Introduction

Trees in agroforestry systems can intercept nitrate (NO3−) from adjacent cropland, making agroforestry systems promising land use systems to reduce excessive reactive N. Riparian tree buffers, situated between terrestrial and aquatic ecosystems, and upland alley-cropping agroforestry (i.e., the spatial alternation of tree rows and crop rows in arable land) have been commonly established for enhancement of ecosystem functions (Mosquera-Losada et al. 2018), such as C sequestration, biodiversity improvement, and NO3− leaching reduction (Smith et al. 2013; Torralba et al. 2016; Veldkamp et al. 2023). The incorporation of trees in agroforestry systems can have positive effects on water retention and nutrient recycling (e.g., through root uptake, biomass production, and litter input), potentially lowering NO3− leaching compared to open croplands (Kay et al. 2019; Veldkamp et al. 2023). Aside from tree root uptake of available N, trees in agroforestry systems can also reduce excess NO3− carried from adjacent cropland (Bergeron et al. 2011; Truax et al. 2017) through enhancing denitrification (e.g., Davis et al. 2008; O’Brien et al. 2017). Above- and belowground litter inputs from trees provide available C which, in turn, enhances denitrification during O2-reduced conditions (i.e., microbial NO3− reduction to NO, N2O, and finally N2). High water content and large C availability (as electron donor) can create reducing soil conditions, during which produced N2O is increasingly converted to non-reactive N2, termed complete denitrification (Chapuis-Lardy et al. 2007; Schlesinger 2013; Ye and Horwath 2016). However, during incomplete denitrification as well as in microaerophilic conditions of nitrification (i.e., organic N and ammonium [NH4+] conversion to NO3− with NO and N2O as by-products), the dominant gaseous product could be N2O (Davidson et al. 2000), a potent greenhouse gas and a major cause of ozone depletion in the stratosphere (Ravishankara et al. 2009). The assessment of different agroforestry systems, such as riparian tree buffer versus alley-cropping tree row, will provide insights into how agroforestry trees influence microbial production of N2O and its reduction to N2 in the soil.

Simultaneous measurements of soil N2O and N2 fluxes under field conditions are challenging due to the high atmospheric background concentration of N2 (Groffman et al. 2006), which hampers the quantification of gross N2O production (N2O + N2 emissions) and consumption (N2O reduction to N2) in soil. Moreover, the limitations of traditional approaches for the field-scale estimation of N2 production (e.g., acetylene inhibition, 15N tracer) are well documented (Groffman et al. 2006; Butterbach-Bahl et al. 2013). The 15N2O pool dilution (15N2O PD) technique was developed as an alternative method to measure gross N2O production and gross N2O consumption under field conditions and without extensive soil disturbance (von Fischer and Hedin 2002). Since the 15N2O PD technique has the drawback of possibly being unable to capture all produced N2 in the soil (e.g., N2 produced in anaerobic microsites) (Yang et al. 2011; Wen et al. 2016), Wen et al. (2016) introduced the more appropriate terms of ‘gross N2O emission’ and ‘gross N2O uptake’ (rather than gross N2O production and consumption) for measurements employing 15N2O PD technique. Hereafter, we use the term gross N2O emission for 15N2O PD-measured N2O + N2 emission and gross N2O uptake for 15N2O PD-measured N2O-to-N2 reduction. To date, there is a gap in the literature for improving the budgets of N2O sources and sinks and for understanding their controlling factors. Despite the need for field-based measurements of soil gross N2O emissions and uptake (Almaraz et al. 2020), the 15N2O PD technique has been applied in a few terrestrial ecosystems only (Yang et al. 2011; Yang and Silver 2016a,b; Wen et al. 2017; Luo et al. 2022). Employment of the 15N2O PD technique in the field can further our mechanistic understanding of the magnitudes of gross N2O and N2 fluxes and their controlling factors to better constrain ecosystem N budgets.

Previous field studies that assessed the role of riparian buffers in NO3− reduction by denitrification have employed static chamber methods to quantify soil N2O fluxes in riparian buffers (e.g., Hinshaw and Dahlgren 2016; Poblador et al. 2017; Kandel et al. 2019). These studies reported that riparian buffers can efficiently reduce NO3− via denitrification with large N2O emissions, but such measurement method can only quantify net N2O efflux and excludes the fraction that is further reduced to non-reactive N2. In contrast, the 15N2O PD technique can quantify both gross N2O emissions and uptake, which can help uncover their controlling factors and improve management to reduce adverse effects of N2O production. Factors regulating gross N2O emissions and uptake in soils, such as mineral N (i.e., electron acceptor), C availability (i.e., electron donor), and moisture content have been documented in a forest stand (Wen et al. 2017), a salt marsh landscape (Yang and Silver 2016a), and a fertilized corn field (Yang and Silver 2016b). In our recent work conducted in alley-cropping agroforestry and open cropland (Luo et al. 2022), we showed that a high mineral N-to-available C ratio increases gross N2O emissions from agroforestry crop row and open cropland whereas a low mineral N-to-available C ratio contributes to a large gross N2O uptake in the agroforestry tree row. These soil factors are highly variable both in space and time and it remains unknown how these factors influence the seasonal and spatial variations of gross N2O emissions and uptake in riparian buffers.

