Introduction

Quantifying the carbon (C) source/sink strength of terrestrial ecosystems is a prerequisite to evaluating the influence of terrestrial ecosystems on atmospheric CO2 concentrations and predicting climate change (Oishi et al. 2013; Luo et al. 2015). The net ecosystem exchange reflects the balance between two large CO2 fluxes, the gross photosynthesis and total ecosystem CO2 efflux (Exbrayat et al. 2013). Soil respiration (R S) accounts for 60–90 % of ecosystem CO2 efflux in forest ecosystems (Liang et al. 2004). Thus, understanding the patterns and controls of R S is of importance.

The temporal variation of R S at different scales is well understood and is affected by the supply of photosynthate, soil temperature (T), and soil water content (Kuzyakov and Gavrichkova 2010; Joo et al. 2012; Hanpattanakit et al. 2015). In contrast, the spatial variability and controlling factors of R S remain poorly understood due to technical limitations and its complex component origin (rhizospheric respiration and heterotrophic respiration) of evolved CO2 (Dore et al. 2014; Prolingheuer et al. 2014). Rhizospheric respiration is mainly controlled by gross primary productivity, the supply of photosynthate to roots from the canopy, root biomass distribution, and root nitrogen (N) concentration (Ruehr and Buchmann 2010; Kuzyakov and Gavrichkova 2010; Hopkins et al. 2013). Heterotrophic respiration is supposed to depend on the activity of soil microbes; substrate availability; litter chemical structure; and the biophysical environment of soil, such as the bulk density, field moisture capacity, and water-filled pore space (Fierer et al. 2009; Karhu et al. 2010; Luan et al. 2012; Tian and Shi 2014; Chen et al. 2015). The role of these controlling factors depends on the spatial scale. Some of these factors may only be appropriate at certain scales, and additional factors may overshadow others as spatial scales change (Martin and Bolstad 2009). For example, at the within-stand level, variation between individual R S measurements is strongly affected by the water-filled pore space, soil C content, soil C:N ratio, fine root biomass, and stand structure, such as tree proximity and leaf area index (Luan et al. 2012; Dore et al. 2014). At the landscape scale, topography, stand development (age), land use change, and plant productivity coupled with substrate availability and the C input to the soil drive R S across the forest types (Saurette et al. 2006; Wang et al. 2006; Martin and Bolstad 2009; Sheng et al. 2010). The quantification and understanding of the controlling factors of the variability of R S at various spatial scales in a study region can reduce errors in scaling up from flux measurements to the stand, landscape, and regional scales (Oishi et al. 2013). However, most studies focus on the variability of R S at a single spatial scale, e.g., at the intra-plot scale (Bréchet et al. 2011), among different plots/treatments in a forest (Ngao et al. 2012; Cheng et al. 2014) or among forest stands at the landscape scale (Sheng et al. 2010; Yoon et al. 2014). Our knowledge of the key factors that regulate the variability of R S at multiple spatial scales is limited.

Over the last few decades, large tracts of primary mixed broad-leaved Korean pine (Pinus koraiensis) forest in northeastern China have been harvested and subsequently transformed into secondary forest and plantations due to human interference or management. The Liangshui National Nature Reserve contains a variety of forest types with different management regimes and thus provides an opportunity to identify the controlling factors of the variability of R S at various scales within a restricted geographic region where air temperature, precipitation, and soil types are similar. In this study, a primary mixed broad-leaved Korean pine forest (BKPF), spruce-fir valley forest (SVF), selective cutting of mixed broad-leaved Korean pine forest (SCF), and Korean pine plantation (KPP) were selected to answer the above questions. Our specific objectives were to (1) quantify and compare the variability of R S at three spatial scales, among collars within a plot, among plots within a specific forest stand, and among forest stands within the landscape, and (2) identify the roles of biotic and abiotic factors that determine the variability of R S within a specific forest stand and among four forest stands.

Materials and methods

Study sites and experimental design

The study was conducted in the Liangshui National Nature Reserve (47° 10′ 50″ N, 128° 53′ 20″ E), Heilongjiang Province, northeastern China. The site has a continental monsoon climate regime, characterized by cold, dry winters and warm, wet summers. The mean annual temperature is −0.3 °C, and the mean annual precipitation is 676 mm, which is concentrated in the summer. The frost-free period is 100–120 days, and the snow period is 130–150 days. The soil is classified as dark-brown forest soil in the Chinese soil classification, which is equivalent to Humaquepts or Cryoboralfs according to the American Soil Taxonomy (Soil Survey Staff 1999).

