Multilevel analysis of factors affecting participants’ land reconversion willingness after the Grain for Green Program

Abstract

Understanding the postprogram land use plans of participants is necessary for the sustainability of the conservation achievements from payments for ecosystem services (PES) programs. Previous studies have analyzed many individual factors affecting participants’ reconversion plans after PES programs. However, whether the regional ecosystem services changes caused by PES programs affect reconversion willingness remains elusive. Here, we used the multilevel linear model to determine the effects of regional ecosystem services changes and individual characteristics on participants’ land reconversion willingness after the Grain for Green Program (GFGP) in the Yanhe watershed of the Loess Plateau. We found that household income, ecological awareness, and employment changes negatively affected reconversion willingness, while nonfarm employment positively affected it at the individual level. At the regional level, the grain production and water yield changes could influence the reconversion willingness of respondents with different individual characteristics. With improved understanding of the factors affecting reconversion willingness, several suggestions to improve the sustainability of the GFGP were proposed. Our study provides a template for analyzing the multilevel factors that affect the sustainability of other PES programs.

Introduction

Human activities have transformed much of the world’s natural land cover (Foley et al. 2005; Song et al. 2018), causing environment degradation and biodiversity loss worldwide (Vitousek et al. 1997; Newbold et al. 2015). Payments for ecosystem services (PES), which are programs that exchange value for land management practices aimed at providing or ensuring ecosystem services (Salzman et al. 2018), arose from the desire to address the trade-offs between ecosystem conservation and socioeconomic development and has become a popular conservation paradigm for bridging conflicts between ecosystem services users and providers (Wunder et al. 2018). Recent decades have witnessed a great increase in the number of PES programs for biodiversity conservation, water purification, and carbon sequestration operating at different scales around the world (Salzman et al. 2018), creating the need to analyze the ecological, economic, and social outcomes and sustainability of PES programs (Tallis et al. 2008; Yang and Lu 2018).

Compared with the outright purchase of land, many PES programs are short term with uncertainty about land use after the programs cease, which may only bring temporary conservation benefits (Chen et al. 2009b). Previous studies have shown that when PES program payments cease, the lands enrolled in the programs are at risk of being converted to crop production and other land use types in some areas, and conservation achievements cannot be sustained without continued conservation payments (Johnson et al. 1997; Roberts and Lubowski 2007; Cao 2011; Song et al. 2014). Considering that farmers are important stakeholders and the implementers in PES programs (Page and Bellotti 2015), studies on the postprogram land use plans of the farmers participating in PES programs and understanding the factors affecting their plans could help design specific policies and strategies to enhance the sustainability of PES programs (Chen et al. 2009b; Deng et al. 2016).

Previous studies have analyzed the factors affecting participants’ reconversion plans after PES programs cease from many aspects. Past studies have suggested that the postprogram land use plans of participants can be determined by their demographic and socioeconomic characteristics and the conditions of the land. Although they differ in different case studies, factors such as gender, education, age, the number of family members, the laborers in the family, economic income, cropland area, the distance from the household to the land, and land yields are found to be potentially related to reconversion plans (Johnson et al. 1997; Cooper and Osborn 1998; Cao et al. 2009). Other studies found that the duration and subsidies of the program should also be considered (Chen et al. 2009b; Yang and Xu 2014). At the neighborhood level, the social norms such as the neighbors’ behavior also have impacts (Chen et al. 2009a). The effects farmers’ attitude toward ecological achievements of PES, perceived social pressure from neighbors, and perceived behavioral control on famers’ intentions were analyzed using the theory of planned behavior (Deng et al. 2016). However, these studies mainly focused on the individual level. Since PES programs also improve regional ecosystem services, whether the regional ecosystem services changes caused by PES programs affect reconversion willingness remains elusive. Studied about how the regional ecosystem services changes and individual characteristics affect reconversion willingness across multiple scales are needed.

