Resilience building of rural livelihoods in PES programmes: A case study in China’s Loess Hills

Abstract

In spite of positive expectations for environmental protection, payments for ecosystem services (PES) can bring about unintended disturbances to rural livelihoods. Based on resilience thinking, this article investigates livelihood resilience building at farm level through the interaction between farm adaptation and disturbances induced by China’s Grain for Green project (GGP). Cluster analysis was conducted to investigate the complexity and diversity of farm adaptation; the crafting of composite indexes was designed to value resilience through disturbance, sensitivity, and adaptability; regression analyses linked the resilience indexes and farm adaptation with access to resources. The results show three adaptation typologies (i.e. reclamation of retired lands, contractive farming, and expansive farming) with distinct land use structures and resilience scores, and highlight the need to improving farmers’ access and endowment of tangible (e.g. farming facilities) and intangible resources (e.g. skill training) for resilience-building practices in light of the GGP. The findings imply that policy interventions combining environmental restrictions with widening resource access to support alternative livelihoods can offset the unintended effects and amplify the success of PES programmes.

Electronic supplementary material

The online version of this article (10.1007/s13280-019-01236-4) contains supplementary material, which is available to authorized users.

Introduction

Rural households in less-developed and transition countries face challenges including rural poverty and environmental degradation, and experience dramatic changes in land use and high variability of incomes (Leonard 1989; World Bank 1992; Dercon 2002). Such challenges occur as society fails to establish institutions and markets to internalize externalities of public goods, to enforce property rights and to reduce transaction costs. In this context, payments for ecosystem services (PES) draw a lot of academic attention as “new institutions designed to enhance or change natural resource managers’ behaviour in relation to ecosystem management through the provision of economic incentives” (Corbera et al. 2009, p. 745). It attempts to tax negative or subsidize positive externalities to reverse the declines in ecosystem services (ES) while conserving natural capital in supporting human well-being (Millenium Ecosystem Assessment Panel 2005). In the past years, PES programmes were underway worldwide, engaging individuals, communities, governments, and international organizations (Schomers and Matzdorf 2013). As an increasingly important tool in shaping planning and governance on ecosystem conservation and regional development, PES schemes should incorporate dynamic, two-way interactions between science and policy (Crouzat et al. 2018) as well as market negotiations between ES buyers and ES providers (Engel et al. 2008). This requires suitable approaches to ensure that the true value of ecosystems and the services provided are taken into account in managing trade-offs and promoting adaptive management.

In terms of valuing ES, previous studies have analysed economic values (Barbier et al. 2011), social values (Gómez-Baggethun and Barton 2013) and cultural values (Chan et al. 2012) covering four main aspects: biodiversity conservation (Bullock et al. 2011), watershed services (Fu et al. 2014), carbon sequestration (Capoor and Ambrosi 2009), and landscape beauty (Milder et al. 2010; Grima et al. 2016). Existing research also evaluated the livelihood benefits to PES participants, particularly in terms of building adaptive capacity and contributing to income (Bebbington 1999; Ariza-Montobbio and Lele 2010; Tacconi et al. 2013). Nevertheless, more research on agronomic and economic viability and livelihood consequences of PES on farm households is sorely needed (Ariza-Montobbio and Lele 2010; Blundo-Canto et al. 2018). A livelihood is viable when it has decent access to resources and generates enough income to sustain its operation while conserving ecosystems and the services. Viability here is crucial for long-term community livelihoods along with natural resources conservation (Pound et al. 2003). Sometimes governments and investors call for simple standardized approaches of environmental governance when implementing PES, which may preclude adaptation (Waylen and Martin-Ortega 2018) and then conflict with requests for livelihood viability. Many of the adverse effects on livelihood viability are mediated by changes in livelihood activities and strategies (e.g. farming and off-farm work) as well as in the allocation of resources, such as land, labour, and capital. Our study therefore focuses in particular on how farm households adapt to PES in relation to their economic viability and resource re-allocation by various strategies. In so doing, it responds to calls for more work on the consequences of PES on farm households’ livelihood viability.

There is broad consensus that PES, by introducing new sources of financing into environmental governance, by inviting new capital into existing agricultural relationships, by altering ecosystem management and shifting land use structure, and by intensifying agricultural production practices and diversifying livelihood activities, often initiates livelihood transition and changes resource allocation of participated farm households. The contemporary land sparing-sharing model have multiple livelihood consequences, as indeed did changes in the diversity of farm household systems (FHSs) and the security of livelihoods (Li et al. 2018). Putting rural livelihoods central, this paper looks at livelihood viability via the resilience (Holling 1973; Walker et al. 2004) and farm adaptation (Darnhofer et al. 2010) in response to PES. We rely on resilience thinking and present a quantitative approach to value the resilience of farm household livelihoods. Our particular attention was given to farm household livelihoods in PES programmes, where there is an immediate and intimate dependency on livelihood resources in terms of human, social, natural, physical, and financial capital. The questions that this paper sets out to answer are what is the typology of adaptive actions and strategies that farm households undertake to build their livelihood resilience in response to the PES? What are the key features of these adaptation typologies? How do the PES scheme and the resultant resource access and endowment affect the value of livelihood resilience and farm households’ adoption of the adaptation typologies? To answer these questions may help rural residents reduce the risk of ecosystem collapse and prevent additional socio-economic degradation of livelihood viability within PES.

The Grain for Green project (GGP) was initiated by the Chinese national government in 1999 as a land set-aside programme, in order to curb environmental degradation and to reduce rural poverty. It is akin to PES programmes where rural households are compensated to restore ecosystem in sloping land (slope exceeding 25°) with an initial target of reducing flood and soil erosion. Meanwhile, intensive farming and diversification into non-farm activities have been encouraged to improve rural livelihoods and alleviate poverty. The introduction of GGP affects the socio-economic and environmental outcomes of rural livelihoods through targeting of land plots, subsidies, inspection regime, promotion of the programme and farmer training, and farmer autonomy (GutiérrezRodríguez et al. 2016). Challenges and criticisms have been noted with regard to its economic inefficiency and environmental ineffectiveness (Uchida et al. 2005; Xu et al. 2009), such as inappropriate forestation and monoculture (Cao et al. 2007), its potential damage to the regional water balance (Zhang et al. 2008; Cao et al. 2009), surplus labour and land abandonment (Peng et al. 2007; Bullock and King 2011), and grazing constraints on animal husbandry (Zhang et al. 2005; Yao et al. 2010). These may lead to a reduction in ES and may aggravate poverty for those with rural livelihoods (Wang et al. 2007; Song et al. 2014), resulting in the “social–ecological trap” (Boonstra and de Boer 2014) or the “ecosystem service curse” (Kronenberg and Hubacek 2013). To avoid such trap or curse, it is necessary to clarify the dynamics and complexity of FHSs and to identify the dysfunctional institutions (Sachs and Warner 1995; Kronenberg and Hubacek 2016). This requires to explore the structural causes of trap or curse and the processes that form the dynamic relationship between livelihood resilience and PES. Also, there is still much to learn about what makes system resilient or not, and how PES intervenes effectively and efficiently.

By considering China’s GGP a PES policy shock to FHSs, the central theme of this paper builds on the ‘Household Resilience-building Strategies Index’ (Li et al. 2016a) but (1) takes a more systematic approach to valuing resilience by capturing farm households’ capacity and sensitivity to persist and adapt to the unintended PES disturbance or perturbation induced by changing conditions; (2) classifies farm households by their adaptive actions on farm efforts (e.g. supplemental irrigation, rent farmland, and grazing) and their long-term plan for adaptation if PES payments cease; and (3) pays particular attention to the notion that limited access to resources constrains the livelihood viability (Bebbington 1999), while researching on the possibility towards improved resilience by adjusting resource allocation (e.g. PES payments, farmland consolidation, and secure tenure) and widening resource access (e.g. access to training, irrigation, and services) to encourage efficient adaptation and livelihood alternatives.

Conceptual and analytical frameworks

Resilience thinking with a focus on the interdependence of social and ecological systems facilitates the analysis and the policy process in striving towards sustainability (Berkes and Ross 2016). Resilience was first mentioned by Holling (1973) as a viewpoint on the behaviour of ecological systems that emphasizes domains of attraction and the need for persistence to change. Walker et al. (2004) defined resilience as “a system’s capacity to absorb disturbance and reorganise while undergoing change, so as to still retain essentially the same function, structure, identity, and feedbacks”, and portrayed its four components in a stability landscape: the maximum change of a system before evolve or collapse (latitude); the difficulty of a system to be changed (resistance); the distance of the current state of a system to an evolving or collapsing threshold (precariousness); and the cross-scale interaction among (sub)systems (panarchy). Thereafter, the complex and contradictory aspects of the concept of resilience were disentangled: as part of resilience, the capacity to deal with external and internal changes within the current regime (adaptability); and the capacity to cross the threshold into new development trajectories due to a radical change in the nature of a system (transformability) (Walker et al. 2004; Folke et al. 2002). These attributes and aspects of resilience clarify the need for a combination of strategies to allow a system to successfully navigate changes and remain functional. It also implies that resilience is relevant for unforeseeable crises, attempts to limit the counterintuitive negative consequences, and can be complementary to ‘sustainability’ which seeks to prevent foreseeable crises through management (Falkenberg 2016).

