The study was conducted in the Dinghushan Biosphere Reserve (DHSBR) in Guangdong Province, southern China (112∘33′ E, 23∘10′ N), with typical subtropical monsoon climate. Mean annual temperature (MAT) and mean annual precipitation (MAP) are 22.2 ∘C and 1927 mm, respectively. The reserve has experienced high rates of atmospheric N deposition (21–38 kg N ha-1 yr-1 as inorganic N in bulk precipitation) since the 1990s (Fang et al., 2008). From 2009 to 2010, total wet N deposition was 34.4 kg N ha-1 yr-1 (Lu et al., 2013). We used two common forest types that grow on relatively steep slopes in the reserve: an old-growth broad-leaved forest (hereafter referred to as BF) and a pine plantation forest (hereafter referred to as PF) (Mo et al., 2006). The BF is a regional climax, mixed broad-leaved forest, which has been protected for at least the last 400 years with minimum human disturbance (Shen et al., 1999). The PF was planted after a clear-cut of the original climax forest in the 1930s and has been subjected to human disturbances such as litter and shrub harvesting until recent years (Mo et al., 2005).

The major canopy species in the BF are Castanopsis chinensis, Machilus chinensis, Schima superba, Cryptocarya chinensis, and Syzygium rehderianum, and the most common understory species is Hemigramma decurrins. Pinus massoniana and Dicranopteris dichotoma are the dominant tree and understory species in the PF, respectively. No N-fixing tree species were found in the plots. The soil in the reserve is classified as lateritic red earth (Oxisol) formed from Devonian sandstone and shale with a thin layer of forest floor litter (0.5–3.0 cm), but the soil depth is variable, ranging from 30 cm in the PF to more than 60 cm in the BF. Probably due to erosion after the clear-cut and continued human disturbance, the PF had lower total soil carbon, N, and phosphorus (P) content than the BF (Table 1).

We used an ongoing N-addition experiment established in both forests in July 2003 (Mo et al., 2006). The experimental plots used for this study consist of control plots and N-addition treatment at 50 kg N ha-1 yr-1 (hereafter named as N plots), each with three replicates in both forests. Each plot is 10 m × 20 m with at least a 10 m wide buffer strip to the next plot. In the N plots, NH4NO3 is mixed with 20 L of water and was added monthly beginning in July 2003 below the canopy using a backpack sprayer, whereas the control plots received 20 L water with no fertilizer. The added N has δ15N of about -3 ‰ of NH4+-N and about 1.8 ‰ of NO3--N, with δ15N of NH4NO3 being -0.7 ‰.

In both forests, major ecosystem compartments, including leaves, twigs, branches, bark and wood of canopy trees, leaves of understory vegetation, fine roots, and 0–30 cm mineral soil, were sampled in January 2013 to determine their N concentration (%) and δ15N (‰). A branch per dominant tree species per plot was cut from the height reached using a pole pruner (ca. 7–8 m), taking advantage of the steep slope, and was separated into leaves, twigs, and small branches. Bark samples were cut off the dominate trees at breast height using a knife. After removing the bark, wood cores were sampled using an increment borer and were separated visually into sapwood (usually the outer 2–3 cm recent wood) and older wood (heartwood). Dominant understory plant species were cut with a knife and kept separate for each species. A total of seven tree species in the canopy and sub-canopy layers and five plant species in the understory layer (young trees, shrubs, herbs, and liana) of the BF were sampled. In the PF, the dominant pine tree and five species in the understory layer were sampled. Mineral soil samples were taken using an auger (5.1 cm in diameter) and were divided into three layers (0–10, 10–20, and 20–30 cm). Two soil cores were sampled and pooled together to form one composite sample for each depth per plot. Living fine roots were hand sorted from the soil samples for each depth, but they were combined into one composite sample for the whole profile (0–30 cm) because the number of fine roots at each depth was too small to grind and analyze separately. Litterfall was collected monthly during July–September 2012 and was pooled together to make one composite sample per plot. The litter was sorted in the laboratory into leaf and others (branches, fruits, flowers, barks), but only leaf values are reported.

