Effects of Urban Forest Types and Traits on Soil Organic Carbon Stock in Beijing

Forests can affect soil organic carbon (SOC) quality and distribution through forest types and traits. However, much less is known about the influence of urban forests on SOC, especially in the effects of different forest types, such as coniferous and broadleaved forests. Our objectives were to assess the effects of urban forest types on the variability of SOC content (SOC concentration (SOCC) and SOC density (SOCD)) and determine the key forest traits influencing SOC. Data from 168 urban forest plots of coniferous or broadleaved forests located in the Beijing urban area were used to predict the effects of forest types and traits on SOC in three different soil layers, 0–10 cm, 10–20 cm, and 20–30 cm. The analysis of variance and multiple comparisons were used to test the differences in SOC between forest types or layers. Partial least squares regression (PLSR) was used to explain the influence of forest traits on SOC and select the significant predictors. Our results showed that in urban forests, the SOCC and SOCD values of the coniferous forest group were both significantly higher than those of the broadleaved group. The SOCC of the surface soil was significantly higher than those of the following two deep layers. In PLSR models, 42.07% of the SOCC variance and 35.83% of the SOCD variance were explained by forest traits. Diameter at breast height was selected as the best predictor variable by comparing variable importance in projection (VIP) scores in the models. The results suggest that forest types and traits could be used as an optional approach to assess the organic carbon stock in urban forest soils. This study found substantial effects of urban forest types and traits on soil organic carbon sequestration, which provides important data support for urban forest planning and management.


Introduction
Urban green spaces, including urban forests, play a pivotal role as substitutes for the lost natural environment in the city's original location [1]. Urban forests provide a large number of ecosystem services [2], such as enhancing amenity values [3], maintaining biodiversity [4], and increasing carbon sequestration [5]. The increase and decrease of soil organic carbon (SOC) may affect climate change greatly [6,7]. Moreover, SOC storage impacts the other ecological functions of soil as well, such as biomass production, nutrient and water-holding capacity, infiltration capacity and resistance to erosion, and providing habitats for biological activity [8,9]. Studies of forest groups and species characteristics affecting soil fertility parameters have been widely carried out in the non-urban environment [10], but less so in an urban context [11]. An in-depth understanding of the relationship between urban forest and SOC is key to maintain and enhance the quality of urban ecosystem services.
There is growing recognition of the top-down and bottom-up regulation between plants and soil organisms [12]. Studies of the effects of aboveground forest types and Based on maps and reachable park directories on the Baidu Map, we located over 200 parks on the map. With the map gridded by 5 km × 5 km, we randomly selected one park in each grid. We discarded parks that were built on former factories to ensure that they were only affected by the typical urban forest environment. After this, we manually added some parks more than 50 years old to make the age of parks more even. Then, 107 parks were selected in total. We tried to sample both coniferous and broadleaved forests in each park to compare the SOC content between forest types. Finally, we sampled 168 plots, which contained 99 broadleaved and 69 coniferous plots. There were 61 parks with both coniferous and broadleaved forest types. The information on sampling sites and species is displayed in Appendix A, Table A1.
We followed five principles to select the sampling plots in parks. Firstly, the plots were set more than 5 m away from the walking path and 20 m away from the traffic artery. Secondly, the plot's understory vegetation was stable and was not to be changed every year to avoid large soil disturbances. Thirdly, plots were selected that were not watered within three days to reduce the short-term impact of human management. Fourthly, the plot's size was required to be bigger than 10 m × 10 m. The sizes of plots ranged from 100 m 2 (10 m × 10 m) to 225 m 2 (15 m × 15 m) in this study. Fifthly, monocultures were sampled to focus on the effects of tree species on SOC. The monocultures were composed of common tree species in Beijing and were artificial single communities.

