Agricultural machinery socialized services as a buffer? Enhancing rural livelihood resilience amid afforestation-driven cropland loss in the Wuling Mountain Area, China

1 Introduction

Achieving carbon neutrality has positioned afforestation as a pivotal nature-based solution to mitigate climate change and reverse ecological degradation. Globally, large-scale ecological restoration programs have substantially altered land-use mosaics, driving a significant transition from arable land to forested ecosystems (Duan et al., 2015; Gutierrez Rodriguez et al., 2016). While the biophysical benefits of afforestation, such as carbon sequestration, biodiversity conservation, and soil erosion control, are well-documented, the socio-economic ramifications, particularly regarding the sustainability of rural livelihoods dependent on shrinking farmland, require deeper investigation within the coupled human–environment systems framework (Deng et al., 2025; Martin et al., 2023).

The conversion of cropland to forest inevitably imposes constraints on agricultural production space, leading to farmland marginalization and fragmentation (Yin and Yin, 2010). These changes pose substantial risks to rural household incomes and food security, potentially creating trade-offs between environmental conservation and rural development goals (Wu et al., 2021). Consequently, understanding how farmers adapt to these ecological transitions is critical. Existing literature has extensively explored subsidy mechanisms and off-farm employment as coping strategies; however, the role of agricultural production factor markets, specifically the emergence of Agricultural Machinery Socialized Services (AMSS), remains underexplored as a buffer against land loss (Qiu et al., 2023; Yang and Li, 2022).

AMSS, which allow farmers to outsource production tasks rather than owning expensive machinery, represent a profound shift in agricultural management paradigms (Chu et al., 2026). In regions undergoing afforestation-driven land system transition, AMSS may serve as a critical adaptive mechanism by reducing operational costs, enhancing labor productivity, and facilitating the reallocation of household resources (Li et al., 2022). The Wuling Mountain Area, spanning the border regions of Hunan, Hubei, Chongqing, and Guizhou provinces, represents one of China’s most ecologically sensitive and economically underdeveloped regions. As a flagship region of the Grain-for-Green Program (GFGP), it offers an ideal natural laboratory to examine the interplay between ecological restoration and rural livelihood adaptation.

Drawing on survey data from the Wuling Mountain Area, this study aims to answer two questions: (1) Does AMSS enhance rural household livelihood resilience in the context of afforestation-driven cropland loss? (2) Does AMSS buffer the negative impact of cropland loss, and through which mechanisms? By addressing these questions, we contribute to the understanding of how factor market interventions can complement ecological governance.

The remainder of the paper is organized as follows. Section 2 develops the theoretical framework and research hypotheses. Section 3 describes the data and methods. Section 4 presents the empirical results. Section 5 discusses the findings and policy implications, and Section 6 concludes.

2 Theoretical framework and hypotheses

2.1 Sustainable livelihoods, resilience, and AMSS

The Sustainable Livelihoods Framework (SLF) provides a systematic lens to analyze how households combine different forms of capital to cope with shocks and adopt livelihood strategies (Scoones, 1998). In the context of the GFGP, cropland shrinkage and fragmentation directly reduce natural capital (farmland), forcing households to reallocate other capitals (e.g., human, financial, social) to sustain their livelihoods. Livelihood resilience, distinct from static livelihood capacity or sustainability, refers to the ability of a household to absorb disturbances (absorptive capacity), adapt to changes (adaptive capacity), and transform its livelihood structure when necessary (transformative capacity) (Speranza et al., 2014; Tanner et al., 2015). Therefore, measuring livelihood resilience requires indicators that capture these multiple capacities, rather than a mere aggregation of livelihood capitals.

AMSS enters this framework as a market-based institution that can restructure the household’s production function. By purchasing machinery services, households effectively substitute capital (hired machinery) for labor and physical capital (own machinery). This substitution has two key consequences. First, it reduces the marginal cost of farming on fragmented, sloped plots—a typical feature of afforested landscapes—thereby maintaining the viability of remaining cropland. Second, it liberates household labor from labor-intensive tasks, enabling reallocation to off-farm employment. Off-farm income not only diversifies the income portfolio but also builds financial buffers against agricultural shocks, thus enhancing overall resilience.

2.2 Land system transition and the need for a buffer

The GFGP and related afforestation policies induce a land system transition characterized by a decline in per-capita cropland and an increase in land fragmentation. This transition constitutes an ecological shock that can erode households’ livelihood base. While direct subsidies provide temporary relief, they often fail to rebuild long-term adaptive capacity. In such a setting, AMSS can serve as a “buffer” by decoupling livelihood resilience from the amount of land owned. That is, households that utilize AMSS are expected to experience a smaller decline in livelihood resilience when faced with a given level of cropland loss than those that do not. This buffering effect is central to understanding how factor markets can mitigate the socio-economic costs of ecological restoration.

