County-level patterns of cropland and their relationships with socio-economic factors in northwestern China

Over the last several decades, the distribution of cropland in China changed as a result of nationwide urbanization, continuous economic development and rapid population growth. Patterns of cropland and the driving factors that have caused changes in those patterns differ substantially among regions. Therefore, landscape classification should be combined with an analysis of the factors that cause land-use change, especially in large areas with complex natural and socio-economic conditions. The study described here was performed in Xinjiang, a typical arid region in northwestern China. After considering the percentage of cropland, the population density and the proportion of the population that practice animal husbandry in 2008, six types of cropland patterns and attributes (TCPAs) were identified using hierarchical clustering analysis. Partial least squares regression (PLSR) models were then developed to determine the main socio-economic indicators for cropland change from 1988 to 2008 within each TCPA. The results indicated that all selected factors were significantly strong drivers of cropland change in the study area. The total power of agricultural machinery, the gross output value of agricultural products and the consumption of chemical fertilizer occurred most frequently in the six PLSR models. However, the population density of minority nationalities, the agricultural population density, the gross output value of forestry and animal husbandry products, and the oil-bearing crops yield were each identified as strong factors only once in the six models. Each variable also had different effects in each of the six groups, and significant differences existed in the composition of the main factors between the groups. The PLSR model partially eliminated co-dependency between variables and facilitated a more unbiased view of the relationships between socio-economic factors and changes in cropland. The approach used here, which combined landscape classification based on related attributes with PLSR models, has important applications in sustainable land-use management, biodiversity conservation, and agricultural land protection.