This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

Farmland abandonment has important impacts on biodiversity and ecosystem recovery, as well as food security and rural sustainable development. Due to rapid urbanization and industrialization, farmland abandonment has become an increasingly important problem in many countries, particularly in China. To promote sustainable land-use management and environmental sustainability, it is important to understand the socioeconomic causes and spatial patterns of farmland abandonment. In this study, we explored the dynamic mechanisms of farmland abandonment in Jiangxi province of China using a spatially explicit economical model. The results show that the variables associated with the agricultural products yield are significantly correlated with farmland abandonment. The increasing opportunity cost of farming labor is the main factor in farmland abandonment in conjunction with a rural labor shortage due to rural-to-urban population migration and regional industrialization. Farmlands are more likely to be abandoned in areas located far from the villages and towns due to higher transportation costs. Additionally, farmers with more land but lower net income are more likely to abandon poor-quality farmland. Our results support the hypothesis that farmland abandonment takes place in locations in which the costs of cultivation are high and the potential crop yield is low. In addition, our study also demonstrates that a spatially explicit economic model is necessary to distinguish between the main driving forces of farmland abandonment. Policy implications are also provided for potential future policy decisions.

Changes in land use and land cover are critically important to understanding global climate change, food security, soil degradation, ecosystem dynamics, and human-environment interactions [

Recently, farmland abandonment has received much attention from various disciplines because of its important impacts on biodiversity and ecosystem recovery, as well as food security and rural sustainable development [

Studies have shown that the determinants of farmland abandonment at the local scale include topographical conditions (e.g., elevation and slope) [

In China, the migration of the labor force from rural to urban areas and the decrease in the total agricultural labor force over the past several decades have been well documented [

This study focuses on gaining an understanding of farmland abandonment in the Jiangxi province of China using a spatially explicitly regression model that describes the dynamical economic drivers of agricultural land use change. The model is based on agricultural land rent theory, logistical regression statistical methods, and Geographic Information System (GIS). Land rent theory, a key theory in land economics, has been commonly applied to explain the main motivation for how and why people use land resources and assess land values. Based on this theory, the main hypothesis in this study is that farmland abandonment at different spatial scales can be explained by land rent theory [

The study area (24°7′N–29°9′N, 114°02′E–118°28′E) is the Jiangxi province, which is located in the southern region of China and covers an area of approximately 166,000 km^{2} (^{2}. The study area is surrounded by three mountains with a chain of undulating hills located in the middle portion. Based on the digital elevation model, 54% of the study area is located at elevations <200 m, 33% at 200–500 m, and 23% at >500 m. The soils are predominantly red soil, yellow soil, and hydromorphic paddy soil. The main vegetation types include subtropical evergreen broadleaf forest, coniferous forest, and broadleaf mixed forests. The main crops in the Jiangxi province include rice, wheat, soybeans, and sweet potatoes. In 2010, the crop planting area covered 3 × 10^{6} ha, and the study area had a population of 44.56 million and a food productivity of 4.00 × 10^{7} t. Recently, many people in the study area have participated in non-farm employment due to the development of urbanization and industrialization, and farmland abandonment has become a common phenomenon in the mountainous areas.

Land use data in this study were derived from the Data Center for Resources and Environmental Sciences of the Chinese Academy of Sciences [

Location map of the Jiangxi Province in China.

Land use and land cover of the study area.

Land use/land cover class | Land use/land cover subclass |
---|---|

Farmland | Paddy field |

Dry field | |

Forest | Woodland |

Shrubland | |

Open woodland | |

Grassland | High covered grass |

Medium covered grass | |

Low covered grass | |

Water area | River and trench |

Lake | |

Reservoir | |

Permanent glacier | |

Beach | |

Bottomland | |

Built areas | City or town region |

Village residential area | |

Rest construct land | |

Other covers | Sand land |

Gobi | |

Salted land | |

Swamp | |

Bare ground | |

Bare rock | |

Rest of used land |

The spatial patterns of farmland changes during two periods (1990–1995 and 1995–2005) are shown in

Observed farmland change of two periods in the study area.

Climate, topopgraphy, and soil are considered as the main natural influencing factors that influence farmland abandonment due to differences in agricultural production conditions. Climatic data were derived from the China Meteorological Data Sharing Service System [

The main topographical factors that influence farm production are slope and altitude [

Socio-economical data at the county level were derived from the Jiangxi statistics yearbook from 1990 to 2006 [

The farmer’s choice of agricultural practice on the farmland is determined by the maximization of net income [

If ^{*}^{*}

According to the above formula, the main determinants of farmland abandonment include the yield of agricultural products, price of agricultural products, wage of agricultural labor, capital, cost of enforcing land property rights, transportation cost,

The yield of agricultural products (

Independent variables used in the study.