All plant and soil samples were oven-dried at 70 ∘C and ground to a fine and homogeneous powder. Mineral soils were sieved (2 mm mesh) to remove non-soil materials, were air-dried at room temperature, and were milled to fine powder. Subsamples were dried at 105 ∘C, and all results are reported on a 105 ∘C basis. Based on their approximate %N, about 4–5 mg of the samples were weighed into tin capsules, and δ15N and N concentration of the samples were determined simultaneously on an isotope ratio mass spectrometer (Isoprime 100, Isoprime Ltd.) coupled to an automatic, online elemental analyzer (vario ISOTOPE cube). An internal standard needle sample from temperate forests, which was analyzed in multiple runs at several laboratories, was used to check reproducibility of the δ15N determination. We analyzed %N and δ15N separately for each dominant tree species per plot, but compartment mean values are reported. Natural abundance δ15N in samples was reported in per mil (‰) relative to the 15N content of atmospheric N2.

Precipitation, throughfall, surface runoff, and soil solution were sampled monthly from September 2012 to February 2013 (except in the dry December and January, when there was not enough precipitation to generate water samples) in the control plots to assess the δ15N of N input and output in the two forests under ambient N deposition. Bulk precipitation was collected at an open area close to the experimental site using an open glass funnel (12 cm in diameter), connected to a 5 L sampling bottle with polypropylene tubes. Throughfall was collected by PVC pipes at five random points within each plot (with a total intercept area of 0.8 m2) at about 1.3 m above the ground in each forest. Each collector was connected to two 50 L buckets with polypropylene tubes. Soil solutions from 20 cm depth (seepage water) were obtained using two zero-tension tray lysimeters (755 cm2 per tray) installed in each plot. Each lysimeter was connected to a 20 L bottle using the steep slope of the sites to facilitate sampling. In both forests, one selected plot for each treatment was delimited hydrologically by placing stable plastic materials and low cement barriers around them. The cement barriers (covered by the plastic material) on the downslope side of these plots were constructed to enable the sampling of the surface runoff in three sections, which were then used as pseudo-replicates.

Natural 15N abundances of NH4+-N and NO3--N in water samples were analyzed after chemical conversion to nitrous oxide (N2O). The NH4+-N was initially oxidized to nitrite (NO2-) by hypobromite (BrO-) and the NO2- was then quantitatively converted into N2O by hydroxylamine (NH2OH) under strongly acidic conditions (Liu et al., 2014). Similarly, a series of chemical reactions of vanadium(III) chloride (VCl3) and sodium azide under acidic conditions was used to convert NO3--N into N2O (Lachouani et al., 2010). The N2O produced was subsequently analyzed for 15N abundance by a purge and trap coupled with an isotope ratio mass spectrometer (PT-IRMS) (Liu et al., 2014).

To evaluate effects of decadal N addition on whole-ecosystem (plant and soil) %N and δ15N, we determined N-pool-weighted plot means of %N and δ15N using N pools for each compartment and tree species contribution quantified in Gurmesa et al. (2016). We excluded the heartwood and sapwood pools in the plant pool calculations for two reasons: first, the low %N in wood samples caused larger uncertainties in the δ15N determinations, and secondly, heartwood and a major part of the sapwood were formed prior to the initiation of the N-addition treatment. We expect the latter to be the explanation that particular heartwoods showed opposite effects of N addition compared to all other compartments.

Differences between the two forests in plot mean %N and δ15N of the different ecosystem compartments and N-pool-weighted plot means in control plots were analyzed using t tests. The effect of N-addition treatment on %N and δ15N of each tree compartments in the BF and understory leaf in both forests was analyzed using mixed-model ANOVA with treatment as an explanatory factor and plant species as a random factor because plant species differed significantly in both parameters (Gurmesa, 2016). All other tests of treatment effects on %N and δ15N were analyzed using simple t tests on plot means.