China has implemented a number of PES programs for nationwide environmental protection (Yang and Lu 2018), among which the Grain for Green Program (GFGP) is one of the world’s biggest PES programs (Yang and Lu 2018; Peng et al. 2019). The GFGP has been implemented since 1999, as Chinese government’s reaction to a series of major droughts and floods in the late 1990s (Salzman et al. 2018). In addition to the main objective of conserving ecosystem services such as reducing soil erosion by increasing vegetation (Ouyang et al. 2016), it also aimed to subsidize rural household incomes and support rural development (Ingram et al. 2014; Li and Zander 2020). From 2000 to 2009, 120 million farmers and 32 million households were paid by the program (Ouyang et al. 2016). After the implementation of the GFGP, the supply of major ecosystem services increased, and its socioeconomic effects are mostly positive (Liu et al. 2008; Lu et al. 2012; Ouyang et al. 2016; Bryan et al. 2018). Given its tremendous size and huge number of participants in the GFGP, the attitude and behavior of the participants when the payments cease will strongly influence the sustainability of the program’s environmental achievements (Cao et al. 2009). Therefore, understanding the influencing factors of participants’ land reconversion willingness is important to provide support for policy making.

Here, we selected the Yanhe watershed (Fig. 1) in the Loess Plateau (LP) of China, where the GFGP has been most intensively implemented (Wu et al. 2019a), to analyze how multiple-scale factors affect reconversion willingness. Our objectives are to determine whether the changes in the regional ecosystem services affect reconversion willingness and to analyze the effects of regional ecosystem services changes and individual characteristics on participants’ land reconversion willingness after the GFGP. To do this, we used the multilevel linear model to analyze the effect of 10 variables at the individual level and 4 variables at the regional level on participants’ reconversion willingness. On the basis of our results, several suggestions to improve the sustainability of the GFGP were provided.

Fig. 1.

Study area. a Location of the Yanhe watershed. b Land use and land cover map of the Yanhe watershed in 2000. c Land use and land cover map of the Yanhe watershed in 2015

Materials and methods

Study area

The Yanhe watershed (36° 21′–37° 19′ N, 108° 38′–110° 29′ E), which covers an area of 7725 km², is located in northern Shaanxi Province at the middle of the LP. This watershed comprises Zhidan, Ansai, Baota, and Yanchang counties of Yan’an City. The elevation of this region varies from 495 to 1975 m. This region is covered by thick loess that is weakly resistant to erosion. The average annual precipitation of the Yanhe watershed is 495 mm and 65% of precipitation falls from June to September. The average annual temperature varies from 8.8 to 10.2 °C (Su et al. 2012). Because of the concentrated rainfall pattern and erosion-prone soil, the Yanhe watershed used to suffer severe soil erosion. To solve this issue, it was selected as a pilot area for the GFGP in 1999. Large scale revegetation measures have been implemented in this watershed, leading to great changes in its land use (Fig. 1). Croplands used for planting corn and wheat were converted to forest and grassland. The duration of subsidies from the GFGP depends on what the cropland is converted to: 2 years for grassland, 5 years for economic forest, and 8 years for ecological forest. After the first phase of the GFGP, the government subsidies are renewed in a second phase, in which the duration of subsides for different cropland conversion is the same as that in the first phase (Wu et al. 2019b). After the implementation of the GFGP, the vegetation coverage of the Yanhe watershed increased, and major ecosystem services improved (Wu et al. 2018). The livelihood activities of the participants in the GFGP were also transformed, changing from farming to nonfarm work such as construction, transportation, and small restaurant and retail business (Wu et al. 2019b).

Household survey

The individual level data were obtained from sampling household surveys in the Yanhe watershed in 2018. The pre-survey was conducted in late July 2018 and the formal survey was conducted by well-trained graduate students in August 2018 as face-to-face interviews with participants in the GFGP. The survey was carried out in 65 randomly selected villages in Zhidan, Ansai, Baota, and Yanchang counties of Yan’an City (Fig. 1). Overall, we collected 293 questionnaires, from which 276 were considered valid after excluding the incomplete questionnaires.

