months (April to September). Winchmore soil is described as Lismore stony silt loam characteristic of that at Tinwald west of Highway 1 and much of the Canterbury Plains.

The Hakatere/Ashburton River has a north and south branch sourced from the Canterbury ranges which converge at the north of the study area. The Hakatere/Ashburton River interacts with local groundwater, losing and gaining water along its length. Flow is lost to groundwater from the South Branch and gains towards the confluence with the North Branch.

Springs and wetlands indicate areas where the water table is naturally close to the surface and groundwater discharge is occurring. Many springs are found in the Hakatere/Ashburton River catchment and often occur in relict river channels (Aitchison-Earl, 2000). In the study area, Carters Creek and Laghmor Creek are both sourced from springs, and there are springs above Lake Hood that flow into the lake (Fig. 1b).

Cropping has been a major land use in the Tinwald area since at least the early 1940s (Fig. 1b; Engelbrecht, 2005). Most of the area is not part of any of the major surface water irrigation schemes, so irrigation was developed from groundwater sources within the area from the 1980s. Cultivation and fertiliser practice in cropping has an impact on the amount of nitrate that is leached from the soil to the groundwater. Winter is the most likely time for leaching to occur due to saturated soils and less nitrogen being used by crops. Nitrogen-fixing clover crops have been used less over time with an increase in commercial fertilisers (predominantly urea). Point sources of nitrate and other contaminants include septic tanks (human effluent), dairy and other animal effluent, and storm water and contaminated water.

Groundwater use in the Tinwald area is mainly for irrigation and for domestic and stock water supply. The Valetta Irrigation Scheme extends to the edge of the study area in the northwest and sources water from the braided alpine Rangitata River to the south (Figs. 2 and 4).

Consistently high nitrate concentrations (greater than MAV of 11.3 mg L-1) in groundwater were first identified in the Tinwald area in 2002 (Hanson, 2002). Maximum recorded nitrate concentrations from all samples between 1990 and 2017 are shown in Fig. 2. In general, lower nitrate concentrations occur close to the rivers and under and downgradient of surface water irrigation schemes (less than 1/2 MAV of 5.6 mg L-1). This is because of dilution by river-sourced water, which is lower in nitrate. Nitrate concentrations are higher in areas with local-groundwater recharge such as Tinwald (>11.3 mg L-1). Nitrate concentrations are lower towards the coast in the old Hinds swamp southeast of State Highway 1 (SH1) (<5.6 mg L-1), where the lower nitrate concentrations are driven by reducing conditions which facilitate denitrification (Hanson and Abraham, 2010). (Note that there are elevated nitrate levels – >11.3 mg L-1 – outside the study area from the east side of Ashburton to the coast, which are the result of historic leaching from meat-processing plants and general agricultural land use. These are not part of this study.)

Nitrate concentrations have increased over time in the Tinwald area, with two long-term monitoring sites (Thews Road and Saleyards wells; Fig. 1b) having statistically significant (p<0.05) upwards trends of 0.44 and 0.29 mg L-1 yr-1 since monitoring began in 1995 (Aitchison-Earl, 2019).

A total of 33 wells were sampled in the study area between 7 February and 14 March 2018 (Fig. 3); 13 of the wells had been sampled in 2004, and others were selected to fill gaps at a range of well depths. Groundwater levels were regionally high at the time of sampling following a sustained period of low levels (Aitchison-Earl, 2019). A large rainfall event of over 100 mm occurred during the sampling period on 21 February, leading to an increase in river flow.

Information on the wells is given in Table 1. Screened intervals and mid-screen depths are given for screened wells; total well depths are given where the wells have no screens. A total of 52 % of the wells had short screens (average 2 m length); 21 % had long screens (average 10 m length); and 27 % had no screens. Field measurements had stabilised for all wells before sampling. A total of 25 wells were purged of at least three well-casing volumes before sampling, and the 8 remaining wells were sampled by low-flow methods (pumps were lowered into the wells, and water was sampled after the pipes had been purged of three pipe volumes).

Samples were analysed for Environment Canterbury's standard suite of major ions through Hills Laboratories (Aitchison-Earl, 2019). Field measurements included dissolved oxygen (DO), pH, conductivity, temperature, and depth to groundwater. A selection of the field quantity and ion concentration results are given in Tables 1 and 2. The samples have been ordered from lowest to highest DO concentrations, and four groups of samples (A to D) are identified to aid discussion. Group A and Group B have low DO values (<4 mg L-1), with A having high δ15N (>15 ‰) and B having moderate δ15N (7 ‰–9 ‰). Group C and Group D have high DO (>8.2 mg L-1), with C having the highest and D having the lowest Cl and SO4 concentrations.