Resilience thinking offers an understanding of the world as an adaptive system in response to unpredictable and qualitatively different types of changes (Darnhofer et al. 2010). In terms of agriculture and rural development, farmers are embedded in a social–ecological system while their livelihoods are a connector between social and natural systems via the management of resource use at farm, community, regional or other levels (Berkes and Ross 2016; Ashkenazy et al. 2017; Li et al. 2018). In the social–ecological system at the farm household level, adaptation is the way of combining and transforming resource use to build resilience to changes (Li et al. 2016a). Putting this definition in the context of resource management, adaptation can expand desirable effects on FHSs and shrink undesirable ones; create new desirable effects or eliminate undesirable ones; and change the current state of the system into a fundamentally new one (Walker et al. 2004; Darnhofer et al. 2010). Nevertheless, the adaptation for resilience and sustainability needs to be ‘context dependent’ (i.e. the details of local and regional contexts and the particular attributes of the systems), because interventions at one level or on some specific components of the systems may lead to undesirable consequences of the wider level or region; and it must itself change over time, because decisions and actions that make desirable effects in the short term may lock them onto a path that generates undesirable consequences in the long term (Walker et al. 2004; Berkes and Ross 2016; Ashkenazy et al. 2017).

Following resilience thinking, we assessed livelihood resilience with a focus on farmers’ adaptation to PES interventions. Placing people in the centre of analysis, it highlights the role of human agency, rights, and capacity to cope with shocks (Tanner et al. 2015). Central to livelihood resilience is the adaptive actions and strategies used by farm households to cope with shocks, navigate uncertainty, and to adapt to changing conditions (Marschke and Berkes 2006). A framework (Fig. 1) demonstrates the role that farm adaptation to PES interventions can play in the rural livelihoods which are embedded in social–ecological systems. The PES influences farmers’ convictions about the role of agriculture and their preferences for livelihoods through ecosystem conservation (e.g. reallocating farmland to forest or grassland), providing subsidies to cover for income loss, and stimulating livelihood diversification (e.g. transferring farm labour to off-farm jobs). It pulls and pushes farmers to adjust their actions and management strategies to smooth incomes and consumptions while conserving ES. The resulting socio-economic values and ES benefits generate certain impacts (or effects) on farm households’ well-being and livelihood viability. It further gives feedback and fine-tunes the PES interventions by explicitly considering the alternative scheme based on non-monetary incentives (e.g. direct use and experience), the full range of benefits that can be derived from ecosystems, as well as who are the beneficiaries and the bearer of costs. Along the feedback cycle, the fine-tuned interventions may in turn support the agency of rural individuals, collectives, and communities to act on their socio-economic values and ES by undertaking a better adaptation of their production activities and resources management strategies. In China’s Loess Hills, there emerged seven types of farm household systems across the valley and riparian areas (Li et al. 2018), i.e. intensive horticulture, contractual employment, intensive livestock, casual employment, intensive highland orchard, horticulture on rented land, and mixed farming; the majority of farm households earned livelihoods while engaging in the stewardship of ecosystems by involving in intensive horticulture, intensive highland orchard, and contractual employment under the GGP PES scheme. The emerging but prevailing intensive farming and non-farm employment remind that careful livelihood reorganization and adaptation support is a necessary condition for successful PES schemes.

Fig. 1.

A framework for the role of farm adaptation to PES interventions played in rural livelihoods

For a thorough understanding, adaptation typology was developed to classify farm households by the range of adaptive actions and strategies in response to the GGP PES scheme. We assumed that farmers’ actions are embedded in an overall strategy which is an expression of personal convictions and preferences in a coherent, targeted, and long-term plan for adaptation. This stems from the premise that “people have a good understanding of the factors that contribute to their ability to anticipate, buffer and adapt to disturbance and change” (Jones and Tanner 2015). The adaptation typology was thus grouped into three broad categories: (i) contractive farming, reducing farm efforts after retiring steep lands, and increasing non-farm diversification if payments cease; (ii) reclamation of retired lands, breaking the grazing prohibition, and planning to reclaim the retired lands to agricultural use if payments cease; (iii) expansive farming, expanding farm efforts, and maintaining farm production and steep-land retirement even if payments cease. All categories are independent of one another. Expansive farming refers to renting cultivated land and use of supplemental irrigation and plans to improve farm production in the remaining farmland even if the payments of GGP compensation cease; contractive farming refers to leasing out cultivated land and null irrigation and plans to shift to non-farm production. These two are positive and the general mainstream, enticing ecosystem conservation through a sustainable transition in rural livelihoods (Gong et al. 2006; Bennett 2008; Grosjean and Kontoleon 2009; Yang et al. 2013). In contrast, land reclamation is negative and refers to actions and plans to reverse GGP efforts. In general, it is unlikely that farms undertaking land reclamation can survive in the longer-term, given the need to protect restored ecosystems and maintain productivity increases. It is important to note that our typology is only a broad indication of farm adaptation to the GGP.

Thereafter, livelihood resilience was assessed from a range of observable indicators representing the GGP PES intervention and the adaptive actions and strategies. Disturbance (or perturbation) represents the stress or undesired effects of the GGP PES intervention on farming systems which put pressure on farmers and impact their livelihoods (Wang et al. 2010). Sensitivity refers to the ease or difficulty of changing the farming system (Walker et al. 2004), and the likelihood of farmers’ livelihoods being disturbed (IPCC 2001; Tian et al. 2015). Adaptability represents the ability of farm households to adapt their production activities and resource management strategies to deal with the intervention within the current regime. Following five types of resource capital (i.e. human, social, natural, physical, and financial) in the rural livelihoods framework (Ellis 2000), farm households’ access to resources determines what adaptation typology they choose and what level of livelihood resilience they obtain. We thus selected variables to represent the five types of resource capital and performed a set of regression analyses to explore the determinants of livelihood resilience and adaptation typologies.

In short, the hypotheses this paper attempted to test are three broad categories of adaptation typology could be set up based on farmers’ actions and long-term plan for adaptation, i.e. expansive farming, contractive farming, and reclamation of retired lands; and farm households’ livelihood resilience and their propensities for the adaptation typology might be determined by their access to resources.

Materials and methods

Study area and data collection

An’sai County in the middle part of China’s Loess Plateau was chosen for a survey to collect primary data. The survey site, Yanhe Township (Fig. 2), was deliberately chosen to accurately represent the range of social and ecological conditions. The township is between two urban areas, 16 km south of the county seat and 25 km northwest of Yanan City, and the topography and semi-arid climate are typical of that region. It has a population density of 73 inhabitants per km², which is representative of the rural areas in the Loess Hills region (Liu et al. 2012). The township covers 210.7 square kilometres, is interlaced with hills, ravines, and plains, and has an average gully density of approximately 4.7 km km² (Xu et al. 2009). Annual precipitation ranges from 296.6 to 645.0 mm (mean 505.3 mm); annual temperatures vary from − 23.6 to 36.8 °C (mean 8.8 °C) (Lu et al. 2003; Lu et al. 2004). The GGP was initiated in 1999 to restore ecosystem in sloping land (slope exceeding 25°) with an initial target of reducing flood and soil erosion. Farmers receive rewards through a two-tier compensation scheme: a grain subsidy of 1500 kg ha⁻¹ year⁻¹ and a cash subsidy of 300 RMB ha⁻¹ year⁻¹. In 2004, the grain compensation was replaced with cash compensation of 2400 RMB ha⁻¹ year⁻¹. Since 2007, the second round of GGP had been implemented at half of the initial compensation rate. Between 1998 and 2009, 53% of the farmland in An’sai County was converted into forests by implementing the GGP (An’sai Statistical Bureau 1999–2013). Agriculture on the remaining farmland was intensified with a reduction of the net sown area (10%) but an increase in irrigated area (21%) and crop-rotation area (3%) along with increasing use of fertilizers (2.4%), mulches (4.5%), pesticides (1.7%), diesel fuel (5.7%), and farming machines (51%) (An’sai Statistical Bureau 1999–2013). Land use for horticulture and orchards increased by 2083 and 2552 ha, while open-field crops and pasture decreased by 2111 and 19 330 ha, respectively; harvest of apples and vegetables increased by 140 and 135%, respectively, whereas the yield of cattle decreased by 67% (An’sai Statistical Bureau 1999–2013). The obvious impact on land use and livelihoods makes Yan’he Township an ideal survey site.

Fig. 2.

