Denitrification as a major regional nitrogen sink in subtropical forest catchments: Evidence from multi-site dual nitrate isotopes

1 INTRODUCTION

Anthropogenic changes to the nitrogen (N) cycle affect the environment both regionally and globally (Galloway et al., 2003; Vitousek et al., 1997) with profound effects on ecosystem functions (Aber et al., 1998; Jaworski, Howarth, & Hetling, 1997), climatic feedbacks (Bernal, Hedin, Likens, Gerber, & Buso, 2012; Shi, Cui, Ju, Cai, & Zhu, 2015; Zaehle, Ciais, Friend, & Prieur, 2011) and human health (Townsend et al., 2003). Reactive N input into the biosphere has doubled in the last century (Cui, Shi, Groffman, Schlesinger, & Zhu, 2013; Gu et al., 2012), and China's relative contribution is growing (Liu et al., 2013). Currently, forests in South China receive up to 60 kg N ha⁻¹ year⁻¹ from the atmosphere (Xu et al., 2015), leading to N saturation with significant nitrate (NO₃⁻) leaching from well-drained soils (Huang, Kang, Mulder, Zhang, & Duan, 2015). Yet, at the catchment scale, large proportions of dissolved inorganic N appear to be attenuated in water-saturated soils just before entering the streams (Duan et al., 2016; Larssen, Duan, & Mulder, 2011).

Biogeochemical N sinks in forest ecosystems include assimilation by plants and microbial biomass, and NO₃⁻ removal by denitrification (Yanai et al., 2013). In the acidic soils of South China, with low availability of phosphorus, N uptake by standing biomass is limited due to slow tree growth (Li et al., 2014; Wang et al., 2007), while net N assimilation into by soil microbes is restricted by low carbon availability (Chen & Mulder, 2007a; MacDonald et al., 2002). Several N mass balance studies have hypothesized that denitrification in near-stream soils acts as a significant N sink on the catchment level (Chen, Mulder, Wang, Zhao, & Xiang, 2004; Larssen et al., 2011; Zhu et al., 2013). Recently, this hypothesis was strengthened by stable isotope studies (Fang et al., 2015). However, the importance of denitrification at the catchment scale is under debate, owing the scarcity of spatially resolved data for soil denitrification (Duncan, Groffman, & Band, 2013).

Previous studies in temperate forests have documented significant, but transient denitrification activities in shallow water-saturated soils (Duncan, Band, Groffman, & Bernhardt, 2015; Duncan et al., 2013; Rose, Sebestyen, Elliott, & Koba, 2014; Wexler, Goodale, McGuire, Bailey, & Groffman, 2014), where NO₃⁻ is derived from soil N cycling rather than from atmospheric sources (Rose, Elliott, & Adams, 2015; Rose et al., 2014; Sabo, Nelson, & Eshleman, 2015). Others have emphasized the importance of coupled nitrification–denitrification, i.e. the simultaneous production and consumption of NO₃⁻ in shallow, fluctuating groundwater bodies (Duncan et al., 2015, 2013; Griffiths et al., 2016). These scenarios differ fundamentally from observations in South China, where large amounts of NO₃⁻ are produced in well-drained top soils on hillslopes and removed in groundwater discharge zones (Larssen et al., 2011). In a previous study in SW China, we observed a robust multi-year ¹⁵N and ¹⁸O pattern in nitrate (NO₃⁻) along a hydrological flow path in a headwater catchment, suggesting that large amounts of NO₃⁻ are produced by nitrification on hillslopes and transported by interflow over argic horizons of common Acrisols to wet soils in groundwater discharge zones, where they are denitrified (Yu, Zhu, Mulder, & Dörsch, 2016). Similar geomorphological and soil-related patterns of N turnover, transport and removal have been recently reported in temperate and subtropical systems (Anderson, Groffman, & Walter, 2015; Griffiths et al., 2016).

Denitrification in soil is difficult to measure directly (Kulkarni, Burgin, Groffman, & Yavitt, 2014), making dual isotopic signatures of NO₃⁻ an attractive tool for integrating temporal and spatial variability of N turnover (Griffiths et al., 2016; Kendall, Elliott, & Wankel, 2007; Schwarz, Oelmann, & Wilcke, 2011; Wexler et al., 2014). Biological transformations fractionate stable isotopes (Fry, 2007; Kendall et al., 2007) and the resulting changes in ¹⁵N and ¹⁸O signatures of residual NO₃⁻ can be used to partition NO₃⁻ sources and sinks as well as to elucidate underlying biological processes. For instance, NO₃⁻ produced through nitrification is often ¹⁵N-depleted compared to NH₄⁺, even though the fractionation effect is small (Kendall et al., 2007). Nitrification-derived NO₃⁻ has usually distinctively smaller δ¹⁸O than NO₃⁻ from precipitation, and this could be used to distinguish soil NO₃⁻ from microbial and atmospheric sources (Barnes, Raymond, & Casciotti, 2008; Rose et al., 2015), as NO₃⁻ produced from NH₄⁺ in soils by nitrification incorporates O from both ¹⁸O-depleted atmospheric O₂ and ¹⁸O-enriched H₂O (Kendall et al., 2007). Microbial denitrification results in enrichment of both ¹⁵N and ¹⁸O in residual NO₃⁻ with a ratio between 1:1 and 2:1. The latter is routinely used to identify denitrification in riparian and aquatic systems (Kendall et al., 2007; Osaka et al., 2010; Wexler et al., 2014).

