Introduction

Forest soils are characterized by considerable spatial variations in soil chemical, physical and biological properties (Gonzalez and Zak 1994; Hoffmanna et al. 2014; Okin et al. 2008; Wang et al. 2013). Such variations inevitably affect biogeochemical processes such as carbon (C), nitrogen (N) and phosphorus (P) cycles in forest ecosystems (Luizão et al. 2004; Yuan et al. 2013). Owing to spatial variations, large uncertainties in soil organic carbon (SOC) stock estimates still remain (Gruba et al. 2015; Yuan et al. 2013). Given the importance of the forest soil C reservoir in the global C cycle, there is a need to improve the accuracy of SOC estimates by considering spatial variations in SOC. Soil N and P are vital components in building soil fertility (Hobbie 1992; Liu et al. 2015). Spatial variations in soil N and P concentrations are important factors affecting the coexistence of plant species by partitioning niches of nutrient utilization (Okin et al. 2008; Xia et al. 2015). Conversely, the coexistence of plant species may also generate spatial variations in soil nutrients by affecting soil nutrient cycling (Blair 2005; Okin et al. 2008). Therefore, a comprehensive understanding of spatial variations in SOC, N and P concentrations is important for accurately estimating SOC storage, and can further our knowledge of interactions between soil and plant as well as plant coexistence mechanisms within forest ecosystems.

The effects of stand characteristics on spatial variations in SOC and soil nutrients (such as N and P) have been addressed in previous studies (Bae and Ryu 2015; Hoffmanna et al. 2014; Xia et al. 2015; Yuan et al. 2013). These influences have been attributed to the fact that forest stands can potentially alter C and nutrients returned to the soil through leaf litter inputs, litter decomposition processes, root uptake, canopy composition, the basal area of trees and root/crowns (Bae and Ryu 2015; Blair 2005; Hoffmanna et al. 2014; Xia et al. 2015; Yuan et al. 2013). Recently, Xia et al. (2015, 2016) reported the exact nutrient inputs of leaf litter production that influence the fine-scale spatial heterogeneity of soil macronutrients and provided evidence that proves the persistence of soil nutrient patches. Previous studies of spatial variations in SOC and soil nutrients involved pure stands or mixed stands in temperate and tropical areas (Cremer et al. 2016; Xia et al. 2015; Yuan et al. 2013). In subtropical areas, however, knowledge of whether and how stand characteristics influence spatial variations in SOC and soil N and P concentrations is still lacking. In addition, within a stand scale (i.e. different forest stands), the influences of stand characteristics on spatial variation patterns of SOC, N and P concentrations were evaluated mostly at a fine sampling scale ranging from one meter to several meters distance (Garten et al. 2007; Xia et al. 2015; Yuan et al. 2013). Variations in SOC, N and P concentrations at a coarse scale of ten meters distance in subtropical forests with diverse tree species have not been fully understood, although the variations at a coarse scale are of particular importance for comprehensively evaluating the influence of stand characteristics.

The distribution of tree species in natural forests tends to follow certain gradients in soil environments. Therefore, the effect of stand characteristics on SOC and soil nutrient concentrations usually varies depending on the topography, soil texture and pH value that exist in pure forest, or within a mix of tree species in temperate and tropical forests (Gruba et al. 2015; Schleuß et al. 2014; Schöttelndreier and Falkengren-Grerup 1999). Topographical factors, including elevation and convexity, can introduce large spatial variations in SOC and soil nutrients (Wang et al. 2007; Sheikh et al. 2009), and this is especially obvious under mountainous conditions, where topographic gradients are variable (Fantappiè et al. 2011; Yuan et al. 2013). Soil type, texture, pH value and soil moisture inherited from parent material are also important contributors to spatial variations in SOC, N and P concentrations (Liu et al. 2015; Rodríguez et al. 2009; Winowiecki et al. 2015). Yuan et al. (2013), for instance, found that 51 % of observed SOC variability could be explained by variations in soil moisture and pH value in a temperate forest. Distinctly, Liu et al. (2015) considered soil texture to be one of the key factors affecting spatial distributions of SOC and soil nutrients under Karst topography. Subtropical secondary forests consist of diverse tree species and are mainly distributed within mountainous and hilly areas. Therefore, stand characteristics, topography, soil texture and pH value may affect spatial variations in SOC, N and P concentrations. However, the relative contributions of the various factors are not fully understood.