Water samples were analysed on an Isoprime mass spectrometer by water equilibration at 25 ∘C using an Aquaprep device for δ18O and by reduction at 1100 ∘C using a EuroVector Chrome HD elemental analyser for δ2H (GNS Science Laboratories). Results are reported with respect to VSMOW2 (Vienna Standard Mean Ocean Water). The analytical precision for this instrument is 0.2 ‰ for δ18O and 2.0 ‰ for δ2H. Results are given in Table 1.

Nitrate samples (NO3) were converted to nitrite (NO2) using cadmium and then to nitrous oxide (N2O) using sodium azide in an acetic acid buffer. The N2O was then extracted from the water sample, passed through a series of chemical traps to remove H2O and CO2, and cryogenically trapped under liquid nitrogen. After being cryofocused in a second trap, the N2O passed through a GC (gas chromatographic) column and into an Isoprime IRMS (isotope ratio mass spectrometer) to determine its isotopic signature of nitrogen and oxygen (GNS Science Laboratories). Our method is modified from McIlvin and Altabet (2005), following personal communication with Mark Altabet. Results are reported with respect to atmospheric N2 standard (AIR) for δ15N and VSMOW for δ18O. The analytical precision for these measurements is 0.3 ‰ for δ15N and δ18ONO3, except for samples below 0.1 mg L-1 NO3 N which may have lower precisions. Results are given in Table 1.

DO concentrations in the Tinwald groundwaters range from 0.18 to 11.8 mg L-1, although the majority are high and indicate relatively oxic groundwater. As noted above, the data in Tables 1 and 2 are ordered from lowest to highest DO values.

Chloride concentrations are useful to distinguish recharge sources. Chloride concentrations are highest in rainfall originating over the sea and near the coast and generally decrease with distance inland. In particular, alpine rivers (with chloride values of 0–5 mg L-1) and coastal rainfall-derived infiltration (with chloride ranging from 10 to 20 mg L-1) can be clearly distinguished (Hayward, 2002; Stewart et al., 2002). However, chloride concentrations in the Tinwald area (Fig. 4a) are greater than expected even for coastal rainfall (most are >15 mg L-1). The values are lower (0–10 mg L-1) to the northeast side of the study area near the Hakatere/Ashburton River and to the southwest.

Sulfate occurs naturally in groundwater and is present in fertilisers and fungicides and so can be an indicator of human influence when concentrations are in excess of background levels as here (see Fig. 4b). As with chloride, levels in alpine rivers and low-altitude rainfall infiltration are very different, but in the case of sulfate the difference is caused by the nature of additions to the soils in the respective catchments rather than the concentrations in rainfall. Concentrations are lower on the northeast and southwest boundaries of the study area. Nitrate concentrations are shown in Fig. 4c. Nitrate concentration exceeds MAV (11.3 mg L-1) in 17 of 33 wells sampled in the study area and is over 20 mg L-1 in 4 wells. The highest nitrate concentrations cluster inland of SH1 to the west and northwest of Tinwald and underlie an area of dominant cropping land use (Fig. 1b). Nitrate is lowest on the northeast boundary of the study area (near the Hakatere/Ashburton River), where it is generally below 1/2 MAV, 5.65 mg L-1, and lower but still over 1/2 MAV on the southwest boundary.

To investigate possible irrigation return flow effects, we compared the concentrations of different solutes and isotopes and include the effect of evaporation as indicated by the stable water isotopes (Fig. 5a–c). Figure 5a shows water δ18O versus chloride. Higher δ18O correlates with higher chloride, but this is not due to evaporation (because the evaporation vector is not parallel to the trend). Instead the main influence is the source of the recharge because both chloride and δ18O are higher in local-groundwater recharge (e.g. Group C samples) and lower in alpine-river recharge (Group D samples). There is no effect due to DO. Sample 21 shows an extra evaporation effect.

Figure 5b shows that chloride and sulfate are well correlated especially when the low-DO samples are excluded. This trend is also due to the recharge sources (see Group C and Group D samples in the figure). Evaporative enrichment is slight. The low-DO samples conform to the overall trend but are more scattered than the other samples.

Figure 5c shows that nitrate and chloride are moderately correlated due to the recharge sources (see Group C and Group D samples), but there are other processes affecting the nitrate concentrations. The low-DO waters have low nitrate concentrations, indicating that they have been partially denitrified. Evaporation has a negligible effect.

