The study was conducted in Ansai County, Shaanxi Province (36°35’21”-36°36’40” N, 109°13’12”-109°16’31” E), the hinterland of the Loess Plateau, where the mean annual temperature and precipitation are 8.8°C and 510 mm, respectively. The rainy season is concentrated between June and September (72.9%). The soils are categorized as Calcic Cambisols (FAO and UNESCO, 1974). Over the past two centuries, the original vegetation has been completely devastated. Secondary forest regeneration began naturally in the 1970s, and plantation efforts have increased significantly since 1999, following the project to return farmland to plantations. Due to soil erosion over a thousand years, the background data on soil properties are quite similar on the Loess Plateau. The initial afforestation densities varied across the the Loess Plateau depending on tree species and site conditions, and later, with the growth of vegetation, the stand density changed. Thus, after 24 years of growth (up to 2022), changes in interspecific relationships and individual differences of plantation vegetation due to stand density are a crucial part of the natural process of succession.

The investigation included five type of plantations on the Loess Hilly Region, and the stand density for each plot was determined through a counting process. There were categorized into two stand types: monoculture plantations of R. pseudoacacia (RP, 1083 ± 42 trees ha^-1), A. davidiana (AD, 733 ± 33 trees ha^-1), and A. sibirica (AS, 700 ± 58 trees ha^-1) and mixed plantations of R. pseudoacacia (1192 ± 22 trees ha^-1) with A. davidiana (325 ± 14 trees ha^-1) (RPAD) and R. pseudoacacia (1025 ± 80 trees ha^-1) with A. sibirica (308 ± 55 trees ha^-1) (RPAS). Specifically, two tree species combinations were designed: RPAD, RP, and AD; and RPAS, RP, and AS. The objectives of this study were to determine which tree species presented greater leaf N and P concentrations and lower nutrient resorption efficiencies in mixed stands than in their corresponding monoculture stands and to identify the factors that influence these phenomena. The plantations were established approximately 16–25 years ago and are situated on loess soil with a southwest orientation. Three 20 m × 20 m plots from each plantation, 15 plots in total, separated by less than 3 km, and a 5-m buffer zone around each plot to avoid edge effects were selected for analysis (see Supplementary Table S1 ).

To ensure that the sample is representative, we select standard wood at the four corners and center of the quadrat. In each plantation stand, green and senescent leaves from pure plantations were collected from five representative trees of each species without withered branches, and samples from mixed plantations were taken from adjacent two different species. The diameter at breast height (DBH, cm) was measured using a tree diameter tape. The average tree height (AH, m) was determined using a height gauge. Meanwhile, the crown area (CA, m²) was determined using a tapeline. When the biomass of plants was at its peak (in mid-August 2022), intact and complete green leaves were collected from the lower, middle, and upper canopies using high branch shears. From November to December 2022, fallen senescent leaves were collected from nylon nets (1 m × 1 m, 0.1 mm) suspended below the canopy. Based on the characteristics of defoliation season of different tree species, the collection time of senescent leaves of R. pseudoacacia and A. davidiana was set in late December 2022, while that of A. sibirica was set in mid-September 2022. Intact 5-10 green leaf samples without pests and diseases were selected and stored in ice bags to measure morphological features such as specific leaf area (SLA), leaf dry matter content (LDMC), and leaf tissue density (LD). Other green and senescent leaves were oven-dried to a constant mass at 105°C for 10 min and at 70°C for 24 h to determine the green leaf nitrogen (GLNC) and phosphorus (GLPC) concentrations and the senescent leaf nitrogen (SLNC) and phosphorus (SLPC) concentrations. The stand characteristics of the monocultures and mixed stands are shown in Supplementary Table S2 .