We proposed two hypotheses: (1) that stand characteristics were the important factors affecting spatial variations in SOC, total N (TN) and total P (TP) concentrations in subtropical forests; and (2) that the magnitude of the effect of stand characteristics on spatial variations varied among SOC, TN and TP because other factors (such as topography, soil texture and pH value) also affected SOC, TN and TP concentrations and C, N and P had different cycling characteristics. If the hypotheses were correct, we would expect to observe differences in the spatial variations in SOC, TN and TP concentrations, and the contributions of stand characteristics, topography, soil texture and pH value to spatial variations would differ among SOC, TN and TP in forests. To test these hypotheses, we collected soil and floor litter samples from two secondary forests in a subtropical area: a Choerospondias axillaris deciduous broadleaved forest and a Lithocarpus glaber–Cyclobalanopsis glauca evergreen broadleaved forest. The C. axillaris forest was regenerated from secondary succession after firewood collection by the local community was outlawed from the late 1950s, while the L. glaber–C. glauca forest was well preserved with less disturbance. The two forests were located in close proximity and had identical climatic conditions and the same parent materials, but consisted of different tree species. The amounts of litterfall and fine root productivity were higher in the C. axillaris forest than in the L. glaber–C. glauca forest (Guo et al. 2015; Liu et al. 2014), and these affected the nutrient input to the soil. SOC, TN and TP concentrations in the soils and litterfalls were all analyzed. Stand characteristics, topography, soil texture and the pH values of the two forests were also measured. We attempted to: (1) determine how stand characteristics affected spatial variations in SOC, TN and TP concentrations in the subtropical forests; and (2) examine whether the magnitude of the effect of stand characteristics varied among SOC, TN and TP owing to their different cycling characteristics.

Materials and methods

Site and stand description

The study was conducted at the Dashanchong Forest Park (28°23′58″ N to 28°24′58″ N, 113°17′46″ E to 113°19′08″ E; 55–217.4 m above mean sea level) in Changsha County, Hunan Province, China. Soil in this area is designated as well-drained clay loam red soil developed from slate and shale rock, classified as Alliti-Udic Ferrosols in the Chinese Soil Taxonomy, which corresponds to Acrisol in the World Reference Base (WRB) for Soil Resources (IUSS Working Group WRB 2006). This region is characterized by a subtropical southeastern monsoon climate, with annual precipitation ranging from 1412 to 1559 mm and an annual mean temperature of 17.3 °C. The mean monthly maximum temperature of 40 °C occurs in July and the mean monthly minimum temperature of −11 °C occurs in January. The forests investigated were stands that originated from secondary succession forests in regions where firewood collection has been outlawed and no other management activities have been carried out since the late 1950s. After decades of afforestation and forest protection, secondary forest vegetations have been developed and preserved, dominated by species that include L. glaber, C. glauca, Castanopsis fargesii, Pinus massoniana and Phyllostachys edulis.

A 1 ha plot (100 × 100 m) was established in each of the two forests investigated, a C. axillaris deciduous broadleaved forest and a L. glaber–C. glauca evergreen broadleaved forest. The locations of individual trees were mapped and tree species, diameter at breast height (DBH), height (H) and crown width were recorded within the two forests. When the individual trees with a DBH > 4 cm in the plots were counted, the C. axillaris forest had lower stand density, average DBH and H and lower basal area (BA) than the L. glaber–C. glauca forest (Table 1). Importance values (IVs) were calculated in an earlier study by Zhao et al. (2013).

Table 1 Stand density, average diameter at breast height (DBH), average height (H), basal area (BA) and importance value (IV) of the top five tree species and whole stands with DBH > 4 cm in the Choerospondias axillaris and Lithocarpus glaber–Cyclobalanopsis glauca forests investigated
Full size table

Contrasting tree species dominated both forests. The top five dominant species ranked by IV in the C. axillaris forest were C. axillaris, Loropetalum chinense, Symplocos setchuenensis, Vernicia montana and Vernicia fordii, while the top five in the L. glaber–C. glauca forest were L. glaber, C. glauca, P. massoniana, C. axillaris and Cleyera japonica. Stand characteristics of the two forests are presented in Table 1.