Our questionnaires contained two parts. The first part recorded the demographic and socioeconomic characteristics of the respondents, including age, gender, education level, the number of laborers in the household, employment structure, income, the proportion of cropland enrolled in the GFGP, and their willingness to reconvert these lands (defined as whether they would like to convert the lands enrolled in the GFGP to cropland when the program cease) (Table 1). These variables were selected based on existing studies that proved that they could affect the reconversion willingness (Chen et al. 2009b; Yuan et al. 2017). The second part recorded the respondents’ attitudes to 10 different statements (Fig. 2). These questions can be divided into three categories: employment and income changes in the respondents’ household (Q1–3), respondents’ perception of environment changes (Q4–7), and respondents’ ecological awareness (Q8–10). The attitudes of the participants were proved to affect their behavior and then the success of the GFGP (Cao et al. 2009; Deng et al. 2016).

Table 1.

Demographic and socioeconomic characteristics of the sample

Variables	Description	Mean	Standard deviation
Age	In years	54.33	9.94
Gender	1 = Male; 0 = female	0.69	0.46
Education	In years	7.7	2.34
Laborers	Number of laborers in the household	2.08	1.04
Nonfarm employment	Ratio of the number of nonfarm laborers to the number of laborers in the household	0.73	0.38
Income	Log-transformed household income (unit: yuan)	10.26	0.87
GFGP	Proportion of cropland enrolled in the GFGP in the household	0.72	0.22
Reconversion willingness	1 = Yes; 0 = no	0.18	0.39

Fig. 2.

Attitudes of the respondents. Response scale “1” to “5” represent “strongly disagree”, “disagree”, “neutral”, “agree”, and “strongly agree”, respectively

Factor analysis and multicollinearity diagnostics

To reduce the number of variables about respondents’ attitudes, we used factor analysis, which is a widely used statistical method that searches for a potentially lower number of independent factors that reflect the variations of a set of observable variables (Wei et al. 2019). The number of factors was determined by a Scree test and the interpretability of the derived factors, and we extracted the first three orthogonal factors (with varimax rotation), which explain 60.15% of the total variance. According to the rotated component matrix (Table 2), the three factors mainly reflect the respondents’ ecological awareness, perception of employment changes, and perception of environment changes.

Table 2.

Rotated component matrix (only coefficients larger than 0.1 are shown)

	Factor 1	Factor 2	Factor 3
Q1		0.862
Q2	− 0.124	0.834	0.132
Q3	0.329	0.548
Q4	0.580		0.322
Q5		0.246	0.784
Q6	0.106		0.865
Q7	0.516		0.433
Q8	0.740		− 0.137
Q9	0.776		0.160
Q10	0.756		0.101

A total of 10 variables, including 7 about demographic and socioeconomic characteristics and 3 about attitudes, were selected to predict the reconversion willingness of respondents at the individual level (Table 3). To test the multicollinearity, we conducted multicollinearity diagnostics using the variance inflation factor (VIF). The VIF of each variable is less than 1.5, which means that the multicollinearity can be ignored and the variables are not related.

Table 3.

The relationships between reconversion willingness and individual characteristics

	Regression coefficient
Intercept	4.643
Individual level
Age	− 0.009
Gender	0.179
Education	− 0.081
Laborers	− 0.294
Nonfarm employment	0.843*
Income	− 0.54**
GFGP	0.122
Ecological awareness	− 0.362**
Employment change	− 0.358**
Perception of environment change	− 0.013
Intraclass correlation coefficient (ICC)
Region level variance proportion	6.7%

**p < 0.05 and *p < 0.1

Regional ecosystem services changes

We chose four ecosystem services that have been usually studied in the LP and the Yanhe watershed (Lu et al. 2012; Wu et al. 2018, 2019a), carbon sequestration, grain production, soil retention, and water provision, to calculate the ecosystem services changes at the regional level. The quantified ecosystem services data of the Yanhe watershed in 2000 and 2015 were derived from the assessment by Wu et al. (2018). They used the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model to quantify water provision (described as water yield) and sediment retention services, used biophysical models to assess agricultural production, and used net primary production (NPP) to assess carbon sequestration. They found that total amount of water yield and agricultural production of the Yanhe watershed declined by 12% and 87%, while soil retention and carbon sequestration increased by 13% and 3% from 2000 to 2015. According to the methods of previous study (Ye et al. 2018), the means of each ecosystem service within a 5 km buffer zone of a respondent’s village location were used to represent the regional ecosystem services provided to residents. The change rates of each ecosystem service from 2000 to 2015 were then calculated.