Soil Sampling and Vegetation Investigation
Soil was sampled in September and October 2018 at the edge of the canopy projection, so the distance to the nearest tree trunk ranged from 1.2 m to 10 m, with an average distance of 4 m from the trunk [40,41]. At each sampling point, we removed the litter layer and collected samples by using a steel push corer with a 3.8-cm inner diameter and a length of 20 cm (Soil Sampler; Hongguang Instrument Inc., Shaoxing, China) to obtain three layers of 0-10 cm, 10-20 cm, and 20-30 cm samples [41,42]. Six samples distributed evenly in each plot were sampled for each vertical soil layer. The soil samples were mixed and transported in a cooler to the laboratory for further analyses.
We investigated the vegetation density (VD) of trees, tree height (H), under crown height (UC), diameter at breast height (DBH, 1.30 m above the ground level), leaf area index (LAI), canopy density (CD), fine root biomass (FRB), canopy area (CA), herbaceous Based on maps and reachable park directories on the Baidu Map, we located over 200 parks on the map. With the map gridded by 5 km × 5 km, we randomly selected one park in each grid. We discarded parks that were built on former factories to ensure that they were only affected by the typical urban forest environment. After this, we manually added some parks more than 50 years old to make the age of parks more even. Then, 107 parks were selected in total. We tried to sample both coniferous and broadleaved forests in each park to compare the SOC content between forest types. Finally, we sampled 168 plots, which contained 99 broadleaved and 69 coniferous plots. There were 61 parks with both coniferous and broadleaved forest types. The information on sampling sites and species is displayed in Appendix A, Table A1.
We followed five principles to select the sampling plots in parks. Firstly, the plots were set more than 5 m away from the walking path and 20 m away from the traffic artery. Secondly, the plot's understory vegetation was stable and was not to be changed every year to avoid large soil disturbances. Thirdly, plots were selected that were not watered within three days to reduce the short-term impact of human management. Fourthly, the plot's size was required to be bigger than 10 m × 10 m. The sizes of plots ranged from 100 m 2 (10 m × 10 m) to 225 m 2 (15 m × 15 m) in this study. Fifthly, monocultures were sampled to focus on the effects of tree species on SOC. The monocultures were composed of common tree species in Beijing and were artificial single communities.

Soil Sampling and Vegetation Investigation
Soil was sampled in September and October 2018 at the edge of the canopy projection, so the distance to the nearest tree trunk ranged from 1.2 m to 10 m, with an average distance of 4 m from the trunk [40,41]. At each sampling point, we removed the litter layer and collected samples by using a steel push corer with a 3.8-cm inner diameter and a length of 20 cm (Soil Sampler; Hongguang Instrument Inc., Shaoxing, China) to obtain three layers of 0-10 cm, 10-20 cm, and 20-30 cm samples [41,42]. Six samples distributed evenly in each plot were sampled for each vertical soil layer. The soil samples were mixed and transported in a cooler to the laboratory for further analyses.
We investigated the vegetation density (VD) of trees, tree height (H), under crown height (UC), diameter at breast height (DBH, 1.30 m above the ground level), leaf area index (LAI), canopy density (CD), fine root biomass (FRB), canopy area (CA), herbaceous vegetation cover (HB), and semi-decomposed litter (LIT) as forest traits. LAI was measured with an LAI-2200C plant canopy analyzer (LI-COR Inc., Lincoln, NE, USA), following the method as described in the study by Thimonier et al. [43]. Plant density of trees was calculated using the number of trees divided by the area of the sample plot. In the herb survey, 5 small subplots of 1 m × 1 m were selected, and the herbaceous vegetation cover and herbaceous species were measured in the small samples. We also collected fine roots of three soil layers in each plot and collected three 10 cm × 10 cm frames of litter samples in each plot to obtain the fine root biomass and semi-decomposed litter quantity. Root samples were collected with the steel corer mentioned above. Soil samples with roots were labeled and put into 1-mm mesh nylon bags and we used a flotation method to separate the roots from the soil [44,45]. The nylon bags were soaked in fresh water for 12 h and washed under regular running water to obtain clean roots [46]. Fine roots were divided as those with a diameter less than 2 mm, based on the cleaned roots. We further differentiated live roots from dead roots based on the shape, color, xylem, and elasticity of the roots and eliminated the roots of shrubs and herbs. Fine root samples were dried at 65 • C until the mass was constant and weighed with an electronic balance (±0.0001 g). We obtained the fine root biomass (g/m 3 ) by dividing the fine root weight by the soil sample volume. The semi-decomposed litter samples were dried to a constant weight at 65 • C and weighed [47]. We used another set of tools to dig the complete core samples to calculate soil moisture content (SMC) and soil bulk density (BD). This set of tools was specially designed for study areas which could not be dug up for soil profiles. The set contains a fixed sleeve in which the ring knife can be put in and used to take intact soil samples.