Sample collection and chemical analysis

Each of the study plots was further divided into a grid of 100 subplots of 10 × 10 m, and samples were obtained from the center of each subplot between May 25 and June 16 in 2014. Within each sampling subplot, floor litter in 50 × 50 cm areas was collected prior to soil sampling. Subsequently, soil samples were retrieved from the 0–10 cm soil layer and taken to the laboratory to measure SOC, TN and TP concentrations, soil texture and pH values. Soil texture (clay, silt and sand contents) were determined by the pipette method (Gee and Bauder 1986). Soil pH values were measured at a soil-to-water (deionized) ratio of 1:2.5 using an FE20 pH meter (Mettler Toledo, Shanghai, China). For SOC, TN and TP measurements, soils were manually sorted to visually remove stones, plant roots and litter, and then sieved through a 0.25 mm mesh. SOC concentrations were determined by the K2Cr2O7/H2SO4 oxidation method. TN concentrations were determined using a semimicro-Kjeldahl method, and TP concentrations were measured by sodium hydroxide (NaOH) fusion and the Mo–Sb colorimetric method. Litter samples collected were oven dried at 65 °C to a constant weight and weighed to determine dry mass.

Data analysis

Descriptive statistics, including maximum value, minimum value, average value, error value and coefficient of variation (CV) of SOC, TN and TP concentrations, were calculated. The differences in SOC, TN and TP concentrations between the two forests were tested by two-way analysis of variance (ANOVA). To improve the precision of the variation function model, outliers were identified using the threshold value method and were replaced with the maximum of the remaining dataset after the outliers were excluded (Tukey 1977). Subsequently, the Kolmogorov–Smirnov (K–S) test was used to examine whether all data conformed to a normal distribution. The data for soil TP concentrations in the L. glaber–C. glauca forest showed a high degree of skew and failed normality tests, so a square root transformation was applied.

Spatial variability of SOC, TN and TP concentrations was determined by geostatistical analysis using GS+ Version 9 software (Gamma Designs Software, Plainville, MI, USA). The geostatistical approach can be used to obtain unbiased estimates and map soil attributes in locations that have not been previously sampled, using relationships between raw data and semivariograms. The latter showed changes in the scatter plot of function values over increases in sample lag. Scatter diagrams could be fitted by four different models, including the exponential model, the spherical model, the Gaussian model and the linear model. Semivariograms use three significant statistical parameters to describe degrees of spatial variation: nugget (C 0), sill (C 0 + C) and range (A 0). C 0 represents random variability on a spatial scale, C 0 + C represents the greatest variation in system properties, C 0/(C 0 + C) reflects the magnitude of the random variation in total variations, and A 0 describes the scale of spatial variation (Okin et al. 2008). Variables have spatial correlations within the A 0 scale, whereas there are no spatial correlations beyond the range of A 0. The semivariance (λ) at a given lag (h) is calculated according to the following equation:

$$ \lambda (h)=\frac{1}{2N(h)}{{\displaystyle \sum_{i=1}^{N(h)}\left[Z\left({x}_i\right)-Z\left({x}_i+h\right)\right]}}^2 $$
(1)

where λ(h) is the semivariance function value, N(h) is the number of pairs separated by lag (h), Z(x i ) is defined as the x i point of the Z sample as it relates to spatial location, and Z(x i  + h) is a regionalized variable on the (x i  + h) point.

Based on the fitness of each model, semivariances of SOC, TN and TP concentrations were described by the linear model for the C. axillaris forest and by the Gaussian model for the L. glaber–C. glauca forest (Table S1). The model selected and the parameters fitted were processed using the kriging procedure to interpolate values for locations that were not sampled. The kriging method is an interpolation method that not only accurately estimates soil properties but also estimates approximated errors and the precision of predicted values for locations that have not been sampled. Spatial distribution maps of SOC, TN and TP concentrations were then generated for each 10 × 10 m subplot using the predicted interpolation values.

