Soil respiration in a subtropical forest of southwestern China: Components, patterns and controls

Partitioning the components of soil respiration is crucial to understand and model carbon cycling in forest ecosystems. In this study, total soil respiration (RS), autotrophic respiration (RA), heterotrophic respiration (RH), litter respiration (RL), litterfall input and environmental factors were synchronously monitored for 2 years in a subtropical Michelia wilsonii forest of southwestern China. RH rates were often higher than RA rates during the two years except for the middle growing season (from July to September). The mean rate of Rs, RA, RH and RL was 1.94 μmol m-1 s-1, 0.85 μmol m-1 s-1, 1.09 μmol m-1 s-1 and 0.65 μmol m-1 s-1, respectively, during the 2-year experiment. Annual CO2 emission derived from RA, RH and RL was 3.26 Mg C ha-1 a-1, 4.67 Mg C ha-1 a-1 and 2.61 Mg C ha-1 a-1, respectively, which accounted for 41.4%, 58.6% and 32.9% of RS. Over the experimental period, the ratio of RA/RS increased with soil temperature but the opposite was true for RH/RS and RL/RS. The Q10 value was 2.01, 4.01, 1.34 and 1.30, respectively, for RS, RA, RH and RL. Path analysis indicated that environmental variables and litterfall production together explained 82.0%, 86.8%, 42.9% and 34.7% variations of monthly fluxes of RS, RA, RH and RL, respectively. Taken together, our results highlight the differential responses of the components of RS to environmental variables.


Introduction
Soil respiration (R S ) is the second largest flux of carbon dioxide (CO 2 ) between terrestrial ecosystems and the atmosphere [1,2]. R S accounts for roughly 80% of ecosystem respiration across global forests [3]. In general, R S is largely controlled by environmental factors, including temperature and moisture [4]. Moreover, R S is complicated by tree growth and the subsequent input of plant litter to soil [5,6].
R S is overwhelmingly the product of respiration by plant roots (autotrophic respiration, R A ) and soil organisms (heterotrophic respiration, R H ) [7,8]. In addition, CO 2 flux derived from decaying litter accounts for a considerable part of R S , which is strongly controlled by the quantity and quality of litter and climate variables [5,8,9]. A recent meta-analysis indicated that aboveground litter removal and root removal declined R S by 22.8% and 34.1%, respectively [6]. Soil temperature is one of the most important factors that control the variations of R S in terrestrial ecosystems but the size of this effect is dependent on system types and climate zones [10,11]. Soil moisture also mediates the temporal and spatial pattern of R S [4,12]. Both R A and R H generally increased with temperature and precipitation across global forest ecosystems [4]. However, moisture effect may be complicated by the effects of soil temperature and other factors [13,14]. Different components of R S could be dominantly mediated by different mechanisms, such as substrate quality, plant traits and environmental factors. Obviously, partitioning the components of R S and exploring the relative importance of biotic and abiotic factors on each component is very helpful for understanding the mechanistic of soil carbon cycling. Chinese subtropical forests have high biomass and productivity, which play an important role in the carbon storage of global terrestrial ecosystem [15]. Over last decades, almost all natural forests in subtropical China have been deforested due to demand for timber, and subsequently often reforested with fast-growing tree species. So far, previous studies focused mainly on the plantations dorminated by non-native fast-growing tree species, such as Pinus massoniana and Cunninghamia lanceolata [16,17]. However, the components of R S and its controls have been scarcely investigated in the restored forests dorminated by native tree species in this region. In this study, we investigate the components of R S for two years using root trenching and litter exclusion techniques in a subtropical Michelia wilsonii (a special native tree species) forest of southwestern China. The specific objectives of this study were (1) to explore the seasonal dynamics of each respiration component; (2) to assess the relative importance of biotic and abiotic factors for temporal patterns of each component.

Ethics statement
We received a permission from the Dujiangyan Bureau of Forestry to conduct this experiment in the studied forest in 2015. In this study, only limited soil samples were collected to measure physical and chemical properties and PVC chambers were set up to monitor soil respiration. Our work had negligible effects on the function of the broader ecosystem. Additionally, this study was carried out in compliance with the laws of the People's Republic of China. This study did not involve measurements of humans or animals, and no endangered or protected plant species were involved.