Study area of Yan’he Township, in China’s GGP PES implementation region

The survey was conducted between February and May 2014 using a semi-structured questionnaire designed for information on GGP PES, household demographics, the availability and utilization of agro-ecological and socio-economic resources, and levels of household income and consumption in 2013. We followed a three-stage stratified sampling procedure. Districts constitute the strata, covering two types of the landscape—floodplain and V-shaped valley. Primary sampling units are villages which were chosen according to their different altitudes and market distances. Heads of households that account for 20% of total households in each village were interviewed. In total, 247 households from the 16 of 28 villages in six districts of Yan’he Township were chosen (Table 1), based on a confidence interval of 5.96 at a 95% confidence level and a population (number of total households) of 3390. The sampled households account for 7.29% of the population of the township. The average household size is 3.57 which is close to the average of An’sai County (i.e. 4.07) (An’sai Statistical Bureau 1999–2013); the rural per capita net income is around 8665 RMB in 2013, which is 483 RMB below the average of An’sai County but close to the average of Yan’an City (i.e. 8681 RMB) (National Bureau of Statistics of China 2014). For the sake of data accuracy and coherence, the entire survey was conducted personally by the first author. The core survey item about farm household livelihoods under the implementation of GGP PES was discussed at the beginning (e.g. perception of impacts on incomes and local environment) and the end (e.g. the perception of livelihoods influenced by GGP PES) of each interview, in order to double-check farmers’ perceptions. Additional farm-level data were generated using farms’ location based on Global Positioning System coordinates and a soil map taken from the Soil Testing and Formulated Fertilisation System (An’sai Agro-Tech Extension and Service Station 2010).

Table 1.

Description of farm household survey sampling

Village	Number (share) of samples (%)	Household size (mean)	Altitude (mean in meters)	Distance to market (mean in km)	Rural per capita net incomea (mean in RMB)
Highlands of V-shaped valley districts
Houjiagou	9 (3.7)	4.00	1111	8.5	7641
Siyaoxian	13 (5.3)	3.31	1197	12.2	7238
Yanta	7 (2.8)	3.71	1115	8.3	8151
Yujiahe	14 (5.7)	2.71	1191	16.6	5923
Zhaiziwan	15 (6.1)	3.20	1175	13.1	9950
Middle elevations of V-shaped valley districts
Gaojiamao	25 (10.2)	3.08	1114	11.7	5968
Yayao	15 (6.1)	3.40	1137	10.5	8574
Lowlands of V-shaped valley districts
Zhuanyaogou	11 (4.5)	3.09	1050	3.1	10 884
Fangjiahe	14 (5.7)	3.29	1068	5.2	17 766
Zhifanggou	6 (2.4)	3.50	1053	9.7	17 863
Upstream of riparian districts
Chafang	20 (8.1)	4.05	1052	7.2	7293
Yunping	19 (7.7)	3.84	1039	8.4	6135
Midstream of riparian districts
Hougoumen	20 (8.1)	3.65	1031	7.2	11 250
Yanjiawan	18 (7.3)	4.72	1023	4.8	10 427
Downstream of riparian districts
Lijiawan	23 (9.4)	3.43	1006	1.7	2444
Yangjiagou	17 (6.9)	3.88	1008	5.7	11 450
Total	246 (100)b	3.57***	1079***	8.2***	8665***

*, **, *** = 0.1, 0.05 and 0.01 levels of significance in differences suggested by non-parametric test

^aRural per capita net income (Appendix S4) is calculated by the aggregate net income of rural residents excluding those people who earn their livelihoods through contractual employment in urban areas

^b247 farm households were surveyed but 1 of them is incomplete

Empirical analysis

After checking for missing values, potential errors, and outliers, a clean dataset of 242 samples was used to analyse the differences in resilience across adaptation typologies, and to explore the linkage of various forms of resources to farm households’ resilience and farmers’ choice on adaptation given local contexts.

Empirical models

The resilience of complex natural and man-made systems cannot be measured directly similar to biodiversity (Jacobs et al. 2015), but can be approximated with a coupled multi-dimensional non-linear equation (Gao et al. 2016):

$$\frac{\text{d}{x}_i}{\text{d}t}=F\left(x_i\right)+\sum _{j=1}^N{A}_{\mathit{ij}}G\left(x_{i,}{x}_j\right)$$ 1

The first term on the right-hand side of Eq. (1) describes the self-dynamics of each component that real systems are composed of, while the second term describes the interactions between component i and its interacting partners i. The non-linear functions F(x_i) and G(x_i, x_j) represent the dynamical laws that govern the system’s components, while the weighted connectivity matrix A_ij captures the interactions between the components (Gao et al. 2016). In evolved farming systems that have been subjected to the disturbance or perturbation of change in the environment, resilience can be explained as systems’ ability to adjust actions and adapt strategies to maintain systems’ fundamental functions and structures (Walker et al. 2004; Darnhofer 2014). Hence, Eq. (2) considers the resilience of farm household livelihoods by capturing farmers’ capacity to adopt adaptive actions and strategies for resilience building (i.e. adaptability) x_i, the disturbance or perturbation x_j induced by changing conditions, and the sensitivity θ that allows farms to persist and adapt to the disturbance and thus reflects the interactions between x_i and x_j (Darnhofer et al. 2010; Li et al. 2016a).

$$\frac{\text{d}{x}_i}{\text{d}t}=f\left(x_i\right)+A_{\mathit{ij}}\left(\theta ,x_j\right)$$ 2

At a certain moment, we assume that disturbance or perturbation negatively interacts with adaptability at a default value A_ij of − 1. Thus, the momentary resilience of farm households can be defined in a reduced form by using the weighting schemes developed by principal component analysis (PCA):

$$R_t=\sum _{i=1}^{N^{\prime }}{a}_i{x}_i-\sum _{k=1}^n{a}_k{\theta }_k·\sum _{j=1}^N{a}_j{x}_j$$ 3

where R_t is the composite index of the resilience of farm household livelihoods at time t; x_i is the value of indicators capturing adaptability while x_j and θ_k are the value of indicators for disturbance and sensitivity, respectively; a is the PCA weighting (Tables S1–S3, Electronic supplementary material) to integrate the specific indicators of disturbance, sensitivity, and adaptability into three composite indexes. Next, the composite index is transformed into values ranging from 0 to 100:

$$R_t^{\text{s}}=\left(\frac{R_t-R_{t\text{min}}}{R_{t\text{max}}-R_{t\text{min}}}\right)\times 100$$ 4

where $R_t^{\text{s}}$ is the normalized value of resilience index at time t, while R_{t max} and R_{t min} are the maximum and minimum values of resilience index at time t for the dataset. A value of 0 indicates the lowest possible degree, while 100 means the highest degree in the dataset. Similarly, the normalized values of disturbance index, sensitivity index, and adaptability index can be also obtained in a range from 0 to 100.

Underlying factors regarding assets and resources determine rural households’ adoption of multiple adaptive strategies and thus play critical roles in livelihood security (Ellis 2000; Li et al. 2018). Based on the livelihoods framework, we hence proposed that farm households’ adaptation choice and their livelihood resilience depends on human capital, natural capital concerning irrigation and land fragmentation, social capital (e.g. land tenure security and access to services), physical capital such as basic infrastructure and production equipment, and financial capital including the PES compensation subsidies. First, we explored the determinants of livelihood resilience by using Ordinary Least Square (OLS) regression models.

$$R_t^{s^{\prime }}=\sum _l{\beta }_l\left(x_{\mathit{kl}}\right)+{\beta }_0+\epsilon l=1,2,3,\dots L;k=1,2,3,\dots K$$ 5

where $R_t^{s^{\prime }}$ is the estimated value of resilience indexes at time t, ε is the error term, β₀ is the constant coefficient, and β₁ is the coefficient of x_kl which is the value of the selected indicators representing the lth form of resources of kth farm households.

Based on logistic models which are convenient for choice-based samples, a multinomial logistic regression model was further developed to explore the determinants of farmers’ adaptation choice. One of the three adaptation types was designated as the reference (Table S4). The probability (P) of choosing other adaptation types is compared to the probability of choosing the reference.

If the first type of adaptation is the reference, then for m = 2, 3.

$$Z_{\mathit{mi}}=ln\frac{P\left(Y_{i=m}\right)}{P\left(Y_{i=1}\right)}={\alpha }_m+\sum _l{\beta }_{\mathit{ml}}\left(x_{\mathit{kl}}\right)+\epsilon l=1,2,3,\dots L;k=1,2,3,\dots K$$ 6

where z_mi is the likelihood of ith type of adaptation predicted from a number of the value x_kl. α indicates the intercept of the regression curve, β is the coefficient of each predictor, and ε represents the error term. Note that when m = 1, the logit is ln(1) = 0 = Z₁₃, and exp(0) = 1.

Key variables

Key variables were selected to quantify adaptation typology (“Adaptation typology” section), livelihood resilience (“Livelihood resilience” section), and livelihood resources (“Livelihood resources” section). This enables our study to explore the consequences of PES on farm household livelihoods, resource re-allocation, and agronomic and economic viability.

Adaptation typology

In local areas of China’s Loess Hills, implementation of GGP PES is considered an effective ecosystem restoration policy. During the last years, local farmers were no longer dependent mainly on land resources by adopting a more diversified livelihood (Hageback et al. 2005). Agricultural practices that include building terraces, returning sloped farmlands to forestland and grassland, and expanding orchards all have had positive and significant impacts on farmers’ livelihood assets, strategies, outcomes, and vulnerabilities, with the reduced dependence upon grain and subsidies income, the diversified strategies for livelihood, and the improved ecological environment (Tang et al. 2013). However, 70% of the interviewed farmers preferred felling the ecological forest and cultivating the land again if the government stopped the cash support (Wang et al. 2010). This implies that in the long run, GGP PES and its ecological compensation system might not be enough to make recovery sustainable without an adequate livelihood transformation strategy.