Elevation and convexity for each 10 × 10 m subplot were also measured. The elevation ranged from 125 to 170 m in the C. axillaris forest and from 71 to 128 m in the L. glaber–C. glauca forest (Fig. S1). Elevation at a subplot center was determined as the average of the elevations of the four corners of the subplot. Convexity was calculated by subtracting the elevation at a subplot center from the mean elevation of the eight surrounding subplots. For edge subplots, convexity was calculated by determining the elevation of each subplot minus the mean altitude of the surrounding subplots. Stand characteristics included tree species number, stand density, crown coverage, average DBH, height, BA, deciduous proportion, evergreen proportion, conifer proportion, Shannon index, average canopy width per tree and litter biomass. In this study, the data for stand characteristics within a 10 × 10 m subplot were only determined for trees higher than the average stand height. Based on leaf morphological and phenological traits, trees were grouped into coniferous, deciduous or evergreen broadleaved species. The summed BA of the coniferous, deciduous and evergreen broadleaved species were divided by the total BA of the entire subplot to obtain their relative proportions.

Since there were weak correlations between topography, soil texture and stand characteristics in the two forests (Tables S2 and S3), stepwise linear regression was applied to determine their effects on SOC, TN and TP for the C. axillaris and L. glaber–C. glauca forests investigated. In the linear regression model, the following variables were included: topography (elevation and convexity), soil pH value and texture (clay, silt and sand) and stand characteristics (tree species number, stand density, crown coverage, average stand DBH, height, BA, conifer proportion, deciduous proportion, evergreen proportion, Shannon index, average canopy width per tree and litter biomass). After application of stepwise linear regression based on Akaike’s Information Criterion (de Micheaux et al. 2013), the variables that had significant correlations with SOC, TN and TP were selected as explanatory variables and grouped into topography, soil texture or stand characteristics (Tables S4 and S5). The relative contributions of stand characteristics, topography and soil texture to variations in SOC, TN and TP concentrations in a forest were calculated by dividing the explanatory (or regression) sum of squares (ESS) of each factor by the total regression sum of squares (TSS) of the variations in the stepwise linear regression. All statistical analyses were conducted using R-3.00 statistical software (R Development Core Team 2009).

Results

Variations in soil SOC, TN and TP concentrations in the two forests

The mean SOC and TN concentrations in the C. axillaris forest were significantly lower than those in the L. glaber–C. glauca forest, whereas the average TP concentration was significantly higher in the C. axillaris forest than in the L. glaber–C. glauca forest (Table 2). The CVs of SOC and of TN concentration in the C. axillaris forest were 31 % and 38 % lower than those in the L. glaber–C. glauca forest, respectively. However, the CV of soil TP concentration exhibited a reverse trend, being 43 % higher in the C. axillaris forest (24 %) than that in the L. glaber–C. glauca forest (17 %).

Table 2 Minimum, maximum, mean, standard deviation (SD) and coefficient of variation (CV) of soil organic carbon (SOC), total nitrogen (TN) and total phosphorus (TP) concentrations at a soil depth of 0–10 cm in the C. axillaris and L. glaber–C. glauca forests investigated
Full size table

Spatial variations in soil SOC, TN and TP concentrations

Geostatistical analyses revealed that the selected models had R 2 values ranging from 0.654 to 0.997 (Fig. 1). In the C. axillaris forest, spatial autocorrelation was weak for SOC and TN concentration but was moderate for TP concentration, as indicated by the C 0/(C 0 + C) ratios that were larger than 0.750 for SOC and TN and between 0.250 and 0.750 for TP. In the L. glaber–C. glauca forest, SOC, TN and TP concentrations showed moderate spatial autocorrelations, with C 0/(C 0 + C) ratios ranging from 0.250 for TP to 0.500 for SOC.

Fig. 1
figure 1

Semivariograms of soil organic carbon (SOC) (a), total nitrogen (TN) (c) and total phosphorus (TP) (e) concentrations in Choerospondias axillaris forest and semivariograms of soil organic carbon (SOC) (b), total nitrogen (TN) (d) and total phosphorus (TP) (f) concentrations in Lithocarpus glaber–Cyclobalanopsis glauca forest. A summary of the semivariogram model parameters for SOC, TN and TP in the two forests is provided under the curve. The selected models for semivariogram analysis (L = linear model; G = Gaussian model) are indicated. The proportion of structural variation C 0/(C 0 + C) was used as an index of the magnitude of spatial dependence

Full size image

Spatial autocorrelation ranges in SOC, TN and TP concentrations varied between the two forests (Fig. 1). In the C. axillaris forest, SOC concentrations had a spatial autocorrelation range of 53.0 m, which was greater than that in the L. glaber–C. glauca forest (23.7 m). Differences in autocorrelation ranges were small for TN and TP between the C. axillaris forest (53.0 m for TN and 53.0 m for TP) and the L. glaber–C. glauca forest (60.8 m for TN and 69.1 m for TP).