Site description
The site is conducted in the Dujiangyan Experimental Forest of Sichuan Agricultural University, southwestern China (103˚37' E, 30˚59' N, 911 m asl). This area is characterized by a midsubtropical, humid, mountainous climate that produces the wet season from May through October and the dry season from November through April. The mean annual temperature and precipitation is 15.2˚C and 1 243 mm, respectively. The dominant tree species is Michelia wilsonii and the understory are Lespedeza bicolor, Pittosporum glabratum, Dranceopteis dichotome and Cyperus rotundus, respectively. The soil is classified as ferralsol with old alluvial yellow loam according to the Chinese Soil Taxonomy (RGCST 2001). The basic topsoil properties (0-20 cm) as determined in August 2016 are as follows: organic C 15.76 g kg −1 , total N 1.92 g kg −1 , total P 0.32 g kg −1 , and pH 5.73. Neither fertilization nor drainage had been carried out since tree establishment in the stand. Moreover, the topography is relative flat (less than slope 10˚) and tree canopy coverage is about 0.9.

Experimental design
In August 2015, five 10 m × 10 m replicate plots were established in the experimental site. There were three treatments: control [C] (undisturbed), No-Roots [NR] (root growth excluded) and No-Litter [NL] (aboveground litter excluded). In each plot, a trench subplot was set up with a dimension of 1 m × 1 m. For the trenched subplots, we dug a trench of 0.2 m width and 0.6 in depth. The polyethylene films (37 μm mesh size) were placed along the bottom and sides of the trenches to prevent roots from entering the trench. The excavated soil was gently backfilled into the trench according to its initial profiles to minimize disturbance. Existing litters in the NL plots (1 m × 1 m) were removed and litterfall was excluded by pyramid-shaped screens (1-mm mesh) placed approximately 1 m aboveground. All litters fell around the NL plots were monthly removed before the measurements of soil respiration.
The trenching method was applied to divide R S into R H and R A , the litter removal was used to calculate the respiration derived from aboveground litter decomposition. We calculate the soil respiration fractions from each source as follows:Heterotrophic respiration (R H ) = NR plots Autotrophic respiration ðR A Þ ¼ Control plots À NR plots Litter respiration ðR L Þ ¼ Control plots À NL plots

Soil respiration measurements
In each treatment plot, one polyvinyl chloride collar (PVC) (diameter of 20 cm and height of 8 cm) was permanently installed 2-3 cm deep into soils to measure soil respiration. All vegetation inside the collar had removed artificially before monitored soil respiration. To minimize the response caused by transient decomposition of dead roots, the initial measurements of soil respiration was carried out 3 months after trenching treatment was conducted. From November 2015 to October 2017, soil respiration was measured every month using an automated soil respiration system (Li-8100, Li-Cor Inc., Lincoln, NE, USA). Soil respiration was measured between 9.00 a.m to 12:00 a.m (Beijing time). At the same time, soil temperature (˚C) and moisture (v/v) at the depth of 5 cm was synchronously measured by the probes connected to Li-8100 system.

Microclimate and litterfall measurements
Rainfall, air temperature (T air ) and soil temperature (T soil ) were constantly measured using HOBO Micro Station Data Loggers (Onset Computer Corporation, Bourne, Massachusetts, USA) located adjacent to the experimental area. Forest aboveground litter was collected using six circular litter traps in each plot. The litter traps were funnel shaped with the diameter was 1 m and the collection area was 0.785 m 2 . During the monitoring period, each trap was collected monthly, then collected litter was oven-dried at 65˚C to a constant mass for 48 h, and weighed.

Statistical analyses
Based on the measured data, an exponential model was performed to describe the relationship between soil respiration fractions and soil temperature [18]: Where R was soil respiration rate (μmol m -2 s -1 ), T was soil temperature (˚C), coefficient α is the intercept of soil respiration when temperature is zero, and coefficient β represents the temperature sensitivity of soil respiration. Based on the continuous soil temperature at the 5 cm depth, monthly and annual soil CO 2 fluxes of each fraction was estimated by integrating CO 2 fluxes for the period from November 2015 to October 2017 using the observed specific response equation between soil respiration fraction and soil temperature. Monthly and annual soil CO 2 fluxes of each fraction was estimated by interpolating measured soil respiration between sampling dates for every day of the year and then computing the sum to obtain the annual or winter values [19].
Structural equation models (SEM) were used to assess the holistic effect of measured variables on monthly fluxes of each respiration component. In the model, monthly soil respiration emission of each component was the response variable. Monthly mean air and soil temperature, monthly precipitation (MP), litterfall production, soil moisture were variables. The normality of data distribution was examined for heteroscedasticity, and all bivariate relationships were checked for signs of nonlinearities before the SEM analysis. These analyses were performed by using the maximum-likelihood estimation. Model fit was considered good when the χ 2 test was low (< 2) and its associated p-value was high (> 0.05). The Bentler's comparative fit index (CFI) and Bentler-Bonett normed fit index (NFI) were used to evaluate the adequacy of fit. For each set of analysis, R 2 values were obtained for each dependent matrix, representing the proportion of total variance explained by the model. All statistical analyses were carried out in SPSS version 20.0 for Window (SPSS Inc, Chicago, Illinois, USA). Graphs were generated using SigmaPlot 12.5 software (Systat Software, Inc., San Jose, CA, USA). Significance was determined at α = 0.05.