Soil samples were collected from 0–20 cm in an S shape with five loci per plot using a 4 cm auger under the representative trees of each plantation, in mid-August 2022. Before the soil passed through the 2 mm mesh screen, we manually removed the roots, stones, and debris. The soil was divided into two parts: one part was air-dried, ground, and passed through a 0.15 mm sieve to determine the soil chemical properties, e.g., soil organic carbon (SOC), total N (TN), total P (TP), available N (AN), and available P (AP) concentrations; the other part was shade-dried and passed through a 1 mm porous sieve to determine the pH and particle composition (i.e., sand content, SA; silt content, SI; clay content, CL). At a depth of 0–20 cm, undisturbed soil was collected from each plot using a ring knife (100 cm³) to determine the bulk density (BD).

The N concentrations of green (GLNC, g kg^-1) and senescent (SLNC, g kg^-1) leaves were determined by the Kjeldahl method (Kjeltec 2300 analyzer unit, Foss Tecator, Hoganas, Sweden) after digestion in H₂SO₄–H₂O₂ solution. The P concentrations of green (GLPC, g kg^-1) and senescent (SLPC, g kg^-1) leaves were determined by the molybdenum yellow method (U-2800 spectrophotometer, Shanghai, China). Specific leaf area (SLA, cm² g^-1), leaf dry matter content (LDMC, g g^-1) and leaf tissue density (LD, g cm^-3) were calculated using the following equation:

where leaf fresh weight (FW) and dry weight (DW) were determined using an electronic balance (precision 0.001 g), leaf thickness (LT, mm) was determined using a digital vernier caliper, and leaf area (LA, cm²) was determined using an Epson Perfection V850 Pro scanner (Epson (China) Co., Ltd. Beijing, China) and ImageJ software.

The soil organic C (SOC, g kg^-1) concentration was determined by the calorific boiling and combustion with the H₂SO₄-K₂Cr₂O₇ method (Nelson and Sommers, 1982). The soil total N (TN, g kg^-1) concentration was determined by the Kjeldahl method (Bremner and Mulvaney, 1982), whereas the available N (AN, mg kg^-1) concentration was measured by alkaline hydrolysis diffusion (Bao, 2000). The soil total P (TP, g kg^-1) and available P (AP, mg kg^-1) concentrations were determined by the molybdenum blue method after extraction with H₂SO₄–HClO₄ and NaHCO₃, respectively (Murphy and Riley, 1962). The soil pH was measured using an automatic acid–base titrator (PB-10 standard pH meter; Sartorius, Göttingen, Germany) with a water/soil ratio of 2.5:1. The cutting ring method was used to determine the soil BD (g cm^-3) (Bao, 2000). The soil particle composition (SA/SI/CL, %) was determined using a Mastersizer 2000 Laser Particle Size Analyzer (Malvern Panalytical, Malvern, UK). The soil physicochemical properties are listed in Supplementary Table S3 .

Data were tested for normality and homogeneity using the Shapiro–Wilk and Levene tests. When necessary, the data were log transformed. One way analysis of variance (ANOVA) followed by Tukey’s test was performed using SPSS 23.0 (SPSS Inc., Chicago, Illinois, USA) to determine the differences in green and senescent leaves concentrations, nutrient resorption, and relative nutrient resorption of R. pseudoacacia among the different stands and the differences among R. pseudoacacia, A. davidiana, and A. sibirica monocultures. Student’s t test was performed using SPSS software to examine differences between monoculture and mixed stands in green and senescent leaves concentrations, nutrient resorption, and its ratio of A. davidiana/A. sibirica, and examine differences between R. pseudoacacia and A. davidiana with A. sibirica in the same mixed stand. For further analysis, two-way ANOVA was performed using SPSS software to determine the effects of stand type, tree species, and their interaction on leaf nutrient concentrations and resorption efficiency. Linear regression analysis was performed using Origin Pro 2024b (OriginLab, Northampton, Massachusetts, USA) to determine the relationships between leaf nutrient concentrations, resorption efficiencies, and soil nutrient concentrations and between leaf nutrient concentrations and resorption efficiencies of specific and all species in monocultures and mixed stands. Pearson correlation analysis was performed using Origin Lab 2024b with “Correlation Plot” applications to confirm the correlations among the leaf nutrient concentrations, resorption efficiencies, and influencing factors (stand characteristics and soil properties) of the individuals and all species in the different stands. After standardizing the data with a Z-Score, random forest model analysis was performed using the “randomForest” package in R 4.2.2 (R Core Team, 2022) to determine the influencing factors that contribute to the leaf nutrient concentrations and resorption efficiencies of individuals and all species in pure and mixed stands. The significance level of the statistical analysis was set at P < 0.05.