The area has a variety of forest stands at different successional stages. It has one of the most concentrated and well-conserved primary mixed broad-leaved Korean pine forests (zonal climax vegetation) in China. The selective cutting of mixed broad-leaved Korean pine forest was regenerated naturally after selective cutting of Korean pine (intensity 30 %) in 1971. The Korean pine plantation was established in 1954 following the clear-cutting of the BKPF. These three forest stands are similar in topography and microclimate, with elevations ranging from 378 to 510 m and slopes ranging from 10 to 15°. The spruce-fir valley forest was the virgin non-zonal climax forest, mainly distributed along the river or streams valley at relatively low altitudes. This area is its southern boundary of natural distribution. The elevation in this forest ranges from 339 to 350 m, and the slope is almost flat. All sites are within a circular area with a radius of 3 km. The basic characteristics and stand compositions of the four forest stands in this study are summarized in Table 1 and 2.

Table 1 Site characteristics for the four forest stands
Full size table
Table 2 Mean soil respiration, soil temperature, soil moisture, and soil physical and chemical characteristics for the four forest stands
Full size table

Three 20 × 30-m survey plots for each forest stand were established in May 2010. Tens to hundreds meters separated plots. A 10 × 10-m subplot was placed within the plot with each R S sample location located at the center of the subplot. A total of six sample locations were selected within each plot.

Soil respiration measurement

The polyvinyl chloride collars (6 cm high and 10.4 cm in diameter) were inserted 4 cm into the soil (including the litter layer) at each sample location at the beginning of May 2013, 1 week before the first measurement campaign. R S was measured approximately every other week during the growing season, from May to October 2013, using an LI-6400 portable CO2 infrared gas analyzer (IRGA; LI-COR Inc., Lincoln, NE, USA) for a total of nine measurement campaigns. The measurements were performed from 10:00 to 16:00 on rainless days to limit the effect of daily fluctuations of soil temperature and rainfall on R S. No measurements were carried out in the winter because the analyzer cannot be used at low temperatures. The soil temperature (T, measured at 5 cm below the ground surface using an Li-6000-09 TC; LI-COR Inc.) and soil water content (measured at 0–10-cm soil depth using a time-domain reflectometry probe; IMKO, Ettlingen, Germany) were measured next to each collar to coincide with R S measurements.

Stand parameters and soil properties

The diameter at breast height (DBH) and position (coordinate) were measured on all trees with a DBH greater than 1 cm within the 20 × 30-m plot and within a 5-m radius outside the plot boundary in May 2010. Based on the DBH and coordinates of all trees and the coordinates of each sampling location (collar), the following stand structural parameters were calculated for each sampling location: mean DBH (mean DBHR), maximum DBH (max DBHR), and cumulative basal area (BAR) of the trees located within 2–10 m (radius) of each R S measurement point (R represent radius).

Samples for fine (<2 mm in diameter) and medium root (2–5 mm in diameter) mass were collected at the end of July and September 2011 and once at the end of each month from May to September 2012 using 10 randomly located soil cores (5-cm diameter × 40-cm depth) in each plot. Next, live fine and medium root in the samples were distinguished, collected, dried at 60 °C to a constant mass, and weighed.

Soil samples were collected using a soil corer (5-cm diameter) within the top 10 cm of soil in September 2013 at three different locations near each sample location. Soil samples were pooled to create one sample to analyze soil chemical factors. The light fraction of soil organic matter was isolated using the method described by Janzen et al. (1992). Subsequently, the bulk soil (soil organic carbon (SOC)) and light fraction organic C contents were analyzed using a multi-N/C 2100 analyzer (Analytik Jena AG, Jena, Germany). The total N (TN) content of the soil was determined by a Hanon K9840 auto Kjeldahl analyzer (Jinan Hanon Instruments Co., Ltd, China). The soil C:N ratio was calculated from the measured SOC and TN. The soil pH was measured in water (1:2.5 w/v). In addition, two soil samples were collected from the top of the soil near each sample location using 100-ml (50.46-mm diameter and 50-mm height) cylinders to analyze the physical characteristics of the soil based on the soil water-retention characteristics (Klute 1986), including bulk density, field moisture capacity, capillary porosity, and non-capillary porosity. The water-filled pore space (WFPS) of the soil was calculated according to the following equation (Linn and Doran 1984):

