Retention of deposited ammonium and nitrate and its impact on the global forest carbon sink

Study sites

The paired ¹⁵N-tracer experiments were conducted in 13 forest sites, of which nine were in China, two in Europe and two in the USA. These sites vary in mean annual precipitation (MAP) from 700 to 2500 mm, in mean annual temperature (MAT) from 3 to > 20 °C, and in soil types (Fig. 1, Supplementary Table 1, Supplementary Table 2). Ambient N deposition (bulk/throughfall NH₄⁺ plus NO₃⁻) at the sites ranged from 6 to 54 kg N ha⁻¹ yr⁻¹. Forest types at the experimental sites include tropical forests in southern China, subtropical forests in central China, and temperate forests in northeastern China, Europe, and the USA. Data from the sites in Europe, the USA, and six of the nine sites in China have been reported previously. Detailed descriptions of these sites and the related data source references are summarized in Supplementary Table 1. Data for forests at the other three sites in China (Xishuangbanna, Wuyishan, and Maoershan) are originally presented here. The Xishuangbanna sites, which is located Xishuangbanna National Forest Reserve in Menglun, Mengla County, Yunnan Province, is a primary mixed forest dominated by the typical tropical forest tree species Terminalia myriocarpa and Pometia tomentosa. The Wuyishan forest, which is located in the Wuyi mountains in Jiangxi Province, is also a mature subtropical forest with Tsuga chinensis var. tchekiangensis as the dominant tree species in the canopy layer. Other common tree species in the forest include Betula luminifera and Cyclobalanopsis multinervis. Maoershan is a relatively young (45 years) larch (Larix gmelinii) plantation located at Laoshan Forest Research Station of Northeast Forestry University, Heilongjiang Province. A few tree species- Juglans mandshurica, Quercus mongolica, and Betula platyphylla- coexist with Larix gmelinii in the canopy. More information about these sites is also presented in Supplementary Table 1.

¹⁵N-tracer experiment

At all sites, small amounts of ¹⁵NH₄⁺ or ¹⁵NO₃⁻ tracers (generally < 1% of the throughfall deposition) were added systematically to forest floors. In all the Chinese sites except Tieshanping, three replicate plots (20 m × 20 m) were each divided into two halves (10 m × 20 m), with one half receiving ¹⁵NH₄⁺ and the other receiving ¹⁵NO₃⁻. At Tieshanping, ¹⁵NH₄⁺ or ¹⁵NO₃⁻ were separately added to three replicate plots (14 m × 14 m). At Wülfersreuth, ¹⁵NH₄⁺ or ¹⁵NO₃⁻ were added to five replicate plots (40–70 m² plots) established in a young Norway spruce forest⁴³. At Solling, three large replicate plots (300 m²) established in a 72-year-old Norway spruce forest were divided into half, and one half was labelled with ¹⁵NH₄⁺ and the other half was labelled with ¹⁵NO₃⁻. In the ¹⁵N-tracer experiments at Harvard Forest in the US (a red pine forest and an oak-dominated deciduous forest), a single large plot (30 m × 30 m) in each forest was split into two, with one half of the plot receiving ¹⁵NH₄⁺ and the other half receiving ¹⁵NO₃⁻. ¹⁵N tracers were added once in all Chinese sites and at Wülfersreuth, and during two growing seasons (1991 and 1992) at Harvard Forest, and over 3 years (2002–2004) at Solling. Details of the ¹⁵N-tracer addition at each site are summarized in Supplementary Table 3.

Sampling

Major ecosystem compartments including trees, understory vegetation, fine roots, and organic and mineral soil layers were sampled before and ~1 year after the ¹⁵N-labelling in each plot. Understory plants and tree compartments including mature leaves, twigs, branches, and stem woods were collected. Organic and mineral soil layers were sampled separately. For mineral soil, sampling depth varied among sites depending on the local soil conditions; 0–40 cm at all Chinese sites except at the Changbai forest (0–15 cm) and Tieshanping (0–30 cm), 0–65 cm at Wülfersreuth, 0–100 cm at Solling, and 0–20 cm at the two Harvard Forest sites.

Chemical analysis

At each site, analyses for ¹⁵N were conducted using isotope-ratio mass spectrometry on the two sets of plant and soil samples taken before and ~1 year after the ¹⁵N-labelling. The preparation of plant and organic layer samples consisted of oven-drying at 50–70 °C, while soil samples were mostly air-dried, but sometimes oven-dried at temperatures varying between 40–70 °C and then sieved (< 2 mm), varying with the site considered. The ¹⁵N content of all samples at all sites were analyzed using elemental analyser-isotope ratio mass spectrometry while using slightly different systems. Details on the variation in temperatures used in the preparation of the plant, organic layer, and soil samples and in measuring systems used are given in the supplementary material and related data source references.