In our study, two-way ANOVA revealed the significant effects of stand type, tree species, and their interaction on leaf nutrient concentrations (P < 0.05, Figure 1 ). Specifically, GLNC of R. pseudoacacia in the RPAS stand was 6.41% lower (P < 0.05; Figure 1a ; Supplementary Table S4 ) and the GLPC of R. pseudoacacia in the RPAD stand was 15.70% higher than those in the RP stand (P < 0.05; Figure 1b ; Supplementary Table S4 ). Additionally, the GLNC and GLPC of A. davidiana in the RPAD stand were 44.48% and 35.23% higher (P < 0.05; Figures 1a, b ; Supplementary Table S4 ), respectively, and those of A. sibirica in the RPAS stand were 28.83% and 14.29% higher than those of their respective monocultures (P < 0.05; Figures 1a, b ; Supplementary Table S4 ).

Furthermore, the SLNC and SLPC of R. pseudoacacia in the RPAS stand were 14.44% and 13.33% lower than those in the RP stand, respectively (P < 0.05; Figures 1c, d ; Supplementary Table S4 ). Moreover, the SLNC and SLPC of A. davidiana in the RPAD stand were 83.08% and 66.67% greater (P < 0.05; Figures 1c, d ; Supplementary Table S4 ), respectively, and those of A. sibirica in the RPAS stand were 42.75% and 22.41% greater than those in the corresponding monoculture stands (P < 0.05; Figures 1c, d ; Supplementary Table S4 ).

Two-way ANOVA revealed the significant effects of stand type, tree species, and their interaction on NRE (Equation 6) and PRE (Equation 7) (P < 0.05, Figure 2 ). Compared with that in the RP stand, the PRE of R. pseudoacacia in RPAD stand was 10.73% greater (P < 0.05; Figure 2b ; Supplementary Table S4 ). The NRE and PRE of A. davidiana in the RPAD stand were 29.93% and 29.35% lower than those in the AD stand, respectively, and those of A. sibirica in the RPAS stand were 12.21% and 9.31% lower than those in the AS stand, respectively (P < 0.05; Figures 2a, b ; Supplementary Table S4 ).

Two-way ANOVA revealed the significant effects of stand type, tree species, and their interaction on relative nutrient efficiency (RRE, Equation 8) (P < 0.05, Figure 3a). Furthermore, the RRE of R. pseudoacacia was less than 100% and 12.05% lower in the RPAD stand than in the RP stand (P < 0.05), but those of A. davidiana and A. sibirica were greater than 100% and did not differ significantly between the pure and mixed plantations (P > 0.05) ( Figure 3a ; Supplementary Table S4 ). Moreover, linear regression analysis revealed that the NREs of A. sibirica and all species in the different stands were significantly positively correlated with PRE (P < 0.05, Figure 3b ).

Pearson’s correlation coefficient was used to determine the relationships between the leaf nutrient concentration, resorption efficiency, and stand characteristics and soil physicochemical properties of individuals and all species ( Figure 4 ). For individual species, the SLPC of R. pseudoacacia was significantly negatively correlated with the soil total phosphorus concentration; PRE was significantly positively correlated with the leaf dry matter content, leaf tissue density, and soil total phosphorous concentration but significantly negatively correlated with the specific leaf area (P < 0.05, Figure 4a ). The GLNC, GLPC, SLNC, and SLPC of A. davidiana and A. sibirica were significantly positively correlated with crown area (P < 0.05, Figures 4b, c ), but the opposite was true for NRE and PRE (P < 0.05, Figures 4b, c ). In all species, GLNC, SLNC, and SLPC were significantly positively correlated with average tree height, soil total phosphorus concentration, and soil available nitrogen and phosphorus concentrations. The SLPC was significantly negatively correlated with diameter at breast height. The NRE and PRE were significantly positively correlated with average tree height and diameter at breast height (P < 0.05, Figure 4d ).