Microclimate and litterfall
Both air and soil temperature showed a clear seasonal dynamic (Fig 1A). Average air temperatures was 14.6˚C and 14.9˚C, respectively, in 2016 and in 2017; likewise, mean soil temperatures was 15.1˚C in 2016 and 17.9˚C in 2017. Rainfall was 872.8 mm and 944.8 mm, respectively, in 2016 and 2017. Similar to air temperature, it was higher in summers but lower in winters ( Fig 1B). However, there was no obvious seasonal variation in soil moisture and a minimum was observed in June of both years (Fig 1B). Annual aboveground litterfall was 312.0 g m -2 a -1 in 2016 and 352.2 g m -2 a -1 in 2017 ( Fig 1C). In general, the amount of litterfall was significantly greater in the winter (November and December) as compared to other seasons ( Fig 1C).

Soil respiration components
The rates of all soil respiration components had obvious seasonality, with the maximum in summer and the minimum in winter (Fig 2A). The mean annual rates of R S , R A and R H was 1.94 μmolÁm -2 Ás -1 , 0.85 μmolÁm -2 Ás -1 and 1.09 μmolÁm -2 Ás -1 , respectively, throughout the experimental period (Table 1). In general, R H was higher than R A over the year. However, the opposite tendency was observed from June to August.
During the two years, the proportion of R A to R S (R A /R S ) ranged from the maximum 70.6% in summer to the minimum 16.6% in winter, whereas the proportion of R H to R S (R H /R S ) showed the opposite pattern (Fig 2B). The proportion of R L to R S (R L /R S ) varied from 14.1% to 67.5%. Annual CO 2 emission derived from R S , R A , R H and R L was 7.93 Mg C ha -1 a -1 , 3.26 Mg C ha -1 a -1 , 4.67 Mg C ha -1 a -1 and 2.61 Mg C ha -1 a -1 , respectively (Table 1). There was a positive logarithmic relationship between R A /R S and soil temperature (Fig 3A, P < 0.001). Conversely, both R H /R S and R L /R S decreased with increasing soil temperature (Fig 3B and 3C, P < 0.01).

Correlations between environmental variables and soil respiration components
R L , R S , R A and R H rates all exhibited an exponential correlation with soil temperature (Fig 4). Correspondingly, the Q 10 values of R S , R A and R H were 2.01, 4.01 and 1.34, respectively ( Fig  4A-4C). Moreover, the Q 10 value of R A was higher than those of R H and R L .

Forest soil respiration in southwestern China
Path analysis showed that measured variables together explained the variation of 82.0%, 86.8%, 43.0% and 34.6%, respectively, in monthly flux of R S , R A , R H and R L (Fig 5). Total, direct and indirect effects of the environmental variations were summarized in Fig 5. T air and rainfall were strongly associated with monthly flux of R S and R L (Fig 5A and 5D). Moreover, T air had an indirect effect on monthly flux of R S (0.26), R A (0.77) and R H (0.41) via T soil , litterfall and moisture (Fig 5A-5C). Soil moisture had a significant negative effect on R S (-0.31), R H (-0.54) and R L (-0.22) (Fig 5A, 5C and 5D).

