Forest Structure Regulates Response of Erosion-Induced Carbon Loss to Rainfall Characteristics

: Forestation is a common measure to control erosion-induced soil and carbon (C) loss, but the effect can vary substantially between different types of forest. Here, we measured event-based runoff, soil, dissolved organic carbon (DOC), particulate organic carbon (POC) and total C loss with runoff plots (20 m × 5 m) in a broad-leaved and a coniferous forest in subtropical China and explored their relationships with rainfall amount, average intensity, maximum 5-min intensity and rainfall erosivity. The broad-leaved forest had a denser canopy but sparse understory vegetation while the coniferous forest had a relatively open canopy but dense understory vegetation. The results showed that runoff, soil, DOC, POC and total C losses were all significantly higher in the broad-leaved forest than the coniferous forest despite the potentially higher canopy interception associated with the greater leaf area index of the broad-leaved forest. The mean runoff in the broad-leaved forest was 3.03 ± 0.20 m 3 ha − 1 event − 1 (mean ± standard error) and 12.49 ± 0.18 m 3 ha − 1 event − 1 in the coniferous forest. The mean soil, DOC, POC and total C loss (kg ha − 1 event − 1 ) was 1.12 ± 0.16, 0.045 ± 0.003, 0.118 ± 0.016 and 0.163 ± 0.017, respectively, in the broad-leaved forest and 0.66 ± 0.09, 0.020 ± 0.002, 0.060 ± 0.009 and 0.081 ± 0.010, respectively, in the coniferous forest. Runoff and DOC losses were driven by rainfall in two forests, but the key rainfall characteristic driving soil, POC and total C losses was different in the broad-leaved forest from that in the coniferous forest due to their different understory patterns. Soil, POC and total C losses were mostly driven by rainfall amount in the broad-leaved forest but by EI 30 in the conifer forest. Our findings highlight that the response of erosion-induced carbon loss to rainfall characteristics differs between different forest types of the same age but contrasting overstory and understory vegetation covers. Moreover, our study underscores the overlooked significance of understory vegetation in regulating these effects. Thus, we call for the inclusion of understory vegetation in the modeling of soil and carbon erosion in forest ecosystems.


Introduction
Soil erosion is of increasing concern globally because it leads not only to land degradation but also soil carbon (C) loss [1].Estimated global terrestrial soil C loss via soil erosion is as high as 0.3-1.0Pg C yr −1 [2], with the potential of weakening the function of soil as a major terrestrial C pool to mitigate climate change [3].Forest restoration has the potential to minimize soil erosion [4], but the effectiveness of forest restoration in reducing soil erosion varies between different forest types [5,6].For instance, differences in gross rainfall partition, a key factor affecting surface runoff, between broad-leaved and coniferous forests [7,8] can lead to differences in soil erosion between the two types of forest.However, our knowledge about the differences of erosion-induced soil C loss between different types of forest remains incomplete.
One widely recognized factor that contributes to differences in soil erosion between broad-leaved forests and coniferous forests is their differences in canopy cover [9,10].A recent global synthesis has concluded that leaf area index (LAI) is the most important factor affecting canopy rainfall interception [11], highlighting the role of leaf area in determining forest water interception and thus surface runoff [12,13].However, while the role of overstory canopy in water interception and soil erosion has been widely recognized, the role of understory vegetation on soil erosion is not well-understood [14][15][16].Understory vegetation not only contributes to throughfall interception but also increases ground roughness [17], which in turn reduces runoff velocity and increases runoff flow path, runoff flow resistance, soil infiltration capacity and sediment trapping [18][19][20][21].It has been illustrated that understory vegetation removal increased erosion-induced soil loss by 8.5 times and C loss by 7.5 times in a rubber plantation of southwestern China [14].More importantly, because understory growth is largely limited by light [22,23], forests with closed overstory canopy typically have sparse understory such that forests with high overstory interception tend to have low understory interception [24].Thus, the differences in surface runoff-induced C loss between forests with different overstory cover may be offset by their difference in understory vegetation cover.Therefore, the often-neglected understory vegetation should be taken into consideration in assessments of how forest restoration mitigates erosion-induced soil and C losses.
In addition to vegetation properties, rainfall characteristics including rainfall amount, erosivity, average intensity and maximum intensity are known to affect soil erosion [25][26][27].The importance of different rainfall characteristics in driving erosion has been reported to be regulated by vegetation [28].In forest ecosystems characterized with multiple vegetation layers, the relation between soil erosion and rainfall characteristics could be very complicated.For instance, forest canopy can substantially reduce rainfall kinetic energy and thus mask the importance of rainfall erosivity on soil erosion [29].Interception by understory vegetation can also reduce surface runoff and delay the generation of saturation overland flow, through which the relation between rainfall amount and soil erosion is disrupted [21].
During the erosion process, C is lost in the form of dissolved organic carbon (DOC), which is dissolved in runoff water, or particulate organic carbon (POC), which is bounded to sediment particles [30].Thus, the transport processes of POC and DOC are different.Studies have explored erosion-induced DOC loss in forest ecosystems [31,32], while studies on forest POC loss are relatively scarce.However, DOC typically accounts for a much smaller proportion of erosion-induced C loss compared with POC [33].Thus, a thorough understanding of erosion-induced C loss must take POC loss into consideration.
In this study, we monitored runoff and soil and C loss of each erosion event in a broad-leaved forest and a coniferous forest, with contrasting overstory and understory vegetation cover for one full year.The broad-leaved forest had a more closed canopy (Figure 1a,b; Table 1), and the coniferous forest had much denser understory vegetation cover (Figure 1c,d; Table 1).Using this unique setting, we aimed to explore whether the controlling factors of soil and C losses were different between different types of forest and if the differences were related to their difference in overstory and understory plant cover.Specifically, we addressed the following questions.First, were erosion-induced soil and C losses lower in the forest with dense understory compared to that with scarce understory?Second, were erosion-induced soil and C losses in the two forests driven by different rainfall characteristics?