Spatial distribution maps for SOC, TN and TP concentrations were generated for the two forests using GS+ Version 9 software (Fig. 2). As shown, SOC, TN and TP concentrations exhibited patchy patterns for both forests. Patch sizes (ranges) for SOC, for example, in the C. axillaris forest (53.0 m) were more than twice as large as those in the L. glaber–C. glauca forest (23.7 m). The patch sizes of TN and TP were found to be larger in the L. glaber–C. glauca forest, which were 1.2 and 1.3 times as large as those in the C. axillaris forest, respectively. High SOC, TN and TP concentrations were observed in valleys in the L. glaber–C. glauca forest, whereas in the C. axillaris forest, high SOC, TN and TP concentrations were found in subplots comprised of trees with large DBH values.

Fig. 2
figure 2

Spatial heterogeneity map of soil organic carbon (SOC), total nitrogen (TN) and total phosphorus (TP) concentrations in the C. axillaris and L. glaber–C. glauca forests investigated

Full size image

Contributions of stand characteristics to variations in SOC, TN and TP concentrations

Stepwise regression analysis revealed that stand characteristics, together with topography and soil texture, were able to explain 20 % and 28 % of SOC variations in the C. axillaris and L. glaber–C. glauca forests, respectively (Fig. 3). Stand characteristics contributed the most to SOC variations in both forests, contributing 8 % in the C. axillaris forest and 21 % in the L. glaber–C. glauca forest (Fig. 3). However, topography contributed the least to variation in SOC, only 5 % in the C. axillaris forest. Soil texture made a modest contribution to variation in SOC, which was 7 % in the C. axillaris forest and 8 % in the L. glaber–C. glauca forest. All selected factors varied greatly in terms of their contributions to SOC variations in the two forests. Elevation, sand content, species number and DBH were significant explanatory variables in the C. axillaris forest (Table S2), whereas silt content, litter biomass, BA and canopy width were significantly related to variations in SOC concentration in the L. glaber–C. glauca forest (Table S3).

Fig. 3
figure 3

Relative contributions of topography, soil texture and stand characteristics to variations in soil organic carbon (SOC), total nitrogen (TN) and total phosphorus (TP) concentrations in the C. axillaris and L. glaber–C. glauca forests investigated

Full size image

Up to 43 % and 40 % of the variation in soil TN in the C. axillaris and L. glaber–C. glauca forests, respectively, were explained by the selected factors, of which stand characteristics explained about 10 % of TN variation in both forests (Fig. 3). Topography contributed to 11 % and 10 % of TN variation in the C. axillaris and L. glaber–C. glauca forests, respectively (Fig. 3). Soil texture was the greatest contributor to variation in TN (Fig. 3). However, the total contributions of soil texture to TN variation were similar in the two forests (Fig. 3). Elevation, clay content, sand content and DBH were significantly related to variations in TN concentration in the C. axillaris forest (Table S2), while significant relationships were detected between soil TN concentration and convexity, sand content, BA and canopy width in the L. glaber–C. glauca forest (Table S3).

The contributions of stand characteristics, topography and soil texture to variations in TP were 80 % in the C. axillaris forest and 60 % in the L. glaber–C. glauca forest (Fig. 3). Among the factors investigated, stand characteristics exhibited a greater contribution to variations in TP than topography and soil texture. The contribution of stand characteristics to variations in TP concentration was higher in the C. axillaris forest (40 %) than in the L. glaber–C. glauca forest (30 %) (Fig. 3). The contributions of topography and soil texture were 20 % and 20 % in the C. axillaris forest and 10 % and 20 % in the L. glaber–C. glauca forest. Elevation, sand content, evergreen proportion and DBH were the significant factors that related to TP concentration in the C. axillaris forest (Table S2). However, in the L. glaber–C. glauca forest, regression analysis indicated that elevation, clay content, silt content, stem, BA and canopy width were significantly related to TP concentration (Table S3).