Several studies have applied ¹⁵N natural abundance models to estimate ecosystem denitrification fluxes (Fang et al., 2015; Houlton & Bai, 2009; Soper et al., 2017). Assuming steady state for ecosystem N cycling, Houlton and Bai (2009) attributed ¹⁵N enrichment in the terrestrial biosphere to solely gaseous (denitrification) and hydrological (leaching) losses. Based on distinct isotopic effects by denitrification and leaching, they estimated the contribution of denitrification to total N loss and from this the denitrification N flux. Recently, Fang et al. (2015) presented a more complex isotopic model based on transformation rates and δ¹⁵N signatures of soil NO₃⁻ pools. This model follows the dynamic turnover of NO₃⁻ and thus provides more realistic estimates of denitrification. When quantifying isotopic effects of N cycling processes, both models depend on ¹⁵N fractionation factors (ε) as reported in the literature. In contrast to the small ε for leaching and biological N uptake, the isotopic effect of denitrification varies over a wide range from 5‰ to 30‰, with a global average of 16‰ (Granger, Sigman, Lehmann, & Tortell, 2008; Houlton & Bai, 2009; Knöller, Vogt, Haupt, Feisthauer, & Richnow, 2011). Under field conditions, the apparent ¹⁵N enrichment is usually smaller, due to complete consumption of NO₃⁻ by denitrification or mixing with NO₃⁻ produced by other processes (Mariotti, Landreau, & Simon, 1988). Thus, apparent ¹⁵N isotopic fractionation effects by riparian and groundwater denitrification have been reported to be around 5‰ (Bottcher, Strebel, Voerkelius, & Schmidt, 1990; Mariotti et al., 1988; Osaka et al., 2010; Søvik & Mørkved, 2008; Spalding, Exner, Martin, & Snow, 1993).

On the global scale, terrestrial denitrification fluxes largely depend on water balance and N load (Bouwman et al., 2013; Seitzinger, Harrison, & Bohlke, 2006). Chinese forests naturally cover a wide range of climate conditions and N deposition rates (Du, Jiang, Fang, & Vries, 2014; Fang & Yoda, 1990), which makes them highly suitable for studying ecosystem-level denitrification. In the present study, we monitored N fluxes in seven Chinese forest catchments for two years. The five southern sites are similar in geomorphology, but differ in N deposition rates. We hypothesized that denitrification in soils of the groundwater discharge zone in warm-humid subtropical forests in South China is a significant N sink, thus contributing to the observed N attenuation at the catchment scale. For comparison, we included two temperate northern sites which strongly differ in climatic and hydrological conditions from the southern. To evaluate the role of denitrification in the different catchments, we conducted sampling campaigns in two rainy summer seasons and analysed dual NO₃⁻ isotopes along the hydrological flow paths, comprising topographically distinct landscape elements like hillslopes and groundwater discharge zones. In addition, we use the N deposition gradient across these sites to assess the question how increased atmospheric N deposition affects the N-sink function of Chinese forested catchments.

2 MATERIALS AND METHODS

2.1 Study sites

Data were collected from five catchments in South China and two catchments in North China (Figure 1). The southern and northern catchments are named sites S1 to S5 and sites N1 to N2, respectively. Details about site location, mean annual temperature and precipitation, vegetation and soil characteristics are given in Table 1. The five southern sites are subtropical and have a monsoonal climate with annual precipitation ranging from 1,028 to 1,591 mm. The two northern sites are in the temperate zone, also affected by monsoonal rains, but have smaller annual precipitation (612–676 mm). At all studied sites, precipitation occurs mostly in summer. The pH of soil O/A horizons (excluding undecomposed litter) at southern sites (4.2–6.3) tends to be lower than at northern sites (7.0 and 6.1 for N1 and N2, respectively). The seven sites represent a gradient of inorganic N flux in throughfall ranging from 5.0 to 48.6 kg N ha⁻¹ year⁻¹ (observed during 2013 and 2014; Table 2).