Study Site
The study was conducted in the Fujian Sanming Forest Ecosystem National Observation and Research Station, in southeastern China (26°19′ N, 117°36′ E).This region has a typical maritime subtropical monsoon climate, with an annual mean temperature of 20.1 °C and mean annual precipitation of 1550 mm in 1960-2011 (National Meteorological Information Center, China Meteorological Administration).The geomorphology of the area is characterized by low mountains and hills, with an average elevation of ~300 m and slope   .The soils developed from biotite granite, with a thickness often exceeding 1 m, can be classified as sandy clay Ferric Acrisol according to the Food and Agriculture Organization classification system, equivalent to Hapludults of the United States Department of Agriculture Soil Taxonomy [34].
The broad-leaved forest is a secondary evergreen forest dominated by Castanopsis carlessi, (Hemsl.)Hayata, C astanopsis fissa (Champ.ex Benth.)Rehder & E.H.Wilson and Schima superba Gardn.& Champ.It is a naturally regenerated forest following clearcutting of a natural forest in 1976.At the time of this study (January 2014), the tree density was approximately 3700 stems per hectare, the canopy was rather closed, and the understory vegetation was scarce, mainly consisting of Gahnia tristis Nees (Table 1; Figure 1a).The coniferous forest is a monoculture Chinese-fir (Cunninghamia lanceolata (Lamb.)Hook.) plantation established after clear cutting the original natural forest in 1976.The planting density was 2192 stems per hectare (Table 1), and the understory vegetation is dominated by a fern species, Dicranopteris pedata (Houtt.)Nakaike (Figure 1d).The two forests have similar slope steepness and aspect and are less than one kilometer apart.
Three 20 m by 5 m runoff plots were established in each of the two forests.The outlet of each runoff plot was connected to a stainless runoff tank which was sheltered with a roof to prevent rain from falling into the tank.

Erosion Measurement and Sample Analyses
Between January and December 2014, water depth in the runoff tank was measured and recorded after each rainfall event.The water volume of each rainfall event was calculated by multiplying water depth by the tank area.After measuring water depth, the water and soil inside each tank were stirred vigorously to thoroughly mix the water and soil.Then, we collected two 1.5 L samples of the water-soil mixture using polyethylene bottles.One sample was used to quantify the soil content.The soil loss of each erosion event was calculated by multiplying the soil content by the runoff volume.The other sample was used for analysis of DOC and POC.The DOC and POC of each mixture sample were separated with a 0.45 µm microporous membrane.The DOC concentration was assayed using a TOC-VCPH/CPN analyzer (Shimadzu Corporation, Kyoto, Japan), and the POC concentration was assayed with an elemental analyzer (Elementar Vario EL III, Elementar, Langenselbold, Germany).The TOC content is the summation of the DOC and POC.