Variable Description | Spatial Resolution | Expected Sign * |
---|---|---|

Yield of agricultural product( |
||

Cumulative temperature above 10 °C (day × °C) | 100 m | − |

Annual precipitation (mm/year) | 100 m | − |

Distance to forest edge (m) | 100 m | − |

Soil depth (cm) | 100 m | − |

Content of soil coarse sand (%) | 100 m | + |

Slope (°) | 100 m | + |

Elevation (m) | 100 m | + |

Wage of agricultural labor( |
||

Proportion of employees in the primary sector (%) | County | − |

Rural labor force participation rate (%) | County | − |

Rate of change of rural labor (%/year) | County | − |

Rate of population urbanization (%) | County | + |

Proportion of secondary sector’s output value (%) | County | + |

GDP per capita (￥/capita) | County | + |

Transportation cost( |
||

Distance to central town(m) | 100 m | − |

Distance to village (m) | 100 m | − |

Distance to primary road (m) | 100 m | − |

Structural characteristics in agriculture | ||

Net income of farmer per capita (￥/capita) | County | ? |

Average agricultural area per farmer (ha/farm) | County | + |

Rate of change of farmer (%/year) | County | ? |

Rate of change of employees in the primary sector (%/year) | County | ? |

Based on the Formulas (1) and (2), the wage of farming labor variables (

Process of farmland abandonment induced by population urbanization and industrialization.

From

According to the agricultural land rent theory, the transportation cost of agricultural products (

For the other determinants in the land rent equation,

Structural variables were included in the analysis model, not for hypothesis testing, but, rather, to obtain additional information on their relationships to farmland abandonment.

The linear regression model is popular in analytical studies, but it is constrained under many situations, especially if the dependent variable is a categorical variable rather than a continuous variable. On the contrary, the logistic regression model is able to address this problem properly. Multivariate logistical regression models generate regression coefficients that are calculated using certain weighted methods that explain the probability of land use change. Thus far, multivariate logistic regression models have been used to study wildlife habitats [

Multivariate logistic regression can determine the effects and strengths of independent variables _{ki}_{i}_{i}_{i}_{1i}, _{2i}, …, _{ki}) is the occurrence probability of farmland abandonment when given _{1i}, _{2i}, …, _{ki}, α is the intercept and β is the slope.

The probability of an event is a nonlinear function constructed by the independent variables and can be expressed as shown:

The odds ratio is used to explain the logistic regression coefficients of the independent variables [_{1}_{1} + _{2}_{2} + … + _{n}_{n})

In this study, we use SPSS [^{2}, and significance of coefficients. A positive coefficient value means that the odds ratio will increase for a unit increase of the independent variable and ^{2} and Deviation D should not be used to estimate MLE. The Akaike information criterion (AIC) statistic becomes a more useful MLE for estimating logistic regression models with continuous independent variables. Therefore, in this study we use the

The goodness-of-fit of the logistic regression model of farmland abandonment can be evaluated using the percent of observations that are predicted correctly (PC), Cohen’s kappa, and the area under the curve (AUC) of the receiver-operating characteristic. The PC method focuses attention on the ratio of correctly predicted cells from the total number of cells, and Cohen’s kappa evaluates the accuracy of location. The indicator AUC is used to measure the performance of the model compared with that of a random model in which the cut-off threshold is varied from zero to one [

To estimate the model, we combine a stepwise selection method and a conceptual model approach. We first choose the independent variables from the conceptual models and use stepwise regression to select the main independent variables before applying the saturated model to analyze which variables contribute to the farmland abandonment.

The prediction maps for farmland abandonment were generated by applying the constructed logistic models from the sampled data to the full dataset and calculating the predicted probabilities in the entire study area. According to the predicted probabilities, a common threshold of 50% is used to judge whether farmland is present or absent [

To adopt a logistic model, we use stratified random sampling to select n observations that are normally distributed. The use of stratified random sampling is intended to avoid spatial autocorrelation. For every observation, we record the value of each dependent and independent variable. In every model, we take 1000 observations and ensure that 0 and 1 have been observed equally for the dependent variables. Unequal numbers of 0 and 1 observations will not affect the coefficient estimation but will affect the constants [_{1} − ln _{2}) where p_{1} and p_{2} are frequencies of 0 and 1 of the dependent variables [

To avoid high multi-collinearity among the explanatory variables, we also should remove the variables that have pair-wise correlation coefficients higher than the threshold value of 0.8 [

Frequency of observations in areas where agriculture has been maintained (absence) and where farmland abandonment has been observed (presence) along the slope, elevation and distance-to-town gradients in the first period (1990–1995).