$$ \mathrm{WFPS}=W/\left(1-\mathrm{B}\mathrm{D}/\mathrm{P}\mathrm{D}\right) $$
(1)

where W is the volumetric water content (cm3/cm3); BD is the soil bulk density (g/cm3); and PD is the soil particle density, which was assumed to be 2.65 g/cm3 (Klute 1986).

Data analysis

The relationship between seasonal R S and T was examined by using an exponential model

$$ {R}_{\mathrm{S}}=\upalpha \times { \exp}^{\beta \times T} $$
(2)

where R S is the soil respiration (μmol CO2 m−2 s−1), T is the soil temperature at a depth of 5 cm, α is the intercept of CO2 efflux when T is zero, and β is the fitted parameter. Q 10 is a quotient of change in R S caused by change in temperature by 10 °C, can be indicative of the sensitivity of R S to temperature, and was calculated as follows:

$$ {Q}_{10}={ \exp}^{10\beta } $$
(3)

The coefficient of variation (CV) was calculated as the standard deviation divided by the mean to represent the relative variations in R S, T, and soil water content. These values were calculated at the following three spatial scales: among collars within a plot (n = 6), among three plots within a specific forest stand (n = 3), and among the four forest stands within this landscape (n = 4).

A regression analysis was performed to examine the similarities in the variability of R S among the nine measurement campaigns. The significant correlations among the R S of nine measurement campaigns indicated that spatial patterns of R S remained constant across the growing season (Tables S1, S2, S3, and S4, Electronic supplementary material). Thus, the R S was averaged over the nine measurement campaigns at each sample location to identify controlling factor determining the variability of R S through the growing season in each forest stand.

The relationships between R S and soil microclimate, stand parameters, and soil properties were examined by linear regression within a specific forest stand and across the four forest stands. Subsequently, all variables that were significantly correlated with R S within a specific forest stand were used in a stepwise multiple regression analysis. Due to non-simultaneous R S and differences in T among the four forest stands, an ANCOVA test was carried out by including T as a covariate with LSD comparisons for the mean R S among the four forest stands. One-way analysis of variance (ANOVA) was performed for comparisons of fine and medium root mass, field moisture capacity, capillary porosity, light fraction organic C contents, SOC, TN, C:N ratio, and pH among the four forest types. Statistical analyses were performed with SPSS software (SPSS 18.0 for Windows, Chicago, USA). Graphs were created using SigmaPlot 12.5 software (Systat Software Inc., San Jose, CA, USA).

Results

Temporal variability of R S

The seasonal variability of R S for all forest stands followed a bell-shaped curve and depended on T at the stand levels (Fig. 1). The exponential models of R S against T explained 52–86 % of the seasonal variability in R S across the four forest stands (Fig. 2). The Q 10 values (the sensitivity of R S to temperature) for R S were 2.46, 2.05, 2.4, and 2.69 in BKPF, SVF, SCF, and KPP, respectively.

Fig. 1
figure 1

Seasonal changes in soil respiration (a), soil temperature (b), and soil water content (c) for the four forest stands. BKPF primary mixed broad-leaved Korean pine forest, SVF spruce-fir valley forest, SCF selective cutting of mixed broad-leaved Korean pine forest, KPP Korean pine plantation. The error bars represent standard errors of means (n = 3)

Full size image
Fig. 2
figure 2

Relationships between soil respiration and soil temperature at a depth of 5 cm for the four forest stands. BKPF primary mixed broad-leaved Korean pine forest, SVF spruce-fir valley forest, SCF selective cutting of mixed broad-leaved Korean pine forest, KPP Korean pine plantation

Full size image

Factors that influences R S within a specific forest stand

The average CVs of R S among plots ranged from 18 % in SCF to 25 % in SVF (Fig. 3). The average CV of R S within plots in KPP (30 %) was significantly lower than that of R S in BKPF (46 %), SVF (44 %), and SCF (52 %; P < 0.01). The average CVs of R S within plots were significantly higher than those of R S among plots in each forest stand (P < 0.05). We found a significant positive correlation among the R S of almost all measurement campaigns for each forest stand (n = 18, P < 0.05; Table S1, S2, S3, and S4, Electronic supplementary material).