Dissolved NH4+-N in water samples in both input (precipitation and throughfall) and output fluxes (surface runoff and soil solution) were 15N-depleted (negative δ15N) in both forests (Table 2). The δ15N of NO3--N was 15N-enriched in precipitation and throughfall and became 15N-depleted in surface runoff and soil solution. However, for dissolved inorganic N (DIN) the concentration-weighted δ15N values (calculated based on data in Table 2 and concentration data in Table S1) were 15N-depleted but slightly increased from precipitation input to soil solution. Mean δ15N of both NH4+-N and NO3--N in input and output fluxes did not significantly differ between the two forests. The temporal variation in δ15N was large (-28 to 2 ‰) for NH4+-N but minor (2 to 5 ‰) for NO3--N (Fig. 1b, d; x axis). The δ15N of NH4+-N values in surface runoff and soil solution were significantly and positively related to the variation in δ15N of NH4+-N in throughfall in both forests (Fig. 1a, b), but the correlation was not significant for NO3--N (Fig 1c, d).

As expected based on the differences in disturbance regime, the BF is more N-rich than the PF. Nitrogen concentrations of plant compartments were significantly higher in the BF than in the PF, except in leaves of canopy trees, litterfall, and fine roots, for which the difference was marginally significant (Table 3). Soil %N was significantly higher in the BF at all depths (Table 3).

Most plant compartments are 15N-depleted with understory and tree leaves, twigs, and branches being most 15N-depleted (below -4 ‰), whereas bark and sapwood were less 15N-depleted within each forest (Table 4). The δ15N values of all plant compartments differ significantly between the two forests, with the PF being more 15N-depleted than the BF (Table 4). Soil δ15N did not a show significant difference between the two forests at any depth (Table 4).

When compared based on N-pool-weighted plot mean, the two forests differed significantly in plant %N and δ15N (Fig. 2a). For the soil, the two forests also differed significantly in N-pool-weighted plot mean %N, with the BF having the higher value, but not in N-pool-weighted plot mean δ15N (Fig. 2b).

Nitrogen concentrations in all measured plant and soil compartments were not significantly affected by N addition in the BF, except in the sapwood (Table 3). In the PF, mean %N values were greater in most plant compartments in fertilized plots, but the change was significant only in leaves of understory plants, whereas soil %N was unchanged (Table 3).

Plant δ15N was negative in both control and N plots in both forests, but N addition significantly increased the δ15N of most plant compartments (Table 4). The changes were more pronounced in the small active plant pools such as leaves of trees and understory plants.

However, the effect of N addition on δ15N was inconsistent in the wood parts (Table 4). For heartwood, the effects were significant for both forests but in different directions than in other plant pools. Due to low %N and challenges in grinding of wood samples, it was difficult to obtain reliable δ15N results for these samples. Much of the sampled wood was also formed prior to the treatment and thus no further evaluation was performed for the wood samples. Nitrogen addition did not cause significant effects on δ15N of litterfall and fine roots. In the BF, there was no correlation between leaf %N and δ15N, but a positive correlation was found for the PF as both %N and δ15N tended to increase in parallel due to N addition (data not shown).

Nitrogen addition tended to decrease soil δ15N in the BF at all depths, but with no significant changes in any layer (Table 4). In the PF, soil δ15N was unchanged by N addition (Table 4).

In summary, the effect of N addition on pool-weighted plot mean plant %N was not significant in either the BF (p= 0.86) or the PF (p= 0.25), but it was more pronounced in the PF (Fig. 2a). However, weighted plot mean plant δ15N values were significantly increased in both forests (p= 0.04 for BF and p= 0.03 for PF) by the N addition. In the soil, where the N pool is obviously larger than in the plants, the effect of the N addition on weighted average %N was not significant in either forest (Fig. 2b). The direction of change in soil δ15N was a decrease, as expected, with incorporation of the added N (δ15N = -0.7 ‰), but the change was again not significant (Fig. 2b).

Deposition N (bulk precipitation and throughfall) was 15N-depleted in NH4+-N and 15N-enriched in NO3--N (Table 2), but since NH4+-N is the dominating N form (Table S1 in the Supplement) DIN deposition is 15N-depleted (-10 to -8 ‰), as also previously reported in the region (Zhang et al., 2008; Koba et al., 2012). The source of the NH4+-N is likely NH3 emissions from activities in the intensively used agricultural land surrounding the DHSBR. Agricultural NH3 emissions are usually 15N-depleted (Bauer et al., 2000). The source of the NO3--N contribution may originate from NOx produced by coal combustion in megacities in Guangdong Province.