Table 1. Background descriptions of the seven sites

Site name	S1 (TSP)	S2 (CJT)	S3 (TMS)	S4 (DGS)	S5 (LGS)	N1 (LPS)	N2 (DLS)
Location	Chongqing	Hunan	Zhejiang	Jiangxi	Guizhou	Ningxia	Beijing
Longitude	106°41′	112°22′	119°26′	114°34′	108°11′	106°20′	115°26′
Latitude	29°38′	27°50′	30°19′	27°35′	26°23′	35°15′	39°58′
Mean annual temperature (oC)	18.2	17.5	11.9	15.8	15.7	5.8	4.8
Mean annual precipitation (mm)	1028	1250	1581	1591	1120	676	612
Annual runoff (mm)a	172	303	414	486	335	68	61
Vegetation	Massone pine-dominated, coniferous-broad leaf mixed forest	Massone pine dominated, coniferous-broad leaf mixed forest	Broad-leaf forest	Evergreen broad-leaf forest	Pinus armandii dominated, coniferous-broad leaf mixed forest	Mixed deciduous broad-leaf and coniferous forest	Mixed, secondary deciduous broad-leaf and coniferous forest
Soil type	Loamy yellow mountain soil (Haplic Acrisol)	Yellow mountain soil (Haplic Acrisol)	Yellow soils (Cambisol)	Red and yellow soils (Alisol)	Yellow mountain soil (Alisol and Acrisol)	Gray cinnamon soil (Luvisol)	Cinnamon soil (Luvisol)
Soil pHb	4.2	4.8	6.3	4.9	4.4	7.0	6.1
Soil C/N ratioc	14.0	14.2	13.2	14.4	12.0	10.4	11.5

• ^a Runoff for the northern sites was estimated by multiplying annual precipitation with runoff coefficient (see Materials and Methods for details).
• ^b Soil pH_H2O of the O/A horizon (0-5 cm).
• ^c Soil C/N ratio of the O/A horizon (0-5 cm).

Table 2. Annual N mass balance (in kg N ha⁻¹ year⁻¹)a

Site no.	N deposition by thoughfall			N export via stream water			N removalb	Nitrification	Denitrification−15NO3− model	Denitrification−15N-soil model	(‰)f
	NH4+-N	NO3−-N	TIN	NH4+-N	NO3−-N	TIN
S1	27.2	21.5	48.6	0.03	9.5	9.5	39.1	108	18.3 (9.3)e	10.1 (1.7)	11.9
S2	15.6	23.2	38.8	0.03	2.7	2.7	36.1	116	21.7 (10.5)	4.8 (1.1)	22.2
S3	4.4	14.9	19.4	0.01	4.8	4.8	14.6	130	14.1 (11.4)	4.7 (0.8)	13.7
S4c	4.7	11.9	16.6	0.02	5.7	5.7	10.9	65	3.8 (4.0)	2.4 (0.3)	6.4
S5	3.9	5.9	9.8	0.02	1.9	1.9	7.8	59	2.4 (5.5)	1.9 (0.3)	6.9
N1d	1.8	3.2	5.0	0.25	0.2	0.5	4.6	56	–g	–	−1.2
N2 d	4.8	6.8	11.6	0.12	6.6	6.8	4.9	–	–	–	–

• ^a Annual fluxes are averages for 2012–2014. Nitrification fluxes are estimated based on ¹⁸O–NO₃⁻ method. Denitrification fluxes are estimated with the ¹⁵NO₃⁻ and the ¹⁵N-soil models, respectively (see Materials and Methods for details).
• ^b N removal refers to the difference between N deposition and N export by stream (input–output)
• ^c Discharge during November to April was estimated based on precipitation. Therefore, N export during this period was also estimated.
• ^d Annual runoff was calculated from the estimated annual discharge and annual average inorganic N concentrations.
• ^e Values in the brackets indicate uncertainty of model results (1 SD).
• ^f represents overall ¹⁵N enrichment in NO₃⁻ along the catchment (–; see Materials and Methods for details).
• ^g Data not available.

All catchments are characterized by well-drained hillslopes (HS) and hydrologically connected, terraced valley bottoms acting as groundwater discharge zones (GDZ). Soil moisture expressed as water filled pore space (WFPS) increases markedly from HS to GDZ (Figure 1). Sodium (Na⁺) is a virtually inert cation in forest soils, and its concentration increased gradually due to evapotranspiration along the water flow path (Figure 1), indicating hydrological continuity from plots A to D at all sites (Osaka et al., 2010). Thus, we could identify hydrological flow paths from throughfall (TF) to hillslope (plots A and B), groundwater discharge zone (plots C and D), and finally to stream water (SW) in both southern and northern catchments (Figure 1). The northern catchments are significantly drier than the southern catchments, as indicated by significantly smaller annual runoff volumes (Table 1) and lower values for soil WFPS (Figure 1), resulting in less developed and discontinuous groundwater discharge zones.