Fig. 3
figure 3

The average coefficient of variation of soil respiration (R S) among and within plots for the four forest stands. Different letters indicate significant differences among different forest stands (P < 0.05). BKPF primary mixed broad-leaved Korean pine forest, SVF spruce-fir valley forest, SCF selective cutting of mixed broad-leaved Korean pine forest, KPP Korean pine plantation

Full size image

There were correlations between R S and forest structural parameters (mean DBH, max DBH, and BA), and the correlations depended on forest stand and the distance between trees and collars (Table 3 and Fig. 4). The maximum significant correlation (R max) was found at a distance of 8 m for the mean DBH, max DBH, and BA (R max = 0.48, 0.75, and 0.69, respectively) in BKPF. The R max was found at a distance of 4 m for the max DBH and BA (R max = 0.62 and 0.61, respectively) in SCF. In contrast, the correlation of R S with mean DBH8 and max DBH8 was strongest and negative (R max = −0.5 and −0.67, respectively) in SVF. The correlation of R S with mean DBH5 was strongest and negative (R max = −0.59) in KPP. Subsequently, the forest structural parameters that had the strongest relationship with R S and the physical and chemical characteristics of the soil that had a significant relationship with R S were used in a stepwise multiple regression analysis (Tables 3 and 4). A general model with both WFPS and mean DBH8 performed best in BKPF. The final model indicated that SOC and mean DBH8 explained 50 % of the variation in R S in SVF. The soil C:N ratio and forest structural parameters, such as BA4 and mean DBH5, explained the variability of R S in SCF and KPP.

Table 3 Correlation coefficients between soil respiration and biotic and abiotic parameters within a specific forest stand
Full size table
Fig. 4
figure 4

Correlations between maximum DBH (max DBH), mean DBH, and basal area (BA) at various distances (from 2 to 10 m) from each measurement point and soil respiration for the four forest stands. BKPF primary mixed broad-leaved Korean pine forest, SVF spruce-fir valley forest, SCF selective cutting of mixed broad-leaved Korean pine forest, KPP Korean pine plantation

Full size image
Table 4 Model coefficients for estimating soil respiration from individual collars using stand structural parameters and soil properties within a specific forest stand
Full size table

Factors influencing R S of the four forest stands

The seasonal changes in the CVs of R S of the four forest stands ranged from 8 to 25 % (Fig. 5). The mean R S in SVF (3.36 μmol CO2 m−2 s−1) was significantly higher than in BKPF (2.16 μmol CO2 m−2 s−1, P < 0.05; Table 2). The mean T differed significantly among forest stands, with the lowest T in SVF (P < 0.05), but the mean soil water content did not differ significantly among the four forest stands (P > 0.05). We found significant differences among forest stands for the physical and chemical properties of soil, such as field moisture capacity, capillary porosity, light fraction organic C contents, SOC, soil C:N ratio, and pH (P < 0.05). In contrast, TN did not differ significantly among the four forest stands (P > 0.05). The correlation analyses between R S and stand characteristics revealed that R S was significantly and linearly correlated with the physical characteristics of the soil, such as field moisture capacity and capillary porosity (P < 0.05; Fig. 6). In addition, the variability of the mean R S across the four forest stands depended on contents of light fraction organic C and SOC, soil C:N ratio, and pH (P < 0.05). The variability of the mean R S across the four forest stands was best explained by the soil C:N ratio (R 2 = 0.63, P = 0.001; Fig. 6c).

