The soil was sampled in January 2017 at the surface layer (0–20 cm) of a control plot at the QualiAgro experimental site (48∘87′ N, 1∘97′ E – https://www6.inrae.fr/valor-pro_eng/Experimental-devices/QualiAgro/QualiAgro-web-site, last access: 2 November 2021). The soil sample was immediately wet sieved at 5 mm and shortly stored at 4 ∘C until microcosm build-up. Aliquots of this sieved soil were used to measure the initial water content in addition to the maximum water holding capacity (WHC) for the further microcosm experiments. This site is located at Feucherolles near Paris, France, and had been designed to evaluate urban compost fertility together with the monitoring of contaminant inputs (Cambier et al., 2019). Soil is a Luvisol with 15 % clay, 78 % silt and 7 % sand; a pH of 6.9; organic carbon (Corg) and total N contents at 10.5 ± 0.2 and 1.00 ± 0.03 gkg-1 soil, respectively; and a catatonic exchange capacity (CEC) of 7.9 ± 0.8 cmol+kg-1 soil. This soil is not contaminated with Cu, and geochemical [Cu] background measured by inductively coupled plasma atomic emission spectroscopy (ICP-AES) after HF-HClO4 extraction was 12 mgCukg-1 soil.

In order to evaluate the impact of soil moisture on the sensitivity of nitrification to Cu toxicity, we carried out a two-step experiment. The first step consisted in initial incubations at five different WHCs over 5 weeks, and the second step was a 3 d bioassay with a spiked Cu gradient (Fig. 1).

For the 5-week initial incubation, five microcosms were built up with about 5 g of sampled soil. Three of them were set up with a constant moisture corresponding to 30 %, 60 % and 90 % of their WHC in order to span, respectively, limiting, optimal and saturating conditions for the microbial activities. These three samples will hereafter be called 30 %, 60 % and 90 %, respectively. Their water contents were verified by weighting every 2 d and water was added if necessary. The two other microcosms were incubated in order to simulate two kinds of drought and dry–rewet cycles. One, hereafter called “Drought” (or DO), started with 1 week at 60 % WHC, and then the soil was left for 3 weeks without added water to mimic a dry period until 10 % of the WHC before rewetting at the initial 60 % WHC. The other, hereafter called “Dry-Rewet” (or DR) encountered two cycles of a 1-week near-saturation period (90 % WHC) followed by a 1-week dry period (10 % of the WHC) and ending with a 1-week near-saturation period. Drying was performed by natural evaporation (gentle air-drying at the laboratory temperature, i.e. 20 ∘C) and controlled by weighting.

At the end of the initial incubation period, we performed a nitrification bioassay using three replicates originating from soils and following an adaptation of the method proposed by Petersen et al. (2012). The bioassay consisted in nitrate production measurement over a short-term aerobic incubation in soil slurries (ratio of soil to solution was 1:10) with ammonium in excess and in the presence of gradients of Cu. Briefly, 3.5 g of fresh soil (approximately 3 g of soil equivalent dry weight) was mixed in 50 mL Falcon^® tubes with 29 mL of a 10 mM HEPES buffer solution (hydroxyethyl piperazineethanesulfonic acid, Sigma-Aldrich, France) to maintain a constant pH under Cu spiking and nitrification activity and containing the substrate (NH4)2SO4 (3 mM) (Sigma-Aldrich, France). Soils were first spiked with a gradient of increasing Cu2+ in the presence of an excess of NH4+, and the resulting potential nitrification activity (PNA) was measured.

The microcosms incubated at constant moisture were kept at their moisture level (30 %, 60 % or 90 % of WHC), whereas those incubated at variable moisture levels were set at 60 % WHC. The NO3-production rates were measured in soil slurries over a short-term aerobic incubation, for each Cu added concentration. Briefly, 1 mL of Cu solution at different concentrations was added in soil slurries to reach added [Cu] of 50, 100, 250, 500, 750, 1000 and 2000 mgCukg-1 of soil (final soil [Cu] of 62, 112, 262, 512, 762, 1012 and 2012 mgCukg-1 of soil and control with 12 mgCukg-1 of soil). The pH was adjusted to 7. Then, microcosms were incubated on a rotary shaker (150 rpm) under aerobic conditions at 25 ∘C until 72 h. After 0, 24 and 72 h of incubation, 2 mL aliquots of 3 g were transferred in Eppendorf vials and centrifuged. The supernatants were collected and stored in microplates at -20 ∘C until analyses of NO3- and NO2- by colorimetric determinations, following the reduction of NO3- in NO2- by vanadium(III) and then the detection of NO2- by the acidic Griess reaction (Miranda et al., 2001). Finally, PNA (µgNO3-Ng-1soilh-1) was calculated on the basis of NO3--N + NO2--N concentrations measured at different time steps. In our bioassay, [NO2-] was negligible and PNA was thus calculated following Eq. (1), by checking the linear production rate of NO3- between 2, 24 and 72 h: 1PNA=[NO3-]Tfinal-[NO3-]TinitialTfinal-Tinitial×Vs÷W, where Vs is volume of solution, W is weight of fresh soil and T is time of incubation.