Riparian buffers are often viewed as hot spots of biogeochemical processes, such as denitrification, generally fuelled by subtle changes in electron acceptor and donor availabilities as well as redox and hydrological conditions (Hefting et al. 2006; Burgin and Groffman 2012; Kachenchart et al. 2012). Riparian zones often present shallow water tables and high soil organic matter content, which favor N2O production (Fisher et al. 2014) and subsequent N2O reduction to N2. N2O production and reduction can both occur during denitrification, a modular pathway encoded by several functional genes harbored by a wide range of bacteria and archaea (Nadeau et al. 2019). Moreover, soil microbial communities and denitrification gene abundance are strongly affected by soil biochemical conditions including soil redox conditions, available C, and mineral N (Beule et al. 2019, 2020; Ma et al. 2020), indicating larger denitrification potential in riparian buffers relative to upland agroforestry.

In riparian buffers, seasonal changes in water level in adjacent aquatic systems result in fluctuating aerobic and anaerobic conditions (Burgin et al. 2010), triggering changes in biochemical and microbial processes and affecting N2O concentrations across soil depths (DeSimone et al. 2010). However, the contribution of subsurface-produced N2O to surface fluxes is still poorly understood. Soil NO3−, C availability, moisture content or redox conditions, and denitrifier community change with soil depth (Yang and Silver 2016a; Han et al. 2020), which are expected to affect N2O emission and uptake in subsoils, and ultimately affect surface gross N2O fluxes. McCarty and Bremner (1992) found that N2O emissions in subsoils are not restricted by denitrifying organisms but by available C, whereas van Groenigen et al. (2005) reported that N2O emissions in subsoils are mainly affected by soil moisture rather than available C or N. However, gross N2O emission and uptake and their underlying controlling factors at different soil depths yet remain to be thoroughly investigated.

Our study aimed to (1) quantify gross N2O emissions and uptake in two contrasting agroforestry systems (riparian tree buffer on poorly drained soil versus alley-cropping tree row on well-drained soil), and (2) determine the relationships between these gross N2O fluxes, substrate levels, and soil microbial community. We hypothesized that following the patterns of available C and mineral N as substrates, (1) the riparian tree buffer will have higher gross N2O emission and uptake than the tree row of alley cropping; the topsoil will have higher gross N2O emission and uptake than the subsoil, and (2) substrate availabilities rather than denitrification gene abundance will control gross N2O fluxes.

Materials and methods

Site description

Our study was conducted within the tree rows of two contrasting agroforestry systems: a riparian tree buffer and an alley-cropping tree row (12-m wide rows of trees alternated with 48-m wide rows of crops) in cropland agroforestry. The riparian tree buffer (Fig. S1a and b) was established on a poorly drained Gleysol soil near the Dramme stream, with a high groundwater level that reached ~ 20-cm depth during high rainfall, located south of Goettingen, Germany (51°26′2″ N, 9°50′10″ E, 227 m above sea level). The alley-cropping tree row (Fig. S1c) was established on a well-drained Phaeozem soil, located near Dornburg, Germany (51°00′40″ N, 11°38′46″ E, 289 m above sea level). Soils at both sites had developed on loess deposits and field examination of the upper 50-cm depth revealed that soil texture at both sites was dominated by silt. The main difference in soil characteristics between these two agroforestry systems was soil drainage (i.e., poorly drained in the riparian buffer in contrast to a well-drained alley-cropping system). Both sites had comparable mean annual temperature and precipitation, with 9.8 ± 0.3 ℃ and 595 ± 34 mm in the riparian tree buffer, and 10.7 ± 0.3 ℃ and 567 ± 32 mm in the alley-cropping tree row (2010–2019; German Meteorological Service). The trees of both agroforestry systems were hybrid poplar clones ‘Max 1’ (Populus nigra × P. maximowiczii) and did not receive any fertilizer application, as commonly practiced in temperate agroforestry systems (Schmidt et al. 2021). The first harvest of the trees of the alley-cropping tree row for biofuel feedstock was in 2015, eight years after planting in 2007, while the trees in the riparian tree buffer were not harvested since planting in 2012. The adjacent crop in the alley-cropping tree row was barley whereas in the riparian tree buffer, the adjacent area was a fallowed cropland.

Sampling design and measurement of gross N2O emission and uptake

Within the tree rows of each agroforestry system, four sampling locations (serving as replicate plots; see Statistical analysis below) were selected. At each replicate plot, four intact cores (5 cm height, 8 cm diameter) were taken from the topsoil (0–5 cm), and one intact core (20 cm height, 8 cm diameter) was taken from the subsoil (40–60 cm). The four topsoil intact cores from each replicate plot were placed together into an air-tight chamber (Fig. S1d); the subsoil intact core was put in another chamber, and all incubations commenced immediately in the field. Gross N2O emission and uptake, net N2O and CO2 fluxes, and other soil factors (see below) were measured separately for the topsoil and subsoil from the intact cores. Measurements were conducted in early spring (April), late spring (June), summer (August), and autumn (October) of 2019.