Stand Characteristics and Soil Physicochemical Properties Measurement
Following a commonly used forest plant survey method [35,36], three 20 m × 20 m quadrats were established in each of the forests.The use of using 20 m × 20 m quadrats is common in forest surveys.The slope of each quadrat was determined using a compass (DQL-12z, Harbin Optical Instrument Factory, Harbin, China).Trees with a diameter at breast height (DBH) larger than 4 cm within each quadrat were recorded to estimate the tree density.The aboveground biomass of all trees with a DBH > 4 cm in each quadrat was estimated with the allometric equations established by Lin et al. [37].
The leaf area index (LAI) of the overstory canopy of each runoff plot was measured using Plant Canopy Analyzer (LAI-2000; Li-Cor, Inc., Lincolin, NE, USA) in October 2013.Within each runoff plot, three 1 m × 1 m quadrats were randomly located and all understory plants inside were harvested and oven-dried at 65 • C to constant weight to estimate understory vegetation biomass.The bulk soil density of 0-10 cm soil of each runoff plot was determined with a known volume (100 cm 3 ) stainless steel cylinder corer.Three random soil columns were sampled for each runoff plot for determination of the bulk soil density.

Rainfall Characteristics Calculation
Rainfall was recorded using an automatic rainfall gauge (Rain root, RR1008, Beijing, China) installed in an open area 50 m apart from the broad-leaved forest.A time interval of six-hour rainless period was used to separate rainfall into different events [38].Average rainfall intensity and maximum 5-min intensity (I 5 ) of each rainfall event were calculated with the following equations: Average rainfall intensity = Rain f all amount Rain f all duration (1) where R 1 , R 2 , . .., R i refers to rainfall amount of the first, second, . .., ith 5-min interval of the rainfall event.
Rainfall erosivity was calculated using the Revised Universal Soil Loss Equation: Rainfall erosivity = E × I 30 (3) where E and I 30 are kinetic energy and maximum 30-min rainfall intensity of a rainfall event.E was calculated with the following equation: where P r is the rainfall amount for the rth increment of a storm hyetograph, which is divided into n parts, e r is the rainfall energy of each increment of the storm, which is determined with the following equation: where i r is the average rainfall intensity for the r th increment.

Statistical Analysis
Incident runoff, soil, DOC loss, POC loss, and total C loss of the two forests were compared with linear mixed-effects models where the sampling date was treated as a random factor.An independent t-test was used to examine the difference of stand density, slope, LAI, aboveground tree biomass, understory vegetation biomass and bulk soil density between the two forests.
We employed linear regression and the Lindeman, Merenda and Gold method to assess the relative importance of rainfall amount, average rainfall intensity, I 5 and EI 30 in predicting incident runoff, soil, DOC, POC and total C losses.Multi-collinearity among variables was examined using the tolerance and variance inflation factor (VIF).All factors with a VIF greater than 5 were gradually removed starting from the factor with the greatest VIF [39,40].None of the four factors had a VIF greater than 5 (rainfall amount 2.24, EI 30 1.13, rainfall intensity 1.94 and I 5 3.36) so that they were all kept in the models.All analysis was performed in R 4.1.3(R Core Team, 2022).

Results
Compared to the coniferous forest, the broad-leaved forest had significantly higher stand density, LAI, and aboveground tree biomass and lower understory vegetation biomass (Table 1).The two forests were similar in stand age, slope and soil bulk density (Table 1).

Erosion Induced C Loss
The mean DOC, POC and total C losses were all significantly higher in the broadleaved forest than the coniferous forest (Figure 4).The mean DOC loss in the broad-leaved forest (0.045 ± 0.003 kg ha −1 event −1 ) was more than twice that in the coniferous forest (0.020 ± 0.002 kg ha event −1 ).The mean POC loss in the broad-leaved forest (0.118 ± 0.016 kg ha −1 event −1 ) was about 97% higher than that in the coniferous forest (0.060 ± 0.009 kg ha −1 event −1 ).Mean total C loss of the broad-leaved forest (0.163 ± 0.017 kg ha −1 event −1 ) was twice that of the coniferous forest (0.081 ± 0.010 kg ha −1 event −1 ).
Thirty-seven out of 52 events in the broad-leaved forest and 33 out of 43 events in the coniferous forest had a DOC to TOC ratio less than 0.5 (Figure 5).In other words, more C was lost in the form of POC than in the form of DOC for more than 70% of the erosion events.Moreover, the ratio of DOC to TOC was not significantly different between the two forests.