In terms of livelihood strategy, it is often identified as three broad types: “‘hanging-in’ strategies, which are concerned to maintain and protect current levels of wealth and welfare in the face of the threats of stresses and shocks; ‘stepping-up’ strategies, which involve investments in assets to expand the scale or productivity of existing assets and activities; and ‘stepping-out’ strategies, involving the accumulation of assets to allow investments or switches into new activities and assets” (Dorward 2009, p. 136). In the context of China’s Loess Hills, our adaptation typology was thus grouped into the three aforementioned categories—expansive farming, contractive farming, and reclamation of retired lands.

In each of the three categories, a comprehensive set of dummy variables (Table 2) was selected to characterize farmers’ adaptation. Variables concerning cultivated land transfer (Sun et al. 2016), use of supplemental irrigation (Zhao et al. 2009), and grazing prohibition (Cao et al. 2009) were selected to study the adaptive actions to GGP PES. On average, 54% of farm households increased irrigation by pumping and rainfall harvesting, whereas 44% of them remained dry farming. Farm households renting or leasing out farmlands accounted for 23 and 33%, respectively; only 3% of them had broken the grazing prohibition of GGP PES. In addition, ‘subjective’ indicators concerning farmers’ plan of their livelihoods (i.e. farming, non-farm diversification, or reclamation of retired lands for agriculture) were derived from qualitative interviews, with an assumption that GGP subsidies would cease (Nolan et al. 2008). When the subsidies cease, around 11% of farm households would reclaim, whereas each of those who would continue farming or shift to non-farm work accounts for about 32%.

Table 2.

Descriptive statistics of variables used for adaptation typologies

Adaptation typology	Variables	Description	Mean ± standard deviation
Expansive category	Rent farmland	Renting land to increase farmland area in response to the current GGP scheme, 1/0	0.227 ± 0.420
	Supplemental irrigation	Supplemental irrigation by pumps and rainfall harvesters to improve farm production in response to the current GGP scheme, 1/0	0.537 ± 0.500
	Farming if no payment	To continue improving farm production even if the payments of GGP compensation cease, 1/0	0.326 ± 0.470
Contractive category	Lease out farmland	Leasing out land to reduce farmland area in response to the current GGP scheme, 1/0	0.331 ± 0.471
	Dry farming	No irrigation in farm production in response to the current GGP scheme, 1/0	0.442 ± 0.498
	Non-farm work if no payment	To shift to non-farm work if the payments of GGP compensation cease, 1/0	0.322 ± 0.468
Land reclamation category	Grazing prohibition	Breaking the grazing prohibition against the GGP scheme, 1/0	0.029 ± 0.168
	Land reclamation if no payment	To reclaim retired lands if the payments of GGP compensation cease, 1/0	0.107 ± 0.310

Thereafter, a composite score of each category was calculated by combining respective variables with an equal weighting scheme. The score was further input into Ward’s method and the K-means cluster analysis (KCA) (Kobrich et al. 2003; Field 2005; Bidogeza et al. 2009). Ward’s method minimizes the variance within clusters and identifies clusters of a relatively equal size, so that it was performed on the dataset to identify the number of clusters. KCA classifies cases into groups that are relatively homogeneous within themselves and heterogeneous between each other, on the basis of the calculated composite score. The agglomeration schedule and dendrogram (Table S5) showed that three homogeneous groups could be generated in the dataset. The three groups are called clusters which present a typology of three types of distinct farm adaptation in China’s Loess Hills.

Livelihood resilience

In this study, we posited that the most robust livelihood system is one displaying high adaptive ability to absorb the PES disturbance and low sensitivity to maintain its basic function and structure (Allison and Ellis 2001; Walker et al. 2004; Quandt 2018). Although there are methods measuring resilience by integrating subjective measures (Jones and Tanner 2015) or drawing from livelihood capital assets (Quandt et al. 2018), a range of observable indicators concerning ‘disturbance’ and ‘sensitivity’ was added to the composite ‘adaptability’ index (Li et al. 2016a). We thereby attempted to systematically investigate livelihood resilience through the interactions between farm adaptation and PES disturbance. The sanctions and monitoring of PES that are perceived by farmers as the control of land use and a limitation of autonomy can crowd out their intrinsic motivations for ecosystem stewardship and disturb their livelihood security (Falk and Kosfeld 2006; Dickinson and Villeval 2008; Ezzine-De-Blas et al. 2015). Thus, indicators about farmland availability and abandonment and labour input were employed to represent such stress and disturbance. Livelihood diversification reflects the precariousness of rural survival, and has both positive attributes and negative connotations for livelihood security depending on the asset basis of livelihoods (Ellis 2000). For instance, productive population (i.e. the population between the ages of 16 and 60) means that nonfarm diversification for income generation can enhance the livelihood security of the very poor who are often more vulnerable to disasters (Husby et al. 2018); on the other hand, high wealth inequality may mean that the promotion by policy of horticulture (e.g. greenhouse) income favours the better off above the poor who can thus perceive a sense of economic unfairness, undermining the effectiveness of environmental stewardship (Starmans et al. 2017; Hamann et al. 2018). Hence, indicators concerning the diversity of rural livelihoods and household wealth (e.g. total assets) were used to represent the sensitivity of farm adaptation to disturbance.

The disturbance was measured by three indicators (Table 3): abandonment of remaining cultivated land, the stress of land set aside on farmland availability, and labour input for monitoring GGP land (Metzger et al. 2006; GutiérrezRodríguez et al. 2016). Our previous studies (Li et al. 2016a, b, 2018) found that at least half of the land should be under cultivation to reorganize and sustain livelihoods. However, the current share of set-aside land reaches 77% (i.e. ratio of GGP land to the total farmland is 3.4 on average) which is far beyond this threshold level. Also, the fixed payments are no longer much higher than the opportunity costs of the set-aside land, as food prices have increased significantly since the implementation of GGP PES (Song et al. 2014). During our survey, 33% of farmers indicated that GGP PES was a source of anxiety and 6% reported the feeling of being boxed in by the restrictions. The mean service days are around 6.3 per year per household. In addition, around 25% of households surveyed had abandoned their remaining farmland (i.e. 0.07 ha per household) mainly due to it being “too far from [their] settlement” or “badly shaped”. These three variables were assumed to be impediments to farm adaptation and livelihood resilience, which might lead to degradation, fragmentation, and unsustainable use of land and eventually into a poverty trap and/or the ecosystem service curse.

Table 3.

Descriptive statistics of variables used for livelihood resilience

Component	Indicator	Description	Mean	Std. deviation
Disturbance (0–1)
	Stress on land use	Ratio of GGP land to the total farmland cultivated by the household, ranges from 0 to 38.33	3.40	4.93
	Service days	Working days carried out by the household on GGP land, ranges from 0 to 270, in man-days (8 h)	6.34	19.31
	Land abandonment	Area of remaining farmland abandoned by the household, ranges from 0 to 1.13, in ha	0.07	0.17
Sensitivity (0–1)
Livelihood activities	Income diversitya	Income diversity indices: 1–5, 1 means a single income activity	1.94	0.74
	Livestock diversitya	Livestock diversity indices: 0–4, 0 means no livestock animals	0.36	0.69
	Crop diversitya	Two-dimensional diversity measurement, 0–1 (0 means the household grows no crops or only one crop)	0.44	0.29
	Dependency ratio	Ratio of the population under 16 and over 60 years of age to the population between 16 and 60 years of age, 0–1 (subtractive)	0.65	0.97
Assets	Total assets (market)	Present value of the total assets that a household possesses calculated by market price, in RMB	230 499	126 948
	Distance to market	Distance from household location to the closest market, in km	8.08	4.02
Adaptability (0–1) (Li et al. 2016a)
Intensification	Irrigation	Annual outlays for irrigation per hectare of total farmland, in RMB/ha	565.44	922.09
	Chemical fertiliser	Annual outlays for chemical fertilisers per hectare of total farmland, in RMB/ha	4380.06	10 025.37
	Organic fertiliser	Annual outlays for manure and other organic fertilisers per hectare of total farmland, in RMB/ha	4021.09	5314.79
	Intercropping	Ratio of the intercropped area to total farmland	0.18	0.37
	Continuous cropping	Ratio of the area with continuous cropping to total farmland	0.19	0.36
Diversification	Crop varieties	Number of crop varieties planted by household	2.61	1.71
	New animal	Number of new animal varieties reared by household	0.26	0.61
	Integration of open-field cropping	Present value of annual income from open-field cropping, in RMB/ha	10 304.69	11 644.69
	Integration of horticulture	Present value of annual income from horticulture, in RMB/ha	105 432.12	170 602.29
	Integration of orchard	Present value of annual income from orchards, in RMB/ha	16 140.22	39 313.14
	Integration of livestock	Present value of annual income from livestock, RMB/LU*	1462.69	5466.30
Alteration	Mulch and bag	Annual outlays for plastic sheeting or organic residues covered on the field and/or bag wrapping fruits, in RMB/ha	732.57	1030.89
	Natural vegetation	Area of grass and forest vegetation owned per household member, in ha per capita	0.30	0.22
	Rainfall harvest	Present value of rainfall harvester, in RMB/ha	1065.78	4143.54
Non-farm labouring	New non-farm employee	Number of new household members engaged in non-farm work	1.21	0.98
	Emigration	Present value of annual payment per household member who is expected to emigrate permanently, in RMB/capita	3971.54	6780.10
	Integration of casual employment	Present value of annual income of casual employment, in RMB/man-day	86.11	111.25
	Integration of contractual employment	Present value of annual income of contractual employment, in RMB/man-day	65.13	84.54