We measured gross N2O emission and uptake using the 15N2O PD technique as described by Wen et al. (2016, 2017). This technique assumes a homogeneous mixture of injected 15N2O with soil-borne N2O in the soil air-filled pores, indicating that the soil pores must be interconnected to the soil surface for homogenous mixing with the introduced 15N2O in the chamber headspace (Wen et al. 2016). Soil cores from the top- and subsoil of each replicate plot were separately placed in air-tight chambers (glass desiccator of 6.6 L volume), equipped with a Luer-lock stopcock on the lid (Fig. S1d). We injected 7 mL of 15N2O label gas (Westfalen AG, Münster, Germany) into the chamber headspace. The label gas contained 100 ppmv 98% single labeled 15N-N2O and 275 ppbv sulfur hexafluoride (SF6, as a tracer for possible physical loss of gases from the chamber headspace), balanced with synthetic air (O2 21%, N2 78.1% and other gases H2O, NOx, CnHm). The injection of 15N-labelled N2O increased the headspace concentration by ~ 125 ppbv N2O with 13.2% 15N initial enrichment and 292 pptv SF6. Gas samples of 100 mL and 23 mL were collected from the chamber headspace after 0.5, 1, 2, and 3 h of on-site incubation and injected into pre-evacuated 100-mL glass bottles and 12-mL glass vials (Exetainer; Labco Limited, Lampeter, UK) with rubber septa, respectively. The 100-mL gas samples were analyzed for 15N-N2O using an isotope ratio mass spectrometer (Finnigan Deltaplus XP, Thermo Electron Corporation, Bremen, Germany). The 23-mL gas samples were analyzed for N2O and SF6 concentrations using a gas chromatograph (SRI 8610C, SRI Instruments Europe GmbH, Bad Honnef, Germany) equipped with an electron capture detector (ECD) and a make-up gas of 5% CO2-95% N2 (v/v; 5.0 purity grade). The CO2 concentrations were determined from the same gas sample using the same gas chromatograph with a methanizer coupled to a flame ionization detector (FID). Concurrent with the field incubation, ambient air samples were collected (23 mL for the determination of ambient N2O concentrations and 100 mL for analysis of the natural abundance of 15N2O signatures), which were needed in the gross N2O flux calculations. Atmospheric N2O concentration was 347.5 ± 0.7 ppbv and 15N natural abundance was 0.3695 ± 0.0001 atom% across sites and sampling dates. Details on the principles and calculations of gross N2O emission and uptake are given in our previous works (Wen et al. 2016, 2017). Net soil N2O and CO2 fluxes were calculated from the linear increase of their concentrations over the incubation period and adjusted with the field-measured air temperature and pressure (e.g., Matson et al. 2017). Gross N2O and net N2O and CO2 fluxes were expressed on the basis of dry mass soil, calculated from subsequently measured gravimetric moisture content of the soil cores. Soil CO2 flux as soil respiration served as a proxy for C utilization activity (Luo et al. 2022).

Soil factors

Immediately following gross N2O flux measurements, the four soil cores of the 5-cm depth topsoil were composited, and the 20-cm long soil core of the subsoil was broken up, obtaining two separate soil samples for the topsoil and subsoil in each replicate plot. Each composite soil sample was thoroughly mixed and measured for water-filled pore space (WFPS), NO3−, NH4+, microbial biomass N (MBN), biodegradable organic C (BDOC), and DNA-based abundance of soil bacteria, fungi, and denitrification genes (for the latter, ~ 20 g fresh soil was placed into a sterile 15-mL polypropylene Falcon tube and frozen at −20 ℃ while in the field). Additionally, soil temperature was measured for each soil depth. WFPS was calculated from the gravimetric moisture content (oven-dried at 105 °C for 24 h), using the measured soil bulk density (determined by the soil core method; Blake and Hartge 1986) and a mineral soil particle density of 2.65 g cm−3. In the field, fresh soil was placed into a pre-weighed extraction bottle containing 150 mL 0.5 M K2SO4 for the determination of extractable mineral N (NH4+, NO3−). Upon arrival at the laboratory, these bottles were shaken for 1 h, filtered through pre-washed (with 0.5 M K2SO4) filter papers, and the extracts were stored at −20 °C until further analysis. MBN was determined by chloroform fumigation-extraction method (Brookes et al. 1985) by fumigating 20 g fresh soil for 5 days followed by extraction with 100 mL 0.5 M K2SO4 as described above. The dry mass of the extracted fresh soils and the fumigated soils was calculated using the gravimetric moisture content. MBN was calculated as the difference in total extractable N between the paired fumigated and un-fumigated samples divided by kN of 0.68 (Shen et al. 1984). Extractable mineral N and total extractable N were analyzed using continuous flow injection colorimetry (SEAL Analytical AA3, SEAL Analytical GmbH, Norderstedt, Germany), where NH4+ was determined by salicylate and dichloroisocyanuric acid reaction, NO3− by cadmium reduction method with NH4Cl buffer, and total extractable N by ultraviolet-persulfate digestion followed by hydrazine sulfate reduction. BDOC was determined by batch incubation, modified from McDowell et al. (2006) and Jones and Willett (2006). Approximately 150 g of fresh soil was shaken in 750 mL of deionized water for 1 h. The extract was centrifuged at 3871×g for 20 min and passed through a 4 µm cellulose filter paper (Whatman 10,311,645). From each filtrate, 50 mL were transferred to one of six 100-mL flasks, covered with perforated parafilm to allow air exchange, and incubated at room temperature in the dark. At 0, 1, 3, 7, 14, and 28 days of incubation, the filtrate of one of the six flasks was passed through a 0.2 µm cellulose acetate filter paper (Sartorius; Goettingen, Germany), and stored at −20 °C until analysis. Dissolved organic carbon (DOC) concentrations were determined using ultraviolet-enhanced persulfate oxidation using a Total Organic Carbon Analyzer (TOC-Vwp, Shimadzu Europa GmbH, Duisburg, Germany). BDOC was determined as the difference between the maximum and minimum DOC concentrations during the incubation period. General soil characteristics (pH, soil organic C [SOC], total N, base saturation, and effective cation exchange capacity [ECEC]) of the topsoil and subsoil of both agroforestry systems were determined at the beginning of our study, using standard methods described previously (Schmidt et al. 2021) (Table 1).