Erosion Induced C Loss
The mean DOC, POC and total C losses were all significantly higher in the broadleaved forest than the coniferous forest (Figure 4).The mean DOC loss in the broad-leaved forest (0.045 ± 0.003 kg ha −1 event −1 ) was more than twice that in the coniferous forest (0.020 ± 0.002 kg ha event −1 ).The mean POC loss in the broad-leaved forest (0.118 ± 0.016 kg ha −1 event −1 ) was about 97% higher than that in the coniferous forest (0.060 ± 0.009 kg ha −1 event −1 ).Mean total C loss of the broad-leaved forest (0.163 ± 0.017 kg ha −1 event −1 ) was twice that of the coniferous forest (0.081 ± 0.010 kg ha −1 event −1 ).

Relative Contributions of Different Rainfall Characteristics to Variation of Runoff, Soil, DOC, POC and TOC Loss as Well as DOC to TOC Ratio
Rainfall amount was the most important predictor of the variation of runoff and soil, DOC and TOC losses in the broad-leaved forest (Figure 6, Table 2).In the coniferous forest, rainfall amount was also the most important predictor of the variation of runoff and DOC loss, while EI 30 was the most important predictor of the variation of soil, POC and TOC losses (Figure 6, Table 2).Average rainfall intensity was the least important predictor explaining the variation of runoff and DOC, POC and TOC losses across the two forests, except for the runoff and DOC loss in the coniferous forest (Figure 6, Table 2).Average rainfall intensity and EI 30 were the most important predictors of the variation of the ratio of DOC loss to TOC loss in the broad-leaved forest and coniferous forest, respectively (Figure 6, Table 2).Thirty-seven out of 52 events in the broad-leaved forest and 33 out of 43 events in the coniferous forest had a DOC to TOC ratio less than 0.5 (Figure 5).In other words, more C was lost in the form of POC than in the form of DOC for more than 70% of the erosion events.Moreover, the ratio of DOC to TOC was not significantly different between the two forests.Rainfall amount was the most important predictor of the variation of runoff and soil, DOC and TOC losses in the broad-leaved forest (Figure 6, Table 2).In the coniferous forest, rainfall amount was also the most important predictor of the variation of runoff and DOC loss, while EI30 was the most important predictor of the variation of soil, POC and TOC losses (Figure 6, Table 2).Average rainfall intensity was the least important predictor explaining the variation of runoff and DOC, POC and TOC losses across the two forests, except for the runoff and DOC loss in the coniferous forest (Figure 6, Table 2).Average rainfall intensity and EI30 were the most important predictors of the variation of the ratio of DOC loss to TOC loss in the broad-leaved forest and coniferous forest, respectively (Figure 6, Table 2).The greater runoff and soil, DOC, POC and TOC losses in the broad-leaved forest cover is not the only factor contributing to the differences in erosion between the two forests.Higher overstory LAI indicates greater overstory canopy interception potential, which would contribute to smaller runoff [41,42].However, water interception occurs not only in the overstory but also in the understory [43].Although LAI measured above the understory vegetation of the broad-leaved forest (4.1) was approximately 1.4 times of the coniferous forest (2.9), the understory vegetation biomass of the coniferous forest (12.1 t ha −1 ) was more than three times the level of the broad-leaved forest (3.7 t ha −1 ) (Table 1).Thus, understory interception of throughfall was likely much greater in the coniferous forest than the broad-leaved forest, offsetting the effect of greater overstory interception on runoff generation and the subsequent erosion-induced C loss.The annual soil loss in the broadleaved and coniferous forest was 58.2 and 34.3 kg ha −1 , respectively.Both are much lower than the losses in other mature plantation forests (0.7-1.5 t ha −1 ) in Southern China, which have a canopy LAI ranging from 1.0 to 2.5, but they did not take understory vegetation cover into consideration [44].This further illustrates that both overstory and understory should be considered when evaluating vegetation effects on erosion.Additionally, the difference in leaf form between the two forests probably also played a role.On the one hand, canopy interception decreases runoff which may contribute to reduce soil erosion.On the other hand, forest canopy accumulates rainwater on the leaf surface and transforms smaller raindrops into bigger water drops that have higher kinetic energy and erosivity than that of smaller raindrops [45,46].This effect of forest canopy on increasing throughfall erosivity has been shown to be more prominent in broader and flatter leaves that can accumulate more rainwater in the form of large water drops than smaller needle leaves [47][48][49].Two studies conducted in banana and rubber forests showed that the kinetic energy of throughfall raindrops was 2 to 36 times more than that of raindrops in the open area [50,51].Thus, the greater erosivity associated with larger water drops of the broad-leaved forest likely also contributed to its greater soil and C losses than that of the coniferous forest.