A random forest model was used to determine the relative importance of influencing factors to leaf nutrient concentrations and their resorption efficiencies for individuals and all species ( Figures 5 , 6 ). Among the individual species, soil available P concentration had large effects on the GLNC of R. pseudoacacia (32.84%), A. davidiana (12.07%), and A. sibirica (11.43%) ( Figures 5a–c ). The clay content had large effects on the GLPC of R. pseudoacacia (24.67%), and pH had effects on A. davidiana (11.54%), and A. sibirica (10.25%). The soil available P concentration had strong effects on the SLNC of R. pseudoacacia (15.44%), and pH had effects on A. davidiana (10.87%), and A. sibirica (14.01%). The soil available P concentration had large effects on the SLPC of R. pseudoacacia (23.22%), A. davidiana (10.87%), and A. sibirica (9.53%) ( Figures 5a–c ). For all species, specific leaf area had strong effects on the GLNC (62.42%), GLPC (47.6%), and SLNC (47.6%), and diameter at breast height on the SLPC (58.93%) ( Figure 5d ).

For individual species, soil sand content (17.36%) had large effects on the NRE of R. pseudoacacia, soil organic C concentration (13.1%) had effects on the NRE of A. davidiana, soil available P concentration (12.73%) had effects on the NRE of A. sibirica ( Figures 6a–c ). The average tree height (11.62%) had strong effects on the NRE of R. pseudoacacia, soil pH (16.15%) had great effects on the NRE of A. davidiana, soil available P (12.73%) concentrations had effects on NRE of A. sibirica. For all species, leaf dry matter content (35.34%) effected the NRE, and average tree height (44.64%) had effects on the PRE ( Figure 6d ).

Linear regression analysis revealed different correlations between leaf nutrient concentrations and nutrient resorption across individuals and all species ( Figure 7 ). For specific species, the PRE of R. pseudoacacia was significantly positively correlated with the GLNC (P < 0.05, Figure 7b ), and the opposite was true between the NRE and SLNC (P < 0.05, Figure 7c ). The NREs of A. davidiana and A. sibirica were significantly negatively correlated with GLNC and SLNC (P < 0.05; Figures 7a, c ) and were the same between PRE and SLPC (P < 0.05, Figure 7d ). For all species, PRE was significantly positively correlated with GLPC (P < 0.05, Figure 7b ) and significantly negatively correlated with SLNC (P < 0.001, Figure 7d ).

Previous studies have reported that N₂-fixing species promoted the expansion of root networks and that mycorrhizal systems increase rhizobial activity, thereby increasing the concentration of soil available N (Forrester et al., 2006; Nygren and Leblanc, 2015; Yuan et al., 2016). Additionally, the decomposition of senescent leaves of R. pseudoacacia can further increase the N resources of A. davidiana and A. sibirica (Forrester et al., 2006; Voigtlaender et al., 2012; Paula et al., 2015). Thus, the soil total N concentration was higher in the mixed stands than in the monoculture stands ( Supplementary Table S3 ). However, the GLNC and specific leaf area were not significantly positively correlated for R. pseudoacacia but were significantly positively correlated for A. davidiana and A. sibirica ( Figures 4a–c ), partially supporting Hypothesis 1. One reasonable explanation is that the canopy of mixed R. pseudoacacia is located above A. davidiana and A. sibirica, the decrease in light intensity is lower by the mixed R. pseudoacacia but higher for mixed A. davidiana and A. sibirica compared to mono species. Particularly, in the mixed stand, the light intensity of R. pseudoacacia was less affected by A. davidiana, while that of which was more affected by A. sibirica. Thus, the GLNC of R. pseudoacacia did not differ between the RPAD and RP stands, whereas lower in the RPAS than in the RP stand ( Figure 1a ), contrary to Hypothesis 1. These findings indicate that the GLNC of R. pseudoacacia is not only affected by the total N content of soil but also by the specific leaf area (Sanz-Perez et al., 2009; Liu et al., 2014; Zhu et al., 2020).

