The Effects of Different Tillage and Straw Return Practices on Soil Organic Carbon Dynamics from 1980 to 2022 in the Mollisol Region of Northeast China

Abstract

Understanding how conservation practices involving tillage and straw return practices affect the soil organic carbon (SOC) in farmland is important for soil carbon sequestration and climate change mitigation. However, limited studies have been conducted to investigate and compare the magnitude and variability of the main conservation practices concentrated in grain-producing regions. In this study, we evaluated the SOC response to the main practices (e.g., no tillage, reduced tillage, deep tillage, and straw return) in the Mollisol region of Northeast China based on collected field data (871 observations) using a combination of meta-analysis and random forest (RF) methods. The results show that the SOC change rate significantly increased from 1980 to 2022, with an average annual increase rate of 0.19–14.92%. Straw return had maximum effects on SOC of 17.44% when the soil pH > 7.5 and 15.22% when the initial SOC < 10 g kg⁻¹. The RF results indicate that the initial SOC is the most important factor for SOC, with relative importance values of 33.4%, 29.4%, 29.0%, and 34.1% for SOC under the four practices, respectively. These findings are essential for the implementation of conservation practices to improve carbon sequestration and grain production in eco-agricultural regions.

1. Introduction

Soil organic carbon (SOC) is an important indicator of soil health and plays a dominant role in the global carbon (C) cycle; it is receiving increasing attention due to its potential in removing greenhouse gases from the atmosphere [1,2]. As a growing number of studies have demonstrated, farm management practices play an important role in mitigating the threat of climate change and ensuring food security by sequestering C [3,4]. Additionally, it is known that grain production can be increased by 30% through improving farm management practices based on the global yield gap analysis results [5]. Therefore, to obtain effective and sustainable agricultural practices for successful SOC sequestration on farmland, it is essential to clarify the mechanisms of SOC dynamics with the changes in diverse management practices over time and space.

Agricultural practices are defined as the practices that increase the productivity of farmlands, including multiple tillage, straw return, fertilizer application, irrigation, and crop rotation regimes [6,7,8]. In recent years, conservation agriculture has been recognized as an eco-friendly approach to enhance SOC, such as different conservation tillage and straw return (SR) practices [9,10]. Conventional tillage (CT) is defined as any tillage activity that uses cultivation, including ploughing and harrowing, to prepare a seedbed [11]. No tillage (NT) is the extreme form of CT practice where seeds are sown in a no-tilled soil without incorporating crop residues [12]. Reduced tillage (RT) is another type of CT that does not turn the soil over, thus avoiding changes in the soil structure [11]. Deep tillage (DT) employs mechanical modification of the soil profile, such as subsoiling, deep ploughing, and mixing soil profiles, which can alleviate high subsoil compaction [13,14]. With the addition of straw, SR has been proven as an effective practice for increasing SOC in farmland [15].

Numerous studies have investigated the effects of various tillage and SR practices on SOC [8,16,17]. Meta-analysis is considered one of the most effective methods for synthesizing and analyzing results from many studies with the same objective to draw general conclusions [18,19,20]. However, some studies have revealed positive effects of management practices on SOC, whereas other studies have reported weak effects or opposing results, even for the same practice. For example, Mehra et al. [21] reported that NT and RT could improve SOC and soil quality, whereas Sauvadet et al. [22] reported that NT did not increase SOC compared with CT practice. Since SOC is subject to many factors, including the soil properties, local environment, and management practices, the effects of agricultural practices on SOC vary according to various climate conditions and soil attributes [20,23].

The complex association between SOC and management practices can lead to gains or losses in SOC [24,25]. Although a meta-analysis was applied to understand the effects of practices on SOC, the magnitude and variability of the combined effects of multiple factors were not sufficiently investigated. Machine learning has recently emerged as a powerful tool for deriving the associations and patterns from complex datasets since it can handle “big data” with greater flexibility compared with traditional statistical analysis methods [26,27]. Meanwhile, its independence regarding experimental designs and prior assumptions can overcome the limitations of meta-analysis methods. Therefore, the combination of meta-analysis and machine learning models can help to understand how tillage and SR practices affect soil C changes [28]. Although some studies also investigated the effects of management practices on SOC using the combined approach [17,29], few studies have revealed the hidden relationships between farmland SOC and conservation practices under the background of climate change and extensive cultivation over a long time scale. Furthermore, most recent studies focused on only one type of practice at the national or global scales, with limited attention given to investigating and comparing the main conservation practices concentrated in the separate ecological or grain-producing regions [16,30]. Understanding how tillage and straw return practices regulate the effects of environmental factors on SOC in an eco-agricultural region is critical for maximizing C accumulation and making management practice optimization more targeted in these regions.