Multilevel linear model

The multilevel linear model (MLM) is a method to process hierarchically structured data (Goldstein 2011). The variation of the dependent variables in this model can be explained partly by the variation at the individual level and partly by the variation at the group level (Zhang et al. 2014). In our study, the dependent variable is the respondents’ reconversion willingness, and the explanatory variables include 10 variables about respondents’ characteristics at the individual level and 4 variables about the ecosystem services changes at the regional level.

Based on previous studies (Zhang et al. 2014; Huang et al. 2020), the modeling process started with a null model that only had an intercept. As the respondents’ reconversion willingness is binary variable (Table 1), a logistic model was chosen as a link function between levels, which was defined as follows:

$$\log \left(\frac{p_{\mathit{ij}}}{1-p_{\mathit{ij}}}\right)={\beta }_{0j}+{\epsilon }_{\mathit{ij}},$$ 1

$${\beta }_{0j}={\gamma }_{00}+u_{0j},$$ 2

where $p_{\mathit{ij}}$ is the occurrence probability of reconversion at the first level, i denotes the index at the first level, j denotes the index at the second level, ${\beta }_{0j}$ is the intercept of group j, ${\epsilon }_{\mathit{ij}}$ is the random variation, ${\gamma }_{00}$ is the overall intercept for both levels, and $u_{0j}$ is the group dependent deviation. Using this null model, the variance of the dependent variable can be decomposed into within-group variance (the first level) and between-group variance (the second level).

The intraclass correlation coefficient (ICC), which is calculated via the null model, is important to an MLM. This coefficient ranges from 0 to 1 and is used to estimate the proportion of the variance among group (higher) levels. A lower ICC means a lower degree of similarity of the measurements within a group. If this coefficient is 0, then no hierarchical structure between variables exists, which means there are no differences among groups and traditional models such as the ordinary least squares can be used (Goldstein 2011). The ICC is calculated as follows:

$$\mathrm{ICC}=({\sigma }_{\mathrm{b}}^2)/({\sigma }_{\mathrm{w}}^2+{\sigma }_{\mathrm{b}}^2),$$ 3

where ${\sigma }_{\mathrm{w}}^2$ is the within-group variance at the first level, and is equal to $\frac{{\pi }^2}{3}$ in a logistic model; and ${\sigma }_{\mathrm{b}}^2$ is the between-group variance at the second level.

Explanatory variables were added to the null model to construct an MLM (Zhang et al. 2014). The mixed model that included the two-level explanatory variables was defined as follows:

$$\log \left(\frac{p_{\mathit{ij}}}{1-p_{\mathit{ij}}}\right)={\beta }_{0j}+{\beta }_{1j}{X}_{\mathit{ij}}+{\epsilon }_{\mathit{ij}},$$ 4

$${\beta }_{0j}={\gamma }_{00}+{\gamma }_{01}{R}_j+u_{0j},$$ 5

$${\beta }_{1j}={\gamma }_{10}+{\gamma }_{11}{R}_j+u_{1j},$$ 6

where $X_{\mathit{ij}}$ refers to the variables about respondents’ characteristics at the individual level (first level); ${\beta }_{0j}$ refers to the intercept of the dependent variable in group j; ${\beta }_{1j}$ refers to the regression coefficient in group j between the first-level predictors and the dependent variable; $R_j$ refers to the ecosystem services changes at the regional level (second level); ${\gamma }_{00}$ and ${\gamma }_{10}$ refer to the fixed effects of the second-level intercept and slope; $u_{0j}$ and $u_{1j}$ refer to random variation; and ${\gamma }_{01}$ and ${\gamma }_{11}$ refer to regression coefficients between the second-level predictors and ${\beta }_{0j}$ and ${\beta }_{1j}$, respectively.

