自然资源学报 ›› 2016, Vol. 31 ›› Issue (6): 1061-1072.doi: 10.11849/zrzyxb.20150639

• 资源研究方法 • 上一篇    

基于贝叶斯网络的基本农田划定方法

关小东, 何建华*   

  1. 武汉大学资源与环境科学学院,武汉 430079
  • 收稿日期:2015-06-08 出版日期:2016-06-20 发布日期:2016-06-20
  • 通讯作者: 何建华(1974- ),男,湖北孝感人,博士,博士生导师,研究方向为复杂地理时空过程分析与模拟、土地利用空间优化决策支持。E-mail: hjianh@126.com
  • 作者简介:关小东(1992-),男,安徽蚌埠人,研究方向为地理过程分析与模拟。E-mail: guanx_d@163.com
  • 基金资助:
    国家自然科学基金资助项目(41471339)。

Prime Farmland Protection Zoning Based on Bayesian Network

GUAN Xiao-dong, HE Jian-hua   

  1. School of Resource and Environmental Sciences, Wuhan University, Wuhan 430079, China
  • Received:2015-06-08 Online:2016-06-20 Published:2016-06-20
  • Supported by:
    National Natural Science Foundation of China, No. 41471339.

摘要: 基本农田划定是落实耕地保护、保证粮食安全的关键问题。目前,基本农田划定的相关研究多基于对耕地适宜性条件的评价,而未涉及对耕地历史变化过程的考虑,因而不能保证划定结果适应实际区域的耕地变化趋势,导致基本农田易被占用,补划与调整频繁。论文综合考虑耕地适宜性条件与历史动态变化状况,提出了一种基于贝叶斯网络模型的基本农田划定方法。以湖北省大冶市这一典型资源型城市为实验区进行了实例研究,划定结果表明,该划定方法能在保证基本农田划定数量与质量的同时,提高其农业用途的稳定性,促进基本农田保护与经济建设、生态保护间协调发展,是一种有效的划定模型。

关键词: 贝叶斯网络, 划定模型, 基本农田

Abstract: Basic farmland protection zoning is the key issue of implementing land protection and ensuring food security. At present, most of the related studies on farmland protection zoning are based on the evaluation of the suitability of cultivated land, with no consideration of the change process of cultivated land in the past. Therefore, they cannot ensure the zoning result to adapt to the trend of cultivated land change, leading to the prime farmland being frequently occupied and adjusted. To guarantee the stability of prime farmland as well as its cultivation quality, we put forward a method of zoning farmland based on Bayesian network (BN) model in this paper, which includes both factors concerning cultivated land suitability condition and its dynamic change. With data of land use status in Daye City at two time points (1997 and 2012), the factors about suitability and changes of cultivated land are obtained. Suitability factors are used to reflect the quality of cultivated land, and the factors about changes are used to study the change disciplinarian of cultivated land. Using the basic farmland potential as the target variable, we defined the structure of the network by expert knowledge. The BN was trained by Maximum Likelihood Method with 57 085 random sample points. The results show that, only 60% farmland in Daye City remains stable from 1997 to 2012. In the three dynamic factors, the influence of building occupancy is the greatest, which accounts for 28.2%, followed by the internal adjustment, 17.4%, the influence of ecological footprint is the least, which accounts for 4.65%. The results of sensitivity analysis also indicate that building occupancy has the greatest influence on the potential value of the prime farmland, with the highest variance reduction of 26.8%. The variance reductions of the distance to the center of city (0.43%) and the center of the town (0.58%) are greater than that of the distance to the road (0.14%) or the railway (0.28%), reflecting that urban expansion has greater impact on the occupation of basic farmland than transportation construction. In natural condition factors, the influence of the distance to the water area is the minimum, which is less than 0.02%, while soil erosion has the greatest variance reduction of 0.34%. It can be deduced that soil erosion, but water, is the most important nature factor to destroy the stability of cultivated land in Daye City. Then, using the values of suitability factors in 2012, we obtained the relative potential value of each cultivated land parcel as the prime farmland through forward reasoning of the trained network, and zoned the farmland with the potential value of the target variable. The results show that, compare to traditional land evaluation method, the zoning model based on BN can improve the stability of agricultural value and ensure the quantity and quality of prime farmland. It is a new and effective model for farmland protection zoning.

Key words: Bayesian network, farmland protection, zoning model

中图分类号: 

  • F323.211