Table 1 Soil characteristics (means ± SE; n = 4 plots) in two agroforestry systems
Full size table

Quantification of bacteria, fungi, and denitrification genes in soil

Part of the composited soil samples from each depth per replicate plot after gross N2O flux measurements was frozen while in the field, stored at −20 °C upon arrival at the laboratory, and subsequently freeze-dried. Freeze-dried samples were homogenized as described earlier (Beule et al. 2019) and DNA was extracted from 200 mg of finely ground soil using a phosphate lysis buffer protocol optimized for subsoils (Guerra et al. 2020). Briefly, soil material was suspended in 1 M phosphate buffer containing 0.5% (w/v) sodium dodecyl sulfate, the mixture was centrifuged, and the supernatant was extracted using phenol and chloroform-isoamyl alcohol. DNA was precipitated using polyethylene glycol-NaCl and the extracts were checked on 0.8% (w/v) agarose gels stained with ethidium bromide. To overcome polymerase chain reaction (PCR) inhibition, the extracts were tested for PCR inhibitors using a DNA amplification inhibition test (Guerra et al. 2020) and diluted accordingly prior to PCR. The abundance of soil bacteria, fungi, and denitrification genes (nitrite reductase genes nirK and nirS, and N2O reductase genes nosZ clade I and II) was analyzed using real-time PCR as described previously (Beule et al. 2020). Briefly, soil DNA extracts were amplified in 4-µL reaction volumes in 384-well microplates in a CFX384 Thermocycler (Bio-Rad, Rüdigheim, Germany). The choice of primers are listed in Table S1. The composition of the reaction volumes and the thermocycling conditions are identical to those used by Beule et al. (2020).

Statistical analysis

The distance between the replicate plots (≥ 15 m) was tested for spatial independence using the first data set of soil gross and net N2O fluxes. Based on von Neumann’s ratio test for randomness (Bartels 1982), these fluxes were not auto-correlated. Thus, our replicate plots were considered biological replicates in our statistical analysis. Each parameter was tested for normal distribution using Shapiro–Wilk’s test and equality of variance using Levene’s test. Seasonal differences for each agroforestry system and soil depth were assessed using one-way ANOVA followed by the Tukey HSD test (if parameters exhibited equal variance and normal distribution) or the Kruskal–Wallis test with multiple comparison extension (if parameters did not meet such assumptions). Considering repeated measurements across seasons, differences between agroforestry systems or between soil depths for each agroforestry were tested using the linear mixed-effects (LME) models, whereby agroforestry system or soil depth was considered the fixed effect and sampling days and replicate plots were random effects (Crawley 2007). Prior to LME analysis, parameters that did not meet normal distribution were either square-root- (i.e., gross N2O emissions, WFPS, total mineral N (NH4+  + NO3−), soil respiration, BDOC) or log-transformed (i.e., net N2O fluxes, MBN, bacteria, fungi, and denitrification gene abundance). The LME models include a variance function (varIdent) to account for heteroscedasticity of the fixed effect and/or a first-order temporal autoregressive function that accounts for decreasing correlation of the measurements with increasing sampling days (Zuur et al. 2009) if this improved the relative goodness of the model fit based on the Akaike information criterion.

As the soil factors commonly exhibited similar distribution (e.g., denitrification genes), we minimized the redundancy of their roles in influencing gross N2O emissions by aggregating distribution-related factors using the principal component analysis (PCA), as used by Barnes et al. (2017). The PCA was conducted on square-root- or log-transformed values (see above) and revealed three groups: (1) indicators of substrate availability (i.e., NH4+, NO3−, soil respiration, and BDOC), (2) microbial community size (i.e., MBN, bacteria, and fungi), and (3) denitrification gene abundance (i.e., nirK, nirS, nosZ clade I, and II). The first component (PC1) explained 68% to 94% of the total variance of each group (Fig. S2). We used the PC1 scores of each replicate plot per sampling day as the new aggregated value. The PCA showed that the lower the PC1 scores, the larger the parameter value; in order to make it intuitive, we used the inverse of PC1 scores so that high PC1 scores directly indicate large parameter values (Fig. S3), as was used by Barnes et al. (2017). Using the PC1 scores, we assessed the causal links of the three groups of controlling factors on soil gross N2O emissions using a structural equation model (SEM; ‘lavaan’ R package) (Barnes et al. 2017). For the SEM, it is necessary to use ordinal values for the categorical variables; i.e., alley-cropping tree row = 1 and riparian tree buffer = 2 for analyzing both agroforestry systems for a specific soil depth, or 40–60 cm = 1 and 0–5 cm = 2 for analyzing both depths of an agroforestry system. The quality of the SEM model was assessed using the chi-square goodness of fit (P > 0.05), standardized root mean square error of approximation (value < 0.08), and comparative fit index (value > 0.9) (Kline 2014). To explore temporal correlations between soil gross N2O fluxes and individual controlling factors, we used Spearman’s rank correlation test. All statistical tests were considered significant at P ≤ 0.05. We conducted all statistical analyses using R version 3.6.3 (R Core Team 2019).