2.5 Rayleigh enrichment factor

The apparent ¹⁵N enrichment factor for denitrification in the groundwater discharge zone was estimated by Rayleigh distillation (Mariotti et al., 1981).

()

where C(NO₃⁻)_s0 and, C(NO₃⁻)_s are initial and residual NO₃⁻ concentrations and δ_s0 and δ_s the δ¹⁵N of initial and residual NO₃⁻, respectively. Assuming that the catchments behave as quasi-closed systems, where no new input of NO₃⁻ (e.g. from nitrification) occurs and progressive consumption of NO₃⁻ (i.e. denitrification) is the only relevant isotopic fractionation process (Wexler et al., 2014), we could then calculate Rayleigh fractionation factors based on the gradual decrease of NO₃⁻ concentration and gradual increase in from hillslopes to groundwater discharge zones.

To describe the overall trend of along the hydrological continuum in the catchments, we defined as

()

where and refer to concentration-weighted averages of in soil water from hillslopes and groundwater discharge zones, respectively.

2.7 Model estimation of denitrification flux: ¹⁵NO₃⁻ model

On the ecosystem scale, assuming that soil NO₃⁻ pools receive continuous inputs and outputs, we estimated denitrification fluxes based on the mass balances of soil NO₃⁻ and (Fang et al., 2015). The model is parameterized with the observed changes of and estimated fluxes of other NO₃⁻ source and sink processes:

()

where F_D, F_A and F_N, represent annual fluxes (kg N ha⁻¹ year⁻¹) of denitrification, atmospheric NO₃⁻ input (throughfall) and gross soil nitrification; refers to the mixed δ¹⁵N signature of nitrification-produced and atmospherically deposited NO₃⁻; refers to δ¹⁵N signature of whole-soil NO₃⁻; refers to δ¹⁵N signature of leached NO₃⁻; ε_D and ε_U refer to isotopic enrichment effects (‰) by denitrification and biological NO₃⁻ uptake, respectively.

This model was applied only to southern site, for which sufficient data were available. F_A was given by NO₃⁻ flux in throughfall. F_N was estimated by partitioning atmosphere-derived NO₃⁻ and soil-produced NO₃⁻ by a two-end member mixing approach based on in throughfall and in hillslope soil water (Barnes et al., 2008). and were determined as concentration-weighted averages of on hillslopes (plots A and B) and across whole catchments (plots A to D), respectively. We did not capture changes of with soil depths, but previous studies of in southern Chinese forest catchments demonstrated the vertical transport and transformation of NO₃⁻ is negligible because of limited infiltration in the argic B horizon (Sørbotten, Stolte, Wang, & Mulder, 2017; Yu et al., 2016). Concentration-weighted averages of in stream water were taken as indicative for . Ranges for ε_D and ε_U were assigned from values reported in the literature, 16‰–20‰ for denitrification (Granger et al., 2008; Houlton & Bai, 2009) and 0‰–4‰ for uptake (Evans, 2001; Granger, Sigman, Rohde, Maldonado, & Tortell, 2010). For more details on model parameterization, see SI - Estimation of denitrification flux by ¹⁵NO₃⁻ model.

2.8 Model estimation of denitrification flux: ¹⁵N-bulk soil model

The outcome of the ¹⁵NO₃⁻ model was compared with denitrification rates estimated by a ¹⁵N-bulk soil model (Houlton & Bai, 2009; Soper et al., 2017). The latter model assumes steady state of all N-transformation processes on the ecosystem level, so that the isotopic effect of internal N cycling (biological uptake and return) can be neglected. The ¹⁵N enrichment of bulk soil is then used to partition between gaseous N loss and N leaching, based on the distinctive isotopic effects of these two processes. Denitrification is calculated as:

()

where f_gas is the fraction of gaseous N loss to total N loss (leaching +gas emission), δ¹⁵N_TB and δ¹⁵N_I are the ¹⁵N isotopic ratios of total biospheric and atmospheric N, respectively, and ε_L and ε_D the isotopic enrichment effects (expressed as ‰) for N leaching and denitrification, respectively.