Drivers of Runoff, Soil, DOC, POC and TOC Losses
Numerous studies have shown that runoff and soil loss are closely related with rainfall amount [52][53][54][55][56].There is also evidence showing that runoff and soil loss are regulated by different rainfall characteristics [57].In our study, variation of runoff and DOC losses were mostly explained by rainfall amount, whereas variation of soil, POC and TOC losses was more explained by EI 30 than rainfall amount in the coniferous forest (Figure 6h,j,k).A study in woodlands in New Mexico reported that soil loss was more correlated with EI 30 than with rainfall amount [58].In addition to confirming the role of EI 30 in predicting soil loss, our study highlights that EI 30 is also important in predicting POC and TOC losses.Moreover, our study shows that DOC and POC losses were mainly regulated by different rainfall characteristics.
On hilly slopes, the saturation overland flow is the dominant runoff generation mechanism [59], in which enough rainfall is needed to saturate the soil pores before overland runoff is generated.This runoff generation mechanism may contribute to the greater explanation power of rainfall amount than other rainfall characteristics for runoff variation in our study.Different from runoff generation, soil loss amount depends mainly on the energy of raindrop splashing and runoff scouring [60][61][62], which were both closely associated with EI 30 [63][64][65].This may explain the dominant role of EI 30 in predicting soil and POC losses in the coniferous forest.However, EI 30 was not the most important predictor of soil and POC loss in the broad-leaved forest; the dense understory vegetation in the coniferous forest likely also played an important role in mediating the relation between rainfall characteristics and erosion (but see Tang et al. [28]).Because of the high soil trapping capacity associated with dense understory vegetation [21,66], transporting the same amount of soil in the coniferous forest required higher runoff velocity (determined mostly by EI 30 ) than in the broad-leaved forest.As a result, the best predictor of soil loss was different between the broad-leaved forest (rainfall amount) and the coniferous forest (EI 30 ).The result illustrates that key factors regulating POC and TOC losses are different between different forest types.
The minor role of rainfall intensity in regulating runoff, soil, DOC, POC and TOC losses in our study (Figure 6; Table 2) was different from results of a large number of rainfall simulation experiments which showed that rainfall intensity was the best predictor of soil erosion [67][68][69][70].The effects of rainfall intensity on soil erosion caused by simulated rainfall may be different from those caused by natural rainfall [71].In most simulations, the rainfall intensity over the entire period of each simulated rainfall event is constant, and the rainfall amount of all simulated rainfall events with different intensities is the same [72].On the other hand, natural rainfall events generally have intervals with different intensities and even some rainless intervals [38].Moreover, two natural rainfall events with similar intensity may have substantially different rainfall amounts.Short-duration but high-intensity rainfall events are common in Southern China [73].The short-duration and high-intensity rainfall events may not generate substantial erosion because the overall rainfall amount is too small to oversaturate the soil pore, especially when the antecedent soil moisture is low.For instance, in our study, the second largest rainfall intensity was 15.4 mm h −1 (Figure 2b), but the event had a total rainfall amount of only 21.8 mm rainfall within 1.4 h, and the event only yielded 0.47 kg ha −1 soil loss in the broad-leaved and 0.30 kg ha −1 in the coniferous forest (Figure 3b).On the contrary, the highest soil loss in the broad-leaved forest and the second highest in the coniferous forest were caused by an event with relatively low intensity but large rainfall amount (1.40 mm h −1 , 66.5 mm rain within 44.5 h).A long-term erosion monitoring experiment in the Mediterranean also demonstrated that event-based soil loss under natural rainfall is not closely related to rainfall intensity [52].Thus, while simulated rainfall experiments help to provide insights into erosion generation processes, the results may not be directly applicable to predict the soil and C losses associated with natural rainfall events.

