基于高光谱的区域土壤质地预测模型建立与评价——以河套灌区解放闸灌域为例  

Establishment and evaluation of model for predicting soil texture based on hyperspectral data——Case study of Jiefangzha irrigation area in Hetao irrigation district

作　　者：张娜[1] 张栋良[1] 李立新 屈忠义[1] 

机构地区：[1]内蒙古农业大学水利与土木建筑工程学院,呼和浩特010018 [2]内蒙古河套灌区解放闸管理局,杭锦后旗015400

出　　处：《干旱区资源与环境》2014年第5期67-72,共6页Journal of Arid Land Resources and Environment

基　　金：国家自然科学基金项目(51069006);内蒙古教育厅青年科技人才计划-A类项目(2012)联合资助

摘　　要：为了运用光谱反射率快速确定土壤质地,对河套灌区6种不同类型土壤质地在室内进行光谱反射率测试,分别运用一元线性回归、逐步多元回归及BP神经网络三种方法建立光谱反射率与土壤砂粒含量及粉粒含量的拟合模型,并利用估测数据对样品进行土壤质地的模拟。结果显示:三种预测模型精度及其预测能力均较为满意,其中BP神经网络的拟合效果最好,砂粒,粉粒估测模型的决定系数R2均为0.86,外部检验决定系数R2分别为0.88,0.90。利用BP神经网络预测得出的粒径含量对样本质地重新判定,发现达到91.74%的样本符合类别分类要求。研究结果为利用高光谱图像大范围确定土壤质地奠定了基础,对于未来区域模型模拟和土壤水力参数推求具有重要指导意义和应用价值。In order to use the hyperspectral reflectance to estimate the soil texture,hyperspectral reflectance of 6different types of soil texture in Hetao irrigation area were tested in laboratory. Predictive models between soil texture（ sand and silt content） and spectroscopy were established with 3 methods-a linear regression,stepwise multiple regression and BP neural network. And the simulation data of the samples were used for estimating the soil texture. Results show that the accuracy and prediction of the models are satisfactory,and the matching result of BP neural network is the best. the coefficient of determination R2of the simulation models on sand and silt content are both 0. 86,while those of the model checking are 0. 88 and 0. 90. The particle size data obtained by BP neural network was used to estimate the soil texture of the samples,with 91. 74% of the samples matching to the soil texture tested in laboratory. The result of study laid the foundation for the estimation of the soil texture in a large scope by using the hyperspectral image. It has an important significance and value for the future regional model simulations and the soil hydraulic parameters calculation.

关 键 词：光谱反射率 土壤质地 逐步多元回归 BP神经网络 河套灌区 

分 类 号：S152.3[农业科学—土壤学]