As one of the four major Mollisol areas around the world, the Mollisol region of Northeast China has been praised as the “Corn Belt” of China, accounting for 41% of the national maize production [31]. However, traditional intensive cultivation has resulted in severe soil degradation, which threatens national food security. Since the extensive land reclamation in the 1950s, significant soil and water erosion has resulted in an average loss in organic matter from 40–60 to 20–30 g kg⁻¹ [32]. In recent years, conservation practices were applied to improve soil C storage, but the effect of different tillage and straw return practices on farmland SOC is not clear in this region. Specifically, direct comparisons between the effects of NT, RT, DT, and SR practices on SOC across different climatic conditions, soil properties, and management techniques have seldom been studied, especially in the Mollisol region of Northeast China. To address these previous limitations, this study clarifies the important factors impacting SOC and suggests optimal management practices for farmland SOC enhancement by using a combination of meta-analysis and a machine learning approach. Therefore, the main objectives of this study were to (1) quantify the SOC changes under different tillage and straw return practices during 1980–2022; (2) evaluate the effects of tillage (no tillage, reduced tillage, and deep tillage) and straw return practices on SOC with different climate conditions, soil properties, and practice durations; and (3) quantitatively identify the contributions of different explanatory factors on SOC in the Mollisol region of Northeast China by using random forest (RF) models.

2. Materials and Methods

2.1. Literature Review and Data Collection

To investigate the effects of tillage and straw return practices on SOC in the Mollisol region of Northeast China, relevant articles were searched for in the available databases, including Web of Science (https://www.webofscience.com/ (accessed on 8 November 2025)), Elsevier (http://www.sciencedirect.com/ (accessed on 8 November 2025)), and China National Knowledge Infrastructure (https://www.cnki.net/ (accessed on 8 November 2025)), with the key words “Northeast China”, and “black soil region” or “Mollisol region”, and “organic carbon” or “soil carbon”, and “straw” or “crop cover”, and “conservation tillage”, and “no tillage”, and “deep tillage”, and “reduced tillage”. The control and experimental treatments of tillage and straw return practices are listed in Table 1. These experiments needed to meet the following inclusion criteria: (1) all studies were conducted on the farmland of Northeast China; (2) provide the means and standard deviations of SOC content or provide three or more duplicates; and (3) provide the test information clearly, such as location, time, experiment replicates, and soil properties. Based on the above criteria, 100 articles with 871 paired observations for the period between 1980 and 2022 were used in this study, and groups of data were available for comparison and summary. The SOC changes in the control and treatment groups during 1980–2022 in the Mollisol region of Northeast China are shown in Figure 1.

Table 1. The control and experimental treatments of tillage and straw return.

Management Practices	Treatment	Tillage Depth and Straw Return	Abbreviation
Tillage	Control treatment	Conventional tillage (with 15–20 cm tillage depth)	CT
	Experimental treatment	No tillage (with 0 cm tillage depth)	NT
		Reduced tillage (with 5–10 cm tillage depth)	RT
		Deep tillage (with 25–40 cm tillage depth)	DT
Straw return	Control treatment	Without straw return	SR0
	Experimental treatment	With straw return	SR

Figure 1. Soil organic carbon (SOC) temporal changes (control + treatment) during 1980–2022 in the Mollisol region of Northeast China. The extremes of the box (×) are Q1 (quartile 1) and Q3 (quartile 3).