Cu in solution was measured by centrifugation of the soil–solution mixture of each bioassay, followed by a determination of Cu in solution by flame atomic absorption spectroscopy. Cu-in-solution values are provided in Table S1 in the Supplement.

Nitrification and denitrification processes are represented following the DNDC model proposed by Li et al. (1992, 2000). In this study, we used a simplified version of DNDC adapted by Zaehle and Friend (2010) initially calibrated for soil WHC > 40 %, which we intended here to test for 30 % of WHC. This simplified version needs less boundary data but keeps a mechanistic description of the main processes. Modelled N species are expressed in amount of N, i.e. NH4-N, NO3-N, NOx-N and N2O-N. To be able to represent both nitrification and denitrification processes occurring in aerobic and anaerobic sites, the soil is split into aerobic and anaerobic fractions based on an empirical relationship linking O2 consumption to soil respiration. In aerobic microsites, nitrification takes place following Eq. (2): 2nitrification=f(SWC)×f(temp)×f(pH)×kNit×(1-anvf)×NH4, with NH4-N being the stock of ammonium (in gNm-2); (1-anvf) the aerobic fraction of the soil described thereafter in Eq. (21); kNit the nitrification rate (d-1); f(SWC), f(temp) and f(pH) three rate modifiers representing the effect of soil water content (m3m-3); and temperature (K) and pH as scalars, respectively. They are described by the following Eqs. (3)–(5). 3f(SWC)=0.0243+0.9975×SWC+5.6358×SWC2+17.651×SWC3+12.904×SWC44f(temp)=0.0233+0.3094×temp+0.2234×temp2+0.1566×temp3+0.0272×temp45f(pH)=1.2314+0.7347×pH+0.0604×pH2

The NH4-N nitrified is transformed into N2O-N, NO-N or NO3-N due to microbial processes and chemonitrification following Eqs. (6)–(8). 6NitrificationN2O=ftv×SWC×kNitrifN2O×nitrification7NitrificationNO=ftv×SWC×kNitrifNO×nitrification+496950×e-1.62×pH×e-31494/(temp×R)×nitrification8NitrificationNO3=nitrification-nitrificationNO-nitrificationN2O Here kNitrifNO and kNitrifN2O are two fixed rates (d-1), ftv a rate modifier controlled by temperature and given in Eq. (9), and R the ideal gas constant. 9ftv=2.7234.6-9615temp

Then, the NO3-N produced during the nitrification process enters the denitrification module where it is reduced sequentially into NOx-N, N2O-N or N2-N following Eqs. (10) to (12). 10DenitrificationNOx=anvf×μNO30.401+0.09×NO3Ntot×B11DenitrificationN2O=anvf×μNOx0.428+0.035×NOxNtot×B12DenitrificationN2=anvf×μN2O0.151+0.079×N2ONtot×B

The anaerobic fraction anvf is described following Eq. (13): 13anvf=0.85×1-psoilO2pairO2, with pairO2 and psoilO2 being the partial pressure in the air and in the soil, respectively. psoilO2 is calculated following Eq. (14). 14∂psoilO2∂t=psoilO2-pO2resp×k×SOC×fCu Here SOC is the soil organic carbon stock (gCm-2), k the decomposition rate, pO2resp the O2 partial pressure related to the respiration and fCu the effect of Cu on CO2 emissions as defined in Eq. (15), following (Sereni et al., 2021; Eq. 5) 15fCuCO2=exp⁡(-0.1-0.1×log⁡(Cu)+0.12×pH).