The clear message from these results is that nitrate, sulfate, and chloride concentrations are increased in areas irrigated by local groundwater compared to those irrigated by alpine-river water.

δ18O values are useful as tracers of the sources of recharge to groundwater, because rainfall recharge and rivers from alpine catchments have different isotope ratio signatures. Scott (2014) reviewed δ18O data for Canterbury and identified isotopic signatures in the Ashburton area. The Hakatere/Ashburton River has very negative δ18O with a mean of -10.1 ‰ for the South Branch and -10.7 ‰ for the North Branch. In contrast, rainfall recharge near the coast is less negative than -8 ‰, although it becomes more negative inland and is typically more negative than -8 ‰ on the upper plains. The Rangitata River, the alpine source of Valetta and other irrigation scheme water, has a mean δ18O of -9.8 ‰ (Taylor et al., 1989).

δ18O data collected in the greater Ashburton area are shown in Fig. 6. The influence of more negative Rangitata River-sourced water can be seen under the irrigation schemes. The δ18O values are less negative (red and orange dots) in the Tinwald study area. More negative values (green dots) occur on the northeast and southwest boundaries of the area, which are related to recharge from the South Branch Hakatere/Ashburton River and the Valetta Irrigation Scheme water. An area south of Tinwald near the coast (Eiffleton Irrigation Scheme) has less negative δ18O values like those observed in the Tinwald study area for probably the same reasons (irrigation return flow).

We also sampled wells for δ2H in 2018. Available evidence supports a local meteoric water line (LMWL) for Canterbury with intercept d=+10 ‰ (Stewart and Taylor, 1981; Taylor et al., 1989; Stewart and Morgenstern, 2001; Scott, 2014; Stewart et al., 2018); i.e. the LMWL is 1δ2H=8.0δ18O+10. Paired δ18O and δ2H data for the Tinwald study are plotted in Fig. 7. The Tinwald data plot below the LMWL and have a linear best-fit line (excluding sample 21, which has been affected by extra evaporation) given by 2δ2H=6.3δ18O-7.2. The slope of less than 8 for this line indicates that the waters have been affected by evaporation. A ratio of about 5 in the 2H and 18O enrichments is expected for evaporation at ambient temperatures (Stewart, 1975). It is likely that the isotopic compositions of the water would have been enriched by evaporation and/or evapotranspiration during the irrigation return flow process. Estimates of the isotopic enrichments required to explain the displacement of the average isotopic compositions of the group from the LMWL are shown by the red arrow with a slope of 5 in Fig. 7. The average isotopic composition of the samples except sample 21 (Table 1) was (-8.58, -61.2), where the bracket represents (δ18O, δ2H). The average initial composition of the samples would then have been (-9.43, -65.4), i.e. where the red arrow with a slope of 5 meets the LMWL (marked by the small red circle in Fig. 7). This degree of isotopic enrichment is obtained by evaporation of approximately 5 % of the water according to both isotopes (see the calculation in Appendix A). Uncertainty in this evaporated fraction is small because the isotopic composition of the remaining water changes rapidly with the degree of evaporation due to the form of the equation, so considerable changes of isotopic composition do not change the fraction evaporated much.

In addition, the difference in the δ18O and δ2H values of Group C and Group D is attributed to their different irrigation sources (local groundwater or alpine-river water) as observed for the chemical compositions. Assuming that both groups are affected by evaporation to the same extent, the difference between the groups compared to the difference between the irrigation sources will give an approximate measure of the irrigation input. The δ18O difference between Group C and Group D is 0.63 ‰ (Table 4), and that between the sources is 1.63 ‰, giving an irrigation input of 39 %. For δ2H it is 4.1 ‰ compared to 10.2 ‰, giving a 40 % irrigation input. These may be slight overestimates because Group C waters may be more affected by evaporation than Group D waters. (The compositions of the irrigation sources are taken as local rainfall (-8.17, -58.7) and alpine river (-9.80, -68.9).)

The nitrate isotope results are given in Table 1 and plotted in Fig. 8a. The samples have symbols depending on their DO concentrations, as in previous figures. The figure displays two important features:

There is a very good linear relationship between the δ15N and δ18ONO3 values of the nitrate.