The GLPC of R. pseudoacacia was greater in the RPAD stand than in the RP stand, but there was no difference between the RPAS and RP stands ( Figure 1b ; Supplementary Table S4 ), which was partially consistent with Hypothesis 1. This could be involved stand characteristics and soil properties interact to affect the GLPC values of R. pseudoacacia in monoculture and mixed stands (Xu et al., 2020; Sardans et al., 2021). For instance, crown area, clay content, soil total P concentration were not significantly correlated with the GLPC values of R. pseudoacacia ( Figure 4a ), but they had great relative contributions to the GLPC of R. pseudoacacia ( Figure 5a ). Consistent with the fact that the GLPC values of A. davidiana and A. sibirica in mixed stands were significantly positively correlated with crown area ( Figures 4b, c ), the GLPC and crown area values of which were greater in the mixed plantations than in the monocultures ( Figure 1b ; Supplementary Table S4 ). It can be inferred that understory trees tend to allocate more P to increase leaf metabolic activity. Additionally, low soil sand content resulted in high soil porosity and water storage (Sardans et al., 2011; Metz et al., 2016), beneficial for root expansion and microbial activity (Nygren and Leblanc, 2015; Ma et al., 2017) and indirectly improves the rapid release of P from organic matter (Blank and Morgan, 2014; Coll et al., 2018; Jiang et al., 2019; Waithaisong et al., 2020). As a result, the soil sand content was lower in the RPAS stand than in the AS stand ( Supplementary Table S3 ).

Decomposing senescent leaves play an important role in nutrient cycling in the plant–soil ecosystem. In particular, mixed plantings can increase the decomposition rate of senescent leaves and rapidly release nutrients into the topsoil through microenvironmental changes such as reduced soil temperature, increased soil moisture, and improved microbial activity (Wang et al., 2014; Paula et al., 2015; Lin et al., 2022). Therefore, the SLNC and SLPC did not differ between the RPAD and RP stands but were lower in the RPAS stand than in the RP stand for R. pseudoacacia ( Figures 1c, d ; Supplementary Table S4 ), whereas the soil total N and P concentrations were greater in the mixed plantations than in the monocultures ( Supplementary Table S3 ). Additionally, senescent leaves with high nutrient concentrations can enhance soil nutrient availability by altering the amount of belowground N and P inputs (Forrester et al., 2006; Voigtlaender et al., 2012; Ma et al., 2017; Waithaisong et al., 2020). Thus, the SLNC and SLPC of A. davidiana and A. sibirica were greater in mixed plantations than in monocultures ( Figures 1c, d ), suggesting that senescent leaves of these mixed species than the monocultures returned more nutrients into the soil. However, whether the amounts of P released by the high-quality senescent leaves of A. davidiana and A. sibirica were much greater than that consumed by rhizobia from R. pseudoacacia is still unclear. First, mixed species growth under water stress conditions may not compete for soil available-P because of the strongly reduced mobility of soil P (Lambers et al., 2008). Second, tree structure can affect the nutrient concentration of senescent leaves in plants (Hayes et al., 2014; Lin et al., 2022). For example, the SLNC and SLPC of A. davidiana and A. sibirica in monocultures and mixed stands were positively associated with the average tree height and crown area ( Figures 4b, c ). It follows that more N and P elements were released from senescent leaves and transported into two shorter, non-N₂-fixing species.