^aPlease see the Electronic supplementary material for details on calculation

Adaptability was measured through intensification, diversification, alteration of practices, and the use of non-farm labour (Table 3) referring to the Household Resilience-building Strategies (HRS) index (Li et al. 2016a). Given the disturbance by GGP PES, “intensification” refers to increasing investment in farmland inputs (i.e. irrigation, chemical fertilisers, organic fertilisers, intercropping, and continuous cropping) to promote the productivity of the remaining farmland (Rockstroem et al. 2010). Per hectare annual outlays for chemical or organic fertilisers were above 4000 RMB, whereas irrigation outlays (565 RMB ha⁻¹ year⁻¹) and the ratio of intercropping (0.18) and continuous cropping (0.19) were relatively low. The diversification of crops, livestock animals, and/or additional income sources can utilize surplus labour (Lin 2011; Kaul and Thornton 2013), diversifying farming income streams and reducing income risk. In the dataset, horticulture brought about the highest annual income (105 432 RMB ha⁻¹), whereas open-field cropping gained the lowest (10,304 RMB ha⁻¹). “Alteration” implies changes in existing practices (i.e. mulching and fruit bagging for weed control and moisture maintenance, natural vegetation for erosion control, and rainwater harvesting for irrigation) in an effort to enhance the stability of production in the remaining croplands (Mutabazi et al. 2015). In 2013, the annual outlays for mulching and fruit bagging were around 732 RMB ha⁻¹; per capita grassland and forest vegetation was about 0.30 ha; the present value of rainfall harvester was 1066 RMB ha⁻¹. Additionally, non-farm employment that was intended to utilize labour resources more efficiently and supplement farming income losses, was taken into account in the form of temporary or permanent migration, casual employment, and contractual employment. Every household had a new member engaged in non-farm work, on average, taking 65 RMB and 86 RMB as the average daily wage for casual employment and contractual employment, respectively. A set of indicators were hence selected according to these four aspects that facilitate livelihood reorganization and income attainment. The sum of these indicators represents farm households’ ability to adapt their production activities and resource management strategies.

The sensitivity of farm adaptation to disturbance was measured with respect to livelihood activities and household assets (Robinson et al. 2015). The sensitivity of livelihood activities (Table 3) to the induced disturbance was represented by income diversity (Appendix S7), livestock diversity (Appendix S8), and crop diversity (Appendix S6) (Shahidullah et al. 2006) as well as the dependency ratio (Hahn et al. 2009). Our data show a high pressure on the productive population as the dependency ratio was 0.65, as well as a low livestock diversity (0.36) and a medium level of crop diversity (0.44) and income diversity (1.94). A certain level of diversity can confer resilience to disturbance by facilitating livelihood diversification, but diversity also means that there are fewer resources available to exploit intensification. It reflects the ease or difficulty of changing the farm household system in which farmers face trade-offs between the resource management of livelihood activities under the GGP PES intervention. The sensitivity of household assets (Table 3) to the intervention was calculated according to total assets (Carter et al. 2004) and distance to market (Li et al. 2016b). It captures the total wealth of farm households to depict the likelihood of their livelihoods being disturbed. We assumed that the livelihoods of wealthier farm households get less likely disturbed. Total assets were the aggregate of the present value of each household’s assets, using the specific market price and according to the depreciation of each item (Appendix S1 and Table S11). Some fixed assets such as grassland and forest were given fixed prices (Table S6) due to the lack of price and imperfect land markets, so the distance to market, which influences their present value, was considered as well. In 2013, the present value of the total assets per household was around 230,499 RMB (1 US$ = 6.19 RMB). Self-insurance and informal insurance arrangements (Dercon 2002) were omitted in this step due to the poorly functioning market and the insignificant difference (e.g. no insurance for climate disaster or damage in agriculture was used).

Livelihood resources

Differences in the calculated resilience indexes (Table 4) as well as in land use structure (Table 5) were demonstrated across the adaptation typologies by using Analysis of Variance (ANOVA) and Non-parametric tests. It implied linkage between land use structure, adaptation typology, and livelihood resilience, raising the question that how access to resources influences farmers’ adaptation choice and their livelihood resilience under the GGP PES scheme. A set of variables representing livelihood resources (Table 6) were selected for regression analyses, based on our previous study (Li et al. 2016a, b; Li et al. 2018) and literature review. In particular, farm households’ access to resources was highlighted by variables like ‘access to irrigation’, ‘access to off-farm training’, and ‘access to agronomic services’; the effect of PES compensation was emphasized by variables, such as ‘GGP subsidy’ and ‘GGP duration’. Improved access to resources, such as access to natural spaces, skills development, and connection to market, can increase ecosystem benefits and promote livelihood resilience (Masterson et al. 2019). PES compensation as an externally imposed economic incentive may present a risk of motivation crowding out (Festré, and Garrouste 2015), and discourage ecosystem conservation when the compensation disappears (Ezzine-De-Blas et al. 2015).

Table 4.

Indexes of disturbance, sensitivity, adaptability and resilience across adaptation typologies (mean ± standard deviation)

Adaptation typologies	Description	Disturbance	Sensitivity	Adaptability	Resilience
1. Reclamation of retired lands (N = 30)	Breaking the grazing prohibition and planning to reclaim the retired lands to agricultural use if payments cease. Average farm size is 1.86 ha per household with 68% compensated by 3069 RMB for ecosystem services in 2013. Income activities are mixed with non-farm work (38%), open-field cropping (22%), horticulture (18%) and livestock (14%)	48.786 ± 13.206	47.357 ± 17.955	30.310 ± 14.340	46.763 ± 21.780
2. Contractive farming (N = 133)	Reducing farm efforts after retiring steep lands and non-farm diversification if payments cease. Average farm size is 1.35 ha per household with 63% set aside by an average payment of 1883 RMB in 2013. Household incomes are mainly from non-farm work (59%), open-field cropping (16%) and horticulture (17%)	47.118 ± 9.406	36.004 ± 17.209	25.968 ± 13.161	48.090 ± 15.536
3. Expansive farming (N = 79)	Expanding farm efforts and maintaining farm production and steep-land retirement even if payments cease. Average farm size is 1.39 ha per household with 60% paid by 2078 RMB in 2013. The major source of household income is horticulture (43%), non-farm work (30%) and open-field cropping (15%)	49.409 ± 6.064	41.894 ± 17.117	31.728 ± 13.245	50.056 ± 14.628
All (N = 242)		48.072 ± 9.083	39.334 ± 17.666***	28.461 ± 13.567***	48.568 ± 16.118

*, **, *** = 0.1, 0.05 and 0.01 levels of significance in differences suggested by non-parametric test

Table 5.

Features of farming systems across adaptation typologies (mean ± standard deviation)

Variable	Description	Adaptation typologies			All (N = 242)
		Reclamation of retired lands (N = 30)	Contractive farming (N = 133)	Expansive farming (N = 79)
Land
Open-field size	Total area of open field with crops, ha	0.274 ± 0.352	0.158 ± 0.252	0.174 ± 0.229	0.177 ± 0.261
Horticulture size	Total area of horticulture with greenhouse, ha	0.079 ± 0.099	0.048 ± 0.078	0.184 ± 0.220	0.096 ± 0.155***
Orchard size	Total area of orchard, ha	0.165 ± 0.271	0.162 ± 0.305	0.196 ± 0.345	0.173 ± 0.314
Farmland rented	Total area of farmland rented, ha	0.087 ± 0.226	0.021 ± 0.077	0.110 ± 0.189	0.058 ± 0.151***
Farmland rented out	Total area of farmland rented out, ha	0.052 ± 0.084	0.104 ± 0.158	0.008 ± 0.043	0.066 ± 0.131***
Laboura
Casual employment	Total active workforce (Appendix S2) in casual employment, man-day	30.500 ± 49.517	67.675 ± 111.569	36.326 ± 57.419	52.833 ± 91.939
Contractual employment	Total active workforce in contractual employment, man-day	178.500 ± 229.212	214.305 ± 233.750	124.241 ± 196.366	180.465 ± 224.493**
Horticultural labour	Total active workforce (family and hired) in horticulture, man-day	96.667 ± 113.937	62.820 ± 111.186	158.443 ± 111.164	98.231 ± 119.220***
Orchard labour	Total active workforce (family and hired) in orchard, man-day	36.000 ± 79.333	38.541 ± 72.038	45.215 ± 76.894	40.405 ± 74.332
Livestock labour	Total active workforce (family and hired) in livestock, man-day	55.000 ± 62.187	9.549 ± 24.328	11.772 ± 29.254	15.909 ± 35.842***
Open-field labour	Total active workforce (family and hired) in open-field cropping, man-day	27.667 ± 33.127	16.504 ± 24.410	15.380 ± 16.404	17.521 ± 23.672
Capital
Rent	Annual outlays for rented land, RMB	202.667 ± 569.082	42.030 ± 230.664	640.278 ± 1354.968	257.240 ± 858.103***
Irrigation	Annual outlays for irrigation water, electricity and equipment, RMB	132.667 ± 174.261	86.316 ± 263.668	236.076 ± 257.469	140.950 ± 260.401***
Mulch and bag	Annual outlays for plastic sheeting or organic residues covered on field and bag wrapping fruits, RMB	349.933 ± 948.900	311.026 ± 693.589	638.696 ± 1150.847	422.816 ± 907.376***
Tillage and ploughing	Annual outlays for ploughing, tillage, and machine harvesting, RMB	160.400 ± 141.351	173.053 ± 191.400	324.063 ± 395.976	220.781 ± 280.044***
Non-farm investment	Annual outlays for non-farm work, RMB	639.333 ± 2213.272	4792.180 ± 18 036.014	2840.759 ± 7319.820	3647.025 ± 14 075.683*
Animal feed	Annual outlays for animal feed, RMB	1662.000 ± 5700.116	546.917 ± 5247.477	458.228 ± 3398.651	656.198 ± 4782.836***
Fertilizer	Annual outlays for fertilizers, RMB	3145.833 ± 2533.899	1649.955 ± 186.211	3420.595 ± 2011.017	2413.413 ± 2184.461***
Pesticides	Annual outlays for pesticides, RMB	408.333 ± 383.157	342.647 ± 414.630	671.456 ± 482.262	458.128 ± 457.917***