2.2. Data Preparation

For each prepared article, alongside the SOC, the following parameters related to the conservation practices and explanatory variables were identified: location (latitude and longitude), mean annual temperature (MAT), mean annual precipitation (MAP), soil pH, bulk density (BD), soil depth (SD), initial SOC, texture, management practices (no tillage (NT), reduced tillage (RT), deep tillage (DT), and straw return (SR)), and the duration. The initial SOC was the SOC amount in the experimental field before the experiment. If some key environmental information or soil properties were lacking in a study, other published articles were also searched to supplement the relevant information from the same site and similar years. Meanwhile, missing data could also be collected from the World Soil Database (Harmonized World Soil Database, HWSD) (https://www.fao.org/ (accessed on 8 November 2025)) and Meteorological Database (https://power.larc.nasa.gov/ (accessed on 8 November 2025)) based on the latitude and longitude information to improve the representativeness of the collected observations. We referenced the soil data of 1985 from HWSD dataset and the climate data during 1980–1990 from Meteorological database. Because most studies provided the soil properties and climate conditions on the experimental sites, the temporal matching error was almost not found. For each observation collected in this study, the mean and SD values of the SOC were obtained from tables or extracted from figures using GetData Graph Digitizer 2.26 software. If only the SE was provided in the article, it was converted to SD using

$$SD=SE\sqrt{n}$$

(1)

When neither SD nor SE was available in a study, the SD was assigned as 1/10th of the mean [33].

SOC content (g kg⁻¹) can be converted to soil organic carbon stocks (SOCS, Mg ha⁻¹) using [34]

$$SOCS=SOC\times BD\times H\times 0.1$$

(2)

where BD is the soil bulk density (g cm⁻³), H is the soil depth (cm), and 0.1 is the unit conversion factor. If soil BD was unavailable, it can be estimated by [35]:

$$BD=1.377exp(-0.0048SOC)$$

(3)

To reveal the SOC response to different conservation practices affected by the explanatory factors, the collected data were classified into different groups according to the data distribution characteristics of the explanatory factors and some previous studies [17,29,36]. All these factors were further divided into subcategories (Table 2). The statistical results of the collected SOC content in different groups of the explanatory variables are presented in Figure S1.

Table 2. Classification of explanatory factors.

Explanatory Factors	Groups
MAT (°C)	<3; 3–6; >6
MAP (mm)	<500; 500–600; >600
Soil pH	<6.5; 6.5–7.5; >7.5
BD (g cm−3)	<1.3; 1.3–1.5; >1.5
Soil depth (cm)	<20; 20–40
Initial SOC (g kg−1)	<10; 10–20; 20–30; >30
Soil texture	Sandy; loam; clay
Duration (y)	<3; 3–10; >10

In addition, the data used in the RF models needed to be compiled into consistent forms, namely SOC, MAT, MAP, soil pH, bulk density, soil depth, initial SOC, soil texture, and duration, for the above four conservation practices. Notably, soil texture was regarded as a category value, while the other variables were continuous data values.

2.3. Data Analysis

First, the C accumulation rate was calculated as follows:

$$∆{SOC}_T={\displaystyle \frac{{SOC}_T-{SOC}_{CK}}{Y}}$$

(4)

where ${SOC}_T$ and ${SOC}_{CK}$ are the SOCS with the conservation and control treatments, respectively, and $Y$ is the time after applying the management practice.

Previously, a standard meta-analysis method was conducted to evaluate the effect of conservation practices on soil organic carbon [37]. In this study, a random effects model was chosen to investigate the impact of conservation practices on SOC. The logarithm (lnRR) was used to represent the impact of farm management on the SOC change, and its effect size was calculated as follows:

$$lnRR=ln(X_t/X_c)$$

(5)

where $X_t$ and $X_c$ are the mean SOC values with and without farm management, respectively. The variance ($v$) of lnRR was calculated as follows:

$$v\left(lnRR\right)={\displaystyle \frac{S_t^2}{n_t{X}_t^2}}+{\displaystyle \frac{S_c^2}{{n_{c}X}_c^2}}$$

(6)

where $s_t$ and $s_c$ are the standard deviations for the treatment and control SOC values, respectively, and $n_t$ and $n_c$ are the sample sizes for the SOC treatment and control groups, respectively.

The overall response ratio of the SOC treatment and control was evaluated using a random effects model with the weight of each observation. The weight factor ($w$) and the mean effect size (${RR}^{*}$) were calculated as follows:

$$w_i={\displaystyle \frac{1}{v_i}}$$

(7)

$${RR}^{*}={\displaystyle \frac{{\sum }_i{w}_{i}ln{RR}_i}{{\sum }_i{w}_i}}$$

(8)

where ${RR}^{*}$ is the weighted effect size, and $i$ represents the $i$th observation.