In general, the process of N fixation by R. pseudoacacia rhizobacteria could increase soil available N concentration (Nygren and Leblanc, 2015). However, the difference in R. pseudoacacia NRE between pure and mixed stands was not significant ( Figure 2a ), inconsistent with Hypothesis 2. Thus, the mixed planted R. pseudoacacia was not changed its demand for N. However, the PRE of R. pseudoacacia was lower in the RP stand than in the RPAD stand, whereas it was not significantly different between the RP and RPAS stands ( Figure 2b ). A possible explanation is that: 1) less P consumed during the N fixation than the soil-available P that absorbed by R. pseudoacacia, 2) senescent leaves were released enough P for R. pseudoacacia, and 2) microclimate could improved the process (Aerts, 1996; Paula et al., 2015; Sardans et al., 2021). For this reason, the decomposition of senescent leaves may release more P to the soil in the RPAS stand than in the RPAD stand. However, the NRE and PRE of A. davidiana and A. sibirica were lower in mixed stands than in monocultures ( Figure 2 ), consistent with Hypothesis 2. These findings suggest that the growth of A. davidiana and A. sibirica in the mixed plantation was less limited by soil N and P than that in the monoculture. Thus, mixed A. davidiana and A. sibirica tend to fully utilize N fixed by rhizobia and reduce NRE (Augusto et al., 2013; Paula et al., 2015; Bu et al., 2019) and therefore better adapted to nutrient-poor soils (Brant and Chen, 2015; Sardans et al., 2017). Additionally, the NRE and PRE of A. davidiana and A. sibirica were negatively correlated with crown area ( Figures 4b, c ), suggesting that the N and P in stems and branches were mainly obtained from the soil but less from reabsorbed nutrients by senescent leaves. Furthermore, the NRE of A. davidiana was negatively correlated with the soil total N concentration ( Figure 4b ), and the PRE of A. sibirica was negatively correlated with the soil total P concentration ( Figure 4c ). Therefore, in mixed plantations, A. davidiana releases N to the soil through senescent leaves, whereas A. sibirica releases P.

Consistent with Hypothesis 2, the relative nutrient rsorption efficiency (RRE) of R. pseudoacacia in pure and mixed plantations was < 100% ( Figure 3a ; Supplementary Table S4 ), which agree with the greater capacity of R. pseudoacaia to uptake and retain P during N fixation (Blank and Morgan, 2014; Liu et al., 2017). Thus, the growth of R. pseudoacacia was mostly limited by soil P (Reed et al., 2012; Han et al., 2013; Cao and Chen, 2017). Particularly, the growth of R. pseudoacacia was more easily limited by soil P in the RPAD stand than in the RP stand ( Figure 3a ; Supplementary Table S4 ). However, inconsistent with Hypothesis 2, A. davidiana and A. sibirica in the pure plantations presented values of RRE > 100%, which were not significantly different between the pure and mixed plantations ( Figure 3a ). These findings suggest that mixed planting of R. pseudoacacia did not significantly alleviate the N limitation imposed on A. davidiana or A. sibirica. This may be related to the fact that the availability of elemental N has a critical effect on the growth and photosynthesis of plants (Elser et al., 2010), especially the higher N requirements of A. davidiana and A. sibirica, which have larger crown areas in mixed plantations than in monocultures ( Supplementary Table S2 ).

The NRE of individuals and all species were positively correlated with PRE ( Figure 3b ), consistent with previous studies (Han et al., 2013; Deng et al., 2019). Particularly, A. sibirica NRE and PRE were significantly positively correlated ( Figure 3b ), indicating the positive regulation of A. sibirica NuRE in pure and mixed plantations by soil N and P limitation. Thus, the N and P distributed in the green/senescent leaves of A. sibirica were derived mainly from the soil. However, the nutrient sources for the green leaves of R. pseudoacacia was also influenced by the rhizobial N fixation (Aerts, 1996; Forrester et al., 2006; Paula et al., 2015). Specifically, rhizobia during N fixation could consume large amounts of soil P, resulting in the differences between soil available N and P concentrations in the rhizosphere (Nygren and Leblanc, 2015; Yuan et al., 2016; Sardans et al., 2021). Thus, NRE and PRE were not significantly positively correlated for R. pseudoacacia in pure and mixed plantations ( Figure 3b ). However, NRE and PRE were significantly positively associated for all species ( Figure 3b ). Therefore, tree mixing is conducive to coordinating the relationship between NRE and PRE for different species through soil N and P availability, although its effect on N₂- and non-N₂-fixing species shows individual variation (Forrester et al., 2006; Coll et al., 2018).