Annual C release and variations through soil respiration components
Previous synthesis showed that annual mean total CO 2 efflux was 3.22 Mg C ha -1 a -1 , 6.62 Mg C ha -1 a -1 and 10.92 Mg C ha -1 a -1 , respectively, for boreal, temperate and tropical forests [2]. The mean annual soil CO 2 effluex of Chinese subtropical forests is 10.64 Mg C ha -1 a -1 , which is comparable to the reported value from global tropical forests [2]. In the current subtropical forest stand, annual Rs emission (7.93 Mg C ha -1 a -1 ) is similar to the value reported in the tropical and subtropical forests (3.45-15.2 Mg C ha -1 a -1 ) [20]. Annual Rs efflux in M. wilsonii forest is higher than those of Cunninghamia lanceolata plantation in Hunan (4.55 Mg C ha -1 a -1 ) and Fujian (4.54 Mg C ha -1 a -1 ) of eastern subtropical China [21,22], and is comparable to those of Castanopsis kawakamii forest (9.34 Mg C ha -1 a -1 ) and Pinus massoniana forest (9.57 Mg C ha -1 a -1 ) [23], but is lower than those of Mytilaria laosensis (10.68 Mg C ha -1 a -1 ), and Castanopsis carlesii (11.18-12.31 Mg C ha -1 a -1 ) and Cunninghamia lanceolata forests (11.99 Mg C ha -1 a -1 ) in subtropical China [9,22,24,25]. These differences may be attributed to climates, vegetation types or substrate quality. Previous studies suggested that mean annual temperature (MAT) and mean annual precipitation (MAP) were positively correlated with Rs [4]. Additionally, there is also a positive relationship (R 2 = 0.35, P = 0.02) between annual respiration flux and MAT in Chinese subtropical forest ecosystems, expect for the Mt. Ailao forest site (S1 Fig; S1 Table). Similarly, annual respiration flux is significantly positively correlated with MAP (R 2 = 0.31, P = 0.03) in subtropical China. In our site, both MAT (15.2˚C) and MAP (1 243 mm) were lower than those reported in Fujian (MAT 20.1˚C and MAP 1 670 mm) [24] and in Jiangxi (MAT 17.9˚C and MAP 1 469 mm) [16], respectively. As a result, lower MAT and MAP may, to some extent, account for our smaller annual Rs flux as compared to other subtropical forests with higher MAT and MAP.

Contribution of components to total soil respiration
The relative contributions of R A and R H to total R S varied with forest types, climate and methods [26][27][28][29]. R A contributed 14%-73% to R S among global forest ecosystems [27]. Likewise,

Forest soil respiration in southwestern China
The R H /R S varied from 10% to 94% among global forest ecosystems [4]. A global synthesis of forest soil respiration showed that the contribution of R A to total R S is higher in deciduous broadleaf forest than in evergreen broadleaf forest and evergreen needle leaf forest [4]. For example in a temperate deciduous forest in northern China, the higher R A /R S (ranged from 61.7%-77%) is closely related to the higher root biomass associated with belowground carbon metabolisms [30]. In this case, mean R A /R S was 41.1%, which was higher than the mean value (30%) of global forest ecosystems [2]. Similarly, the mean R A /R S value (41.1%) estimated in this evergreen broad-leaved forest is comparable to the results observed in a monsoon evergreen broad-leaved forest (tree species name) (44.52%) in southern China [31], but is lower than the findings investigated in an evergreen needle forest (Pinus massoniana) (55-63%) or in a deciduous broadleaf forest (tree species name) (54-59%) [32]. As a consequence, the contribution of R A to R S might, to large extent, be dependent on forest functional type that determines the belowground root growth and activities [11,[30][31][32]. Additionally, in this study, the relative contribution of R A and R H to R S largely varied with season. Over the growing seasons (from July to August) of two years, R A contributed more to R S as compared to R H . This is due to the fact that R A and R H are dominated by different mechanisms [10]. R A is closely linked to root activity and photosynthesis, while R H is the respiratory product of soil organic matter decomposition that mainly controlled by substrate and temperature [26]. As a result, there is a significant increase in R A during the growing seasons in M. wilsonii forest as a result of fast root growth and large root secretions. In addition, R H /R A increased with soil temperature but R A /R A decreased with soil temperature. R A was more sensitive to temperature relative to R H . Such results also can, to some extent, account for the higher R H /R A in the growing season noted in this study.
Litter respiration is an important source of CO 2 emission. In our study, annual CO 2 emission from litter layer was 2.61 Mg C ha -1 a -1 , which value is higher than the results observed in Cunninghamia lanceolata (1.15 Mg C ha -1 a -1 ) and Castanopsis kawakamii (1.17 Mg C ha -1 a -1 ) forests in subtropical China [5,22], but is lower than the values reported in a Castanopsis carlesii forest (4.34 Mg C ha -1 a -1 ) [5], indicating R L is mainly regulated by tree-associated litter quantity and quality [8]. During the experimental period, aboveground litter accounted for 14.1%-67.5% of total R S , with a mean value of 32.9%, which is close to the results found in secondary Castanopsis carlesii forest (34.4%) [5]. Numerous studies have evidenced that litter manipulation could markedly change total soil respiration [5,8,33]. Li et al. [5] found that litter addition could enhance annual CO 2 flux by approximately 12.5%, but decreased by 15.1% when the litter was removed.