*, **, *** = 0.1, 0.05 and 0.01 levels of significance in differences suggested by non-parametric tests

^aPlease see the Electronic supplementary material for details on calculation

Table 6.

Determinants of resilience indexes under the Grain for Green Project

Variables	Description	Descriptive		Resilience	Adaptability	Sensitivity	Disturbance
		Mean	Std. deviation
Household size	Number of household members	3.562	1.201	2.086 (3.375)***	3.554 (4.3331)***	− 0.287 (− 0.338)	− 0.999 (− 1.595)
Age	Age of head of household	48.116	9.322	− 0.921 (− 1.417)	0.039 (0.045)	0.958 (1.071)	− 0.419 (− 0.635)
Schooling	Number of years the head of household spent at school, in years	5.872	3.388	0.082 (0.136)	0.012 (0.014)	− 1.868 (− 2.265)**	0.443 (0.729)
Experience	Number of years of experience in current occupation in terms of horticulture, orchard, livestock or non-farm work, in years	13.438	6.704	1.513 (2.461)**	2.254 (2.761)***	0.922 (1.090)	− 1.018 (− 1.635)
Secure tenure	Perception of holding a secure tenure of set-aside land or not, 0/1	0.343	0.476	0.347 (0.537)	0.268 (0.313)	− 0.882 (− 0.993)	0.669 (1.023)
Access for female	Females participated in training and official meetings representing the household or not: 1/0	0.781	0.414	1.142 (1.976)**	1.963 (2.558)**	0.621 (0.781)	− 0.091 (− 0.156)
Off-farm training	Access to off-farm training or not: 1/0	0.438	0.497	1.403 (2.302)**	0.824 (1.018)	− 2.514 (− 2.997)***	− 0.082 (− 0.133)
Agronomic services	Access to agronomic services, such as pest control, or not: 1/0	0.310	0.463	− 0.074 (− 0.121)	1.659 (2.041)**	1.645 (1.952)*	0.129 (0.208)
Farming machine	Present value of farming machine (e.g. tractors) that a household had, in RMB	9226.277	18 127.044	1.885 (2.878)***	3.024 (3.477)***	1.978 (2.194)**	0.191 (0.287)
Livestock holdings	Stock of livestock animals that a household had, in standard livestock units (Appendix S3)a	0.696	2.441	− 1.045 (− 1.735)*	− 0.225 (− 0.281)	2.005 (2.418)**	0.265 (0.434)
Orchard	Present value of orchards (e.g. apple orchard) that a household had, in RMB	310.083	768.540	1.465 (2.340)**	1.580 (1.900)*	0.611 (0.709)	0.596 (0.940)
Greenhouse	Present value of greenhouse that a household had, in RMB	11 933.884	45 303.721	0.962 (1.585)	2.118 (2.629)***	− 1.448 (− 1.734)*	− 0.453 (− 0.737)
Land fragmentationa	0–1, 0 means that a household farms a single, contiguous plot of the land fragment and that all farmland is completely consolidated	0.610	0.428	− 2.873 (− 4.060)***	0.491 (0.523)	4.994 (5.129)***	2.601 (3.628)***
Access to irrigation	Access for arable land to irrigation water or not: 1/0	0.781	0.414	2.390 (3.781)***	1.992 (2.374)**	− 2.775 (− 3.191)***	− 0.976 (− 1.525)
GGP subsidy	Ratio of GGP subsidy to the food expense of a household in 2013	0.231	0.211	− 0.171 (− 0.259)	0.114 (0.129)	3.923 (4.306)***	− 1.887 (− 2.813)***
GGP duration	Duration of households participating in the GGP, in years	12.719	2.709	0.274 (0.458)	0.078 (0.099)	− 0.324 (− 0.393)	− 1.028 (− 1.695)*
Constant				39.173	28.461	39.334	48.073
Degrees of freedom				16	16	16	16

*, **, *** = 0.1, 0.05 and 0.01 levels of significance, respectively; figures in parenthesis indicate t-values; observations: 242, resilience model with F = 9.867; prob > F = 0.0001; R² = 0.412; adaptability model with F = 7.371; prob > F = 0.0001; R² = 0.344; sensitivity model with F = 19.768; prob > F = 0.0001; R² = 0.584; Disturbance model with F = 2.431; prob > F = 0.002; R² = 0.147

^aPlease see the Electronic supplementary material for collinearity test (Table S7) and model tests (Tables S8, S9 and Figure S1)

The livelihood resources were represented in five forms which often complement each other, and a balance between them can increase the ability of farm households to respond to shocks (Jacobs et al. 2015). Human capital refers to the skills, education, labour ability, and labour availability of farm households (Scoones 1998). Household size and household head’s age, education, and experience in current occupation (e.g. horticulture, orchard, livestock, or non-farm work) were hence employed (Canagarajah et al. 2001; Dzanku 2015; Gautam and Andersen 2016). The labour availability and ability of farm households may increase the value of livelihood resilience and affect the adaptation adoption. Social capital that encompasses networks, informal safety nets, and access to services (Adato and Meizen-Dick 2002) was measured by land tenure security, access to agronomic services and off-farm training, and access for female members to training programmes (Pagiola et al. 2005; Groom et al. 2008; Bennett et al. 2011; Mullan et al. 2011; Li et al. 2016a). Secure tenure and networks may increase trust and ability to work enhancing resilience. Physical capital was reflected from basic infrastructure and production equipment such as farming machine, orchard, greenhouse, and animal stock (Ellis 2000; Dercon 2002; Nix 2003). Better infrastructure and agricultural implements are associated with livelihood adaptation and resilience building. Natural capital that deals with natural resources such as water resources, soil fertility, and land quality and quantity (Campbell et al. 2001), was observed from access for arable land to irrigation water and a Simpson index describing fragmentation of landholdings (Appendix S5). Lack of irrigation and land fragmentation are major obstacles to agricultural productivity growth and food security in China (Brown and Halweil 1998; Tan et al. 2006; Monchuk et al. 2010). Irrigation availability would increase resilience values, whereas land fragmentation might impair resilience building. With respect to financial capital, GGP compensation subsidies were emphasized. It was calculated as the ratio of a household’s annual compensation to its food expense, in order to reflect the subsidy effect on reducing households’ survival stress for developing alternative livelihoods (Nolan et al. 2008) and deal with the variation of subsidy levels by locale (Rodriguez et al. 2016). In addition, the duration of farm households participating in GGP PES was used to control the effect of the different amount of compensations over various participating time on the results.

Results and discussion

Putting livelihood resilience central allows us to investigate livelihood viability in relation to PES programmes, in the diversity across farm adaptation (Table 4). Farm households’ access to resources associated with differentiated values of livelihood resilience (Table 6) and adaptation propensities (Table 7). This provides implications to fine-tune PES schemes, so that ES can be safeguarded to sustain rural livelihoods.

Table 7.