The 95% CI was calculated using

$$95\%CI=lnRR\pm 1.96S\left(lnRR\right)$$

(9)

where $S\left(lnRR\right)$ is the SE of $lnRR$, which was calculated using

$$S\left(lnRR\right)=\sqrt{1/{\sum }_i{w}_i}$$

(10)

The meta-analysis was performed in MetaWin 3.08 software. A bootstrap approach with 999 iterations was employed to calculate the mean values and their 95% confidence intervals (CIs). If the 95% CI overlapped 0, the explanatory factor did not affect the SOC; if not, the explanatory factor had a significant effect on the SOC. Publication bias was assessed by using Egger’s regression test, which presented no significant publication bias (Table S1).

2.4. Machine Learning

The RF model is an integrated machine learning algorithm that was employed to predict the SOC under the different management practices, climatic variables, and soil properties in the analyzed studies. The advantages of an RF model include high accuracy, reduced overfitting, no feature normalization requirement, and good noise immunity [38]. The RF model contains three user-defined parameters, namely the number of trees (ntree), the number of variables used as predictors for each tree (mtry), and the minimum size in each terminal mode (node size) [39]. In this study, ntree was set to range from 500 to 2000 to train the model. The default value of mtry is generally set to one-third of the number of selected predictors. Node size was also tested with different values, i.e., 1, 2, 3, 4, and 5. During the regression process, several regression trees were built on the basis of a unique bootstrap sample, and the averaged trees were used to estimate the SOC. The model ensemble reduced the variance and mitigated the overfitting risk, which is common with individual decision trees. In addition, the RF model could calculate the relative importance of each explainable factor (e.g., conservation practices, climatic variables, and soil properties) and reveal the underlying results between the predictors and explainable factors.

Here, an RF model was used to quantify the relative importance of the explanatory factors on the SOC dynamics collected from the published articles. Several explanatory factors, namely, MAT, MAP, soil pH, bulk density, soil depth, initial SOC, soil texture, and duration, were selected to establish the RF models for each conservation practice. For RF model validation, we split the SOC sampling sites into ten blocks of equal size across the entire study region. Then, 10-fold cross-validation was employed to evaluate the RF model performance, and the predictive performance of the model was assessed using the determination coefficient (R²) and root mean square error (RMSE). The RF model was implemented using the randomForest package in R 4.5.0 software (https://www.r-project.org/ (accessed on 8 November 2025)).

3. Results

3.1. SOC Changes and Accumulation Rate During 1980–2022

NT, RT, DT, and SR practices generally increased the SOC and accumulation rate in the Mollisol region of Northeast China during 1980–2022 (Table 3). The increased SOC under SR was higher than that with different tillage measures. For the 0–20 cm soil depth, SR increased the SOC by 14.92 ± 1.42% compared with SR0. Compared with CT, the increases in SOC in the NT, RT, and DT systems were 8.27 ± 1.13%, 7.89 ± 1.85%, and 4.07 ± 1.57% at the 0–20 cm depth, respectively. The SOC accumulation rate changes were not as similar as the SOC changes in response to different practices at the 0–20 and 20–40 cm depths. The greatest increase of 2.19 ± 0.32 Mg ha⁻¹ yr⁻¹ was observed when RT was applied at the 20–40 cm depth, followed by an increase of 1.95 ± 0.09 Mg ha⁻¹ yr⁻¹ when SR was practiced at the 0–20 cm depth. The increase in SOC accumulation rate presented no obvious trends between the 0–20 and 20–40 cm depths.

Table 3. Average changes in SOC and accumulation rates under different conservation practices at 0–20 and 20–40 cm soil depths in Northeast China.

Conservation Practices	Soil Depth (cm)	SOC Change (Mg ha−1)			SOC Accumulation Rate (Mg ha−1 yr−1)
		n	Mean (%)	SE	n	Mean	SE
CT to NT	0–20	154	8.27	1.13	154	0.77	0.07
	20–40	35	0.19	2.31	35	0.03	0.04
CT to RT	0–20	50	7.89	1.85	50	1.13	0.13
	20–40	20	9.13	4.25	20	2.19	0.32
CT to DT	0–20	61	4.07	1.57	61	0.85	0.15
	20–40	24	7.56	2.21	24	1.87	0.28
SR0 to SR	0–20	380	14.92	1.42	380	1.95	0.09
	20–40	117	11.25	1.25	117	1.07	0.06

Note: CT, conventional tillage; NT, no tillage; RT, reduced tillage; DT, deep tillage; SR0, no straw retained; SR, straw return. n represents the pairs of mean and standard deviation values for the data samples; SE, data standard error.