The low δ15N of NH4+-N in the soil solution of both forests resembles that in precipitation and throughfall (Table 2), and it is likely due to transport of 15N-depleted throughfall N through macrospores, as supported by the positive relationship between δ15N of NH4+-N in soil solution and that in throughfall (Fig. 1b). Further 15N depletion of NH4+-N (6 to 7 ‰) from throughfall to soil solution may occur by preferential retention of the heavier 15N isotope by cation exchange on soil surfaces (e.g., Karamanos and Rennie, 1978), although preferential nitrification of the lighter isotope could work in the opposite direction. This fractionation effect of nitrification (leaving the substrate NH4+-N 15N-enriched and the product NO3--N 15N-depleted; Högberg, 1997) may explain the relative 15N enrichment of NH4+-N (2 to 6 ‰) from throughfall to surface runoff in both forests (Table 2). A contribution of NO3--N from nitrification of 15N-depleted throughfall NH4+-N during surface runoff passing through the biological active litter layer may also explain the 4 to 6 ‰ 15N depletion of NO3--N from throughfall to surface runoff (Table 2).

While NO3--N is the dominant N form in soil solution (Table S1) and the N-leaching fluxes are almost as large as the N inputs by deposition in both forests (Fang et al., 2009), nitrification is an important process in the soils at the DHSBR. However, as soil-solution NO3--N was 15N-enriched (-1 ‰) relative to the 15N-depleted throughfall NH4+-N (-15 ‰), this cannot be the main substrate for nitrification in the soil. Also, the relatively narrow temporal variation of δ15N for soil solution NO3--N (Fig. 1d) indicates dominance of a substrate for nitrification with stable δ15N content such as soil organic N and/or adsorbed NH4+. Conversely, gaseous losses of 15N-depleted N by denitrification would enrich 15N in soil N as well as soil-solution NO3--N (Houlton et al., 2006). For the BF, denitrification N losses were estimated to be as high as 2.6 kg N ha-1 yr-1 as N2O (Zhang et al., 2008) and 30 kg N ha-1 yr-1 as N2 (Fang et al., 2015). This may explain why DIN in soil solution is slightly 15N-enriched relative to the DIN input (bulk precipitation or throughfall), despite the apparent importance of fractionation via nitrification in the soils of both forests.

Climate is important in regulating global patterns of δ15N in plants and soils (Amundson et al., 2003; Craine et al., 2009, 2015b). Based on the relationships between plant and soil δ15N and climate parameters (MAT and MAP) established by Amundson et al. (2003), the expected δ15N value at the DHSBR is 0.4 ‰ for plants and 5.2 ‰ for the top 10 cm soil. In a global synthesis for forests, Martinelli et al. (1999) reported an average leaf δ15N of 3.7 ± 3.5 ‰ for tropical forests, and a major recent survey across the Amazonian basin observed similar 15N-enriched leaf δ15N levels (3.1 ± 2.3 ‰) (Nardoto et al., 2014). For tropical forest soils, Martinelli et al. (1999) reported 9.3 ± 1.8 ‰ for the top 10 cm. However, the observed leaf δ15N values at the DHSBR were much lower, between -4 and -6 ‰ for the two forests (Table 4). Similar low leaf δ15N values (-2 to -5 ‰) were found in other (sub)tropical forests in eastern Asia (Fang et al., 2011a; Wang et al., 2014; Kitayama and Iwamoto, 2001). The top 10 cm soil δ15N values at the DHSBR (2.2 to 2.6 ‰, Table 4) were again lower than expected from local climate or observations in other tropical forests. Apparently, the ecosystem δ15N values at the DHSBR are closer to the values reported for temperate forests by Martinelli et al. (1999) for leaves (-2.8 ± 1.8 ‰) and for soil (1.6 ± 3.6 ‰), as well as those reported from N-saturated temperate forests (Koopmans et al. 1997; Sah and Brumme, 2003).