The second feature of the figure is surprising. Denitrification is only expected to take place where DO levels are very low (e.g. <0.5 mg L-1; McMahon and Chapelle, 2008). But here denitrification effects are observed even when the DO concentrations in the groundwaters are much higher. The lowest DO range in Table 1 (with DO < 4 mg L-1, samples 01–09) includes Group A waters which have the lowest DO values and most marked denitrification effects (with δ15N values from 15 ‰ to 20 ‰), Group B waters with DO from 1.36 to 3.39 mg L-1 and δ15N values from 7 ‰ to 9 ‰, and one other well (07) that in contrast shows only minor denitrification (δ15N is 5 ‰) despite its relatively low DO (2.68 mg L-1). The intermediate-DO group (4 < DO < 8.2 mg L-1) has intermediate nitrate concentrations and shows smaller denitrification effects (δ15N values from 3.5 ‰ to 6.0 ‰, samples 10–14). The highest-DO group (with DO > 8.2 mg L-1) is nearly saturated with oxygen and shows minimal denitrification effects (δ15N values from 1.7 ‰ to 4.8 ‰, samples 15–33).

Figure 8b shows the natural log of the nitrate concentrations versus δ15N values; the natural log is used because the denitrification line will be linear on this type of plot according to the Rayleigh formula (Appendix B). The grey bands show approximate values of the δ15N values of possible nitrate sources (i.e. natural soil with δ15N of -3 ‰ to 7 ‰, inorganic fertiliser with -3 ‰ to 3 ‰, and effluent with 9 ‰ to 25 ‰; Fogg et al., 1998; Stewart et al., 2011). The Ln(NO3) values they are plotted at are schematic; we normally expect background nitrate concentrations from natural sources in soil to be about <0.7 to 1 mg L-1 (Close et al., 2001; Daughney and Reeves, 2005). Nitrogen-fixing clover is a possible source of nitrate with an isotopic composition like that of soil nitrate in pasture in Canterbury (Trevis, 2012), but we think its contribution is not large because the abundance of clover has decreased over the years as fertiliser use (particularly urea) has increased.

The enrichment factors producing the denitrification lines in Fig. 8b and c are ε(15N)=-3.0 ‰ and ε(18O)=-2.1 ‰ (Table 3). These are similar to the ranges determined by Clague et al. (2015) (ε(15N)=-1.1 ‰ to -9.6 ‰ and ε(18O)=-1.0 ‰ to -7.2 ‰) and values by Stenger et al. (2018) (ε(15N)=-2.0 ‰ and ε(18O)=-1.3 ‰), while Mariotti et al. (1988) gave an ε(15N) range from -5 ‰ to -8 ‰. Other authors (Kendall, 1998, and references therein) gave much larger negative values. Mariotti et al. (1988) suggested that low values may occur if denitrification occurs in dead-end pores causing a non-fractionating sink for nitrate by diffusion. Stenger et al. (2018) considered that small-scale physical heterogeneity, including the localised distribution of resident electron donors and the effect of lateral flows, was a more likely cause with their coarse-textured ignimbrite materials.

Figure 8c shows the natural log of the nitrate concentrations versus δ18ONO3 values. As with Fig. 8a and b, the denitrification line through Group C and Group B wells project to sample 01. This well is located south of Tinwald near Laghmor Creek in the Hinds swamp denitrification area. Group A wells show the greatest denitrification effects; the other samples in the group (02, 03, and 06) lie to the right of the denitrification line in Fig. 8b, indicating that they have larger proportions of effluent than the rest of the samples. Samples 02 and 03 occur downgradient of the old Tinwald Saleyards; sample 06 is northwest of SH1 in an area of lifestyle blocks adjacent to the major cropping area. It is probable that the effluent source is providing a source of dissolved organic carbon to fuel denitrification reactions in Group A wells.

Group B wells (showing moderate denitrification) are located closest to the Hakatere/Ashburton River. One sample (04) is in the cropping area, and the others (05, 08, and 09) are in areas with lifestyle blocks, which could contribute both septic tank and animal effluent to assist denitrification.

Group C wells (representative of wells irrigated by local groundwater) are distributed through the central part of the high-nitrate hotspot. They plot in the upper part of the cluster of green points in Fig. 8a and to the right in Fig. 8b and c.

Group D wells (representative of wells irrigated more by alpine-river water) are located on the southwest boundary of the study area. They plot in the lower part of the green point cluster in Fig. 8a and to the left in Fig. 8b and c.

The green points in Fig. 8 have δ15N values that are mostly within the soil nitrate or inorganic-fertiliser ranges and show little evidence of denitrification. Natural soil nitrate alone does not account for the slightly elevated nitrate concentrations in these wells, making inorganic fertiliser (or rather organic N derived from it; see discussion below) the likely dominant source.