Fig. 5
figure 5

Seasonal changes in the coefficients of variation of soil temperature, soil moisture, and soil respiration among the four forest stands

Full size image
Fig. 6
figure 6

Relationships between soil respiration and the soil organic C content (a), light fraction organic C content (b), C:N ratio (c), field moisture capacity (d), capillary porosity (e), and pH (f) at a depth of 0–10 cm across the four forest stands. BKPF primary mixed broad-leaved Korean pine forest, SVF spruce-fir valley forest, SCF selective cutting of mixed broad-leaved Korean pine forest, KPP Korean pine plantation

Full size image

Discussion

Sources of the variability in R S within a specific forest stand

Our results showed that the minimum CV of R S within plots occurred in KPP. The CVs of forest structure parameters (e.g., max DHB4, mean DBH5, and mean DBH8) for BKPF, SVF, and SCF were higher than that for KPP (Table S5, Electronic supplementary material). This result indicates that the forest structures of the two primary forests (BKPF and SVF) and the selective cutting of mixed forest (SCF) were highly heterogeneous compared to that of plantation (KPP), which could cause higher spatial variability of the rhizospheric component of R S through underground C allocation (Katayama et al. 2009; Luan et al. 2012).

Although there was large temporal differences for R S of a given sample location, the variability of R S remained constant across the growing season within a specific forest stand. This result indicated that some relatively stable factors drive the typically high or low R S of a given sample location. R S originates from rhizospheric respiration and heterotrophic respiration, and the two components are controlled independently by different biotic and abiotic factors (Scott-Denton et al. 2006). These various controlling factors affect the variability of R S (Prolingheuer et al. 2014). Our previous studies and other reports have shown that rhizospheric respiration was correlated with medium or fine root biomass (Pregitzer et al. 2008; Shi et al. 2015). However, root biomass measurements cannot be conducted easily. In contrast, the tree size distribution may account for the variability in R S due to significant relationships between root biomass and forest structure parameters (mean DBH, max DBH, and BA at a certain radius) (Katayama et al. 2009; Bréchet et al. 2011; Luan et al. 2012). In this study, the final model for explaining the variability of R S in BKPF and SCF confirmed this inference. The final model indicated that the forest structure parameters explained the variability of R S in BKPF and SCF. The reason could be that the tree size distribution regulated fine and medium root distribution and biomass, further controlling the variability of R S within the forest stand. Furthermore, we found that R S was negatively correlated with forest structure parameters, such as mean DBH8 and mean DBH5 in SVF and in KPP. This result indicated that larger mean DBH within a few meters (radius) of R S measurement point reduced R S, which appears contradictory to previous inference. But, it is noticeable that the dominant tree species, such as spruce and fir in SVF and Korean pine in KPP, were evergreen coniferous tree species. Thus, the amount of solar radiation reaching the litter and soil layers is small. The sizes of forest gaps or leaf area index may affect R S more strongly when compared to fine and medium root distribution in SVF and KPP. Oishi et al. (2013) reported that within a Pinus taeda plantation, higher leaf area resulted in lower R S, which was attributed to the fact that increased solar radiation associated with lower leaf area index, along with increased T that appears to lead to increases in R S. However, we did not find significant relationships between R S and T (P > 0.05) in SVF and KPP. Several other ecologically driven processes could explain an inverse relationship between R S and forest structure parameters in SVF and KPP, and more research is needed to explain the association. Although the study has several limitations, our results that R S had a significant relationship with forest structural parameters enable us to extrapolate R S spatially using forest structural parameters and have implications for an optimum R S sampling scheme. For example, the selection of R S sampling position should take into account the spatial distribution of tree size.

The heterotrophic component of R S is mainly controlled by several physical and chemical factors of soil and the substrate availability (Luan et al. 2011; Olsson et al. 2012; Jagadamma et al. 2014). Our results showed that WFPS is a good proxy for the physical factors of soil that explain the variability of R S in BKPF. The variability of R S in BKPF depended on the physical properties of soil, whose variability was partly due to the high heterogeneity of microtopography compared to that of the other forest types. The high WFPS that not only inhibits microbial activity as a result of the lack of oxygen (Wiaux et al. 2014) but also limits CO2 emissions due to water filling the soil pores (Ball 2013) could explain the negative relationship between R S with WFPS. Positive relationships were found between the R S and soil C:N ratio in the topsoil in SVF, SCF, and KPP. Our result agrees with that reported by Ngao et al. (2012), who found that the positive relationship between basal soil respiration at 10 °C and the C:N ratio at the topsoil organo-mineral layer probably reflects better C consumption efficiency with increasing C:N ratios rather than the effect of the C supply under non-limiting N conditions. This is due to the fact that litter decomposes more completely under non-limiting N conditions when decomposers are sufficiently supplied with labile C from the humus than when they are C limited (d’Annunzio et al. 2008). All of the above findings suggest that controlling factors for regulating R S depend on forest stands and sites.