Thus, our results reject our first hypothesis that ecosystem δ15N at DHSBR would compare with other observations from warm and humid climates. Furthermore, DHSBR forests were not more 15N-enriched than temperate forests. Martinelli et al. (1999) discussed reasons for the 15N enrichment of tropical forests (relative to temperate forest) and concluded it could result from open N cycles in tropical forests, with fractionation during microbial activities resulting in losses of isotopically light 14N forms, which leave isotopically heavier 15N to cycle internally within tropical ecosystems. Despite noticeable fractionation processes in the soil at DHSBR (Sect. 4.1) and high N availability leading to considerable N losses, there is no evident ecosystem 15N enrichment at the DHSBR or in other Chinese forests with high N deposition (Fang et al., 2011a; Wang et al., 2014).

We suspect this phenomenon to be an imprint from the high and 15N-depleted N deposition (Table 2). The 15N signature of deposition N can alter plant δ15N by direct uptake in the canopy and by altering the signature of available N in the soil (Craine et al., 2015a) (as it is noticeable for NH4+-N in soil solution; Fig. 1b). A similar mechanism involving preferential uptake of particularly 15N-depleted NH4+-N could also explain the occurrence of 15N-depleted plants in tropical rainforests in southern China (Wang et al., 2014). Such influence of deposition N can be region-specific, as shown for some forests in Europe that appear to follow a different trajectory for increasing leaf δ15N with N deposition from that of forests in the USA (Pardo et al., 2006).

The conclusion that plant δ15N is influenced by the 15N-depleted N deposition is further supported by the result that tree-ring δ15N of Pinus massoniana at the DHSBR (sampled near the PF plots) decreased from 2 ‰ in the 1960s to -1 ‰ in the late 1990s and that the decrease was found to coincide with the increasing deposition of 15N-depleted N over the last 50 years (Sun et al., 2010). In line with this finding, long-lived plant compartments (bark and wood) were less 15N-depleted than short-lived compartments (leaves, twigs, and branches) in both forests (Table 4).

The lower soil δ15N in the DHSBR relative to the global average for tropical forest soils may in part also be an imprint from the 15N-depleted N deposition. However, with a N pool of ∼ 2400 kg N ha-1 (equal to more than 60 years of N deposition) in the top 10 cm alone (Gurmesa et al., 2016), the influence should be minor compared to that in short-lived plant compartments that hold a N pool of 1 order of magnitude less.

The steep slopes at the DHSBR may contribute slightly to lower the soil δ15N because steeper slopes promote non-fractionating erosional losses of soil organic matter and decrease the residence time of soil N compared to forests on more gentle slopes, which may conversely have more fractionation from denitrification due to greater soil moisture (Amundson et al., 2003; Hilton et al., 2013; Perakis et al., 2015).

Nitrogen addition increases N availability and is thus expected to increase plant δ15N as a result of fractionation during N uptake and cycling, as discussed above. Several N-addition experiments in temperate forests indeed observed this effect (Högberg et al., 2011, 2014; Korontzi et al., 2000; McNulty et al., 2005; Näsholm et al., 1997). Accordingly, plant δ15N values in both forests at DHSBR were increased by N addition (Table 4, Fig. 2a). The changes in δ15N occurred in small and short-lived plant compartments (e.g., leaves and roots) that are responsive to contemporary N input manipulation (Fang et al., 2006; Johannisson and Hogberg, 1994; Pardo et al., 2002) compared to the large, long-lived, and less responsive compartments (e.g., bark and wood). Such changes in plant δ15N could be a result of fractionation processes, but alternatively it may originate from uptake and incorporation of the added N fertilizer, which had an enriched 15N signature (-0.7 ‰) relative to δ15N of the plants (e.g., -4 to -6 ‰ in leaves).