2.3 Research hypotheses

Based on the above theoretical framework, the following hypotheses are proposed:

H1: AMSS positively enhances rural household livelihood resilience.

AMSS improve agricultural production efficiency, reduce household labor burdens, and stabilize production under land fragmentation and labor shortages. These functions help strengthen households’ ability to cope with external shocks and maintain livelihood stability.

H2: AMSS improve livelihood resilience by facilitating household labor reallocation toward non-agricultural employment.

By outsourcing labor-intensive agricultural operations to mechanized service providers, rural households can release labor resources from farming activities and increase participation in off-farm employment, thereby diversifying income sources and improving adaptive capacity.

H3: The effect of AMSS on livelihood resilience varies with the intensity of afforestation-induced cropland transition.

As cropland conversion intensifies, agricultural production conditions become increasingly constrained. Under such conditions, AMSS may play a more important role in mitigating production difficulties and supporting livelihood adaptation.

The conceptual framework is illustrated in Figure 1.

3 Materials and methods

3.1 Study area

The Wuling Mountain Area (108°–110°E, 27°–30°N) is a mountainous region spanning four provinces. Since the early 2000s, it has been a focal area for the GFGP, with steep croplands converted to forest. The region is characterized by ethnic minority populations, limited off-farm employment opportunities, and a high dependence on smallholder agriculture. Three representative counties were selected: Longshan County (Hunan), Lichuan City (Hubei), and Wulong District (Chongqing). These counties exhibit varying degrees of afforestation intensity, enabling comparative analysis.

3.2 Data collection

A household survey was conducted in July–August 2023. A stratified random sampling strategy was adopted: three townships per county (stratified by high, medium, and low afforestation intensity), and 2–4 villages per township. A total of 360 valid questionnaires were obtained. The sample includes both households that participated in the GFGP (i.e., have a positive returned ratio) and those that did not (returned ratio = 0), allowing us to exploit variation in cropland loss intensity. The survey captured demographic characteristics, land tenure, income composition, and detailed AMSS utilization. Written informed consent was obtained from all participants, and the protocol was approved by the Ethical Committee of the Agriculture Equipment Institute of Hunan.

3.3 Variable construction

3.3.1 Dependent variable: Livelihood Resilience Index (LRI)

Drawing on the resilience measurement literature (Speranza et al., 2014; Cutter et al., 2008) and the SLF, we constructed a multidimensional LRI that reflects three essential capacities: buffering capacity (absorptive capacity), self-organization (adaptive capacity), and learning and innovation (transformative capacity). These three dimensions correspond, respectively, to the short-term ability to withstand shocks, the medium-term ability to reorganize resources, and the long-term ability to learn and innovate. While they do not map one-to-one to the five livelihood capitals, each dimension draws upon multiple capitals. For example, buffering capacity incorporates financial capital (savings) and physical capital (insurance); self-organization includes human capital (education) and social capital (cadre/party membership, loan channels); learning and innovation emphasizes human capital (skills) and information access.

The specific indicators, their definitions, and the entropy weights are presented in Table 1. All indicators were normalized to [0, 1] before weighting. The entropy weight method was used to avoid subjective bias (Zhao et al., 2023). The resulting LRI ranges from 0 to 1, with higher values indicating greater resilience.

Table 1. LRI measurement system.

Weights are calculated via the entropy weight method.

3.3.2 Core independent variable: AMSS

The key independent variable is the participation in AMSS. We measure this using the intensity of service utilization, defined as the logarithm of total expenditure on outsourced machinery services (in CNY) plus one: AMSS_cost_log = ln(AMSS_expenditure + 1). In robustness checks, we also employ a dummy variable (AMSS_dummy) indicating whether the household purchased any machinery services during the cropping season.

3.3.3 Mediating variable: off-farm employment

Measured as total person-months of off-farm employment per household during the survey year.

3.3.4 Moderating variable: cropland loss intensity

The proportion of originally farmed cropland that was returned to forest under the GFGP (returned_ratio), ranging from 0 to 0.611.

3.3.5 Control variables

At the household level: age of head, education of head, household size, number of laborers, per capita remaining cropland (mu), land fragmentation (number of plots). At the village level: distance to nearest township market (km). Village fixed effects are included in all models to control for unobserved local heterogeneity.

