| 区域土壤水盐空间分布信息的BP神经网络模型研究 |
| |
| 引用本文： | 姚荣江,杨劲松,邹平,刘广明.区域土壤水盐空间分布信息的BP神经网络模型研究[J].土壤学报,2009,46(5):788-794. |
| |
| 作者姓名： | 姚荣江 杨劲松 邹平 刘广明 |
| |
| 作者单位： | 中国科学院南京土壤研究所,南京,210008 |
| |
| 基金项目： | 国家863计划重点项目课题，国家科技支撑计划，中国科学院南京土壤研究所创新领域前沿项目，中国科学院知识创新工程重要方向项目 |
| |
| 摘 要： | 针对黄河下游三角洲盐渍区土壤水盐动态的复杂性和空间的变异性,将人工神经网络引入土壤水盐信息的模拟和预测中,探讨了3层网络结构下隐含层神经元数对网络训练和预测的影响,建立了0～20cm表土层水盐含量及其空间分布的BP神经网络模型。结果表明:研究区域表土层水盐与土壤容重和地下水性质间均具极显著的关联性,表土层盐分BP网络的输入因子以经、纬度坐标、土壤容重、地下水埋深和矿化度5个变量为宜,含水量则为经、纬度坐标、土壤容重和地下水埋深4个变量;隐含层神经元数过多会导致"过拟合",当表层土壤盐分和含水量BP网络的拓扑结构分别为5:8:1和4:6:1时,其预测精度最高;表土层水盐BP神经网络模拟值与观测值得到的分布图表现出相似的空间格局,BP神经网络模型有效地模拟了表土层水盐含量及其空间分布特征。该研究为黄河三角洲地区土壤盐渍化的发生、发展及演变规律分析提供理论基础,并为盐渍土地的水盐调控与科学管理提供决策依据。
|
| 关 键 词： | 土壤水盐 空间分布 BP神经网络 黄河三角洲 |
BP neural network model for spatial distribution of regional soil water and salinity |
| |
| Yao Rongjiang,Yang Jingsong,Zou Ping and Liu Guangming.BP neural network model for spatial distribution of regional soil water and salinity[J].Acta Pedologica Sinica,2009,46(5):788-794. |
| |
| Authors: | Yao Rongjiang Yang Jingsong Zou Ping and Liu Guangming |
| |
| Institution: | Institute of Soil Science, Chinese Academy of Sciences;Institute of Soil Science, Chinese Academy of Sciences;Institute of Soil Science, Chinese Academy of Sciences;Institute of Soil Science, Chinese Academy of Sciences |
| |
| Abstract: | Aiming at the complexity and spatial variability of the dynamic soil water and salinity in the saline region of the Lower Yellow River Delta, artificial neural network was introduced for modeling and prediction of soil water and salinity. Influence of the number of neurons in the hidden layer on training and forecasting was discussed for the three-layered network, and Back Propagation Neural Network (BPNN) models were established for modeling contents of water and salinity and their spatial distribution in the surface soil 0～20 cm in depth. Results indicate that the water and salinity in the surface soil was significantly correlated with soil bulk density and groundwater properties across the study area. For surface soil salinity, it is advisable to have the five variables, i.e. longitude and latitude of the site, soil bulk density, and depth and mineralization of groundwater cited as input vectors, while for soil moisture, the four variables, i.e. longitude, latitude, bulk density, and groundwater depth. An excessive number of neurons in the hidden layer would result in overfitting. Considering forecasting precision, the topological structure of the BP network was defined as 5∶ 8∶ 1 and 4∶ 6∶ 1 for salinity and moisture in the surface soil, respectively. Distribution maps of the observed surface soil water and salinity and their BPNN simulation displayed similarity in spatial pattern, and the BPNN effectively simulated contents of water and salinity and their spatial distribution in the surface soil with high accuracy. The findings of the study can serve as a theoretical basis for analyzing the occurrence, development and evolvement regularities of soil salinization in the Yellow River Delta, and provide a scientific basis for decision-making in regulating soil water and salt regulation and implementing scientific management of saline soils. |
| |
| Keywords: | Soil water and salinity Spatial distribution BP neural network Yellow River Delta |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《土壤学报》浏览原始摘要信息 |
| 点击此处可从《土壤学报》下载免费的PDF全文 |
|
