自然资源学报 ›› 2014, Vol. 29 ›› Issue (5): 744-756.doi: 10.11849/zrzyxb.2014.05.002

• 资源利用与管理 • 上一篇    下一篇

基于BP-CA的海滨湿地利用空间格局优化模拟研究——以大丰海滨湿地为例

欧维新1,2, 肖锦成1, 李文昊1   

  1. 1. 南京农业大学土地管理学院, 南京210095;
    2. 农村土地资源利用与整治国家地方联合工程研究中心, 南京210095
  • 收稿日期:2013-04-11 修回日期:2013-08-20 出版日期:2014-05-20 发布日期:2014-05-20
  • 作者简介:欧维新(1974-),男,湖南益阳人,副教授,硕士生导师,中国自然资源学会会员(S300001377M),主要从事湿地资源利用效应与管理研究。E-mail:owx@njau.edu.cn
  • 基金资助:
    中央高校基本科研业务费自主创新重点研究项目“区域土地利用的生态保护空间网络研究”(KYZ201166)。

Spatial Pattern Optimization Simulation of Coastal Wetland Use Based on BP Neural Network and Cellular Automata—A Case of Dafeng Coastal Wetland

OU Wei-xin1,2, XIAO Jin-cheng1, LI Wen-hao1   

  1. 1. College of Land Management, Nanjing Agricultural University, Nanjing 210095, China;
    2. National & Local Joint Engineering Research Center for Rural Land Resources Use and Consolidation, Nanjing 210095, China
  • Received:2013-04-11 Revised:2013-08-20 Online:2014-05-20 Published:2014-05-20
  • About author:10.11849/zrzyxb.2014.05.002

摘要: 论文运用BP神经网络与CA耦合模型,遴选出驱动滩涂开发的自然、经济、社会方面的23 个空间变量,借助Matlab 平台,构建了海滨湿地利用空间格局模拟模型。利用盐城大丰海滨1988、2002 和2009 年三期遥感影像,及相应年份人口经济数据,根据格局优化模拟思路,先使用湿地功能生态位模型诊断出历史过度开发区域,再训练合理开发转换规则,最后仿真出2016 年大丰海滨湿地利用的空间优化格局。模拟中依据目标年覆被数量结构,限定训练样本随机抽取比例,提高了1 万个样本的有效信息量;通过对比不同BP隐藏层节点数时的模拟精度,确定“23-13-8”和“23-17-8”分别为两组实验的最优BP结构;使得模拟效率提升,仿真总体精度高于93.6%。研究表明,近些年大丰海滨湿地覆被类型间的演替趋于频繁,自然湿地年开发速度明显加快,开发重心不断东移,呈现出向海“滚动开发”的特征。建议放缓两大保护区核心区周边的无序开发,推进大丰港附近滩涂的有序围垦。

Abstract: In this paper, using BP neural network model and cellular automata (CA) model, 23 spatial variables including natural, economic and social aspects which drive the beach reclamation of coastal vegetation are selected and the spatial pattern simulation model of wetland is established by using MATLAB R2010b software. With remote sensing image of Landsat TM/ ETM+ in Dafeng, Yancheng coastal wetland in three periods of 1988, 2002 and 2009, and the population and economic data of the respective year and based on the simulation thought of optimizing pattern, firstly, we use the wetland function niche model to diagnose historical excessively exploiting areas. Then we train the conversion rules of CA model belonging to the rational development rule. Finally, we simulate the space optimization pattern of Dafeng coastal wetland in 2016. To improve operational efficiency and accuracy of the model, we use two new methods in this study. One method is based on the cover quantity structure in the target year, which limits the proportion of training samples randomly, so as to improve 10000 samples of the effective information amount. The second is through comparing with simulating accuracy under different hiding layer nodes of BP neural network,"23-13-8" and "23-17-8" are determined as the optimal BP structures in two experiments. These two improvements enhance the efficiency of simulation, and the overall accuracy is higher than 93.6%. This research shows that, during these years, the succession among the cover types of Dafeng coastal wetland tends to be frequent, and the speed of reclamational year of natural wetland is obviously faster. At the same time, the development focus is constantly moving eastward, which shows the characteristics of "rolling development" to the sea. Finally, it is recommended to slacken the disorderly development in the surrounding core areas of Jiangsu Yancheng National Nature Reserve and Dafeng Milu National Nature Reserve, and to promote the orderly reclamation of tidal flats near Dafeng Port.

中图分类号: 

  • F301.24