3.2. Responses of SOC to Management Practices

3.2.1. Tillage Management

The SOC responses to its explanatory factors under NT are presented in Figure 2a. The significantly positive responses of SOC to NT were observed for the MAT, MAP (500–600 mm and >600 mm), pH (6.5–7.5), BD (<1.3 g cm⁻³ and >1.5 g cm⁻³), SD, initial SOC (<10 g kg⁻¹ and 10–20 g kg⁻¹), and texture (sandy). The largest increase in SOC reached 13.90% in the BD (>1.5 g cm⁻³), followed by the 13.82% SOC increase in the SD (0–20 cm). The increase in SOC showed no increasing trend with the increase in the MAT, while the larger increase reached 9.29% in the MAT (<3 °C) group. No distinct variations were exhibited between the MAP groups in the response of the SOC to NT, where the effect size ranged from 6.10% (500–600 mm) to 7.70% (>600 mm). Regarding the soil properties, significant increases in the SOC were observed in the initial SOC (6.78–7.75%), SD (12.46–13.82%), BD (8.60–13.90%), pH (7.50%), and soil texture (8.30%) groups.

Figure 2. Responses of the SOC to explanatory variables under different tillage management practices: (a) no tillage (NT), (b) reduced tillage (RT), and (c) deep tillage (DT). The numbers in the graph represent the number of comparisons (control + treatment). The points and error bars represent the mean and 95% confidence intervals. Abbreviations: SOC, soil organic carbon; MAT, mean annual temperature (°C); MAP, mean annual precipitation (mm); BD, bulk density (g cm⁻³); SD, soil depth (cm); initial SOC, initial soil organic carbon (g kg⁻¹). * represents significant differences at p < 0.05.

The SOC responses to its explanatory factors under RT are presented in Figure 2b. Significant positive responses of SOC to RT were observed for the MAT, MAP (<500 mm and 500–600 mm), pH (6.5–7.5 and >7.5), BD (<1.3 g cm⁻³ and 1.3–1.5 g cm⁻³), SD, initial SOC, texture, and duration (<3 y and 3–10 y). The largest increase in SOC reached 15.20% in the soil texture (clay) group, followed by 15.17% in the soil BD (<1.3 g cm⁻³) group. The pH showed no distinct variations in different groups for the SOC response to RT, and the effect sizes were 11.32% and 12.52% in the 6.5–7.5 and >7.5 groups, respectively. The SOC significantly increased by 13.92% in the <3 °C group, while it increased by 11.84% and 9.36% in the MAP 500–600 mm and <500 mm groups, respectively.

After RT, the increase in SOC gradually decreased with the increase in the initial SOC, where it increased by 14.95%, 10.97%, and 10.35% in the <10, 10–20, and >20 g kg⁻¹ groups, respectively. In addition, a positive relationship was observed between the soil depth and the SOC response to RT, with the effect sizes of 13.49% and 10.71% in the 0–20 and 20–40 cm groups, respectively.

The SOC responses to its explanatory variables under DT are presented in Figure 2c. The significantly positive responses of SOC to DT were observed for the MAT (<3 and 3–6 °C), MAP (<500 and 500–600 mm), pH (6.5–7.5), BD, SD, initial SOC, texture (loam), and duration. Regarding the climatic factors, the increase in SOC increased with the increase in the MAT, where it increased by 4.87% and 11.18% in the <3 and 3–6 °C groups, respectively. Meanwhile, there was little variation between the MAP groups in the SOC responses to DT, with effect sizes of 8.62% and 9.59% in the <500 and 500–600 mm groups, respectively. For the soil properties, the largest increase in the SOC reached 13.13% in the initial SOC (<10 g kg⁻¹) group, followed by 11.27% in the soil BD (>1.5 g cm⁻³) group. After DT, the higher increase in SOC was found in the 3–10 y duration group, with an effect size of 10.28%.