Temperature sensitivity of soil respiration components
Over past decades, the temperature sensitivity of R S (hereafter referred to as apparent temperature sensitivity) has gained more attention due to its importance for climate-carbon feedback in terrestrial ecosystems. In this study, the temperature sensitivity (Q 10 value) of R S was 2.01, which is similar to the mean value (2.51) of Chinese forest ecosystems [12], but is lower than the mean value (3.4) estimated for global forest ecosystems [4]. Different components of R S may response to soil temperature differently [34]. For an example, the Q 10 value for R A and R H was 3.74 and 1.92 in a Castanopsis carlesii plantation, respectively [24]. A synthesis also indicated that the Q 10 value of R A is significantly higher than that of R H in global forest ecosystems [4]. Similarly, the Q 10 value of R A and R H was 4.01 and 1.34, respectively in the present study, implying that R A is more temperature-dependent as compared to R H . The C release derived from litter layer is easier to be affected precipitation as compared to R A and R H . There was a negative relationship between R L and soil moisture in this study. Thus, the temperature dependence of R L could, to large extent, be offset and complicated by negative effect of soil moisture. Therefore, no significant correlation was detected between R L and soil temperature in the present study. A significant positive linear correlation was found between R A /R H and Q 10 of R A in Chinese subtropical forests (R 2 = 0.72, P < 0.01, S2 Fig). Such result showed that a greater T air : monthly mean air temperature, T soil : monthly mean soil temperature, Rainfall: monthly rainfall amount, Litterfall: monthly aboveground litter production, Moisture: soil moisture. Solid lines are shown positive correlations path, dashed lines are shown negative paths, and gray lines mean removal paths. Standardized coefficients are listed on each path. " Ã " mean significant difference at 0.05 level, " ÃÃ " mean significant difference at 0.01 level. " ÃÃÃ " mean significant difference at 0.001 level. contribution of R A to R S may produce a higher Q 10 of R A , suggesting that root activities might more sensitive to future warming [4].

Effects of environmental factors on soil respiration components
R A is mainly influenced by tree species and root activity [14]. However, both R H and R L are mainly influenced by soil temperature and substrate properties. The path analysis indicated that measured variables together explained 82% and 87% variations monthly fluxes of R S and R A, respectively (Fig 5A and 5B). Soil temperature is a key factor regulating the variations of two components. Contrastingly, all factors together only accounted for 43% variations in monthly flux of R H . This is attributed to the fact that season-associated changes, such as moisture or microbes, counteract the dominant effect of temperature on R H . Similar phenomena have recently been observed in grassland ecosystem [35].
Similar to R H , litter layer is stored on the surface of the forest ground. The CO 2 emission from litter decomposition is easy to be affected directly and/or indirectly by multiple factors, including temperature, moisture, microorganisms. Path analysis showed that all factors together only explained 35% variation in monthly flux of R L . Sufficient soil moisture associated with frequent rainfall in the study area throughout the year may overshadow temperature effect [36]. A recent study also suggested that soil temperature had no significant impacts on R L in a subtropical mixed forest [37].
Trees species-induced variation in the quantity and quality of litterfall might impact the relationship between R L and litterfall [17]. For an instance, there was a significant relationship between R L and litter mass of the current month in Cunninghamia lanceolata forest. However, R L was significantly related to the litterfall of two months ago in Castanopsis carlesii forest [5]. In our site, M. wilsonii is a broad-leaved evergreen tree species, whose litterfall had a peak in winter and a sub-peak in spring. Several results have suggested a potentially lagged effect of litterfall on R L in subtropical forests [5,17]. Similar finding was observed in our study. There was a significant correlation between R L rate (4 months behind) and monthly litterfall in the M. wilsonii forest (R 2 = 0.46, P = 0.04).

Conclusions
We separated R S into different components (R A , R H and R L ) using trenching and litter removal techniques in a subtropical M. wilsonii forest of southwestern China. The contribution of each component to total R S varied with seasons. R A /R S increased with increasing soil temperature, whilst both R H /R S and R L /R S declined with increasing soil temperature. Path analysis showed that monthly fluxes of each component were dominated by different factors. T soil and other factors can well-predict the seasonal dynamics of R S and R A . Measured variables did not show a good correlation with R H and R L . The results noted in this study highlight the important implication of rainfall and root phenology for soil respiration in subtropical forests in this specific region.  Table. Values of forest CO 2 efflux from partial Chinese subtropical forests in literature. (DOCX)