Results

Soil gross N2O fluxes, soil factors, and microbial population size

Gross N2O emissions and uptake in the topsoil were higher in riparian tree buffer than in alley-cropping tree row (P < 0.03; Fig. 1a and c) whereas gross N2O fluxes in the subsoil did not differ between agroforestry systems (P > 0.06). Comparing soil depths at each agroforestry system, the gross N2O emissions were marginally higher (P = 0.06) in the topsoil than the subsoil in the riparian tree buffer but there was no difference (P = 0.53) between soil depths in the alley-cropping tree row (Fig. 1a and b). Gross N2O uptake was two to eight times higher than the measured net negative N2O fluxes (cf. Figure 1c and e).

Fig. 1
figure 1

Soil gross N2O emission (a, b), gross N2O uptake (c, d), and net N2O flux (e, f) at depths of 0–5 cm (topsoil) and 40–60 cm (subsoil) in riparian tree buffer and alley-cropping tree row. Means ± SE across seasons in panels a, c, and e (n = 4 plots) followed by different capital letters indicate significant differences between agroforestry systems for the topsoil and different small letters for the subsoil (linear mixed-effects models at P ≤ 0.05)

Full size image

Riparian tree buffer had higher WFPS, total mineral N, soil respiration, and BDOC than alley-cropping tree row (P < 0.05; Fig. 2a, c, e, g). In the riparian tree buffer, the topsoil had higher mineral N and BDOC than the subsoil (P < 0.02; Fig. 2c and g). Similarly, in the alley-cropping tree row, the topsoil had lower WFPS but higher mineral N and soil respiration than the subsoil (P < 0.04; Fig. 2a, c, e).

Fig. 2
figure 2

Soil water-filled pore space (a, b), total mineral N (c, d), soil respiration (e, f), and biodegradable organic C (g, h) at depths of 0–5 cm (topsoil) and 40–60 cm (subsoil) in riparian tree buffer and alley-cropping tree row. Means ± SE across seasons in panels a, c, e, and g (n = 4 plots) followed by different capital letters indicate significant differences between agroforestry systems for the topsoil and different small letters for the subsoil (linear mixed-effects models at P ≤ 0.05)

Full size image

Topsoil MBN and bacterial abundance were greater in riparian tree buffer than in alley-cropping tree row (P < 0.01; Fig. 3a and c). Conversely, subsoil MBN, bacterial, and fungal abundance were higher in alley-cropping tree row than in riparian tree buffer (P < 0.04; Fig. 3a, c, e). All three measures of the microbial community size (MBN, bacterial 16S rRNA, and fungal 18S rRNA gene abundance) decreased from the topsoil to the subsoil (P < 0.01; Fig. 3a, c, e).

Fig. 3
figure 3

Soil microbial biomass N (a, b), bacterial 16S rRNA (c, d), and fungal 18S rRNA gene abundance (e, f) at depths of 0–5 cm (topsoil) and 40–60 cm (subsoil) in riparian tree buffer and alley-cropping tree row. Means ± SE across seasons in panels a, c, and e (n = 4 plots) followed by different capital letters indicate significant differences between agroforestry systems for the topsoil and different small letters for the subsoil (linear mixed-effects models at P ≤ 0.05)

Full size image

Comparing agroforestry systems, the abundance of nirK, nirS, and nosZ clade I genes in the topsoil were greater in riparian tree buffer than in alley-cropping tree row (P < 0.01; Fig. 4a, c, e). In the subsoil, the riparian tree buffer harbored more microorganisms that carried nirS genes than the alley-cropping tree row (P = 0.01; Fig. 4c). Conversely, nirK and nosZ clade II gene abundance in the subsoil were greater in alley-cropping tree row than riparian tree buffer (P < 0.01; Fig. 4a and g). Comparing soil depths, the abundance of denitrification genes decreased from the topsoil to the subsoil in both agroforestry systems (P < 0.01; Fig. 4a, c, e, g).

Fig. 4
figure 4

Soil denitrification gene abundance: nirK (a, b), nirS (c, d), nosZ clade I (e, f), and nosZ clade II (g, h) at depths of 0–5 cm (topsoil) and 40–60 cm (subsoil) in riparian tree buffer and alley-cropping tree row. Means ± SE across seasons in panels a, c, e, and g (n = 4 plots) followed by different capital letters indicate significant differences between agroforestry systems for the topsoil and different small letters for the subsoil (linear mixed-effects models at P ≤ 0.05)