3.2.2. Straw Return

The SOC’s responses to its explanatory factors under SR are shown in Figure 3. Except for the MAT (<3 °C) group, a significantly positive relationship was identified between all the other explanatory variables and the SOC response to SR management. The largest increase in the SOC reached 17.44% in the soil pH (>7.5) group, followed by 15.22% in the initial SOC (<10 g kg⁻¹) group, which were the greatest value increases across all the conservation practices in this study. Climatic factors, i.e., MAT and MAP, indicated positive increased effects on the SOC response to SR with the increase in the two factors. For instance, the increase in SOC increased from 10.57% to 13.92% in the <500 and 500–600 mm MAP groups, respectively. Meanwhile, the increase in SOC increased with the increase in the soil pH values, where the effect sizes were 11.73%, 11.98%, and 17.44% in the <6.5, 6.5–7.5, and >7.5 groups, respectively. After SR, the larger increase in SOC reached 11.82% in the BD (1.3–1.5 g cm⁻³) group, 13.64% and 12.49% in the soil depth (0–20 cm and 20–40 cm) groups, and 11.76% in the soil texture (loam) group. Regarding the duration, the largest increase in the SOC showed no increased trend with the increase in the duration, while it reached 11.17% in the <3 y group, followed by 10.83% and 10.75% in the >10 y and 3–10 y groups, respectively.

Figure 3. Responses of the SOC to the explanatory variables under different straw return management practices. The numbers in the graph represent the number of comparisons (control + treatment). The points and error bars represent the mean and 95% confidence intervals. Abbreviations: SOC, soil organic carbon; MAT, mean annual temperature (°C); MAP, mean annual precipitation (mm); BD, bulk density (g cm⁻³); SD, soil depth (cm); initial SOC, initial soil organic carbon (g kg⁻¹). * represents significant differences at p < 0.05.

3.3. SOC Prediction Using RF Model

Eight explanatory factors, namely, MAT, MAP, soil pH, BD, soil depth, initial SOC, texture, and duration, were selected to establish the RF models for SOC prediction. Figure 4 shows that the fit coefficients between the observed and predicted SOC for the test dataset were 0.76, 0.85, 0.93, and 0.87, and the corresponding RMSEs were 2.61, 2.16, 2.07, and 2.23 g kg⁻¹ for NT, RT, DT, and SR, respectively. The scatter plots were all distributed around the 1:1 line for the four practices, which indicates that the RF models had a high prediction accuracy and good performance.

Figure 4. Scatter plots of the predicted SOC versus observed SOC using the RF model under different management practices: (a) no tillage (NT), (b) reduced tillage (RT), (c) deep tillage (DT), and (d) straw return (SR). The dashed lines represent the fitted curves, and the shaded area represents the 95% confidence intervals. The solid line represents the 1:1 line.

3.4. Relative Importance of Explanatory Factors

The relative importance of each explanatory variable for the SOC increases in response to different practices is presented in Figure 5. The initial SOC was identified as the most important factor for the SOC, with the relative importance values of 33.4%, 29.4%, 29.0%, and 34.1% under NT, RT, DT, and SR, respectively. The MAT also had an important impact on the SOC under the four practices, where the importance scores ranged from 12.2% to 14.9%. In addition, the BD significantly enhanced the SOC under NT and SR, with importance scores of 14.4% and 14.2%, respectively. The soil texture had the smallest impact on the SOC across the different practices, with scores of 4.0%, 5.5%, 4.0%, and 5.3% for NT, RT, DT, and SR, respectively.

Figure 5. Relative importance of the explanatory factors under the different management practices: (a) no tillage (NT), (b) reduced tillage (RT), (c) deep tillage (DT), and (d) straw return (SR).