Assuming that fractionation effects are minor, the decadal N addition with a δ15N value of -0.7 ‰ can be viewed as a tracer addition since it differs from the abundance in the major ecosystem pools. Based on a 15N mass balance calculation (Nadelhoffer and Fry, 1994), and using the control plots as reference, the fraction of added N that was incorporated into plants could be estimated (Table S2). Since the calculation relies on the difference in δ15N between the control and the N plots in the target pool, it is only meaningful when this difference is significant. Thus, the fraction of added N incorporated could only be estimated for the total plant N pool but not for the soil (Fig. 2). The results showed that ∼ 15 % of the total 500 kg N ha-1 added over the last decade was incorporated into plant pools in both forests. For the BF this was less than the estimated fate (24 % to plants) of a stronger tracer (Gurmesa et al., 2016). Nevertheless, it indicates substantial incorporation of input N into plants in the BF even though the N addition did not increase the net uptake in the forest, i.e., no change in %N in plant compartments at the BF.

For soils, N addition tended to decrease δ15N, opposite to results in other long-term experimental N addition (Högberg, 1991; Högberg et al., 1996, 2011), where soil δ15N increased after addition of N. The authors explained that the increase was the result of fertilizer-induced fractionation due to increased N transformation rates. In our study, fractionation may also occur, but the decreasing tendency of soil δ15N indicates that the incorporation of the isotopically lighter added N (relative to the soil) is likely, as discussed by Högberg et al. (2014).

The result supports our second hypothesis that the added N is incorporated into the ecosystem N pools, with plant (and soil) δ15N changing slowly toward the 15N signature of the decadal N addition. This again highlights the importance of the 15N signature of input N in controlling ecosystem δ15N.

As expected from previous studies, the BF is more N-rich than the PF, as indicated by higher %N in major ecosystem pools in the BF (Table 3). Accordingly, plant %N in short-lived compartments (and in the pool-weighted plant pools) did not respond to the decadal N addition in the BF, whereas plant %N in the PF tended to increase, though only significantly in understory plants (Table 3, Fig. 2a). In the BF, the plant tissues were apparently saturated with N, while the PF could still retain part of the addition (Fang et al., 2009). Most plant compartments in the BF are more 15N-enriched than the PF (Table 4) and the change in plant δ15N after decadal N addition was most pronounced in the PF (Fig. 2a). This again could hint at a difference in N status, where the larger changes in plant δ15N in the PF indicate larger incorporation of added N into plants in the PF than in the BF, which is in agreement with our third hypothesis.

The difference under ambient conditions may in part be related to higher N cycling rates and subsequent losses of the lighter 14N in the BF through fractionation and subsequent plant uptake of 15N-enriched soil N (Magill et al., 2000; Zhang et al., 2008; Nadelhoffer and Fry, 1994). Conversely, leaf δ15N in the PF could be more affected by 15N-depleted deposition as the forest is still expanding in biomass and has lower N availability; thus, it might depend more on the 15N-depleted atmospheric N input than the BF does. An additional explanation could be that the PF is dominated by Pinus massoniana, which has ectomycorrhizal fungi, whereas the majority of the plants in the BF have arbuscular mycorrhizal association (Gurmesa, 2016), and ectomycorrhizal plants are found to be more 15N-depleted than arbuscular mycorrhizal plants (Craine et al., 2009; 2015a).

Soil δ15N did not significantly differ between the BF and PF (Table 4, Fig. 2b), although we expected soil to be more 15N-enriched in the BF than in the PF. Soil δ15N values are reported to increase with organic matter age (Bauer et al., 2000), and we expect soil organic matter of the top soil to be older in the BF because this layer might have been lost by erosion in the PF, as could be noted from the lower C, N, and P concentration (Table 1) and lack of depth pattern of soil δ15N in the PF (Fig. 2b). A common feature in soil profiles is 15N enrichment with soil depth (Bauer et al., 2000; Emmet et al., 1998; Koba et al., 2010; Boeckx et al., 2005), as observed in the undisturbed BF, but not in the disturbed PF (Table 4). The absence of 15N enrichment with soil depth may again be an effect of erosion and soil mixing from human disturbances that may shape soil N and δ15N patterns over ecosystem succession (Perakis et al., 2015). The 15N enrichment with depth is known to occur as a result of fractionation followed by removal of lighter 14N by plants, microbes, or through leaching following decomposition, whilst the 15N-enriched N fraction is transported and accumulated at deeper soil profiles (Högberg et al., 2011; Hobbie and Högberg, 2012; Nadelhoffer et al., 1988).