Discussion

Effects of stand characteristics on variations in soil SOC, TN and TP concentrations

Average SOC and TN concentrations were significant lower but average TP concentration was significantly higher in the C. axillaris forest than those in the L. glaber–C. glauca forest (Table 2). This supported our first hypothesis that stand characteristics affected soil SOC, TN and TP concentrations. Our results were consistent with the study by Ding et al. (2015) that SOC and soil N concentrations were significantly higher in an evergreen broadleaved forest than in a deciduous broadleaved forest in subtropical areas. This could be explained by the fact that evergreen broadleaved forest accumulated more soil organic matter (Ding et al. 2015). It was reported that L. glaber–C. glauca forest accumulated more litter and had a lower decomposition rate of leaf litter (Guo et al. 2015) and fine roots (Tong et al. 2012) compared to C. axillaris forest. However, the result that soil TP concentrations were higher in the C. axillaris forest than in the L. glaber–C. glauca forest differs from that of Ding et al. (2015). It could be attributed to more P input (litterfall and fine root productivity) from trees and rapid litter turnover in the C. axillaris forest (Guo et al. 2015; Liu et al. 2014; Tong et al. 2012). In contrast, Yang et al. (2014) found relatively higher SOC and N but slightly lower P in a deciduous broadleaved forest than in an evergreen broadleaved forest. Average SOC concentration observed for broadleaved forests in this study was lower than that reported by Zhang et al. (2014) for broadleaved forest in a tropical area, but TN and TP concentrations were much higher for broadleaved forest in our study than those found by Zhang et al. (2014). These differences between the studies are due to the fact that Yang et al. (2014) studied a planted forest and Zhang et al. (2014) studied a tropical area, while the forests in our study were naturally regenerated in a subtropical area.

The C. axillaris forest exhibited lower variations (CV) in SOC and TN concentrations but higher variations in TP concentration compared to the L. glaber–C. glauca forest (Table 2). The variations were comparable to the results obtained in temperate forests (Wang et al. 2015) and in tropical forests (Robert et al. 2007; Xia et al. 2015). Moreover, the two forests investigated in this study showed different spatial variation patterns. Autocorrelations for SOC, TN and TP concentrations in the C. axillaris forest were weak or moderate and showed large autocorrelation ranges (Fig. 1) and large grain spatial distribution (Fig. 2). In contrast, the L. glaber–C. glauca forest showed moderate autocorrelation for SOC, TN and TP concentrations, but autocorrelation ranges and spatial grain varied among SOC, TN and TP concentrations. Stepwise linear regression analysis (Tables S2 and S3) showed that the variations in SOC, TN and TP in the forests investigated were positively and significantly correlated with stand characteristics, including BA, DBH and canopy width. This indicates a potential influence of stand characteristics on the spatial heterogeneity of soil nutrients in these forests (Hirobe et al. 2001). A previous study also found influences of BA and the canopy structure of trees on spatial variations in soil nutrients in temperate forest soil (Yuan et al. 2013). Canopy structure may affect the temperature and moisture content of forest floor soil, which are important factors affecting the litter decomposition processes; thus, the tree canopy may influence spatial variations in soil nutrients (Sariyildiz 2008). Xia et al. (2015) observed that the distribution of giant (large DBH) trees was related to nutrient conditions and that differences in the DBH of trees may result in different inputs of litterfall to the soil surface and help create spatial variations in soil nutrient concentrations.

Contributions of stand characteristics to soil nutrients in the two forests

Although stand characteristics were important factors affecting spatial variability in SOC, TN and TP concentrations, their ability to explain spatial variations differed for SOC, TN and TP concentrations, which was higher for TP in comparison to SOC and TN in both forests. This supported our second hypothesis that the magnitude of the effect of stand characteristics on spatial variations varied among SOC, TN and TP concentrations. This could have been due to different cycling characteristics among nutrient elements, which were more open for C and N cycles, whereas they were relatively closed for P. Such differences in the magnitude of the effect of stand characteristics on different cycling elements were also observed in tropical forests at similar or smaller study scales. Xia et al. (2015) reported a different effect of stand characteristics (litterfall nutrients) on fine-scale spatial variations of N and P within a 1 ha tropical forest. Changes in vegetation cover were shown to affect the magnitudes of spatial variations among SOC, N and P (Blair, 2005). It is worth noting that the majority of total P in soils is inorganic P, and this proportion of P should not be affected by stand characteristics, suggesting that the latter may affect mainly spatial variations in available P in forest soils.