The importance of scale on the variability of R S

The seasonal patterns of R S in the four forest stands were strongly related to changes in T, which was consistent with previous studies (Ma et al. 2014). The Q 10 values for R S ranged from 2.05 to 2.69 and showed the varied response of R S to temperature among the four forest stands. The temporal variability of R S is straightforward to capture using an automated soil chamber system at diurnal, seasonal, and annual timescales (Wang et al. 2013; Hanpattanakit et al. 2015). However, the spatial variability of R S and its driving factors are under-researched and complicated at different spatial scales (Buczko et al. 2015). In the present study, the average CVs of R S within plots (30–52 %) were significantly higher than those of the R S among plots (20–25 %) in each forest type (P < 0.05), and this is similar to previous findings (Wang et al. 2006; Tang et al. 2009). Tang et al. (2009) found that the variations in R S within a stand were higher than among stands (four stand replications for the young, intermediate, and mature forests) in a Great Lakes forest chronosequence. These results indicated that finer-scaled heterogeneity in R S is important and should be of concern.

Many of the drivers of R S at small scales may be overpowered by stronger influences at larger scales. Martin and Bolstad (2009) reported that at the scales of 1–10 m, variation between individual R S measurements could be explained by positive relationships between the forest floor mass, root mass, C and N pools, or root N concentration, whereas the topography strongly influenced soil moisture and soil properties and created spatial patterns of R S on the landscape. Oishi et al. (2013) showed that spatially, among three proximate temperate forest ecosystems, R S increased with leaf production, whereas within a P. taeda plantation, R S decreased with increasing leaf production. In this study, forest structure parameters (mean DBH, max DBH, and BA at a certain radius) and physical and chemical factors of soil were strong determinants of the variability in R S within a specific forest stand. However, no significant correlation between the mean R S and the BA, mean DBH, and fine and medium root mass was found across the four forest stands (P > 0.05). It is probable that the forest structure regulated fine and medium root distribution and amount of solar radiation reaching the forest floor (Hilker et al. 2008; Bréchet et al. 2011), further influencing the variability of R S at fine scales within a specific forest stand. However, at larger scales, the fine and medium root biomass did not differ significantly among the four forest stands. The variation of R S across the four forest stands was explained by physical and chemical factors of soil. These results indicate that at fine scales, within a specific forest stand, the biotic and abiotic factors, such as at a distance of a few meters for the mean DBH, max DBH, and BA and soil C:N ratio, determine the variability of R S together, whereas at larger scales, the factors controlling the variability of R S are mostly abiotic in nature (e.g., SOC, soil C:N ratio, and field moisture capacity). In addition, the soil pH was significantly and negatively correlated with R S across the four forest stands (P < 0.05), while there was no significant correlation between soil pH and R S within a specific forest stand, except within SCF as a result of the minor range of soil pH within a specific forest stand. These findings suggest that spatial scales have an important impact on the measuring and modeling of R S.

Conclusions

Our results showed that the variability of R S did not show significant changes with the season within each forest stand. The variability of R S within a specific forest stand depended on the forest structure parameters (i.e., the mean DBH, max DBH, and BA within a few meters (radius) of R S measurement point) and physical and chemical properties of soil, such as WFPS, SOC, and soil C:N ratio. The physical and chemical properties of soil, such as SOC, soil C:N ratio, and field moisture capacity, contributed to the variation of R S across the four forest stands. Furthermore, the variability of R S within plots was higher than that of R S among plots in each forest stand, and thus, fine-scale heterogeneity in R S should be of greater concern. These results imply that the selection of R S sampling position should take into account the spatial distribution of tree size and that sampling scheme with a small number of collars are liable to generate biased average R S within the plot. The sufficient number of replicates that is required to obtain an unbiased average R S within the plot needs to be further studied. Quantifying and understanding the variability of R S at different spatial scales can help reduce errors in the estimates of R S when scaling up from plot-scale measurements to stand, landscape, and regional scales.