Calculation of ¹⁵N-tracer recoveries

Percent recoveries (¹⁵N_rec) of the added ¹⁵N tracers in each ecosystem compartment were estimated based on N pool size estimates and changes in ¹⁵N contents of ecosystem pools according to the principle of ¹⁵N mass balance⁴⁴ as shown by Eq. (1) below:

Where N_added is the mass of ¹⁵N we experimentally added (kg N ha⁻¹); N_pool-before and N_pool-after are the N pool (kg N ha⁻¹) in each ecosystem compartment before and 1 year after ¹⁵N labelling, respectively; %¹⁵N_atom-before and %¹⁵N_atom-after are ¹⁵N abundance (%) before and after ¹⁵N labelling, respectively; ¹⁵N_lost is the mass of the ¹⁵N experimentally added that is lost from the ecosystem.

From Eq. (1), we can derive Eq. (2) to calculate the mass of ¹⁵N recovered (¹⁵N_rec) at ecosystem level as:

Assuming that N pool did not change significantly over the study period, we can get Eq. (3) to calculate the ¹⁵N_rec as per cent of the added ¹⁵N as follows:

We used equation (3) to calculate the ¹⁵N-tracer recovery as we also did not account for  net N increment in both plant and soil compartments change in N pool over the relatively short experimental period (about 1 year). Such a small net change in N pool is difficult to detect using the traditional inventory method, which requires repeated measurement on N pool during a longer time (usually every 5 years), especially for soil N pool due to its large background N pool and spatial heterogeneity.

Upscaling of ¹⁵N recovery to global forest N retention

We established conceptual pathways of ¹⁵N retention and partitioning after the ¹⁵N-labelling (Supplementary Fig. 1) and assumed that the unrecovered ¹⁵N is lost by leaching and denitrification. To predict global N retention from results of ¹⁵N recoveries in our paired ¹⁵N-labelling experiments, we considered nine potential predictors from factors that influence ecosystem N retention as suggested in the literature^{9,15,20,30,45} including variables that define climate (MAT, MAP), ecosystem N status or soil fertility (soil C/N ratio, soil clay content, leaf C/N ratio, N deposition), N pool (soil organic mass, wood biomass), and annual net primary production (NPP). Then, we fitted all possible regressions of ¹⁵N retention in plant and soil pools and the ¹⁵N loss fraction with the set of predictor variables across the 13 sites (Supplementary Table 4) using the glmulti package in R. Then, the best regression model was selected based on the minimum corrected Akaike information criterion⁴⁶, constrained by the cutoff of variance inflation factor (VIF) > 3 to avoid multicollinearity among predictors. Global maps of partitioning of deposited NH_x and NO_y (Fig. 1a) to plants, woody biomass and soil as well as loss patterns (Fig. 3, Supplementary Fig. 7) were derived based on the best regression models summarized in Supplementary Table 4, using globally gridded products of the corresponding predictors. In addition, the key variables with significant and the highest explanations for predicting variations in ¹⁵N allocation to plant and soil pools and ¹⁵N loss fraction across the 13 sites are shown in Fig. 2.

Data sources for scaling up N retention

A global map of mean annual NPP was obtained from MODIS NPP product (MOD17, version 5.5) for the period from 2000 to 2015, with a spatial resolution of 1 km⁴⁷. We used C/N ratios of mineral soil at 0–10 cm from three global soil databases, (1) the Harmonized World Soil Database⁴⁸, (2) the gridded Global Soil Dataset for Earth System Modeling (GSDE) of Beijing Normal University (BNU)⁴⁹, and (3) the Global Observation-based Land-ecosystems Utilization Model of Carbon, Nitrogen, and Phosphorus (GOLUM-CNP v1.0) database⁵⁰. Litter mass and C/N ratios of wood, organic, and mineral soil layers were obtained from GOLUM-CNP v1.0. Soil clay content data was also obtained from BNU.

Global N deposition map

The global map of average NH_x and NO_y deposition to forests between the years 2000 and 2010 was created using forest-specific values of NH_x and NO_y inputs obtained from four different models (Supplementary Table 5). This map of global N deposition was combined with a global forest cover map with a spatial resolution of 1 km that is derived from a consensus land-cover product³⁴. Here, a forest cover fraction > 20% in a 1-km pixel was defined as forest. Based on this, we estimated the total global forest area to be ≈42 million km².