δ¹⁵N_TB was approximated with δ¹⁵N of bulk soils down to 50 cm depth. Since δ¹⁵N was only measured in surface soils, we applied a correction factor of +2.5‰ to account for the typical ¹⁵N enrichment with soil depths (Fang et al., 2015; Hobbie & Ouimette, 2009). For site S5 where soils were sampled at 0–10 cm, a smaller correction factor of +1.5‰ was applied. For isotopic fractionation effects, we used ranges of 0‰–0.8‰ for ε_L and 16‰–20‰ for ε_D according to previous literature data (Houlton & Bai, 2009). For further details concerning model parameterization, see SI - Estimation of denitrification flux by ¹⁵N-bulk soil model.

3 RESULTS

3.1 Soil chemical properties and mineral N concentrations

At the five southern sites, soil C and N contents and C/N ratios generally decreased from hillslopes (A and B plots) to groundwater discharge zones (C and D plots), while no clear trend was found at the two northern sites (Table S1). Despite similar C/N ratios, S4 and the northern sites had lower soil C and N contents than the southern sites. Soil pH increased from hillslopes to groundwater discharge zones, at all sites except N2. Site S1 had the lowest soil pH.

On hillslopes, soil water NH₄⁺ concentrations were generally <0.1 mg N L⁻¹ (except for N1, Figure S1), which was significantly smaller than in throughfall, while soil water NO₃⁻ concentrations were significantly larger than in throughfall (Figure 2). Along the hydrological continua from hillslopes to groundwater discharge zones and streams, NH₄⁺ concentrations in soil water remained small, while NO₃⁻ declined (Figure 2). Only at sites S2 and N₂, NO₃⁻ concentrations in the streams exceeded those in the soil of the groundwater discharge zones. Across all catchments, sites S1 and S2 had the highest NO₃⁻ concentrations on the hillslope, whereas site N1 had the lowest NO₃⁻ concentrations, both for hillslopes and groundwater discharge zones (Figure 2).

The samples collected during the summer campaign for isotope analysis showed similar spatial patterns of inorganic N concentrations (Figure S2), as the long-term data (Figure 2 and Figure S1). However, at site S4, NH₄⁺ and NO₃⁻ concentrations were unexpectedly high in the groundwater discharge zone compared to those on the hillslope (Figure S2), possibly due to strong evaporation in the groundwater discharge zone, similar to what Yu et al. (2016) found during a drought period at S1 in 2013. However, these elevated N concentrations play little role for the annual NO₃⁻ flux, as water transport was small.

To account for concentration changes due to evapotranspiration, we normalized NO₃⁻ concentrations against Na⁺ concentrations along the flow path. The spatial pattern of the NO₃⁻/Na⁺ ratios showed a sharp increase from throughfall to hillslopes and a subsequent pronounced decrease from hillslopes to groundwater discharge zones (Figure S3b), confirming that the observed change of NO₃⁻ concentration was due to net production and consumption and not due to evapotranspiration.

3.2 Isotopic composition of soil water NO₃⁻ and bulk soil along water flow paths

Mean in soil water was ≤0‰ on hillslope, which is smaller than in throughfall, at all sites except for plot A at S5 (Figure 3 and Figure S4). The negative signatures on hillslopes were associated with low NH₄⁺ and high NO₃⁻ concentrations (relative to those in throughfall; Figure 2, Figures S1 and S2). The signals increased significantly from hillslopes to groundwater discharge zones at the five southern sites (Figure 3 and Figure S4a), while NO₃⁻ concentrations decreased significantly (Figure 2 and Figure S2b). The values in throughfall were distinctively higher than in soil water (Figure 3 and Figure S4b). Similar to , increased from hillslopes to groundwater discharge zones at the five southern sites. in stream runoff was in the range of 15‰ to 20‰, not significantly different from that in the groundwater discharge zone.

At site N1, both and decreased gradually from throughfall to soil water on the hillslope and further to the groundwater discharge zone (Figure 3 and Figure S4). At site N2, with only a single observation at plot D, and were high compared to observations in ground water discharge zones of other sites. This strong ¹⁵N enrichment in NO₃⁻ occurred simultaneously with a sharp decrease in NO₃⁻ concentration (Figure S2b).

Rayleigh fractionation factors for denitrification were calculated for southern sites except S4 (Figure 4), yielding a range from −6.24‰ to −2.97‰. , representing the overall difference in between hillslopes and groundwater discharge zones, was positive in all southern sites and negative in the northern site N1 (Table 2). In the southern sites, varied in a range from 6.9 to 22.2‰, with site S2 having the highest value.

At the southern sites (except S5), bulk soil δ¹⁵N values ranged from −3‰ to +3‰, gradually increasing from hillslopes to groundwater discharge zones (Figure S5). At site S5, bulk soil δ¹⁵N was relatively constant along the hydrological continuum (4‰ to 5.5‰). Across the southern sites, bulk soil δ¹⁵N was negatively correlated with the soil C/N ratio (Figure S6).

