关键词: 耕地生产力, 自组织特征映射神经网络, 可视化, 聚类, 分区, 中国

Abstract: 【Objective】The productivity of arable land is critical to food security in China. Investigating the factors controlling the productivity in different regions and influence mechanism is of importance to decision-making about reasonable utilization of arable land resources, policy regulation for remediation of contaminated land, constructing high-quality farmland, and accordingly, the national food security. 【Method】Firstly, the index system of natural and socio-economic factors was developed. These factors include soil fertility, climatic conditions, topography, economic investment and effective feedback. Then, the arable land in China was divided into regions with different productivities based on previous studies and land productivity gradations. Because the relationships between arable land productivity and its influencing factors are very complex, which are non-linear and multiple dimensional, self-organizing map was employed, which is a powerful approach for pattern analysis. U-matrix and component planes were used in order to visually interpret diversity and similarity of arable land productivity, and intuitively reflect the sensitive factors of regional changes. In addition, pressure index of arable land was used for correction of the clustered results. Finally, sub-regions with different arable land productivities were determined based on the resulted clusters, arable land productivity gradation, and the principles (including comprehensive consideration, relative consistency, regional conjugation and administrative units). 【Result】 Results show that spatial visibilities of soil fertility, climatic conditions , topography, economic investment and effective feedback in provinces were significant and these factors were developed non-equilibrium. The natural elements and socio-economic factors were both clustered in six groups. By using the aforementioned method, the arable land was divided into 10 regions and 24 sub-regions in China with different productivities. The 10 regions are as follows: Northeast plains region, Northern China plain and hills regions, the Loess Plateau regions, Shichuan basins, the Middle and lower reaches of the Yangtze River plain regions, Southern China hills regions, Yunnan-Guizhou Plateau regions, Inner Mongolian Plateau regions, Northwestern lands regions, and Qinghai-Tibet Plateau regions. Sub-regions were briefly described for arable land productivity regions with advantages and constraints. Finally, China arable land productivity zoning system was established. 【Conclusion】Taking natural factors, socio-economic factors and pressure index of arable land into consideration, zoning results are in accordance with the objective reality, indicating that self-organizing map is an effective method to monitor and identify novel relationships between arable land productivity and factors with an intricate visualization process, as well as extract and determine zoning of arable land productivity. In addition, government officers can readily identify existing advantages and limiting factors of arable land productivities in different regions in China.

Key words: arable land productivity, self-organizing map, visualization, clustering, zoning, China