Soil Preparation and Analysis
We analyzed the soil moisture content (SMC), soil pH, soil bulk density (BD), soil organic matter concentration (SOCC), and density (SOCD). In brief, all laboratory tests followed standard methods. SMC was determined by gravimetric methods. To measure soil pH, water and field-moist soil were mixed in a 1:1 volumetric ratio, allowed to stand for 10 min, and pH was then estimated in the supernatant using a bench-top pH meter, reflecting the soil acidity. To test BD, the intact soil cores were dried at a temperature of 105 • C and soil BD was obtained by dividing the dry soil weight by the volume of the intact soil core (100 cm 3 ). The SOCC (g kg −1 ) was measured using a SOC analyzer (Multi N/C 3100, Analytik, Jena, Germany). SOCD (kg m −2 ) within each sample was calculated according to Equation [48]: where BD is the bulk density (g/cm 3 ), SOC is the soil organic carbon content (g/kg), and h is the depth of each soil layer (0.1 m).

Statistical Analyses
In this study, statistical analyses were performed using the statistical software R 4.0.3 (R Core Team, 2016) to reveal the relationships between the urban forest and soil variables. All datasets were checked for normality using the Shapiro-Wilk test (p > 0.05) prior to analysis and then normalized or scaled if necessary.
Soil parameter differences between the two forest types (coniferous and broadleaved) in three layers (0-10 cm, 10-20 cm, and 20-30 cm) were tested by two-way ANOVA, whereas no interaction effect was found in this study. Therefore, a one-way ANOVA test was separately performed on forest types and layers. The Levene test was performed to test the homogeneity of variance. The Kruskal-Wallis test was performed on pH and SMC values as these data failed for normality, and other soil variables (SOCC, SOCD, and BD) were analyzed using the parameter approach. The Steel-Dwass test and the Tukey HSD (honest significant difference) post-hoc tests were used to analyze differences among three layers.
Partial least squares regression (PLSR) was used to evaluate the influence of forest traits on SOCC and SOCD. PLSR integrates the advantages of principal component analysis (PCA), canonical correlation analysis (CCA), and multiple linear regression. PLSR has been widely applied to the study of forest ecosystems and has been proven to reveal the relationship between environmental factors and forest structure. In PLSR, forest traits and soil variables were chosen as predictors. Ten forest traits and three soil variables were chosen as predictor variables for PLSR-H, FRB, LAI, HC, LIT, VD, UC, CD, soil pH, soil moisture content, and bulk density. SOCC and SOCD were chosen as response variables. All variables were inspected for outliers before modeling. Leave-one-out cross-validation was used to select the optimum number of components. The root mean squared error of prediction (RMSEP) and the coefficient of determination (R 2 ) were used to evaluate the model's performance. Variable importance in projection (VIP) scores were used to evaluate the predictors' contribution to PLSR [49]. Predictors with VIP scores greater than 1 were considered important to the models. The R packages 'pls' and 'plsVarSel' were used to perform the analyses.

Variance of Soil Properties under Urban Forests
The soil organic carbon of urban forests varied significantly. The average value of SOCC and SOCD across all the parks was 8.02 ± 0.12 g kg −1 (mean ± standard error) and 1.21 ± 0.02 kg m −2 (mean ± standard error), respectively (Table 1). SOCC showed significant difference between forest types (F-value = 6.556, p < 0.05) and so did SOCD (F-value = 5.264, p < 0.05). Soils under coniferous trees had higher SOCC and SOCD than those under broadleaved trees (Table 1). There was a significant difference in SOCC (p < 0.01) among different layers but not in SOCD. According to multiple comparisons, the 0-10 cm layer held significantly higher SOCC than both the 10-20 cm and 20-30 cm layers. Bulk density (1.53 ± 0.19 g cm −3 ) and pH (7.82 ± 0.55) showed no significant difference between the coniferous and broadleaved plant groups (p > 0.05). SMC (12 ± 5%) was significantly different between these two forest types. The soil under coniferous trees was moister than that under broadleaved trees. There were significant differences among the three layers only in BD. Table 1. SOCC (g kg −1 ) and SOCD (kg m −2 ) in the coniferous and broadleaved groups in three layers. Results are presented as means ± standard error. Abbreviations: SOCC, soil organic carbon concentration; SOCD, soil organic carbon density. Different letters in the same row of SOCC and SOCD indicate significant differences between layers at the 0.05 level.