3.3 Catchment-scale N mass balance

TIN deposition (throughfall) in the studied forests varied from 5.0 to 48.6 kg N ha⁻¹ year⁻¹ (Table 2), covering the typical range of N deposition rates in Chinese forests (Du et al., 2014). At sites S1 and S2, about half of the N was deposited as NH₄⁺. At the other sites, inorganic N fluxes in throughfall were dominated by NO₃⁻. Compared with earlier throughfall data for the period 2001–2004 (Chen & Mulder, 2007b), this suggests that the relative importance of NO₃⁻ in N deposition has increased significantly during the last 10 years (Huang et al., 2015).

Inorganic N export by stream water was dominated by NO₃⁻ at all sites except N1, where TIN runoff was very small (Table 2). Although increased N deposition caused increased annual stream N export, this relationship was weak and stream runoff only accounted for a small fraction of the observed N loss, similar to what has been reported previously for southern Chinese catchments (Figure 5a) (Larssen et al., 2011; Yu et al., 2017). Catchment N removal rates, computed as the differences between N deposition and stream N export, scaled positively with N input (Table 2). For the southern sites, more than 65% of annual N input was removed within their catchments, with S2 removing most (93%). The remainder was exported from the catchments, dissolved in stream water. The northern site N1 removed most of its N input while site N2 was poorest in attenuating N runoff (42%).

Whole-catchment denitrification fluxes for the five southern sites estimated by the ¹⁵NO₃⁻ model ranged from 2.4 to 21.7 kg N ha⁻¹ year⁻¹ (Table 2). These fluxes accounted for 31% to 97% of the total N sinks (Figure 5b) and were positively related with both N deposition and removal rates. Gross nitrification rates estimated for parameterizing the ¹⁵NO₃⁻ model ranged from 59 to 130 kg N ha⁻¹ year⁻¹ (Table 2). High gross nitrification rates were found at sites with high N deposition rates. Denitrification fluxes estimated with the ¹⁵N-bulk soil model were generally lower (1.9 to 10.1 kg N ha⁻¹ year⁻¹, Table 2).

4 DISCUSSION

Soils on hillslopes seemed to efficiently nitrify NH₄⁺ (Figure 2 and Figure S1), as indicated by the low and values observed during summers (Figure S4). Especially, the significantly smaller in soil water than in throughfall suggests that NO₃⁻ in soil water mostly originated from nitrification and not from atmospheric deposition (Curtis, Evans, Goodale, & Heaton, 2011; Rose et al., 2015). Dual NO₃⁻ isotopic signatures on hillslopes generally fell into a box range of −10‰ to +5‰ and −10‰ to +10‰ for and , respectively (Figure 3), which has previously been used to identify nitrification-derived NO₃⁻ in soils (Kendall et al., 2007; Wexler et al., 2014).

At all southern sites, the decrease in NO₃⁻ concentrations from hillslopes to groundwater discharge zones (Figure 2) was associated with a significant increase of both and (Figure 3). Given the negligible isotopic fractionation by assimilation (Mariotti, Germon, & Leclerc, 1982), this suggests that denitrification in near-stream groundwater discharge zones acts as a significant N sink, similar to what has been reported for temperate forest ecosystems (Osaka et al., 2010) and agroecosystems (Billy, Billen, Sebilo, Birgand, & Tournebize, 2010). A regression of against values for each site showed slopes between 0.5 and 1 (Figure 3), which are well within the range considered diagnostic for denitrification (Kendall et al., 2007; Lehmann, Reichert, Bernasconi, Barbieri, & McKenzie, 2003; Mayer et al., 2002; Wexler et al., 2014). The apparent isotopic effects of denitrification determined with Rayleigh equations generally agreed within the southern sites (not assessed for S4), with an average absolute ¹⁵N enrichment effect of 5‰ (Figure 4), which is similar to the values reported for groundwater denitrification (Bottcher et al., 1990; Mariotti et al., 1988; Spalding et al., 1993). Spatial patterns of dual NO₃⁻ isotopic signatures found in the southern catchments were largely in line with our previous long-term study at S1 (Yu et al., 2016), suggesting efficient N turnover and NO₃⁻ removal in hydrologically connected landscapes. Our finding of similar patterns at four additional subtropical catchments in South China confirms that denitrification in near-stream groundwater discharge zones is quantitatively important and a widespread phenomenon in monsoonal, subtropical forest catchments.