The relative growth rate of NO3-N, NOx-N and N2O-N denitrifiers is described, respectively, by μNO3, μNOx and μN2O following Eqs. (16)–(18). 16μNO3=0.67×fdenit(temp)×fdenitNO3(pH)×NO3NO3+16617μNOx=0.34×fdenit(temp)×fdenitNOx(pH)×NOxNOx+16618μN2O=0.34×fdenit(temp)×fdenitN20(pH)×N2ON2O+166 Here fdenit(temp), fdenitNO3(pH), fdenitNOx(pH), and fdenitN2O(pH) are rate modifiers depending on air temperature and soil pH described in Eqs. (19) to (22). 19fdenit(temp)=2(temp-22.5)/1020fdenitNO3(pH)=1-11+e(4.25×pH)/0.521fdenitNOx(pH)=1-11+e(5.25×pH)22fdenitN2O(pH)=1-11+e(6.25×pH)/1.5

The denitrifier biomass dynamic B (kgm-2) is described following Eq. (23). 23∂B∂t=anvf×(μNO3+μNOx+μN2O)-3.82×10-3×B

Finally, all the gaseous forms of mineral N are emitted into the atmosphere. It is important to note that we did not directly use the DNDC model but a simplified version adapted by Zaehle and Friend (2010). The original code was in Fortran, and we translated it into R to facilitate its manipulation. The time step of the model was 30 min, and most of the parameters were kept to the original values of Li et al. (1992, 2000), except k_nit, which was modified to 0.1743 instead of 0.2 to better fit the data from the control. Furthermore, the amounts of NH4-N fixed to the clay were reduced to 0 as the bioassay was performed in excess of NH4-N (see 2.2.0).

We used measures of N species at the end of the initial incubation period as initial values of N species for DNDC (Table 1a and Fig. 2). To estimate the anaerobic volume fraction during the 3 d bioassay, we used a C mineralisation rate k (Eq. 14) determined on the basis of measurements performed on the same soil (Annabi et al., 2007) and ran DNDC for a 45 d equilibrium period. We then extracted the initial anaerobic volume fraction and partial O2 pressure.

The dose–response curves of PNA during the bioassay to Cu gradient were plotted and tested with linear, quadratic or cubic functions as fitting models. Our aim was to find, if possible, a similar modelling fit function for all initial moisture incubation treatments. Thus, for each moisture treatment, the two best functions of fit were selected through Akaike information criterion (AIC) and R2 criteria and compared with ANOVA. After selection of a common type of functions, the permutability of the different function parameters was tested with the Chow test (gap v.1.2.2 package, which tested regression 1 on the basis of sample 2 and vice versa). If the p.v. (p value) exceeds its critical values, regressions cannot be considered equal (Zhao, 2007).

To estimate the effect of [Cu] and soil moisture on the different variables measured, we used the nonparametric Kruskal–Wallis test. The fits between the model and the data of soil nitrate concentration during the bioassays were measured using root-mean-square error (RMSE, Eq. 2): 24RMSE=1N∑i=1N(Xi-Yi)2, where i is the number of observations (1 to N), X is the predicted value and Y is the observed value. RMSE was decomposed in standard bias (Eq. 25), non-unity slope (Eq. 26) and lack of correlation (Eq. 27) components following Gauch et al. (2003), with X‾ and Y‾ the mean modelled and observed values, b the slope of the least-square regression of Y on X and r2 the square of the correlation. 25SB=X‾-Y‾226NU=(1-b)2×∑xn2N27LC=(1-r2)×∑yn2N

All analyses were done with R 3.2.3 (R Core Team, 2015).

The soil N species measured at the end of the soil initial incubations in each soil moisture treatment were used to initialise the DNDC model (Table 1). Two anomalous points leading to anomalous calculated NO2-N values were excluded from the experimental results because of technical problems during measurements (the C replicates in the DR and DO cases).

The bioassay experiments performed at the end of the soil initial incubations allowed us to determine the rate of nitrate production as a function of soil [Cu] for each soil moisture level (Fig. 1). In all cases, the PNA values were found to decrease with the increase in soil [Cu] but at different rates depending on the moisture treatment. Based on AIC values (Table S2 in the Supplement), we first selected one model per moisture incubation that better fit the data. For 30 % and 60 % of WHC, a quadratic model was found to provide the better compromise between the number of parameters and the prediction capacity. For 90 % WHC, no significant difference was found between the cubic and the quadratic models (ANOVA, p.v = 0.07). For DR, no significant difference was found between linear and quadratic models (Table S2a and b), whereas for DO the cubic model provided a substantially better fit than the quadratic model (AIC and adj. R2 score, Table S2). Finally, we found that the quadratic model fitted all the sets of data correctly, allowing homogeneity across the initial moisture incubation treatments (Fig. 2b). The quadratic function was thus chosen to quantify the Cu effect on PNA including the DO treatment.

