Simulation Study On Decision-Making Of Farmers’ Abandonment Of Terraced Fields ——A Case Study Of 474 Households In 22 Typical Villages In Fujian Province

Terrace is an important type of cultivated land in Hilly and mountainous areas in China.Although our government has strengthened the protection of terraces,there are still obvious abandonment in many places.Due to the wide distribution of terraced fields,it is difficult for investigators to grasp the actual abandonment in a wide range.Farmers are the micro behavior subjects of land use.It is of great theoretical and practical significance to explore the mechanism of terrace abandonment from the scale of farmers and then simulate the situation of terrace abandonment.Therefore,this study uses the indicators that have a significant impact on Farmers’ decision-making to build a abandonment judgment model to simulate farmers’ terrace abandonment behavior,hoping to grasp the current situation of terrace utilization from the farmers’ scale through the model,so as to better protect and utilize terrace resources.The research idea of this study is: the first step is to carry out a field questionnaire survey in 22 typical terrace villages in 8 counties of Fujian Province,obtain the information of farmers’ family population,income and terrace utilization status,and sort out 474 effective questionnaires.The second step is to construct the logistic regression model to determine the factors that have a significant impact on the decision-making of terrace abandonment.The third step is to do principal component analysis on the indicators with significant impact,and obtain the principal components that can reflect the original variables and are not related to each other to the greatest extent through dimension reduction.The fourth step is to construct multilayer perceptron neural network and PCA-MLP mlp neural network model based on the obtained principal components.Finally,using 30%（142 households）sample data to test the accuracy of binary logistic regression model and PCA-MLP neural network model in farmers’ decision-making,and compare their simulation effects.The main conclusions are as follows:（1）The results of binary logistic regression model show that the average land area,the proportion of female labor force,the proportion of family support and care population and total agricultural income have a significant negative impact on Farmers’ terrace abandonment decision,while commuting time,irrigation conditions and the proportion of non-agricultural income in total household income have a significant positive impact on Farmers’ terrace abandonment decision.The convenience of terraced field planting,the dependence of farmers’ families on agricultural income,the burden of farmers’ families and the ability of migrant workers will all play an important role in the abandonment of terraced fields.（2）The simulation results of farmers’ terrace abandonment behavior show that PCA-MLP neural network model is better than binary logistic regression model in judging whether farmers abandon terraces.When the number of hidden layers of neural network is 1and the number of nodes is 6,and the activation function is set as sigmoid function,the judgment accuracy of 142 farmers is 88.7%,which is higher than 73.9% of binary logistic regression model.（3）（1）Improve the convenience of terrace planting: improve agricultural infrastructure and increase investment in agricultural machinery popularization and innovation.（2）Improve the economy of terrace planting: reduce the investment cost of terrace agriculture and increase the income of terrace planting.（3）Follow the principle of multi selectivity of farmers’ Employment: improve non-agricultural employability and reduce life pressure.（4）Maximize model value: promote model use,optimize model design,and mine model functions.