Full size image

Factors controlling gross N2O fluxes

The SEM was used to disentangle the three groups of controlling factors (i.e., substrate availability, denitrification gene abundance, and microbial population size) on their influences on gross N2O emissions, in cases where there existed differences between agroforestry systems or between soil depths. As gross N2O emissions only differed between riparian tree buffer and agroforestry tree row in the topsoil (Fig. 1a and b), the SEM was hence carried out separately for the topsoil with both agroforestry systems included (Fig. 5a). Likewise, gross N2O emissions differed between depths only in the riparian tree buffer (Fig. 1a and b), and thus the SEM was conducted for this system with both depths included (Fig. 5b). In the topsoil (Fig. 5a), the positive path coefficients of the three soil factor groups illustrated their congruent patterns that increased from alley-cropping tree row to riparian tree buffer (Fig. S3a). Similarly, in the riparian tree buffer (Fig. 5b), the positive SEM path coefficients of the three groups of soil factors indicated their consistent increase from the subsoil to the topsoil (Fig. S3b). The SEM paths revealed that among these three groups of soil factors, substrate availability rather than denitrification gene abundance or microbial population size dominantly influenced gross N2O emissions in the topsoil of both agroforestry systems (Fig. 5a) and in both soil depths of the riparian tree buffer (Fig. 5b). As expected, there was a positive link between microbial population size and denitrification gene abundance (Fig. 5a and b) since denitrification genes are widespread in soil microorganisms.

Fig. 5
figure 5

Structural equation modeling (SEM) of the causal links of soil factors controlling gross N2O emissions in 0–5 cm depth (topsoil) of both agroforestry systems across seasons (a), and in 0 – 5 cm and 40–60 cm depths (subsoil) of riparian tree buffer across seasons (b). Substrate availability is the principal component (PC)1 score, composited from soil NH4+, NO3⋅− concentrations, soil respiration, and biodegradable organic C (BDOC). Denitrification gene abundance is the PC1 scores, aggregated from nirK, nirS, nosZ clade I, and nosZ clade II. The microbial community is the PC1 scores, composited from bacterial 16S and fungal 18S rRNA gene abundance, and microbial biomass N (MBN). Values next to the lines are standardized path coefficients (at *P < 0.05, **P < 0.01, ***P < 0.001); green lines represent positive pathways. Grey dashed lines represent non-significant pathways. Single-headed and double-headed arrows refer to unidirectional and bidirectional relationships, respectively. The quality of models was assessed by the chi-square goodness of fit (P > 0.05), standardized root mean residual (SRMR, value < 0.08), and comparative fit index (CFI, value > 0.9). Agroforestry system and soil depth were ordinal categorical variables in SEM: alley-cropping tree row and 40–60 cm depth were encoded by a value of 1; riparian tree buffer and 0–5 cm depth were assigned by a value of 2

Full size image

Gross N2O emissions were highly correlated with gross N2O uptake across agroforestry systems, soil depths, and seasons (rho = 0.96, n = 64, P < 0.01; Fig. S4a). As to the relationships with soil factors, gross N2O emissions were correlated with total mineral N, BDOC, and ratios of nosZ clade I-to-clade II genes in the topsoil of both agroforestry systems (rho = 0.44–0.52, n = 32, P < 0.01; Fig. S4b, c, d). Among these factors, autocorrelations were observed between the ratio of nosZ clade I-to-clade II genes and total mineral N (rho = 0.75, n = 32, P < 0.01; Fig. S4e) and BDOC (rho = 0.52, n = 32, P < 0.01; Fig. S4f). In the subsoil of both agroforestry systems, gross N2O emissions were positively correlated only to total mineral N (rho = 0.45, n = 32, P = 0.01; Fig. 6c).

Fig. 6
figure 6

Spearman’s rank correlations between gross N2O uptake, soil respiration (a) and nirK gene abundance (b) in the topsoils of riparian tree buffer and alley-cropping tree row, and between gross N2O emission, gross N2O uptake and total mineral N (c, d) in the subsoils of riparian tree buffer and alley-cropping tree row

Full size image

Gross N2O uptake in the topsoil of the riparian tree buffer and alley-cropping tree row was positively correlated with soil respiration and nirK gene abundance (rho = 0.38–0.40, n = 32, P < 0.033; Fig. 6a and b). In the subsoil, gross N2O uptake in riparian tree buffer and alley-cropping tree row showed a positive relationship with total mineral N (rho = 0.35, n = 32, P = 0.05; Fig. 6d). We detected no other significant correlations among measured parameters.

Discussion

Gross N2O emissions

The larger gross N2O emissions (i.e., encompassing the emitted N2O and the N2O reduced to N2) from the topsoil of riparian tree buffer than alley-cropping tree row (Fig. 1a) were mainly driven by the larger substrate availabilities (Figs. 2c, e, g) and higher WFPS (Fig. 2a) in the poorly drained riparian buffer. This was also illustrated by the positive correlations of gross N2O emissions with available C (electron donor; Fig. S4b) and mineral N (electron acceptor; Fig. S4c), suggesting that the N2O- and N2-producing processes (i.e., nitrification and denitrification) were enhanced in the riparian buffer where soil moisture was high (Fig. 2a and b). In early spring as poplar leaves sprouted (Fig. S5a) and presumably tree roots had still low N uptake, soil mineral N peaked (Fig. 2d). Towards summer with flourishing leaves (Fig. S5b and c) and possibly increasing tree root uptake, soil mineral N declined and remained low in fall (Fig. 2d) when leaves senesced (Fig. S5d); these suggest indirect effect of tree root uptake, via soil mineral N as a substrate, on microbial N2O- and N2-producing processes.