Consistent with Hypothesis 3, the negative correlation between NRE and the GLNC of different plantations were not significant for R. pseudoacacia, but they were significant for A. davidiana and A. sibirica ( Figure 7a ). These findings suggest that rhizobial activity can significantly affect the NRE of R. pseudoacacia but not those of A. davidiana and A. sibirica. The reason for this may be that the mixed R. pseudoacacia maintains NRE by enhancing rhizobial N fixation and fully utilizes the fixed N, allowing A. davidiana and A. sibirica to benefit from the N fixation of R. pseudoacacia, which does not compete with the other plants for soil available N (Augusto et al., 2013; Bu et al., 2019; Chen and Chen, 2022). Additionally, the GLNC of mixed A. davidiana and A. sibirica can also originate from other sources, such as N deposition, available N and minerals. However, soil P is the main resource of the GLPC for different species (Blank and Morgan, 2014); thus, soil-available P is the main nutrient limiting plant growth, whether it is an N₂-fixing species or not. Thus, the PRE and GLPC were positively associated in R. pseudoacacia but negatively associated in A. davidiana and all species ( Figure 7b ), indicating a difference in P uptake strategies between R. pseudoacacia and A. davidiana and A. sibirica. Sardans et al. (2012), Liu et al. (2017), and Li et al. (2018) reported that plant strategies for nutrient uptake were based on energy allocation to leaves and roots, with plants preferring to absorb more nutrients at a lower energy cost. On this basis, the energy cost of the green leaves of R. pseudoacacia was lower to retain P to compensate for soil P deficiency, whereas the energy cost of the roots of A. davidiana was lower during root nutrient uptake.

According to the theory of nutrient uplift, plants growth were mostly dependented on nutrients in the topsoil (Jobbagy and Jackson, 2001), and therefore the high nutrient concentrations in senescent leaves were associated with low nutrient resorption efficiency but high soil nutrient concentration (Aerts and Chapin, 1999; Sardans et al., 2017). Thus, the NRE of individual species was significantly negatively correlated with SLNC ( Figure 7c ), which is consistent with Hypothesis 3 and suggests an inverse control of SLNC on NRE. Furthermore, plants growing under high soil total N concentrations tend to withdraw a lower percentage of N from their leaves, resulted in more N returning to the topsoil via SLNC (Jobbagy and Jackson, 2001; Hayes et al., 2014; Brant and Chen, 2015). However, the negative relationship between PRE and SLPC was not significant for R. pseudoacacia, but the relationships were significant for A. davidiana and A. sibirica ( Figure 7d ), consistent with Hypothesis 3. These results indicate that other factors influence the process of P supply to R. pseudoacacia, such as P consumption by rhizobia during N fixation; however, tree mixture has a weak effect on P cycling in A. davidiana and A. sibirica (Sardans et al., 2012; Cierjacks et al., 2013). Therefore, the decomposition of senescent leaves can provide sufficient N for plant growth, but the P concentration does not meet the requirements of R. pseudoacacia, although it is sufficient for A. davidiana and A. sibirica (Jobbagy and Jackson, 2001; Ma et al., 2017).

Over, the results demonstrated that A. davidiana and A. sibirica are suitable mixing species to be planted with R. pseudoacacia to improve N and P cycling. In future forest management and ecosystem restoration planning, appropriate mixed species compositions and their effects on stand characteristics should be the first considerations to maintain the nutrient cycling of temperate forests.