In addition, other factors may be related to variations in SOC, TN and TP concentrations. In our study, the contribution of topography to spatial variations was higher for TN and TP concentrations than for SOC concentration. Topography is known to affect local microclimates, litter decomposition and the leaching of soil surface nutrients (Baldeck et al. 2012; Xia et al. 2015). However, these processes may result in different impacts among SOC, TN and TP concentrations. Different studies have addressed the influence of topography on variations in soil SOC and nutrient concentrations. For example, Xia et al. (2015) found a possible link between N (NO3-N), but not P, and topographic gradients.

In contrast to SOC and TP, soil texture showed the highest contribution to variations in TN, supporting the finding of Rodríguez et al. (2009) that soil texture determined the magnitude of the plant effect on spatial variations in forest soil N concentration. However, soil texture contributed similarly to SOC, TN and TP variations in the two forests investigated (Tables S2 and S3; Fig. 3), indicating that the strength of the influence of soil texture on spatial variations in SOC, TN and TP concentrations did not vary among different forests. This could be attributed to the fact that soil texture was relatively stable (Rodríguez et al. 2009) and resistant to modification by stand characteristics over short time periods (Augusto et al. 2002; Hagen-Thorn et al. 2004; Pohl et al. 2009). Our findings extended previous general results in temperate or tropical forests to subtropical forests, which showed that SOC and soil nutrient variations in forests were highly dependent on stand characteristics. However, influences from abiotic factors, including topography and soil texture as well as the cycling characteristics of elements, were also important (Gruba et al. 2015; Paluch and Gruba 2010; Schulp et al. 2008; Xia et al. 2015).

In this study, the high proportion of unexplained variations in SOC concentration could be ascribed to the fact that potentially important factors, such as soil respiration, were not taken into account. Topsoil carbon can easily diffuse into the atmosphere as CO2 via soil respiration, and such loss of carbon to the atmosphere can be strongly affected by the quantity and quality of litter inputs to soils (Fanin et al. 2011; Vesterdal et al. 2012). The composition of different species contributing to litter production was found to be highly diverse in both forests. Therefore, variations in litter quantity and quality may have led to spatially high variations in soil respiration (Guo et al. 2015), which may consequently have caused a spatial variation in SOC concentration. Subtropical forests in southern China are undergoing increasing N deposition (Chen and Mulder 2007). This increased N deposition may exceed the forest soil N retention capacity and thereby cause substantial leaching of N to surface runoff (Akselsson et al. 2010). The heterogeneous composition of tree species may interact with soil conditions to influence N leaching processes, which, in our study, resulted in an unexplained variation in TN concentration between the two forests investigated. Stand characteristics, topography and soil texture in this study were better able to explain the variations in soil TP concentration than those of SOC and TN concentrations. This is probably because the P was relatively stable and could not be lost via processes such as soil CO2 respiration or N leaching.

Conclusions

The C. axillaris deciduous broadleaved forest investigated in our study exhibited lower average concentrations and coefficients of variation of SOC and TN than the L. glaber–C. glauca evergreen broadleaved forest studied, whereas TP showed a reverse trend. Spatial autocorrelations of SOC, TN and TP concentrations were either weak or moderate in the C. axillaris forest but were moderate in the L. glaber–C. glauca forest. The different patterns of SOC, TN and TP in the two forests reflected the more diverse tree species and complex structure of the L. glaber–C. glauca forest compared to the C. axillaris forest. In addition, topography and soil texture were related to variations in SOC, TN and TP. Stand characteristics contributed the most to spatial variations in SOC and TP, while soil texture made the largest contribution to variations in TN. Topography contributed the least to variations in SOC, TN and TP. The contribution of stand characteristics differed for SOC, TN and TP due to their different cycling characteristics.