Variance of Forest Traits
The results showed that two popular families, Salicaceae and Pinaceae, dominated the forests. The Wilcoxon test for paired samples was performed to identify the differences in forest traits between plant groups. There were significant differences in forest traits between coniferous and broadleaved groups (Figure 2). The results showed that there were significant differences in most forest traits between the two forest types. Among them, LAI, FRB, VD, HB, and CD values were significantly higher in coniferous forests. In contrast, H, UC, DBH and CA were significantly higher in broadleaved trees. Meanwhile, the semi-decomposed litter quantity showed no significant difference between forest types, which may be due to clearing and leaf management in parks.

Effect of Forest Traits on SOC
The performance of the PLSR models is shown in Table 2. For the total SOCC and SOCD models, higher R 2 and lower RMSEP values were observed in the total SOCC model. In the total SOCC model, the components explained 42.07% of the variance in SOCC; in the 10-20 cm model, the components explained 45.50% of the variance in SOCC, which was the highest level of explanation among the three layers. In the total SOCD model, the components explained 35.83% of the variance in SOCD; in the 10-20 cm model, the components explained 45.83%. These results showed that forest traits had a more substantial explanatory power for total SOCC. Simultaneously, three components were observed in both the best performing SOCC and SOCD models, and the addition of more components did not improve the explanatory power of the model. The following components did not have a strong correlation with the residuals of the predictors. Different models in different layers analysis showed that forest traits had more substantial explanatory power for the SOCC and SOCD models at a depth of 10-20 cm. In comparison, the explanatory power for the 0-10 cm and 20-30 cm models was slightly weaker than that of the 10-20 cm model. Concerning the VIP scores of each of the forest traits, DBH had an important contribution in all models ( Table 3). The traits of LAI, DBH, CA, HC, and CD showed important contributions to the models of both the total SOCC and SOCD, indicating that forest traits associated with forest biomass had a strong relationship with SOC in urban forests. Moreover, comparing the results of different layer models showed that more forest traits contributed to the SOCC and SOCD models at the 10-20 cm layer than the other two layers. FRB's VIP score was greater than 1 only in the 10-20 cm layer model, which showed a relationship between forest fine roots and SOCC and SOCD in urban forests. The contribution of forest traits was the least in the 0-10 cm model. Additionally, based on the regression coefficient of each forest trait, it was found that except for the weak negative correlation coefficient between CA and SOC in the 0-10 cm layer model, the other forest traits positively correlated with SOC. The VIP scores of LIT, VD, and UC in all models were less than 1. The results showed that these three forest traits had no statistically significant relationships with SOCC and SOCD in urban forests. Overall, these results showed that forest traits positively affected SOCC and SOCD in urban forests.