Results

Brief summary of the respondents

Table 1 shows the summary of the demographic and socioeconomic characteristics of the respondents. Of our respondents, 69% were male, and the average age of the respondents was 54. Our respondents had low levels of education (mean = 7.7 years). Each surveyed household had an average of 2.08 laborers, and 73% of them worked in nonfarm jobs. The average household income was 41 213 yuan in 2017 (1 USD = 6.75 yuan in 2017). An average of 72% of each household’s cropland was enrolled in the GFGP. Only 18% of the respondents tended to reconvert, which is similar to the results of other studies (Chen et al. 2009b).

As for their attitudes, more than 80% of the respondents said the employment in their household changed after the GFGP. More laborers were released from agriculture and engaged in more nonfarm work; however, only 57.6% of the households had increased incomes (Fig. 2). More than 90% of the respondents agreed that the local environment had improved after the GFGP: soil erosion was controlled, and floods from rivers and the number of days of sand dust decreased. Regarding ecological awareness, 71.0% of the respondents agreed that the environment is as important as the economy, 84.8% of them thought everyone should protect the environment, and 63.4% of them would like to introduce the importance of environmental conservation to others.

Impacts of individual characteristics on land reconversion willingness

The ICC shows that 6.7% of the variance can be attributed to the regional level (Table 3), which indicates that the differences in respondents’ land reconversion willingness could be divided into individual and regional levels, and the MLM could be applied in our study (Huang et al. 2020). The individual characteristics had the greatest impact on respondents’ reconversion willingness and accounted for more than 90% of the variance.

At the individual level, nonfarm employment had a positive effect on reconversion willingness (p < 0.1), which means the households with a high ratio of laborers working in nonfarm jobs were more likely to reconvert the GFGP land to crop production. Income, ecological awareness, and employment changes had negative effects on reconversion willingness (p < 0.05), which means that the respondents with higher household income, higher ecological awareness, and more significant employment changes in their households were more likely to keep the land in the GFGP rather than reconvert it.

Impact of ecosystem services changes on land reconversion willingness

We established the MLM using significant indicators (p < 0.1) at individual level and changes of ecosystem services at regional level, and found that the changes in carbon sequestration, grain production, and water yield at the regional level could influence the reconversion willingness of respondents with different individual characteristics (Table 4). The relationships between the nonfarm employment, income, ecological awareness and reconversion willingness of respondents were affected by the changes of these ecosystem services.

Table 4.

Regression coefficients between ecosystem services changes and slope of reconversion willingness to individual factors

	Regression coefficient
	Carbon sequestration	Grain production	Soil retention	Water yield
Nonfarm employment (slope)	41.22*	− 22.43**	19.68	22.74**
Income (slope)	− 25.75**	7.43	1.62	− 12.36***
Ecological awareness (slope)	− 7.75	− 6.32	4.30	6.19***
Employment change (slope)	2.19	− 3.87	− 8.08	0.26
Intercept	222.77**	− 67.45	− 31.57	107.84***

***p < 0.01, **p < 0.05, and *p < 0.1. Only significant indicators (p < 0.1) at individual level were included in the MLM

To test the effects of ecosystem services changes on regional variation of relationships between individual factors and reconversion willingness, we examined reduction in unconditional model variance as each predictor added (Table 5). Then the proportion of variance in slope of reconversion willingness to each individual factor explained by respective ecosystem service change was calculated. Specifically, the grain production service change could explain 18.9% of the regional variance in reconversion willingness with different nonfarm employment, and the water yield service change could explain 1.9% of the regional variance in reconversion willingness with different incomes (Table 5). An increased grain production service change would weaken the positive impact of nonfarm employment on reconversion willingness, and its regression coefficient is significant at the 0.05 level (Tables 3, 4). An increase in the water yield service change would enhance the negative impact of income on reconversion willingness, and its regression coefficient is significant at the 0.01 level (Tables 3, 4). Some other regression coefficients are also statistically significant (Table 4); however, the explained variance ratios are negative (Table 5), which indicates that they cannot be used to explain the regional variance in reconversion willingness with different individual characteristics (Huang et al. 2020).