4. Discussion

4.1. Changes in SOC Under Different Practices

The overall SOC changes in response to different practices ranged from 4.07% to 14.92% generally (Table 3). This result was similar to previous studies, which showed the ranges of 5–19% with tillage practices [40,41] and 11–16% with straw return practices [18,42]. The largest SOC change was observed in SR (14.92%) at the 0–20 cm depth, followed by SR (11.25%) at the 20–40 cm depth. This is because straw cover contributes to biomass production and microbial activities, which can promote soil aggregation and enhance the water-holding capacity, and thus, it has a significant effect on the soil C input [43]. The maximum SOC accumulation rate was observed in RT (2.19 Mg ha⁻¹ yr⁻¹) at the 20–40 cm depth, followed by SR (1.95 Mg ha⁻¹ yr⁻¹) at the 0–20 cm depth and DT (1.87 Mg ha⁻¹ yr⁻¹) at the 20–40 cm depth. It was indicated that RT and DT have a great potential to improve the SOC accumulation rate at deeper soil depths. The results also show that NT enhanced the SOC by 8.27%, followed by RT (7.89%) and DT (4.07%) at the 0–20 cm depth. Conservation tillage measures, such as NT and RT, can reduce soil disturbance and increase the straw C input, and thus, enhance SOC accumulation [44,45]. In addition, DT showed a higher potential to increase the SOC at the 20–40 cm depth than at the 0–20 cm depth; this is because DT tills deeper, burying the straw in the topsoil into the 20–40 cm deep layer. Meanwhile, DT can facilitate root development and increase the root-derived C input into the deep soil layer, thus enhancing the C storage relative to the topsoil C [46].

4.2. Effect of Climatic Factors on SOC

Climatic factors, such as the MAT and MAP, can influence SOC accumulation and storage through affecting the decomposition rate of organic matter, biomass production, and microbial activities [47]. The Mollisol region of Northeast China is mostly located in high-latitude areas, where the heat resource is the main constraint factor for soil C accumulation. This study’s results demonstrate that the climatic factors had a greater effect on the increase in SOC under SR than under the tillage practices. This result is consistent with previous studies, such as Liu et al. [48] and Qin et al. [49]. With an increase in air temperature and precipitation, the microbial decomposition of the straw cover increases and enhances SOC accumulation; meanwhile, high rainfall also promotes crop residue C input into the soil. Compared with other explanatory factors (e.g., soil properties and duration), the climatic factors did not show a distinct advantage regarding enhancing SOC under the tillage and straw return practices since the climatic conditions vary differently across the Mollisol region of Northeast China and the increase in SOC also depends on the balance of water and heat resources. The temperature and precipitation in such high-latitude regions generally become the constraint factors for soil C accumulation in the study area.

4.3. Effect of Soil Properties on SOC

The soil properties, e.g., soil pH, BD, soil depth, initial SOC, and texture, all affected the SOC under different tillage and straw return practices. Regarding SR, all soil properties across the different groups showed significantly positive effects that increased the SOC. Therefore, it is suggested that SR is an effective conservation practice that is suitable for most soil conditions in the Mollisol region of Northeast China. The results regarding the SOC responses to soil properties reveal that the maximum SOC improvement occurred when the soil pH was >7.5, followed by when the initial SOC was <10 g kg⁻¹, where the SOC increased by 17.44% and 15.22%, respectively. These findings are consistent with those of previous studies [17,50]. Microbial activity may be restricted by acidic soil conditions, which could influence the decomposition of organic matter, thus leading to a reduced increase in the SOC of acidic soils [51]. Denoncourt et al. [52] also reported that microorganisms have better C use efficiency in alkaline soils, leading to a greater increase in SOC with higher soil pH values [53,54]. Lower BD values promote water infiltration, aeration, and root development, all of which will support microbial activity and SOC accumulation [55]. This could explain our results from using RT and SR practices, in which lower BD had a greater effect on SOC enhancement than higher BD. After the application of DT practice, the topsoil SOC can be leached into deep layers of soil, and meanwhile, soil BD can be changed by DT [46]. SOC accumulation would be regulated by the BD after DT, not the initial BD. Thus, lower BD did not show a greater effect on SOC than high BD under DT here, due to the initial BD values employed in our study.

Soils with fine textures, such as clay, generally store more SOC; soils with coarse textures, such as sand, generally store less SOC [56,57]. The relationships between the soil texture and organic carbon are variable and complicated since they depend on several factors, such as farm management, soil type, and climate [58]. Our meta-analysis revealed that the SOC response to soil texture differed under the tillage and straw return practices. For example, NT practice could be recommended for improving soil C sequestration in sandy soil with low initial SOC contents. SR was more effective at enhancing the SOC in loam soils, and it had a lower effect on SOC accumulation in the clay group. This was consistent with the results of Wang [59], which indicated that SOC increased more in sand or loam soil conditions. That was because the physicochemical properties of sand and loam soil can be changed easily after the application of SR practice. There is low initial SOC in such soils, and they hold more carbon and have significant potential to reach SOC saturation. Soil depth also has different effects on SOC accumulation under various tillage practices. In our work, compared with CT, NT and RT had significantly positive effects on the SOC accumulation, especially at topsoil depths. This finding is consistent with previous studies [60,61], which indicate that the SOC content decreased with increasing soil depth under NT/RT practices. This is because NT and RT offer no or less disturbance in the surface of soil compared with CT and provide more opportunity to store C and reduce C mineralization to the atmosphere [62].

