基于可见光/近红外技术的干旱区绿洲土壤盐分空间分布识别  

Monitoring the Spatial Variability of Soil Salinity in the Oasis Arid Areas with Visible-near-infrared(VNIR) Spectroscopy

作　　者：宁娟[1,2] 丁建丽[1,2] 杨爱霞[1,2] 邓凯[1,2] 

机构地区：[1]新疆大学资源与环境科学学院,乌鲁木齐830046 [2]绿洲生态教育部重点实验室,乌鲁木齐830046

出　　处：《中国农村水利水电》2017年第1期43-48,共6页China Rural Water and Hydropower

基　　金：国家自然科学基金资助项目(U1303381;41261090;41161063);自治区科技支疆项目(201591101);自治区重点实验室专项基金(2016D03001;2014KL005)

摘　　要：土壤盐渍化是制约干旱区经济发展的重要因素,准确获取土壤盐分信息成为治理干旱区土壤盐渍化的前提。利用采集自渭干河-库车河绿洲的67个表层(0~10 cm)土壤样品,通过可见光/近红外(visible-near infrared spectroscopy,VNIR)光谱技术结合实验室测定的土壤盐分数据,在相关分析提取敏感波段的基础上,运用偏最小二乘回归(PLSR)和支持向量机(SVM)方法进行土壤含盐量的建模并对比分析;此外,利用地统计方法,将最优模型预测得到的土壤含盐量数据进行空间插值,最终获取土壤含盐量的空间分异规律。结果表明:1利用原始一阶微分经支持向量机方法(SVM)建立的土壤含盐量模型,预测集决定系数R2为0.853,均方根误差RMSE为0.381,该模型具有较高的预测精度和较好的稳健性,可以作为有效手段估算干旱区绿洲土壤含盐量。2土壤表层含盐量属于中等空间变异,且空间结构比小于25%,具有强空间自相关。3利用支持向量机得到的土壤含盐量盐分分布图中,土壤含盐量预测值和实测值的决定系数R2为0.786,均方根误差(RMSE)为0.528,显示了较高的预测精度,该研究为快速获取干旱区绿洲土壤含盐量空间分异规律提供了一种新方法。Soil salinization is an important factor restricting economic development in arid areas,accurating obtain of soil salt information become the premise of governance soil salinization in arid.In this contribution,based on sensitive wave bands selected by the spectral correlation with soil salinity that measured by combined with visible-near-infrared（VNIR） spectroscopy and salt data.Using partial least squares（PLSR） analysis and support vector machine（SVM） method to establish models to estimate the soil salinity in the oasis arid areas of Weigan and Kuqa Rivers.In addition,predicting the spatial variability of the soil salinity through using Universal Kriging method.The results show that the support vector machine（SVM） model pre-processed with first order differential has the highest estimation accuracy.The determination coefficient（R2） of support vector machine model is 0.853,the root mean square error（RMSE） is 0.381.The soil salinity has the strong spatial autocorrelation,it belongs to medium spatial variation and the spatial structure less than 25%.The spatial variability of soil salinity shows a higher prediction precision.The determination coefficient（R2） of soil salinity predicted and measured values is 0.786 and the root mean square error（RMSE） is 0.528.This study demonstrates a new method to acquire spatial variability of soil salinity in the oasis arid areas with considerable success.

关 键 词：土壤含盐量 干旱区 高光谱 空间异质性 

分 类 号：S156.46[农业科学—土壤学] TV93[农业科学—农业基础科学]