Calculation of N-induced C sink

The N-induced C sink was estimated via the stoichiometric upscaling method¹⁹, i.e., by multiplying the N retention in woody tissues of stems, branches, and coarse roots and in the soil with the C/N ratios in these compartments. The C sink due to NH_x and or NO_y deposition was calculated separately using Eq. (4) as follows:

where N_dep is NH_x or NO_y deposition (kg N ha⁻¹ yr⁻¹); ({}^{15}{{{{{{rm{N}}}}}}}_{{{{{{rm{org}}}}}}}^{{{{{{rm{R}}}}}}}), ({}^{15}{{{{{{rm{N}}}}}}}_{{{min }}}^{{{{{{rm{R}}}}}}}) and ({}^{15}{{{{{{rm{N}}}}}}}_{{{{{{rm{wood}}}}}}}^{{{{{{rm{R}}}}}}}) indicate the fraction of deposited NH_x or NO_y allocated to organic layer, mineral soil, and woody biomass, respectively; and ({frac{{{{{{rm{C}}}}}}}{{{{{{rm{N}}}}}}}}_{{{{{{rm{org}}}}}}}), ({frac{{{{{{rm{C}}}}}}}{{{{{{rm{N}}}}}}}}_{{{min }}}), and ({frac{{{{{{rm{C}}}}}}}{{{{{{rm{N}}}}}}}}_{{{{{{rm{wood}}}}}}}) indicate C/N ratios in the soil organic layer, soil mineral layer and woody plant biomass, respectively. f is the fraction we applied to account for flexible C/N in response to elevated N deposition. At elevated N deposition, wood C/N ratio may decrease, and N accumulates without stimulating additional ecosystem C storage. To account for this scenario, we adopted a flexible stoichiometry⁵¹, in which the effects of N deposition on wood C/N ratios are accounted for by multiplying the C/N ratios of wood with a fraction f (from 1 to 0) depending on plant growth response to different rates of N deposition level (kg N ha⁻¹ yr⁻¹). Results of growth responses to experimental N addition and field N gradient studies show plant growth increased with increasing N deposition, flattening near 15–30 kg N ha⁻¹ yr⁻¹ and a reversal toward no enhanced growth response at about 100 kg N ha⁻¹ yr⁻¹ (ref. ^36,52). Therefore, for N deposition < 15 kg N ha⁻¹ yr⁻¹, we assumed that N deposition has no effect on C/N ratios of wood (f = 1). Then, we assumed that f decreases to 0.5 when N deposition reaches 30 kg N ha⁻¹ yr⁻¹ and assumed that f decreases to 0 when N deposition reached 100 kg N ha⁻¹ yr⁻¹ (i.e., N deposition will not increase tree growth anymore, and no new C is gained due to N deposition).

In the soil, deposited N can be retained through assimilation into microbial biomass, immobilization in soil organic matter (SOM), and abiotic immobilization in the soil. It is the fraction of soil retained N that are immobilized in the persistent pool of SOM (as organic N) that can contribute to long-term C sinks in the soil. We assume the fraction to be ≈ 80% based on results from previous ¹⁵N-tracer studies^53,54,55,56.

We used four sets of global N deposition (Supplementary Table 5), six different wood C/N values (Supplementary Table 8), and three different soil C/N values that varied with plant functional type⁵⁰ in our stochastic calculations of the N-induced C sink.

Analysis of uncertainties in global N retention and associated C sink

To estimate uncertainty of the global N retention map, we considered the uncertainty of regression coefficients and the input data for upscaling in Supplementary Fig. 8 for each grid. We randomly sampled 10 out of the 13 sites to do the regression and then upscaled with one randomly selected input dataset for 1000 times using Monte Carlo methods to generate 1000 sets of global maps for retention or loss of deposited NH_x and NO_y (Fig. 3). The standard deviations of these 1000 sets of global maps are defined as the uncertainty of retention maps shown in Supplementary Fig. 8. For the uncertainties of the N-induced C sink, we considered the uncertainty of global retention maps (1000 sets), the uncertainty of global N deposition (four global N deposition maps; Supplementary Table 5), and the uncertainty of wood C/N ratios (six sets; Supplementary Table 8). Thus, we derived 24000 maps of N-induced C sink (4 global N deposition maps × 1000 sets of retention maps × 6 wood C/N ratios × 1 soil organic C/N ratio) for both deposited NH_x and NO_y. The total uncertainty of the N-induced C sink is defined by 95% confidence intervals.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Source: Ecology - nature.com