Marginal effects of reporting the determinants of adaptation typologies under the Grain for Green Project

Variables	Description	Descriptive		Reclamation of retired lands	Contractive farming	Expansive farming
		Mean	Std. deviation
Household size	Number of household members	3.562	1.201	0.061** (0.024)	− 0.060 (0.044)	− 0.002 (0.040)
Age	Age of head of household	48.116	9.322	0.009 (0.026)	− 0.020 (0.045)	0.011 (0.042)
Schooling	Number of years the head of household spent at school, in years	5.872	3.388	− 0.027 (0.023)	0.040 (0.042)	− 0.013 (0.039)
Experience	Number of years of experience in current occupation in terms of horticulture, orchard, livestock or non-farm, in years	13.438	6.704	0.007 (0.023)	0.049 (0.041)	− 0.057 (0.040)
Secure tenure	Perception of holding a secure tenure of set-aside land or not, 1/0	0.343	0.476	0.035 (0.026)	− 0.088** (0.042)	0.053 (0.038)
Access for female	Females participated in training and official meetings representing the household or not: 1/0	0.781	0.414	0.021 (0.025)	− 0.047 (0.040)	0.026 (0.038)
Off-farm training	Access to off-farm training or not: 1/0	0.438	0.497	− 0.035 (0.023)	0.086** (0.040)	− 0.051 (0.037)
Agronomic services	Access to agronomic services, such as pest control, or not: 1/0	0.310	0.463	− 0.003 (0.026)	− 0.069* (0.041)	0.072** (0.036)
Farming machine	Present value of farming machine (e.g. tractors) that a household had, in RMB	9.226.277	18.127.044	− 0.040 (0.027)	− 0.185*** (0.060)	0.145*** (0.050)
Livestock holdings	Stock of livestock animals that a household had, in standard livestock units (Appendix S3)a	0.696	2.441	0.168*** (0.053)	− 0.254** (0.120)	0.085 (0.102)
Orchard	Present value of orchards (e.g. apple orchard) that a household had, in RMB	310.083	768.540	− 0.051 (0.045)	0.065 (0.051)	− 0.014 (0.040)
Greenhouse	Present value of greenhouse that a household had, in RMB	11.933.884	45.303.721	− 0.004 (0.038)	0.013 (0.040)	− 0.008 (0.032)
Land fragmentationa	0–1, 0 means that a household farms a single, contiguous plot of the land fragment and that all farmland is completely consolidated	0.610	0.428	− 0.030 (0.029)	− 0.075 (0.047)	0.105** (0.042)
Access to irrigation	Access for arable land to irrigation water or not: 1/0	0.781	0.414	− 0.022 (0.025)	− 0.275*** (0.059)	0.297*** (0.058)
GGP subsidy	Ratio of GGP subsidy to the food expense of a household in 2013	0.231	0.211	0.046** (0.023)	− 0.053 (0.049)	0.007 (0.048)
GGP duration	Duration of households participating in the GGP, in years	12.719	2.709	0.006 (0.031)	0.041 (0.041)	− 0.047 (0.036)

^aPlease see the Electronic supplementary material for collinearity test (Table S7) and model test (Table S10)

*, **, *** = 0.1, 0.05 and 0.01 levels of significance, respectively; figures in parenthesis indicate standard errors; observations: 242, with degrees of freedom = 32; prob > F = 0.0001; Chi square = 119.223

Resilience comparison across adaptation typologies

Adaptation typologies were formed for the dataset, while it shows differences in calculated indexes (Table 4). The score of resilience index was low (48.568 ± 16.118) in the dataset with insignificant differences across the adaptation typologies. It reveals a low level of livelihood resilience of farm households—the low capacity to maintain or improve their livelihoods, though they had derived three adaptation typologies (Fig. 3) in response to the GGP PES intervention. FHSs assigned to ‘reclamation of retired lands’, which were characterized by dependence on GGP subsidy for food and a mixed form of income activities with high engagement in livestock production, got the lowest level of resilience. In contrast, ‘expansive farming’ which combines horticulture and open-field cropping with non-farm work, granted farm households a relatively higher level of resilience. The disturbance induced by GGP PES showed insignificant differences across the adaptation typologies, while significant differences were demonstrated in the index of adaptability and sensitivity. FHSs assigned to ‘contractive farming’ predominantly by non-farm work (59%) had the lowest level of adaptability, while expansive-farming farms got the largest adaptive capacity. With regard to the sensitivity, reclamation farms were most likely affected and contractive-farming farms were the least.

Fig. 3.

Adaptation typologies of farm-household livelihoods in response to the GGP PES intervention

In addition, the land use structure in terms of land, labour, and capital differed significantly across the adaptation typologies (Table 5). The expansive farming had the largest horticulture size (0.184 ha) and the biggest area of farmland rented (0.110 ha), while the contractive farming was characterized by its smallest farm size but the largest area of farmland rented out (0.104 ha). In terms of family labour distribution, the contractive farming invested the most in contractual employment (214 man-days), while the expansive farming input the most in horticulture (158 man-days). In contrast, reclamation farms had much higher input in livestock production (55 man-days). With regard to capital (US$ 1 was equal to RMB 6.19 in 2013), the expansive farming invested more in rent, irrigation, mulching, ploughing and tillage, fertilizers, and pesticides than the others; contractive farming had the largest investments in non-farm work (4792 RMB) while the reclamation farms invested the most in animal feed (1662 RMB). The variation of land use structure is in line with the defined adaptation typologies, which tested the validity and reliability of our typology analysis.

In our dataset, only 12% of farms were assigned to ‘reclamation of retired lands’, but there is evidence that some farmers were already reverting to or stated preferences to previous cropping and grazing practices. This implies that farmers who have no sufficient livelihood alternatives will be likely to convert lands back to agriculture when subsidies end. This is in accord with the statement of Wang et al. (2010) about farmers’ preference to felling forestation and re-cultivation if the government cash support stopped. The contractive farming accounting for 55% of surveyed farm households was the major farm adaptation type in our dataset. Contractual employment outside agriculture enabled rural households to smooth incomes and compensate for income losses caused by GGP PES (Newman and Canagarajah 2000; Hoang et al. 2014). But non-farm diversification that exacerbates a rural exodus may lead to land degradation and undermine agricultural sustainability (Falkenberg 2016). Moreover, contractual employment is often in urban areas so that the availability and mobility of skilled labour becomes an entry barrier (Du et al. 2005), limiting its effects on rural poverty alleviation and resilience building. In contrast, the expansive farming, which increases farm size by renting land and integrates horticulture with casual employment, gave farmers greater capabilities to cope with the disturbance and to reorganize their livelihoods for economic equilibrium. Given the local context, it can avoid the potential threat that non-farm diversification poses for agricultural sustainability, and help to prevent rural households reverting back to the former vicious cycle between rural poverty and environmental degradation when GGP payments cease (Groom et al. 2008). Nevertheless, decisions and actions on adaptation at the individual and household level is a competitive process, subtly differentiated by context, adaptive capacity, and perception of risk (Osbahr et al. 2010). The complex adaptive patterns of FHSs, and hence their resilience, are influenced by combinations of drivers (Berkes and Ross 2016).

Resource access for resilience building

Linkage of resource access to farm households’ resilience

Resource access needs to be improved to support farmers’ resilience building under GGP PES. The linkage of five forms of resource capital to livelihood resilience is shown in Table 6. Overall, the OLS models significantly fit the data as the P value of the F-test was less than 0.002 with independent variables being statistically significant at levels of 0.1, 0.05, and 0.01. Among these variables, land fragmentation and livestock holdings significantly decreased the score of resilience index. Keeping all other variables constant, land fragmentation reduced the resilience index (ranging from 0 to 100) by 2.873, while livestock holdings reduced the index by 1.045. This can be offset by variables including household size, the experience of the head of household, females’ access to training, off-farm training, and irrigation, as well as possession of agricultural facilities such as farming machine and greenhouse. The results indicate that livestock rearing households with fragmented farmland, limited family labour, and restricted access to irrigation often have low resilience under the implementation of GGP PES. Fragmentation of landholdings and a lack of irrigation often jeopardize agricultural production growth due to the time lost in travel and the efficiency lost in irrigation management and machine use in numerous plots of an irregular shape (Nguyen et al. 1996; Tan et al. 2006). Livestock rearing is an obstacle to livelihood reorganization and agricultural production probably due to the high costs of feed purchases and the grazing constraints imposed by GGP (Zhang et al. 2005; Yao et al. 2010; Li et al. 2018). Also, farmland is more likely to be abandoned in areas with less rural labour nor irrigation access but steeper slopes (Xie et al. 2014). These restrain farmers to have sufficient livelihood alternatives and undermine the livelihood resilience. It can be released by learning experience in current occupation and widening access to agricultural facilities (e.g. farming machine, orchard, and greenhouse) and access to off-farm training, especially for females. The tangible and intangible assets can help rural households overcome the entry barriers of livelihood alternatives (e.g. non-farm employment, horticulture, and orchard cultivation), reducing the inequality between rich and poor and enhancing rural welfare outcomes (Woldenhanna and Oskam 2001). Moreover, a new land management system is required to improve farmers’ access to high-quality farmland by planning and designing land consolidation and land transfer, addressing the existing type of land tenure (Brandt et al. 2002; Carter and Olinto 2003), undervalued farmland prices (You 2012), and imperfect markets (Kung 2002; Grosjean and Kontoleon 2009).