3.4 Econometric model

3.4.1 Baseline model

Given that the LRI is a continuous variable, OLS is a suitable and widely used baseline estimator (e.g., Fang et al., 2016; Fang et al., 2018). The model is specified as (Equation 1):

LR I i = α O + α 1 AMS S i + α 2 returned _ rati o i + ∑ β j X ji + ∑ γ k Z ki + ε i (1)

where LRI_*i* represents the livelihood resilience index of household i; AMSS_*i* denotes the intensity of machinery service use; X_*ji* is a vector of household-level control variables; Z_*ki* represents village fixed effects to control for unobserved heterogeneity across regions; and ε_*i* is the error term. Robust standard errors are clustered at the village level.

3.4.2 Testing the buffering effect

To test whether AMSS buffers the negative impact of cropland loss, we extend the baseline model by adding the interaction between AMSS and returned_ratio (Equation 2):

LRI i = α 0 + α 1 AMS S i + α 2 returned _ rati o i + α 3 ( AMS S i × returned _ rati o i ) + ∑ β j Χ ji + ∑ γ k Z ki + ε i (2)

If α 3 > 0 and significant, it indicates that as cropland loss intensifies, the marginal effect of AMSS on resilience increases;, i.e., AMSS buffers the adverse shock. The marginal effect of AMSS at different levels of returned_ratio is further explored through simple slope analysis.

3.4.3 Mediation analysis

We employ the Sobel test and bootstrapping (1,000 replications) to assess the indirect effect of AMSS through off-farm employment.

3.4.4 Addressing endogeneity: instrumental variable approach

Potential endogeneity arises from omitted variables (e.g., unobserved managerial ability) and reverse causality (more resilient households may invest more in AMSS). We instrument AMSS_log with the average slope gradient of the village (IV_slope). Steeper slopes hinder machinery operation, thus affecting service supply, but do not directly influence household livelihood resilience except through the mechanization channel. The first-stage F-statistic is 12.14 (>10), and the Cragg–Donald Wald F-statistic exceeds the Stock–Yogo 10% critical value (16.38), supporting the instrument’s relevance. Overidentification is not tested because the model is exactly identified. The exclusion restriction is plausible: slope gradient is a geographical feature not directly determined by household resilience. We use 2SLS to estimate the local average treatment effect (LATE).

3.5 Robustness checks

We perform several robustness checks: (1) replacing LRI with log of total household income; (2) using only village fixed effects without household controls; (3) employing an alternative weighting method (principal component analysis); and (4) using AMSS_dummy as the independent variable.

4 Results

4.1 Descriptive statistics

Table 2 presents the descriptive statistics of the main variables. The average age of household heads was 43.3 years (SD = 11.1), with an average of 6.8 years of education. Households experienced an average cropland reduction of 22.1% (returned_ratio), with substantial variation (SD = 0.126, range 0–0.611). The average expenditure on AMSS was 407.5 CNY, with a highly skewed distribution (SD = 555.3). The mean off-farm employment duration was 1.48 months per year, and the average livelihood resilience index was 0.375 (SD = 0.173, range 0–1.0), indicating considerable heterogeneity in household adaptive capacity.

Table 2. Descriptive statistics of main variables (n = 360).

1 mu ≈ 0.067 ha.

4.2 Baseline regression and buffering effect

Table 3 presents the main results. Column (1) shows the OLS baseline without the interaction term. AMSS_log has a coefficient of 0.011 (p < 0.001), supporting H1. Returned_ratio is negative but not statistically significant, likely because its effect is heterogeneous and absorbed by the interaction term.

Table 3. Impact of AMSS on livelihood resilience: baseline, interaction, and IV estimates.

Robust standard errors clustered at village level in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. Bold values highlight the focal variables of this study (AMSS utilization and its interaction with the returned cropland ratio)

Column (2) adds the interaction term AMSS_log × returned_ratio. The interaction coefficient is 0.074 (p = 0.022), indicating that for each unit increase in returned ratio, the marginal effect of AMSS on LRI increases by 0.074. Simple slope (Figure 2) analysis shows that at a high level of cropland loss (e.g., returned_ratio = 0.5), the marginal effect of AMSS is approximately 0.046 (p < 0.01), while at low cropland loss (returned_ratio = 0) it is only 0.009 (p < 0.05), consistent with a buffering effect (H3).

Column (3) reports the 2SLS estimate. The first-stage F-statistic is 12.14, and the Cragg–Donald Wald F-statistic is 15.42 (>10% Stock–Yogo critical value). The interaction term remains positive and statistically significant (β = 0.095, p = 0.028), confirming that the buffering effect is robust to endogeneity correction. The coefficient of AMSS_log in the IV model is smaller and not significant, likely because the LATE captures compliers for whom the direct effect is heterogeneous.