The parameters of the five quadratic functions (one for each moisture treatment) were found to be different from each other, except for 60 % and 90 % WHC (p.v. = 0.001, Chow test). A single function was thus used to adjust PNA to soil [Cu] curves at 60 % and 90 % WHC but with different intercepts for these two WHC treatments (Table S3 in the Supplement and Fig. 2).

The final four quadratic equations are as follow: Eq. (28) for 30 % WHC, Eq. (29) for 60 % and 90 % WHC, Eq. (30) for DR, and Eq. (31) for DO (Fig. 2). 28FCu30=0.782-0.000451×Cu+9.49×10-8×Cu229FCu60/90=b-0.000342×Cu+4.30×10-8×Cu2 with b = 0.795 for 60 % WHC and b = 0.796 for 90 % WHC 30FCuDR=0.552-0.000164×Cu+6.09×10-8×Cu231FCuDO=0.625-0.000192×Cu+2.82×10-8×Cu2

According to the fitted equations, the decrease in nitrate production rates as a function of soil [Cu] depended on initial incubation treatment. Decreases were found to be steeper following 30 % WHC > 60 %–90 % WHC > DO > DR.

These four equations were then added in the DNDC model, allowing us to adjust Eq. (2), which regulates the nitrate production to soil Cu contents.

Thanks to our laboratory experiments, we were able to define a function modulating the soil NO3-N production rates in relation with soil [Cu] and depending on soil moisture. Our results showed that soil nitrate decreases with an increase in soil [Cu]. Initial incubation treatment significantly affects the response of soil nitrate production rate to subsequent Cu stress, with a steeper decrease on the order 30 % WHC > 60 %–90 % WHC > DO > DR for the Cu range studied. The lowest sensitivity of Cu in soils initially incubated with dry–rewet events suggests that it might have selected more resistant communities (Barnard et al., 2013; Gleeson et al., 2008). More complex dose–response functions have been used in Sereni et al. (2022) to assess thresholds and loss of function after such a double stress. These results are in relatively good agreement with those presented here using the quadratic fit, especially for the highest half of [Cu]. However, they also presented a limited increase in nitrification rate for small Cu input that we were not able to emphasise in the present study. In the present article we used simple functions of fit to describe the response of soil nitrate production to Cu gradient after the first moisture stress as they further have to be included in the DNDC model. After implementing these quadratic Cu modulating functions into the DNDC-Cu model, we were however able to reproduce the observed soil nitrate stock, particularly for the soils incubated at 60 % and 90 % WHC. The variability around the mean due to the Cu effect was also reproduced by our DNDC-Cu version at 30 % of WHC despite strong underestimation of mean soil nitrate stocks due to model moisture limit (Li et al., 1992). In the case of the DR and DO incubated soils, the so-called “Cu function” also accounted for the effect of drought stress. In fact, our Cu functions were defined on the basis of soil nitrate production against the whole gradient of Cu, thus also considering the control without Cu. However, the double-stress effect was also well reproduced in nitrate production.

Based on nitrate production measurements, we modelled a decrease in denitrifying activities with an increase in soil [Cu] as a consequence of the decrease in soil nitrate stocks. However, the experiments performed here did not allow us to determine if the soil Cu contamination rather affects nitrifying bacteria (e.g. decrease in nitrifying activity and in NO3-N production) or denitrifying bacteria (e.g. increase in denitrifying activities and NO3-N consumption). The effect of soil contamination on N2O-N production is debated because (i) microbial species involved are not clearly identified (Wrage-Mönnig et al., 2018), (ii) species richness is not necessarily related to soil functions (Ruyters et al., 2013) and (iii) denitrifying communities could have different sensitivities than nitrification to soil contamination (Hund-Rinke and Simon, 2008; Vásquez-Murrieta et al., 2006). Also, our approach to model N2O-N, N2-N and NOx-N production in the contaminated context could have been more constrained with measurement of denitrification rate to assess the effect of Cu on proportion of production and consumption of NO3-N.

Based on our simulations, the soil Cu contamination was expected to substantially modify the proportion of available N in soils with the increase in NH4-N stock at the expense of NO3-N. NH4-N accumulation and the large expected decrease in NO3-N/NH4-N ratio in contaminated soils (around 50 % for the 60 % WHC) may lead to a shift in plant community structures with different preferences in N assimilation (Cui and Song, 2007; Peacock et al., 2001). Therefore, Cu stress could not only have implications in microbial community patterns as a stressor but could also induce further shifts due to N species redistributions in soils.