Despite larger microbial population size (Fig. 3a and c) and denitrification gene abundance in the topsoil of riparian tree buffer than alley-cropping tree row (Fig. 4a, c, e), these microbial factors did not correlate with the gross N2O emissions. This was caused by the seasonally invariable microbial population and denitrification gene abundance which did not reflect the seasonal variations of gross N2O emissions. Although there was a positive correlation of gross N2O emissions with the ratio of nosZ (encoding for N2O reductase) clade I-to-nosZ clade II (Fig. S4d), this ratio was autocorrelated with mineral N and BDOC (Fig. S4e and f), suggesting that the effect of nosZ gene abundance on gross N2O emissions was mainly facilitated by soil N and C availabilities. Thus, in agroforestry systems in which high levels of WFPS can occur, substrate availabilities promote gross N2O emissions, which agrees with previous findings (Yang and Silver 2016a, b; Wen et al. 2017). For the subsoil, despite larger available C (Fig. 2e and g) and higher WFPS (Fig. 2a) in riparian buffer than alley-cropping tree row, gross N2O emissions of both agroforestry systems were low (Fig. 1a), which was driven by their low mineral N (Fig. 2c and d). These findings imply that the riparian tree buffer performed better than the alley-cropping tree row in reducing surficial NO3− to N2O and N2. Furthermore, the subsoil under riparian tree buffer with its large available C and high WFPS has substantial potential to reduce NO3− when this is transported to the subsoil.

There are only few studies that concurrently investigated denitrification gene abundance and gross N2O fluxes (e.g., Luo et al. 2022; Stuchiner and von Fischer 2022). It is worth exploring how denitrifying populations adapt to environmental differences between two contrasting agroforestry systems. The positive correlations of nosZ clade I-to-nosZ clade II ratio with mineral N and BDOC (Fig. S4e and f) suggested a niche differentiation of the two clades of nosZ-carrying microorganisms in response to the available substrates. Specifically, nosZ clade I, dominating in riparian tree buffer (Fig. 4e), corresponded to high N and C availabilities particularly in early spring (Fig. S4e and f) whereas nosZ clade II, dominating in topsoil of alley-cropping tree row (Fig. 4g), corresponded to low N and C availabilities (< 3.5 mg N kg−1, < 15 mg C kg−1 of BDOC, and < 1.4 mg C kg−1 h−1 of soil respiration; Fig. 2c, e, g). This suggests that nosZ clade I denitrifiers in topsoils are responsive to seasonal changes in N and C availability (Tsiknia et al. 2015) whereas conditions of low N and C availabilities seem to favor the expression of nosZ clade II genes (Semedo et al. 2021). Moreover, at our time of sampling, trees of the riparian tree buffer were already seven years old while the trees of alley-cropping tree row were four years after their first harvest of aboveground wood biomass. The longer period of plant litter addition, together with the inherently larger topsoil SOC content at the riparian site (Eglin et al. 2008; Olsson et al. 2009) probably caused the larger available C (Fig. 2e and g), which may have supported the larger topsoil microbial population size (Fig. 3a and c) and denitrification gene abundance (Fig. 4a, c, e) in the riparian tree buffer compared to the alley-cropping tree row. In the subsoil, however, the microbial population size (Fig. 3a, c, e) and denitrification gene abundance (Fig. 4a and g) showed an opposite pattern as in the topsoil—i.e., alley cropping > riparian buffer—which may be due to the more favorable mid-range WFPS in the well-drained, alley-cropping tree row as opposed to the commonly high WFPS in the riparian subsoil (Fig. 2a and b). These findings suggest that microbial and denitrifier population size equilibrated to environmental conditions (e.g., available C, soil moisture), and their DNA-based measurements cannot serve as a reliable indicator for biogeochemical process rates. Wei et al. (2023) employed an in-situ 15N labeling of mineral N and quantified soil surface net N2O flux as well as 15N-N2O site preference. They found that soil O2 status, rather than functional gene abundance, influences soil N2O production in undisturbed soil. In our earlier study (Luo et al. 2022), whereby gross N2O emissions and soil factors were measured over 1.5 years, we also found that substrate availabilities are more important than functional gene abundance in regulating gross N2O emissions. Other studies also concluded that the gene abundances only indicate the population size of the microbial community or functional microbial groups and not whether these genes are actively transcribed and, subsequently, translated into actual process rates (Rocca et al. 2015; Tomasek et al. 2017).

By aggregating distribution-related soil factors into three main groups (i.e., substrate availability, denitrification gene abundance, and microbial population size; Figs. S2 and S3), we were able to disentangle the influence of substrate availabilities from that of microbial population size or functional gene abundance. The dominant control of substrate availabilities on gross N2O emissions from the topsoils (Fig. 5a) was illustrated by their congruent differences between agroforestry systems (i.e., riparian buffer > alley-cropping tree row; Figs. 1a, 2c, e, g) and consistent seasonal patterns (i.e., spring > summer or fall; Figs. 1b, 2d, f, h, S3a top panels). Moreover, substrate availabilities also predominantly influenced gross N2O emissions from the riparian tree buffer (Fig. 5b) caused by their similar soil depth and seasonal patterns (Figs. 1a, b, 2c, d, g, h, S3b top panels). Since the seasonal patterns of substrate availabilities coincided with the seasonal pattern of WFPS (Fig. 2b, d, f, h), whereas microbial factors mostly remained similar across seasons (Figs. 3 and 4 right panels), these findings support our earlier study that DNA-based gene and microbial abundance are not reliable indicators of gross N2O emissions (Luo et al. 2022). Our results support a recent literature review which concluded that the degree of control of microbial communities over biogeochemical process rates has been overstated (Yang et al. 2023).