Limitations and Implications of This Study
Although our results clearly showed that the loss of DOC and POC in the broad-leaved and coniferous forest were driven by distinct rainfall characteristics, it has limitations.One limitation of our study is the lack of a thorough examination of the potential impact of soil properties on surface runoff and soil and C losses.The two forests were converted from the same natural broad-leaved forests so that the soil properties should be similar between the two forests.However, some soil properties such as pH could still be different and taking them into consideration would provide more complete information on the factors leading to the differences in soil and C losses between the two forests.Another major limitation of our study is that we had only one forest for each forest type.However, our study is among the first to examine the role of forest structure/type in mediating the role of different rainfall characteristics in regulating runoff and soil and C losses.Coniferous plantations are replacing natural broad-leaved forests in Southern China and many parts of the world [74][75][76][77][78][79][80].Our result illustrates that such alternation in forest type/structure could have major implications for the forest hydrological cycle and C budget.Thus, we urge more studies to examine how such forest replacement may affect soil and C losses.
Soil erosion models are increasingly used to predict future soil erosion, with the Universal Soil Loss Equation (USLE) or revised USLE being most commonly used to model soil erosion in forest ecosystems [81][82][83][84][85].To model regional-scale soil erosion in forest ecosystems, the crop management C factor is generally estimated based on the normalized difference vegetation index (NDVI) [80,86], which is mostly determined by overstory vegetation [81,87].However, our study indicates that overstory cover alone cannot explain the differences in soil erosion between the two forests.Thus, we call for attention to be paid to the largely neglected understory vegetation in soil erosion modelling.Moreover, our results also suggest that different rainfall characteristics (i.e., rainfall amount or EI 30 ) should be used to predict future C loss in contrasting forest types.

Conclusions
This study illustrates that runoff, soil, DOC, POC and total C losses differ significantly in forests with contrasting overstory and understory structures.Possibly owing to the variations in understory biomass between the broad-leaved and conifer forest, their soil and total C losses are driven by two different rainfall characteristics, rainfall and EI 30 , respectively.Different driving factors of soil and C losses in contrasting forests imply that different rainfall characteristics should be used in the erosion models to predict future soil and C loss of forest types with contrasting overstory and understory vegetation cover.Moreover, our study also highlights the potential role of overlooked understory in not only controlling soil and C losses but also mediating their responses to rainfall characteristics.

Figure 1 .
Figure 1.Hemispherical canopy images (a,b) and understory vegetation (c,d) of the studied broadleaved forest (a, c) and coniferous forest (b, d).The broad-leaved forest has more closed overstory canopy but lower understory vegetation cover compared to the coniferous forest.

Figure 1 .
Figure 1.Hemispherical canopy images (a,b) and understory vegetation (c,d) of the studied broadleaved forest (a,c) and coniferous forest (b,d).The broad-leaved forest has more closed overstory canopy but lower understory vegetation cover compared to the coniferous forest.

Figure 3 .
Figure 3. Runoff (a) and soil loss (b) in the broad-leaved forest and the coniferous forest.The inserted half-violin plots show the differences between two forests.Sca er points with error bars are mean ± standard error.

Figure 3 .
Figure 3. Runoff (a) and soil loss (b) in the broad-leaved forest and the coniferous forest.The inserted half-violin plots show the differences between two forests.Scatter points with error bars are mean ± standard error.

Figure 4 .
Figure 4. Dissolved organic carbon (DOC) (a), particle organic carbon (POC) (b) and total C loss of each rainfall event (c) in the broad-leaved forest and the coniferous forest.The inserted half-violin plots show the differences between two forests.Sca er points with error bars are means ± standard errors.

Figure 4 .
Figure 4. Dissolved organic carbon (DOC) (a), particle organic carbon (POC) (b) and total C loss of each rainfall event (c) in the broad-leaved forest and the coniferous forest.The inserted half-violin plots show the differences between two forests.Scatter points with error bars are means ± standard errors.

Figure 5 .
Figure 5.The ratio of DOC to TOC loss of each event in the broad-leaved forest and the coniferous forest.The inserted half-violin plots show the differences between two forests.Sca er points with error bars are means ± standard errors.

3. 3 .
Relative Contributions of Different Rainfall Characteristics to Variation of Runoff, Soil, DOC, POC and TOC Loss as Well as DOC to TOC Ratio

Figure 5 . 17 Figure 6 .
Figure 5.The ratio of DOC to TOC loss of each event in the broad-leaved forest and the coniferous forest.The inserted half-violin plots show the differences between two forests.Scatter points with error bars are means ± standard errors.Forests 2024, 15, x FOR PEER REVIEW 11 of 17

Figure 6 .
Figure 6.The relative contributions of predictors, based on LMG scores, to model fit of the multiple linear regression models of the runoff (a,g), soil (b,h), DOC (c,i), POC (d,j) and TOC (e,k) losses as

Table 1 .
Stand characteristics and soil bulk density (mean ± standard error) of the broad-leaved forest and the coniferous forest.Numbers of the same row sharing no common le er are significantly different at p < 0.05.