In the present study, we predicted the highest soil N2-N, N2O-N and NOx-N stocks in the moistest treatments. Indeed these species are produced by the denitrifying bacteria expected to behave optimally at 90 % WHC or after DR cycles (Li et al., 1992; Homyak et al., 2017). However, N2O-N and NOx-N emissions were modelled as higher in the driest soils, whereas numerous studies (Dobbie and Smith 2003; Xiong et al. 2007; Manzoni et al. 2012) reported high measured N2O-N emissions with high moisture. In the present version of DNDC-Cu, the soil N emissions were directly controlled by their diffusion in soil, calculated on the basis of clay and soil moisture content. The diffusion of each species would hence be 11 times smaller under the 90 % WHC (D_s =  0.00357) than under the 60 % WHC treatment (D_s = 0.0306) because the model described the diffusion as a whole and did not separate pores with or without water. Diffusion was hence slower in the water than in the air. Thus, the weighted mean diffusion was lower in the high-moisture treatment. In the absence of Cu soil nitrous stocks being roughly 1.6 times and soil N2-N stocks 11.1 times larger under 90 % WHC treatment than the other, the emissions of N2O-N were larger under the driest treatment even if stocks were smaller.

Several studies also reported flushing events with Dry–rewet cycles which would enhance C mineralisation, known as the Birch effect (Birch, 1958; Göransson et al., 2013), hence reducing soil O2 concentration. Moreover, soil [O2] is closely related to the pore size distribution, being of major importance in nitrification–denitrification control (Khalil et al., 2004), with a dominating nitrification for aggregates up to 0.25 cm (Kremen et al., 2005). Pore size distribution under dry–rewet events is controlled by cracking, (des)aggregation (Cosentino et al., 2006; Denef et al., 2001) or gas displacement (Kemper et al., 1985) that we were not able to take into account in the present study. In DNDC, the calculation of denitrification rate and diffusion was based on a rough description of the anaerobic zone with approximation of soil pore space distribution (Blagodatsky et al., 2011; Li et al., 2000). The soil pore space distribution approach has been demonstrated to be more generally applicable (Arah and Vinten 1995; Schurgers et al. 2006), whereas soil aggregates have been shown to control the extent of nitrification and denitrification (Kremen et al., 2005; Schlüter et al., 2018). However, if models have been proposed to take O2 availability at the aggregate size into account in the nitrous oxide production (Kremen et al., 2005; Leffelaar, 1988), they also point out the difficulty in parameterisation that needs a large panel of soil measurements. Moreover, they are rarely transposable at the mesoscale and regional scale due to high spatial variations in soil structure (Butterbach-Bahl et al., 2013). The DNDC-Cu version we used here in particular pointed out the difficulty in dealing with a biogeochemistry model with physical processes, with large discrepancies between modelled soil stocks and emissions. The validation we performed focused on soil nitrate stocks, and a second step to go further on would be the measure of gaseous species to ensure that emissions were also impacted by soil treatment. Moreover, we assumed here that soil [Cu] affected the C mineralisation with a decrease in soil O2 production leading to an increase in denitrification and N2O-N and NOx-N. Nevertheless, the present DNDC-Cu version did not take into account the retroaction between C and N cycles. Further research would thus be required to include Cu contamination in interacting C and N cycles.

It is well known that climate change and rainfall patterns could substantially modify the soil N balance and its GHG emissions (Galloway et al. 2003, 2008; Butterbach-Bahl et al. 2013). Despite limitation in DNDC accuracy for nitrous emissions (Foltz et al., 2019), our results tend to show that increased Cu contamination might also affect soil N emissions, producing the smallest emissions of NOx-N and N2O-N. These two gases are of major importance in GHG mitigation with a warming potential per mass 300 and 40 times greater than CO2, respectively. Agricultural soils being the dominating source of N2O-N (Beauchamp, 1997; Signor and Cerri, 2013), even a limited decrease in their emissions could have major implications for climate. Based on our modelling, the combined effect of soil moisture and [Cu] was particularly important with larger differences in N2O-N and NOx-N emissions between rainfall patterns at high [Cu] (Sect. 3.3.2). Sereni et al. (2022) also showed that soil Cu contamination differently affects soil nitrification depending on primary soil moisture stress. Here we showed that the N2O-N and NOx-N emission variations are significantly more sensitive to the combined effect of Cu and precipitation regime than the nitrate stock. Based on these results, Cu inputs to moist soils would lead to larger decreases in soil N2O-N and NOx-N emissions compared to drier soils, with the strongest effects likely to occur on soils subjected to abrupt and intense shifts in rainfall patterns.