Unlike the southern sites, the northern site N1, having the lowest N deposition rates of all catchments, showed a more continuous decrease of and along the flow path (Figure 3 and Figure S4), suggesting that denitrification is less important for N retention in this catchment. In addition, dual NO₃⁻ isotopic signals indicated that nitrification occurred in the groundwater discharge zone, particularly during dry periods. Besides the smallest N deposition rates (Table 2), N1 has least NO₃⁻ leaching from hillslope soils (Figure 2) and lacks significant changes in NO₃⁻ concentration along the flow path. This indicates that N1 is N-limited, with N largely being assimilated by plants and immobilized in soils, similar to what is commonly seen in N-limited temperate forest ecosystems in Northeastern America (Goodale, 2017; Nadelhoffer, Colman, Currie, Magill, & Aber, 2004).

Even though high (23.4‰) and (30.0‰) were found in pooled soil water samples from the groundwater discharge zone at site N2 (Figure 3), indicating strong denitrification, NO₃⁻ export via stream water was high relative to its input (Table 2). This suggests that the N2 catchment has an overall weak N sink strength. As already indicated the northern sites have drier conditions, both in terms of annual runoff volume and in soil moisture contents of the groundwater discharge zone (Table 1; Figure 1). Thus, we hypothesize that the groundwater discharge zones at the northern sites are less developed and consequently less effective in removing NO₃⁻ through denitrification. This is supported by end-member-mixing analyses (Figure S7 and Table S2), which showed that stream runoff at N2 receives more than 50% of its NO₃⁻-rich water from the hillslope, compared with significantly smaller contributions (<20%) at the southern sites.

Although the southern sites showed a common pattern of NO₃⁻ attenuation (Figure 2; Table 2), the relationship between stream N export and atmospheric N deposition could be further elucidated by catchment hydrology. At the sites with largest precipitation and runoff (S3 and S4), we observed highest N export by stream relative to N deposition (19% and 22%; Table S2), indicating that NO₃⁻-rich hillslope water by-passes the groundwater discharge zone during heavy precipitation events, and directly feeds into the stream (Rose et al., 2014). These findings lead to the conclusion that the efficiency of catchments to remove NO₃⁻ depends on the extent to which groundwater discharge zones act as a conduit for water as it passes from hillslope to stream.

Distinct, hydrologically connected landscape elements with prevailing oxidative or reductive conditions appear to be crucial for N retention observed at the catchment scale. Argic horizons with restricted vertical hydraulic conductivity are widespread in Acrisols of subtropical China and favor the transport of NO₃⁻ produced on hillslopes by lateral “interflow” to groundwater discharge zones in which NO₃⁻ is denitrified. By contrast, direct seepage of NO₃⁻ on the hillslopes to deeper groundwater plays a minor role (Sørbotten et al., 2017). Topographical control on denitrification is well-known (Anderson et al., 2015; Duncan et al., 2013), but seems to be favored by the high degree of hydrological connectivity in southern Chinese forest catchments. This is also supported by increasing bulk soil-δ¹⁵N and decreasing C/N ratios along the topographical gradient of our southern sites (Figures S5 and S6). As ¹⁵N enrichment in soils is often related to losses of ¹⁵N-depleted N (e.g. as NO₃⁻ and N₂; Amundson et al., 2003; Billings & Richter, 2006), the observed ¹⁵N enrichment in groundwater discharge zones likely reflects long-term denitrification activity, as indicated by isotopic enrichment in NO₃⁻ (Figure 3).

Annual N mass balances for the seven catchments indicated that up to 93% of N input via throughfall was removed without being exported by the stream (Table 2). An exception was the northern site N2, which exported more than half of the N received by throughfall to the stream. Possible N sinks comprise N uptake by the biosphere and removal as gaseous N by denitrification (Yanai et al., 2013). Given the excessive amounts of NO₃⁻ leached from well-drained hillslope soils but moderate N export by the streams (Figure 2; Table 2), gaseous N losses from the water-saturated soils appear to be important for the N budget of southern Chinese forests (Van Breemen et al., 2002; Sudduth, Perakis, & Bernhardt, 2013). We therefore estimated catchment denitrification fluxes based on two alternative models, using either ¹⁵NO₃⁻ (Fang et al., 2015) or ¹⁵N-bulk soil (Houlton & Bai, 2009). The estimated fluxes varied from 1.9 to 21.7 kg N ha⁻¹ year⁻¹ (Table 2). Although the two isotopic approaches differed fundamentally in their parameterization, the estimated denitrification fluxes showed a consistent pattern across the five sites, which correlated positively with N deposition rates. Comparing the denitrification rates estimated by the two models, the ¹⁵N-bulk soil approach yielded 22% to 78% lower estimates than the ¹⁵NO₃⁻ model. This has also been found in other studies (Fang et al., 2015; Soper et al., 2017), and attributed to the fact that ¹⁵N signature of bulk soil carries only diluted ¹⁵N effects of denitrification especially on large spatial scales and does not account for the isotopic effect during complete NO₃⁻ consumption by denitrification (Houlton & Bai, 2009). Thus, the very low denitrification fluxes estimated by the ¹⁵N-bulk soil model for site S2 reflects the moderate ¹⁵N enrichment of bulk soil in its groundwater discharge zone (Figure S5), while it showed the highest in soil water (Table 2 and Figure S4) among all sites during summer. This calls for more systematic studies into the in forest catchments to scrutinize temporal patterns, as has been done for site S1 (Yu et al., 2016).