In the second period, the frequency of abandoned farmland (presence) is obviously higher than that of maintained agriculture (absence) with a distance to village of more than 8000 m, at slopes above 10 degrees and

Frequency of observations in locations where agriculture has been maintained (absence) and where farmland abandonment has been observed (presence) along the slope, distance-to-village and cumulative-temperatures-above-10-degrees gradients in the second period (1995–2005).

As shown by the Wald-Chi-square tests (^{2} can be used to judge the importance of independent variables. The more the value of Wald’s χ^{2} for an independent variable, the more important the independent variable is. The results from the spatially explicit logistic regressions for farmland abandonment in the first period indicate that topographic characteristics, level of population urbanization, status of rural labor supply and market accessibility are the most important spatial determinants of farmland abandonment (

Final prediction model of farmland abandonment based on sampling observations in the first period (1990–1995) (Observation: n = 1000).

Variables | Estimator (β) | Standard Error (SE) | Wald χ^{2} Statistics |
EXP (β) | |
---|---|---|---|---|---|

Constant | 21.209 | 5.887 | 12.981 | 0.000 | 2 × 10^{9} |

Cumulative temperature above 10 degrees | −0.001 | 0.000 | 12.439 | 0.000 ^{***} |
0.999 |

Annual precipitation | −0.009 | 0.004 | 5.852 | 0.016 ^{*} |
0.991 |

Distance to forest edge | −0.001 | 0.000 | 9.273 | 0.002 ^{**} |
0.999 |

Soil depth | −0.005 | 0.011 | 0.191 | 0.662 | 0.995 |

Content of soil coarse sand | 0.056 | 0.012 | 22.604 | 0.000 ^{***} |
1.058 |

Slope | 0.201 | 0.025 | 66.942 | 0.000 ^{***} |
1.223 |

Elevation | 0.003 | 0.001 | 8.126 | 0.004 ^{**} |
1.003 |

Proportion of employees in the primary sector | −0.013 | 0.007 | 3.203 | 0.074 | 0.987 |

Rural labor force participation rate | −0.008 | 0.004 | 2.959 | 0.085 | 0.993 |

Rate of change of rural labor | −0.090 | 0.048 | 3.611 | 0.050 ^{*} |
0.914 |

Rate of urbanization | 0.058 | 0.009 | 41.730 | 0.000 ^{***} |
1.060 |

Distance to town | 7.0 × 10^{−5} |
0.000 | 17.243 | 0.000 ^{***} |
1.000 |

Distance to village | 8.0 × 10^{−5} |
0.000 | 1.476 | 0.224 | 1.000 |

Distance to primary road | 2.0 × 10^{−5} |
0.000 | 1.256 | 0.262 | 1.000 |

Net income of farmer per capita | −0.003 | 0.001 | 15.816 | 0.000 ^{***} |
0.997 |

Average agricultural area per farm | 2.404 | 1.203 | 3.993 | 0.046 ^{*} |
11.068 |

Rate of change of employees in the primary sector | 0.059 | 0.042 | 1.931 | 0.165 | 1.060 |

*

According to the Wald-Chi-square tests (see ^{2},

From the results of two logistic regression models, we conclude that some bio-physical and transportation cost-related variables are the relatively important determinants of farmland abandonment,

Final prediction model of farmland abandonment based on sampling observations in the second period (1995–2005) (Observation: n = 1000)

Variables | Estimator (β) | Standard Error (SE) | Waldχ^{2} Statistics |
EXP (β) | |
---|---|---|---|---|---|

Constant | 12.046 | 2.372 | 25.787 | 0.000 | 1.7× 10^{5} |

Cumulative temperature above 10 degrees | −0.001 | 0.000 | 17.084 | 0.000 ^{***} |
0.999 |

Annual precipitation | −1.9 × 10^{−4} |
0.001 | 0.032 | 0.859 | 1.000 |

Distance to forest edge | 2.6 × 10^{−4} |
0.000 | 6.556 | 0.010 ^{**} |
1.000 |

Soil depth | −0.051 | 0.011 | 23.670 | 0.000 ^{***} |
0.950 |

Content of soil coarse sand | −0.027 | 0.015 | 3.168 | 0.075 | 0.973 |

Slope | 0.080 | 0.022 | 13.382 | 0.000 ^{***} |
1.083 |

Elevation | −0.001 | 0.001 | 1.184 | 0.277 | 0.999 |

Proportion of employees in the primary sector | −0.015 | 0.006 | 5.736 | 0.017 ^{*} |
0.985 |

Rural labor force participation rate | −0.014 | 0.005 | 8.751 | 0.003 ^{**} |
0.986 |