4.3 Mediating effect of off-farm employment

The Sobel test statistic is 3.99 (p < 0.001), and the bootstrap confidence interval for the indirect effect does not contain zero. The indirect effect through off-farm employment accounts for approximately 35% of the total effect, confirming H2. In the interaction model, the mediation pathway remains significant across all levels of returned_ratio. Table 4 presents the results.

Table 4. Mediating effect of off-farm employment.

*p < 0.01, **p < 0.05, *p < 0.1.

4.4 Heterogeneity analysis

To further explore the buffering effect, we split the sample by afforestation intensity. In the high-intensity group (returned_ratio > 0.3), the interaction term was 0.042 (p = 0.38), suggesting that when cropland loss is extremely severe, mechanization provides limited buffer. In the medium-intensity group (0.15 ≤ returned_ratio ≤ 0.3), the interaction term was 0.103 (p < 0.01), indicating a strong buffering effect where AMSS helps optimize remaining land. In the low-intensity group (returned_ratio < 0.15), the interaction term was 0.059 (p = 0.11), which is suggestive but not significant. These results support the “threshold effect” interpretation. Table 5 presents the results.

Table 5. Heterogeneous buffering effect.

*p < 0.01, **p < 0.05, *p < 0.1.

4.5 Robustness checks

We conducted three robustness checks. First, we replaced the dependent variable with the logarithm of total household income (a single-dimensional welfare metric). The AMSS coefficient remained positive and significant (β = 0.031, t = 3.29, p < 0.01). Second, we employed village fixed effects only (omitting household-level controls), and the result held (β = 0.012, t = 4.26, p < 0.001). Third, we used the alternative entropy weight calculation, and the significance pattern was unchanged. These results confirm the robustness of our baseline findings. Table 6 presents the main results.

Table 6. Robustness checks for the buffering effect.

*p < 0.01, **p < 0.05, *p < 0.1.

5 Conclusion and discussion

5.1 Principal findings

This study demonstrates that AMSS significantly enhance rural household livelihood resilience in the Wuling Mountain Area, and that they serve as a buffer against the adverse effects of afforestation-driven cropland loss. The labor liberation effect, through which mechanization enables off-farm employment, is a key mechanism. The buffering effect is most pronounced under moderate cropland loss, suggesting the presence of a viability threshold. We argue that building institutional support for agricultural service markets should be an integral part of ecological governance strategies under carbon neutrality goals.

5.2 Discussion

Our findings confirm that AMSS significantly enhances household livelihood resilience in afforested landscapes. More importantly, the positive interaction term between AMSS and cropland loss intensity provides novel evidence that AMSS acts as a buffer—it mitigates the negative impact of cropland loss on resilience. This result aligns with the theoretical notion that socialized machinery services can decouple livelihood security from land size by improving labor productivity and enabling income diversification.

The mechanism of labor reallocation (H2) is strongly supported: by outsourcing farming tasks, households free up labor for off-farm employment, which in turn builds financial buffers and skills. This pathway is especially crucial in the Wuling Mountain Area, where land is limited and off-farm opportunities are the primary route out of poverty.

The heterogeneity analysis reveals a non-linear buffering pattern: the effect weakens in extremely high afforestation intensity. This “threshold effect” echoes the concept of marginalization traps in land system science (Martin et al., 2023). When cropland falls below a critical level, even mechanized farming becomes unviable, and households may require different policy support (e.g., full non-farm transition). Therefore, while AMSS is effective for many, it is not a panacea for the most severely affected households.

5.3 Policy implications and limitations

These findings have important policy implications. Integrating support for agricultural service markets into ecological compensation packages could enhance the cost-effectiveness of programs like GFGP. Governments could provide subsidies or low-interest loans to machinery cooperatives in afforestation regions, and prioritize training programs that equip farmers with off-farm skills. Such market-oriented instruments would complement direct payments and foster a synergistic pathway where ecological restoration and rural wellbeing are mutually reinforcing.

This study has limitations. The cross-sectional design limits causal inference despite the IV approach. Future research should leverage panel data or quasi-experimental designs. Additionally, while we focused on economic resilience, integrating biophysical indicators would provide a more holistic assessment of social-ecological feedbacks.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Mengcheng Wang, Huazhong Agricultural University, China

Reviewed by: Chao Liu, Central China Normal University, China

Chenyujing Yang, Beijing Forestry University, China