Gross N2O uptake

The strong positive relationship between gross N2O uptake and emissions (Fig. S4a) suggests that part of the emitted N2O was further converted to N2 (Yang et al. 2011; Wen et al. 2016; Yang and Silver 2016b). However, such a relationship was not observed in our earlier studies in Germany in forests under high N depositions (Wen et al. 2017) and in fertilized croplands (Luo et al. 2022), which both reported large soil mineral N levels (i.e., > 20 mg N kg−1). Instead, Wen et al. (2017) found gross N2O uptake to correlate with available C. High water content and large C availability (as electron donor) can create reducing soil conditions, favoring N2O uptake (Chapuis-Lardy et al. 2007; Schlesinger 2013; Ye and Horwath 2016). As expected, the riparian tree buffer exhibited higher topsoil gross N2O uptake (Fig. 1c) under higher WFPS and larger C availability compared to alley-cropping tree row (Fig. 2e and g). Such conditions also favored a larger population size of nirK- and nirS-denitrifiers and nosZ clade I-N2O reducers (Fig. 4a, c, e), altogether resulting in larger gross N2O uptake in the topsoil of riparian tree buffer than the alley-cropping tree row. The positive relationships of topsoil gross N2O uptake with soil respiration and nirK gene abundance (Fig. 6a and b) further manifested the critical role of nirK-denitrifiers and C availability in governing gross N2O uptake. The larger population size of nirK-denitrifiers was accompanied by larger available C, WFPS, microbial population size (Fig. 3a and c) in the riparian tree buffer than in the alley-cropping tree row both at low mineral N levels (i.e., < 10 mg N kg−1; Fig. 2c). This led to an increased reduction of N2O as the last electron acceptor of the denitrification sequence (Davidson and Seitzinger 2006). Additionally, increased electron donor (C availability) and population size of nosZ clade I-carrying microorganisms likely favored N2O reduction to N2 under moist soil conditions in the riparian tree buffer.

An earlier study suggested that N2O reduction is highly dependent on soil water content and C availability (Chapuis-Lardy et al. 2007). Hence, we expected that gross N2O uptake in the subsoil would be higher in the riparian buffer than in alley-cropping tree row because the former had higher subsoil WFPS (Fig. 2a) and larger subsoil C availability than the latter (Fig. 2e and g). Contrary to our expectation, both agroforestry systems showed similarly low subsoil gross N2O uptake (Fig. 1c), despite high WFPS and large available C in riparian buffer subsoil. The possible reason was because of extremely low total mineral N concentrations in the subsoils of both agroforestry systems (< 1 mg kg−1; Fig. 2c); this very low mineral N signified a very low substrate for N2O production and hence very low N2O that can be reduced to N2. This suggests N availability rather than C availability limited the gross N2O uptake in the subsoils, presumably via the control of available N as a substrate for gross N2O emissions (Figs. 6c, S4a). This was possibly the underlying mechanism explaining the positive correlation of gross N2O uptake and total mineral N in the subsoil (Fig. 6d), which also suggested a high potential of mineral N removal in the subsoil where C availability was large (Fig. 2e and g). The low mineral N contents in the subsoil of these agroforestry systems may also be related to the deep roots of fast-growing hybrid poplar trees, which can effectively take up subsoil NO3− (Fortier et al. 2010; Rood et al. 2011) and incorporate it in their biomass (Truax et al. 2017).

Implications

This study contributes to the knowledge gap on the capacity of the top- and subsoil under contrasting agroforestry systems to produce and concurrently consume N2O, which can only be quantified using the 15N2O pool dilution technique. Our findings of larger gross N2O emissions and gross N2O uptake under moist soil conditions in the topsoil of the riparian tree buffer than the alley-cropping tree row resulted mainly from larger N and C availabilities rather than from functional gene abundance or microbial population size. Although we did not detect differences in net N2O fluxes between depths within each agroforestry system, the high net N2O fluxes in the topsoil of riparian tree buffer indicated that net N2O emissions mainly originated from surface soil. The substantial gross N2O uptake of subsoils suggests that the presence of unfertilized tree rows of alley-cropping agroforestry system and riparian tree buffer can effectively reduce NO3− that is transported from adjacent cropland with high N load. In the alley-cropping tree row, the unmatched fluxes between net N2O uptake and gross N2O uptake indicate an underestimation of the soil N2O sink when the budget is only based on net N2O-flux measurements. However, the production of N2O within the soil profile (e.g., Koehler et al. 2012; Corre et al. 2014) and the occurrence of pulse events (e.g., dry-wetting and freeze–thaw cycles) could result in an underestimation of the role of trees in agroforestry systems as a mediator, since our present study only had seasonal measurements and not contiguous sampling depths. Therefore, further research should consider measurements at increased temporal frequency and more depth increments for field quantification of gross N2O fluxes for an in-depth understanding of the dynamics of gross N2O fluxes in the soil profile of agroforestry systems. Our results contribute to the growing evidence that agroforestry stimulates important ecosystem functions compared to open cropland (Smith et al. 2013; Veldkamp et al. 2023). Quantification of these ecosystem functions will be an important step towards inclusion in the European Common Agriculture Policy (European Commission 2021).