Discussion
Our results showed that the coniferous and broadleaved urban forest groups significantly affected soil organic carbon. Specifically, SOCC and SOCD were higher in the coniferous group than in the broadleaved group. This result is similar to findings in a natural context by Błońska and Gruba [50] and findings in urban parks in Finland by Setälä et al. [41]. Here, even in an urban context, SOC of the coniferous type forest was higher than that of the broadleaved forest, similar to forest communities in natural ecosystems-for instance, the carbon concentration under a spruce forest was found to be higher than under broadleaves [51,52]. This trend in SOC content of different forest types may be because conifers have a higher leaf area index, even though broadleaved trees have larger crowns. Our result is similar to Bae and Ryu's findings that complex canopy structure may boost organic carbon in forest soils [53]. The FRB of coniferous trees is also higher than that of broadleaved trees, and fine roots are another factor that directly affects soil organic carbon [54]. Some studies on urban forest roots have revealed that the roots of urban trees can maintain or enhance organic matter in soil [55]. In the three soil layers, SOCC in 0-10 cm was significantly higher than the other two layers, whereas there was no difference in SOCD among layers. Edmondson reported similar results, demonstrating that not only SOCC but also SOCD of urban forest differed among different layers [17]. In our study, the top layer in the coniferous group held the highest SOCC (Table 1). Due to the input of nutrients from vegetation and artificial management, the content of organic carbon in the surface layer of soil is usually high, which exhibits a dynamic interaction with biological and anthropogenic activities [56,57]. Therefore, in cities, soils under coniferous trees, especially the upper layer, can provide greater potential for organic carbon storage.
The PLSR of SOCC and SOCD models selected significant variables with VIP scores > 1. DBH was the only variable in all models showing VIP > 1 and the highest VIP, which indicated that DBH was the most important predictor for SOC content in our study. Our result is similar to the study of Eni, which found that DBH was the only canonical variable revealing soil-vegetation interrelationships [58]. In the two forest types, the DBH of broadleaves was higher than that of conifers, whereas the soil organic carbon under broadleaves was lower than that of conifers. Moreover, the result of PLSR showed that there was a positive correlation between DBH and SOCC. This means that the difference in coniferous and broadleaved forest soil organic carbon is likely to be caused by many factors. Moreover, we found that all variables of crown characteristics-LAI, canopy area, and canopy density-were also important for the prediction of urban forest soil organic carbon, with VIP scores > 1 in at least seven models (Table 3). Thus, the results indicate that the canopy structure of urban forest communities has an important impact on soil organic carbon. The difference in forest traits among forest types showed that the LAI and canopy density of conifers were higher than those of broadleaved trees, whereas canopy area displayed the opposite result. These results, combined with the PLSR results, may emphasize that LAI and canopy density have a stronger effect on soil organic carbon and are more suitable to be predictors. On the one hand, a complete and complex canopy structure in urban forests can increase the input of litter in the area, which becomes an important source for accumulating organic carbon in the soil [59]. According to the significant difference between forest types in the leaf area index and the lack of differences between them in litter, we can infer that human activities strongly influence this mechanism, such as litter removal through management practices [60][61][62]. On the other hand, changes in canopy structure affect the microclimate of the soil habitat, including environmental factors such as temperature, light, wind speed, and precipitation [63]. The environments of forests can affect the activities of fauna and organisms in the soil [64,65], thus changing the overall soil respiration intensity and affecting the decomposition and transformation of soil organic carbon [66]. Based on the models of total SOCC and SOCD, the forest traits mentioned showed a positive correlation with urban soil organic carbon. This positive correlation revealed a correlation between tree biomass and soil organic carbon content, suggesting that urban forests with more complex canopies and higher leaf areas may have higher organic carbon stock in their soils. In addition, our results also showed that forest type did not account for all variance in SOC, indicating that more factors need to be taken into account, such as urban forest age and position. Many studies have shown that urban forest age has a positive effect on SOC and SOC [11,67,68]. Habitats with a more mature and stable urban forest structure can provide good shelter and necessary food sources for aboveground organisms [68] and play a positive role in the underground ecological environment [21].
In addition, we used forest traits to predict the overall SOCC and SOCD and studied the relationship between SOC content and forest traits at different soil layers. The results showed that SOC content in 10-20 cm layer was most closely related to forest traits, with the highest model interpretation and stability (Table 2). Notably, the VIP score of fine root biomass was greater than 1 in the models both of SOCC and SOCD in the 10-20 cm layer, which indicated the contribution of fine roots to soil organic carbon in this layer. Huyler et al. also found that tree roots played a role in maintaining the carbon level below the soil surface [69]. Fine roots are important plant organs used to absorb water and nutrients [70], and fine roots are analogous to leaves and central organic matter inputs, given that the sloughed roots are added to the soil humus pool [68]. Due to the rapid renewal of fine roots, the annual return of carbon, nutrients, and energy from fine roots to the soil is even higher than that of ground litter [71], which was confirmed by the study of Hui et al. [72]. At 0-10 cm, the models showed that the contribution of forest traits to SOC content was the weakest. The contribution of litter content to SOC was not found in any model, which indicates that the organic carbon in urban forest soil was affected by other factors than litter input in our study. This study's sampling area was concentrated in urban parks, and the sampling area was disturbed by human factors to varying degrees. At present, most of the litter management in parks in Beijing area is still dominated by regular cleaning, which leads to the litter being cleared away and degrading soil organic carbon. The high-intensity manual management pattern in urban parks destroys surface soil structure and accelerates the decomposition of organic carbon and therefore SOC cannot accumulate effectively [73,74]. In contrast, green space management in Paris may help to increase SOC stock in open soil [75]. However, researchers of urban forest soil organic carbon need to learn more from the research experience of natural forest soil and carry out more systematic and comprehensive research in the world [76].

Conclusions
A better understanding of the effects of urban forests on SOC content is crucial to advancing urban ecosystem services, even though quantifying and predicting carbon in urban forest soil remains difficult. Thus, our results demonstrated that urban forests affect the concentration and density of soil organic carbon through forest types and forest traits in temperate climates. We found that there was a significant difference in SOC content between coniferous and broadleaved types in urban forests in Beijing. Using PLSR, we selected diameter at breast height, leaf area index, crown area, and canopy density as the significant predictors of soil organic carbon in urban forests. The data show that analyzing forest traits could be an optional approach for cost-effectively predicting urban forest SOC with higher accuracy and practicability.