Table 5.

Variances in slope of reconversion willingness to individual factors explained by ecosystem services changes

	Unconditional variance	Conditional variance	Explained variance ratio (%)
Carbon sequestration
Nonfarm employment (slope)	1.564	1.671	− 6.85
Income (slope)	0.390	0.511	− 30.87
Intercept	28.731	37.798	− 31.56
Grain production
Nonfarm employment (slope)	1.564	1.269	18.89
Water yield
Nonfarm employment (slope)	1.564	1.643	− 5.03
Income (slope)	0.390	0.383	1.90
Ecological awareness (slope)	0.198	0.231	− 16.69
Intercept	28.731	27.914	2.84

Discussion

Understanding the postprogram land use plans of farmers participating in the GFGP is necessary for the sustainability of the conservation achievements of the GFGP (Cao et al. 2009; Chen et al. 2009b). We used the MLM to analyze the effects of individual characteristics and regional ecosystem services changes on participants’ land reconversion willingness after the GFGP in the Yanhe watershed. Our study shows that although individual characteristics had major impacts on respondents’ reconversion willingness, the differences in the ecosystem services changes at the regional level could explain 6.7% of the total variance.

At the individual level, we found household income, ecological awareness, and employment changes negatively affected reconversion willingness, while nonfarm employment positively affected it (Table 3), which are consistent with previous studies (Chen et al. 2009a; Yang and Xu 2014; Deng et al. 2017). However, the relationships between reconversion willingness and age, gender, education, number of laborers, proportion of cropland enrolled in the GFGP, and perception of environment change are non-significant (Table 3). Our findings indicate that respondents who had lower household incomes, fewer household employment changes, lower household involvement in crop production, and lower ecological awareness were more likely to reconvert their GFGP land back to agriculture. The negative relationships between reconversion willingness and income and employment change (p < 0.05, Table 3) are because the households that had less household income and less abilities, incentives, or opportunities to transfer their labor to nonfarm production had a strong dependence on cropping income (Yang and Xu 2014). Thus, they had a higher propensity to convert GFGP land to cropping. The positive effect of nonfarm employment on reconversion willingness (p < 0.1, Table 3) is mainly because that GFGP lands are usually marginal for growing crops, and the GFGP participants would not reconvert those lands to agriculture as long as they already have adequate land for farming (Chen et al. 2009a). Therefore, the households involved more in crop production and less in nonfarm employment had a lower willingness to reconvert their GFGP land to crop production. In addition to these socioeconomic characteristics, we found that the ecological awareness of respondents also played an important role in reconversion willingness, which is consistent with previous studies that indicate that farmers’ pro-ecological intentions are crucial to the sustainability of ecological conservation achievements (Deng et al. 2017).

Previous ecosystem services assessments showed that grain production and water yields in the Yanhe watershed decreased (Zheng et al. 2016; Wu et al. 2018); thus, the changes of the two services are negative, and increased changes mean that the services decreased less. At the regional level, we found that the positive impact of nonfarm employment on reconversion willingness in areas where grain production decreased less was lower than that in areas where grain production decreased more, which means that the differences in the reconversion willingness of households with different levels of involvement in crop production were less in areas where grain production decreased less. As discussed above, as long as the GFGP participants already have adequate land for farming, they would not reconvert GFGP land to agriculture (Chen et al. 2009a), which may explain the differences between regions with different grain production changes. We also found that the negative impact of income on reconversion willingness in areas where water yields decreased less was higher than that in areas where water yields decreased more, which means that the differences in the reconversion willingness of households with different incomes were less in areas where water yields decreased more. In a water-scarce area such as the LP, crop production is mainly limited by water availability (Kang et al. 2002). Greater decreases in the water yield indicate that the potential benefits from reconverting GFGP land to crop production is relatively low; thus, the differences in the reconversion willingness of households with different incomes were less.