Rate of change of rural labor | −1.113 | 0.724 | 2.364 | 0.124 | 0.329 |

Rate of urbanization | 0.031 | 0.011 | 7.639 | 0.006 ^{**} |
1.032 |

Proportion of secondary sector’s output value | 0.031 | 0.015 | 4.561 | 0.033 ^{*} |
1.032 |

GDP per capita | 7.1 × 10^{−}^{5} |
0.000 | 6.279 | 0.012 ^{*} |
1.000 |

Distance to town | 4.4 × 10^{−}^{5} |
0.000 | 11.315 | 0.001 ^{**} |
1.000 |

Distance to village | 3.5 × 10^{−}^{5} |
0.000 | 26.371 | 0.000 ^{***} |
1.000 |

Distance to primary road | 1.5 × 10^{−}^{5} |
0.000 | 1.102 | 0.294 | 1.000 |

Net income of farmer per capita | −0.002 | 0.001 | 17.766 | 0.000 ^{***} |
0.998 |

Average agricultural area per farm | 2.851 | 0.909 | 9.835 | 0.002 ^{**} |
17.297 |

Rate of change of farm | 0.263 | 0.172 | 2.330 | 0.127 | 1.301 |

Rate of change of employees in the primary sector | −2.380 | 0.618 | 14.838 | 0.000 ^{***} |
0.093 |

*

The wages for farm labor variables are significantly related to farmland abandonment in the both model (see

The goodness-of-fit of the two period logistic regression models was observed to differ (

Goodness-of-fit of logistic regression models for farmland abandonment in two periods.

Model | AIC | PC | AUC | Kappa |
---|---|---|---|---|

First period model | 0.89 | 0.81 | 0.80 | 0.45 |

Second period model | 1.22 | 0.70 | 0.70 | 0.41 |

Our study shows that farmland abandonment was highly correlated with yield potential-related variables, which corroborates the results of previous studies [

The paper also shows that climate factors impacted farmland abandonment. The reason why lands with lower

The wage for farming labor contributes to farmland abandonment in both models. Farmland is more likely to be abandoned in those regions with a lower rural labor force participation rate. The main reason for this result is that the lower the

The reason that variable

The variables

With respect to the transportation cost of agricultural practices, the variables

For the determinants of the structural characteristics in agriculture, the variables

Compared the models of two time periods, we indicate that there are almost the same driving forces and consistent relationship with farmland abandonment. However, there are some differences in the important variables for the models of two periods. In the first period, the important variables are slope, rate of urbanization and content of soil coarse sand. However, the distance to village and the net income of farmer per capita are important for farmland abandonment in the second time period. This is because China’s reform and opening up has made large amounts of rural laborers move into cities since the late 1990s, which increases the opportunity cost of farming labor.

These results generally proved the study hypothesis as they suggest that farmland abandonment took place in locations where the agricultural cultivation costs were high and yield potential was low. Although the important drivers differ between the two periods, the construction of two spatially explicit economic models in this study revealed the same dynamic mechanisms of farmland abandonment. In other words, farmland abandonment can be effectively explained by agricultural land rent theory in the study area.

Although a portion of important driving forces of farmland abandonment were identified through the land economic model in this study, the mechanisms of land use changes cannot be comprehensively and systematically reflected, if only from an economics angle [

Based on the perspective of “economic man”, this study explored the mechanisms of farmland abandonment. While this explanation may seem appealing for the sake of simplicity, the reality is usually not as simple. On the one hand, agricultural production is very often carried out in disregard of economic profitability, simply because of not being market-oriented but, rather, aimed at the subsistence of the family. On the other hand, lack of generational succession is a common factor for farmland abandonment in developed countries, even in highly productive areas. Although the latter could be partially explained as cost of opportunity for labor, intangible aspects like prestige, public recognition, or pride of farming as a profession also play a significant role.

At the province scale, there is no spatial difference in some variables,

From a European perspective, there are a lot of policy measures actually in place which should prevent land abandonment on marginal sites [

The spatially explicit economic model used in this study is able to identify the main driving forces of farmland abandonment. The yield of agricultural product-related variables,

We thank four anonymous reviewers for their constructive comments. This study was supported by the National Natural Science Foundation of China (No. 41361111), the Major Research Plan of National Social Science Foundation of China (No. 12&ZD213), the Natural Science Foundation of Jiangxi Province (No. 20122BAB203025), the Social Science Foundation of Jiangxi Province (No. 13GL05 and No. 13YJ53), the China Postdoctoral Science Foundation (No. 2012M521286 and No. 2013T60647).

Hualin Xie and Peng Wang had the original idea for the study. Peng Wang was responsible for data collecting. Hualin Xie and Guanrong Yao carried out the analyses. All the authors drafted the manuscript, and approved the final one.

The authors declare no conflict of interest.