Based on the RF model results, the relative importance of the initial SOC on the increase in SOC under different tillage and straw return practices was the greatest across all the explanatory factors. Furthermore, all the tillage and straw return practices had greater SOC enhancement with a low initial SOC than with a high initial SOC in our work. This result is consistent with Georgiou and Wang [50,59]. It is easy to understand that soil with a low initial SOC has more increasing potential to change to a high-SOC soil compared with soil with a higher initial SOC under the same management practices.

4.4. Effect of Duration on SOC

The duration of the management practices demonstrated various effects on the soil C accumulation in this study. Previous studies stated that the soil properties require several decades to respond to management practices [63,64]. However, in our study, it can be observed that the management practices had a positive effect on the SOC within a much shorter period, with an increase in SOC of 11.22% under RT and 10.28% under DT, both of which were within the short term (3–10 y). Our results are similar to previous studies [65,66], in which the results demonstrated that management practices would take 3–7 years to achieve the soil C changes. This is because SOC accumulation is a dynamic process with new C input and old C output, and organic matter decomposition and SOC mineralization both require several years to produce any noticeable changes, especially in high-latitude regions. However, our meta-analysis found that the responses of SOC to straw return were all significant for the <3 y, 3–10 y, and >10 y periods, and there was no distinct difference across the durations. This is consistent with the findings of Xin et al. [17]. Although a process is needed to decompose the crop residue, SOC enhancement would not be obtained with additional years of SR. This phenomenon can be explained by SOC saturation and the favorable environment for pests or pathogens after long-term SR, thus decreasing the soil fertility [67,68]. In addition, soil erosion caused by high-intensity agricultural practices and extreme weather conditions is also considered a primary factor affecting soil C storage and soil degradation in Northeast China [69,70]. The straw cover can help to protect the soil from water and wind erosion in the initial periods of straw return implementation [71,72]. This explains the significantly positive effect of straw return on the SOC in such a short period (<3 y) in our results.

In this meta-analysis, the collection of studies on the effect of tillage and straw return practices on farmland SOC was hindered by the lack of reported data, wide variability in management patterns, and the range of environmental variables. For example, the straw return depth was not introduced here, which may lead to uncertainty in the SR results across many environmental conditions. Meanwhile, the interaction between tillage and SR practices also needs to be explored in future work. Another potential limitation of our study is the assumption of 1/10th of the mean SOC amount for estimating its standard deviations, which may not hold true across the entire data set and could introduce bias. This will be addressed in future work by employing measured uncertainty data where available. Furthermore, agricultural practices also change with local policies and advanced technologies, especially in high-intensity agricultural areas. These changes also affect the SOC response to tillage and straw return practices under different environmental conditions over time, and this topic should be considered in future research.

5. Conclusions

In this study, the combination of meta-analysis and an RF model was demonstrated to be a powerful approach for investigating farmland SOC responses to different tillage and straw return practices and their interactions with climatic conditions and soil properties. The findings show that the conservation practices (e.g., no tillage, reduced tillage, deep tillage, and straw return) caused different increases in SOC in the Mollisol region of Northeast China. The meta-analysis results confirm that the conservation practices can significantly enhance SOC. The straw return practice was proven to provide the greatest increase in SOC, especially for soil with a higher pH or lower initial SOC. RT and DT had a greater positive effect on SOC over 3–10 y, and SR had an increased effect with much shorter periods (<3 y). The results from the RF models indicate that the initial SOC was the most important factor affecting SOC, with SOC increasing with the increase in the initial SOC when the initial SOC was lower than 10 g kg⁻¹. In the future, optimized management practices need to be assessed for different environmental conditions, which would provide effective methods to achieve the 4‰ target through suitable conservation practices.