The sustainability of PES programs can be achieved only if participants are willing to maintain conservation benefits even after the programs end (Uchida et al. 2005). Based on the multilevel analysis of the factors affecting participants’ reconversion willingness after the GFGP, policy makers can now improve the sustainability of GFGP from the following aspects. First, diversifying and improving household income are essential to minimizing the risk of reconversion of GFGP land (Bryan et al. 2018). Local governments and policy makers should support the transition of rural labor to nonfarm work to lessen households’ reliance on farm profitability and reduce labor surpluses (Song et al. 2014). This transition can be achieved by creating more local nonfarm jobs (Cao 2011), providing training to develop new labor skills (Yang et al. 2018), providing information services on new employment opportunities (Yin et al. 2014), and providing financial and technical support for labor migration and business (Wu et al. 2019b). Second, since the GFGP is a traditional top-down style program that does not engage farmers regarding their interest and willingness to some degree (Yin and Yin 2010; Song et al. 2014), it is important to enhance the participants’ ecological awareness and recognition of the ecological benefits of the GFGP through a wide range of publicity, training, and education programs to reduce their reconversion willingness after the GFGP. Third, key regions that are at risk of losing sustainability can be targeted by understanding the reconversion willingness of participants (Chan et al. 2006). By analyzing the effects of regional ecosystem services changes on reconversion willingness, place-based solutions and prior protection measures for ecosystem services can be provided.

Our study has several limitations. First, because many young laborers in our study area temporally migrated to cities and other areas for work and men were more willing to accept the interviews during our survey, the average age of the respondents was 54 and male respondents accounted for a high proportion (Table 1), which may affect the representativeness of the samples. Second, due to the data availability, household survey data from 2018 and data of ecosystem services changes between 2000 and 2015 were used in our analysis. The temporal mismatch of these data may influence the accuracy of the estimation (Huang et al. 2020).

Conclusion

Our study used the MLM to determine how regional ecosystem services changes and individual characteristics affected participants’ land reconversion willingness after the GFGP in the Yanhe watershed. We found that nonfarm employment had positive effect on reconversion willingness, while income, ecological awareness, and employment changes had negative effects on it at the individual level. At the regional level, the grain production and water yield changes could influence the reconversion willingness of respondents with different individual characteristics: a lesser decrease in grain production would weaken the positive impact of nonfarm employment on reconversion willingness, and a lesser decrease in water yield would enhance the negative impact of income on reconversion willingness.

Our study provides a template for analyzing the multilevel factors affecting the sustainability of PES programs, which has significant policy implications. With improved understanding of the factors affecting reconversion willingness, we proposed several suggestions to improve the sustainability of the GFGP: diversifying and improving participants’ household income, enhancing the participants’ ecological awareness and recognition of the ecological benefits of the GFGP, and providing place-based solutions and prior protection measures of ecosystem services for key regions that are at risk of losing sustainability.

Biographies

Xutong Wu

is a PhD Candidate in the College of Urban and Environmental Sciences, Peking University. His research interests include ecosystem services and social–ecological systems.

Shuai Wang

is a Professor in the State Key Laboratory of Earth Surface Processes and Resource Ecology, Faculty of Geographical Science, Beijing Normal University. His expertise is focused on human–land coupling system, socio-hydrology, ecosystem services, and ecosystem adaptive management.

Bojie Fu

is an Academician of Chinese Academy of Sciences, Fellow of the Academy of Sciences for Developing World (TWAS), Corresponding Fellow of the Royal Society, Edinburgh, UK, and a Member of American Academy of Arts and Sciences. His research areas are land use and land cover change, landscape pattern and ecological processes, ecosystem services and management. He has published more than 400 scientific papers and 10 books, including Science, Nature Geoscience and Nature Climate Change.