In terms of adaptability index, variables including household size, the experience of the head of household, access to training for females, access to agronomic services and irrigation, as well as possession of farming machine, orchard, and greenhouse, significantly increased the score. With regard to the sensitivity index, schooling, access to off-farm training and irrigation, and possession of greenhouse significantly reduced the score, whereas access to agronomic services, possession of farming machine and livestock animals, as well as land fragmentation and GGP subsidy enhanced the score significantly. In particular, keeping all other variables constant, education level of the head of household, access to off-farm training and irrigation, and the possession of greenhouse decreased the score by 1.868, 2.514, 2.775, and 1.448, respectively. In terms of the disturbance index, keeping all other variables constant, duration and provision of GGP subsidy reduced the score by 1.028 and 1.887 while land fragmentation increased the score by 2.601 at a significant level of 0.01. The results imply that farm households with more family labour and more educated and experienced head had a high level of adaptability and less sensitivity to the GGP PES disturbance. In terms of social capital, access to agronomic services can contribute to farmers’ adaptability as well as access to off-farming training can alleviate livelihood sensitivity to the disturbance for resilience building. This can be explained by the fact that agronomic and skill training enables farmers to develop livelihood alternatives in response to the changes. In particular, access to training for women can help disseminate knowledge and accumulate experience in developing production and building resilience (Li et al. 2016a). With regard to physical capital, orchard and horticulture facilities and farming machines can facilitate the development of intensive farming, reducing livelihood turbulence and enhancing livelihood resilience under GGP PES. In contrast, households with more livestock holdings were restrained by the grazing constraints (Zhang et al. 2005; Yao et al. 2010; Li et al. 2018). In terms of natural and financial capital, it is necessary to increase access to irrigation, reduce land fragmentation, and adjust the scale and magnitude of subsidies to alleviate sensitivity and promote adaptability to the disturbance. Fragmented farmland should be consolidated to avert a gradual reduction in farm size (Niroula and Thapa 2005) and to encourage the use of machinery and appropriate irrigation technologies (Rembold 2003). Although subsidies are a stable income source of local farmers, they seem to undermine farmers’ long-term adaptability and resilience by reducing their motivation. The GGP subsidies should more target disadvantaged farmers rather than political preferences, in order to ensure distributional equity and encourage farmers to seek alternative livelihoods (Rodriguez et al. 2016).

Farm households’ adaptation propensities

Using the same independent variables, a multinomial logistic model explored the determinants of farmers’ adaptation choice. The results explained the likelihood at a rate of 66.1% (Chi square = 119.2; P < 0.0001) with ‘reclamation of retired lands’ being the base comparison group (Table S4). To interpret the results from a practical standpoint, marginal effects are reported in Table 7. Farm households with more livestock holdings, household members, and larger share of GGP payments on food expense were more likely to take up land reclamation than contractive or expensive farming. The propensity would increase by around 17, 6, and 5% points, if the stock of livestock animals, the number of household members, and the ratio of GGP subsidy to food expense increased by 1 unit, respectively. It implies that animal husbandry on remained farmland was an inefficient livelihood alternative given the current GGP PES constraints on grazing and cropping. One study carried out in the same province found that more farmers stated the propensity for land reclamation in intensively disturbed areas than in moderately disturbed or nature reserve areas (Zhang et al. 2013). This reminds that the intervention of PES schemes would generate undesired or negative outcomes, due to the problems like rent-seeking, unequal bargaining power, and volatility of payments (Kronenberg and Hubacek 2016). In China’s Loess Hills, farm households’ propensity for animal husbandry and land reclamation can be explained by path dependence. The set of decisions farmers face to adapt their livelihoods is limited by the decisions they have made or experienced prior to the GGP PES due to the restricted resource access and lack of promotion. Subsidies that compensate farmers’ economic loss facilitating the GGP implementation for ES, may allow farmers especially the poor to depend on past knowledge trajectory and livelihood strategies. The socio-ecological system might hence revert back to the former cycle between poverty and ecosystem degradation. This highlights that the current scheme in terms of government cash support cannot maintain the ecosystem conservation achievement while reducing rural poverty in long term. GGP PES should integrate land conservation efforts with an optimized share of land set aside to farmland as well as a careful support plan for farmers to undertake livelihood adaptation.

Adaptation typologies with higher values of resilience need to be promoted by GGP PES for a sustainable livelihood transition. Farm households who inclined to adopt intensive farming often had access to irrigation and agronomic services as well as a farming machine in fragmented farmland. As shown in Table 7, access to irrigation and access to agronomic services increase the propensity by 30 and 7% points, respectively. Farm households that had more farming machines and fragmented farmland were more likely to take up expansive farming by 15 and 11% points, respectively. In comparison, the lack of irrigation, agronomic service provision or farming machines, as well as the owning of off-farm training or insecure land tenure increase farm households’ propensity for contractive farming along with non-farm diversification. Access to off-farm training increases the propensity by 9% points, whereas access to irrigation, secure land tenure, and access to agronomic services decrease the propensity by 28, 9, and 7% points. In addition, 1 unit of farming machines or livestock holdings would decrease the propensity for contractive farming by about 19 and 25% points. In China’s Loess Hills where rural financial markets are imperfect, off-farm income can partly be invested in agriculture promoting farm production and household income. Nevertheless, non-farm diversification may lead to cropland abandonment and impair agricultural production due to the labour competition between farming and off-farm works. Given the rich labour pool but low education level shown in Table 7 (average household size is 3.6 but schooling year is only 5.9 years), promotion of education and training regarding agronomy, machine use, and off-farm skills can improve labour productivity and hence reduce labour competition. It can hence offset the undesired effects of non-farm diversification on rural development. The results also show that provision of irrigation can dramatically direct farm households’ decision between expansive farming and contractive farming in that semi-arid hilly area. Secure land tenure and consolidation of the remaining fragmented croplands can facilitate expansive farming in smallholder agriculture through increasing machine use and family labour productivity. The previous study also claimed to improve the skills of migrant farmers via continuing education, reinforce employment guidance, and improve the management of labour markets while constructing a mature trade market for farmers to tap into and stimulating rural financial systems (Wang et al. 2010). A similar study in Vietnam reminded that insecure land tenure, high transaction costs, and high opportunity costs reduce the benefits of PES in long term and undermine the livelihood viability of local households (To et al. 2012).

Institutional constraints and uncertain system dynamics need to get tackled to implement adequate strategies such as ecosystem stewardship in the Anthropocene (Hansen 2014; Hermanns and Li 2018). The government ought to improve the institutional framework and functioning mechanism of factor markets with regard to rural labour, land, and finance, in aligning with the GGP PES scheme as well as a technical support plan for livelihood adaptation. The GGP PES scheme should contain a variety of payment options apart from money (e.g. agricultural investment plans and continuous education and training); payments should be only given for the additional provision of services at the minimum costs (Engel et al., 2008; Ezzine-De-Blas et al. 2015; Burivalova et al. 2019). Differentiated payments by efforts and performances along with transparent control may limit free-riding and increase participants’ feeling of competence and fairness (Dickinson and Villeval 2008; Hamann et al. 2018). This can promote the motivation of individuals for adopting conservation practices and increasing conservation effort. In addition, to align cooperation and collective action can prevent individuals from free-riding on others’ attempts to sustainably manage and use the shared ecosystem services (Hamann et al. 2018). The above-mentioned approaches may improve the efficiency and effectiveness of the GGP PES scheme, reduce the inequality between groups of people, and promote the livelihood resilience of FHSs.

Conclusions

We can conclude that the PES schemes for ES bring about opportunities for reducing poverty while promoting environmental conservation, but also generate certain disturbance or perturbation for livelihood viability. The PES schemes should promote efficient livelihood alternatives and enrich the poor in resources to enable adaptation towards resilience and sustainability. Farmers with resource equity and efficient livelihood alternatives would willingly choose to engage in the stewardship of ecosystems and are less likely to reclaim retired lands to agricultural use after subsidies end. On the other hand, if disadvantaged farmers were not effectively targeted or farmers had limited access to efficient livelihood alternatives, this would be a deterrent for the resilience building of rural livelihoods and the success of PES programmes.

In the current context of China’s Loess Hills, animal husbandry is not an efficient livelihood alternative. It gained a low degree of resilience and a high likeliness of reversing GGP PES and reclaiming the retired lands after payments cease. Horticulture farming integrated with non-farm employment, which is dominant in the adaptation of “expansive farming”, seems to be an efficient alternative. It can provide farmers with a stable income source and high ability to remain livelihoods on the farm to withstand the GGP PES disturbance. Nevertheless, the score of resilience index was low in the dataset and needed to be enhanced. It is vital to improving the endowment of tangible (e.g. farming facilities) and intangible resources (e.g. skill training), supporting farmers to undertake adequate adaptive actions and strategies. In addition, it requires to strengthen the rationale for the rural development under PES programmes by following the coupled social–ecological systems approach and resilience thinking.

This paper attempted to quantitatively value livelihood resilience and to explore the possibility towards improved resilience, with specific attention to the interactions between farm adaptation and PES disturbance. It may facilitate the understanding of PES impacts on livelihood viability as well as the planning and governance on ecosystem conservation and rural development. However, the methodology of valuing resilience needs to be crosschecked and further improved. It is necessary to conduct a dynamic analysis across multiple scales, especially addressing the dynamic interactions between adaptability and sensitivity and disturbance, in near future with quality panel data. Also, there is a need for evidence of how different PES schemes influence human behaviour for socio-economic values and ecosystem service benefits, and whether resource equity can reduce socio-economic inequality and thus promote ecosystem stewardship actions and resilience-building strategies.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Biographies

Qirui Li

is a research associate at Leibniz Institute of Ecological Urban and Regional Development (IOER), Germany. His research and publications concern farm economics and ecosystem services, livelihoods resilience of social–ecological systems, urban–rural transformation, and impact assessment of land use change.

Peter Zander

is head of the research area “Economics of Sustainable Land Use” within the Institute of Socio-Economics at Leibniz Centre for Agricultural Landscape Research (ZALF), Germany. His research interests are economics of sustainable land use, bio-economic modelling of farming systems, technology, and policy impact assessment.