Denitrification estimates by isotopic model approaches are fraught by uncertainties adhering to isotopic fractionation effects for N loss processes (Soper et al., 2017). While isotopic effects for N uptake and leaching show little variation (Evans, 2001; Granger et al., 2010), isotopic effects of denitrification itself vary over large ranges. Isotopic effects have been described ranging from 5‰ to 30‰ for pure culture (Granger et al., 2008; Mariotti et al., 1981) and 6‰ to 65‰ for natural soils (Houlton & Bai, 2009; Wang et al., 2018). The apparent average isotopic effect of denitrification determined with our field data was ~5‰ (Figure 4), which is at the low end of the global reported range. However, this value likely underestimates the actual isotopic effect of in situ denitrification due to complete NO₃⁻ consumption by denitrification or mixing of NO₃⁻ from other sources (Fang et al., 2015). Therefore, we adopted a range of 16‰ to 20‰ for both isotopic models. This range is close to the global average reported for denitrification, and has been used in previous model studies (Fang et al., 2015; Houlton & Bai, 2009). Model uncertainties given in Table 2 show that the ¹⁵NO₃⁻ model is more sensitive to variations in isotopic effects than the ¹⁵N-bulk soil model.

Global isotopic models indicate that fluxes of N gas emission from tropical forests are in the range of 6 to 26 kg N ha⁻¹ year⁻¹ (Bai, Houlton, & Wang, 2012). Site-specific model estimates based on field isotopic measurements reported smaller denitrification fluxes, ranging from 1 to 7.5 kg N ha⁻¹ year⁻¹ (Bai & Houlton, 2009; Soper et al., 2017; Weintraub, Cole, Schmitt, & All, 2016), which may be attributed to a conservative N cycle in these forests with low N deposition (Soper et al., 2017). By contrast, based on the ¹⁵NO₃⁻ model, Fang et al. (2015) estimated denitrification fluxes of 5.6 to 30.1 kg N ha⁻¹ year⁻¹ for a few tropical forests in southern China and a few temperate forests in Japan. Having similar inorganic N deposition rates as our sites (Table 1), their estimated denitrification fluxes were close to or larger than TIN input, while our estimates based on the ¹⁵NO₃⁻ model accounted for 23%–73% of the annual input (Table 2). The larger N loss (gaseous and leaching) than N input may indicate that these forest ecosystems are not at steady state, i.e. that they lose N from organic pools in soil and plant biomass over time which is leached or denitrified. Another reason for the relatively smaller N removal in our study may be that our ¹⁵NO₃⁻ model was based on the isotopic difference between NO₃⁻ found along the lateral water flow path (concentration-weighted whole-soil NO₃⁻) and NO₃⁻ on hillslopes (input), while Fang et al. (2015) considered the isotopic difference between NO₃⁻ in stream water and that of whole soil profiles (Equation 3). Hence, our estimates may miss denitrification in the hyporheic zone and are strictly speaking only valid for NO₃⁻ removal along the hydrological flow path from the hillslope to the groundwater discharge zone. We substantiated N removal by denitrification by plotting the overall change of along the hydrological flow path in each southern catchment (, Table 2) against estimated denitrification fluxes, and found a significant positive correlation (Figure 5c). This validates the idea that obtained along hydrological flow paths may serve as a proxy for the denitrification N sink strength of hydrologically connected landscapes.

Our study provides evidence for significant denitrification in subtropical forests on a regional scale. N mass balances, process identification by isotopic box ranges and model estimates of nitrification and denitrification largely agreed and support our hypothesis that denitrification in groundwater discharge zones acts as a strong N sink. The estimated denitrification fluxes help us to close the N budget for subtropical Chinese forest catchments, which have long been reported to be strong N sinks (Figure 5a; Chen & Mulder, 2007a; Larssen et al., 2011). Denitrification estimated by the ¹⁵NO₃⁻ model accounted for 31% to 97% of total N removal (Figure 5b), illustrating that denitrification may indeed account for a large fraction of the missing N. In addition, denitrification scaled proportionally (slope = 0.45) with N deposition (Figure 5c), implying that subtropical Chinese forests may play an important role in attenuating the ever-increasing atmospheric N loads in this region (Liu et al., 2013) before entering into water courses.