A mixing curve between two nitrate source end members (soil nitrate and fertiliser plus effluent mixture nitrate) has been fitted to the solid green points (Appendix C). The best-fitting curves give δ15N=4.1 ‰ and δ18ONO3=0.0 ‰ for the blended nitrate source. This indicates that the source is dominated by inorganic fertiliser but has a small proportion of effluent source based on 15N (it is assumed that the highest nitrate concentrations are little affected by denitrification).

Fertilisers have been applied to much of the area between the Ashburton River and Hinds River, not just to the Tinwald study area, and rainfall applies to the whole area with contours of the δ values in rainfall decreasing inland from the coast (Stewart et al., 2002). Yet the Tinwald area shows elevated nitrate concentrations (and chloride, sulfate, etc.) compared to the surrounding areas (see Fig. 2). The difference is that the Tinwald study area is irrigated by groundwater from shallow local wells with high solute concentrations, whereas much of the rest of the area is irrigated by alpine-river water with low solute concentrations.

The irrigation return flow process is illustrated schematically in Fig. 9. Dewandel et al. (2007) defined an irrigation return flow coefficient C equal to the recharge from irrigation (i.e. irrigation return flow, IRF, divided by the irrigation flow itself, I, so that C=IRF/I). C is also equal to the overall recharge rate for rainfall and irrigation in our system (we omit surface runoff and lateral seepage in this treatment because both are expected to be small). (Note that drains in the area are fed mainly by groundwater.) C is used to quantify the effect of irrigation return flow on the water balance.

The average chemical concentrations of Group C and Group D are taken as representative of the Tinwald hotspot and outside groundwaters respectively (average values and standard deviations for each chemical are given in Table 4). The table also gives the enrichment factors and irrigation return flow coefficients between Group C and Group D derived from each chemical; the enrichment factor is equal to 1/C assuming that chemicals input via rainfall and irrigation are concentrated into the recharge fraction, i.e. concentrated by the loss of actual evapotranspiration (AET). Cl mass balance has often been used to estimate recharge because it is conservative; here the enrichment factor is 3.2, and coefficient C obtained is 31 %. The SO4 enrichment factor is very large (7.0), suggesting greater fertiliser input into the Group C area soil than into the Group D area soil. The HCO3 factor is small (1.2) probably because of chemical re-equilibration as water passes through the soil in both areas. The average enrichment factor is 2.4, and coefficient C is 42 % for all of the chemicals except SO4 and HCO3. Values of coefficient C can be compared with lysimeter measurements of the recharge fraction from the nearby research station of Winchmore (Thorpe and Scott, 1999, Fig. 2); the recharge fraction is recharge divided by input (i.e. rainfall plus irrigation). Thorpe and Scott (1999) found that the recharge fraction at Winchmore was about 39 % for the average irrigation input of 200 mm yr-1, in good agreement with the chemical results above.

The δ18O and δ2H values of Group C and Group D are affected by the different irrigation water sources (local groundwater or alpine-river water) and by evaporation as described in Sect. 4.2. An irrigation input of up to 40 % is indicated by the mean isotopic compositions of Group C and Group D. Evaporation is indicated by displacement of the sample points from the LMWL in Fig. 7. An approximate calculation given in Appendix A shows that evaporation of about 5 % of the water can explain the average displacement of the points.

Irrigation return flow has important implications for management of nitrate in agriculture. An important aspect of water resource management is understanding and modelling of nitrate transport in water systems (in this case groundwater). Much effort is being expended to model the effects of nitrate produced by farming practices in order to introduce and substantiate appropriate controls on farming to protect the water supplies of downstream communities (e.g. Environmental Canterbury, 2020). Irrigation return flow can seriously distort such modelling by lengthening the timescale of nitrate transport by abstracting water from groundwater downstream and applying it upstream and by adding nitrate on a second pass through the soil.

Another effect of irrigation return flow is distortion of tracer age dating results. Tritium concentrations will not be reset by interaction with the atmosphere when irrigation water is applied to the soil, so the tritium residence times of groundwater affected by irrigation return flow will appear to be older than they really are. In contrast, CFC and SF6 residence times will be reset to zero in the soil, and groundwater residence times will reflect time since recharge. This appears to be the case for data in the Tinwald area, although data are scarce (Stewart et al., 2002).

A practical consideration is that if irrigation water already contains nitrate then too much fertiliser could be applied leading to unnecessary economic cost and greater nitrate leaching potential, if the nitrate in the groundwater is not accounted for by nutrient budgeting (e.g. Flintoft, 2015).