博斯腾湖湖滨绿洲土壤电导率高光谱估算模型  

Hyperspectral estimation model of the soil conductivity in the lakeside oasis of Bosten Lake,China

作　　者：赵慧 李新国[1,2] 牛芳鹏 靳万贵[1,2] 麦麦提吐尔逊·艾则孜[1,2] ZHAO Hui;LI Xin-guo;NIU Fang-peng;JIN Wan-gui;Mamattursun·Eziz(College of Geographic Sciences and Tourism,Xinjiang Normal University,Urumqi Xinjiang 830054;Xinjiang Laboratory of Lake Environment and Resources in Arid Zone,Urumqi Xinjiang 830054)

机构地区：[1]新疆师范大学地理科学与旅游学院,新疆乌鲁木齐830054 [2]新疆干旱区湖泊环境与资源实验室,新疆乌鲁木齐830054

出　　处：《中国土壤与肥料》2021年第2期289-295,共7页Soil and Fertilizer Sciences in China

基　　金：国家自然科学基金项目(41661047,41561073)。

摘　　要：以博斯腾湖湖滨绿洲为研究区,采用分数阶微分对光谱指数进行波段优化,筛选高光谱数据的特征波段,利用偏最小二乘回归(PLSR)和支持向量机(SVM)构建土壤电导率高光谱数据的估算模型。研究结果表明:(1)分数阶微分的高光谱数据与土壤电导率的相关性:随着分数阶微分阶数的增加,特征波段数呈现逐渐增加的趋势,2阶是特征波段数量最多的阶数,特征波段数量为335(P=0.01),相关系数绝对值最大为0.64。(2)分数阶微分优化光谱指数的高光谱数据:随着分数阶微分阶数的增加,光谱矩阵图表现为相关系数在正负值之间波动较大,0.8阶在光谱指数DSI的相关系数绝对值最大为0.75;平方根、对数、倒数的相关系数绝对值最大为0.64。(3)基于PLSR和SVM构建土壤电导率估算模型:基于0.8阶微分和光谱指数DSI筛选的特征波段建立的估算模型估算效果较好,其中SVM构建的估算模型最优,模型精度为R_(SVMc)~2=0.89,RMSE_(SVMc)=0.03,R_(SVMv)~2=0.80,RMSE_(SVMv)=1.12。利用SVM估算模型可以有效地对研究区土壤电导率进行定量估算。Taking the oasis of lake Bosten Lake as the research area,the spectral index was optimized by fractional differential,the characteristic bands of the hyperspectral data were screened,and the estimation model of the soil conductivity hyperspectral data was constructed by partial least square regression(PLSR)and support vector machine(SVM).The results show that:(1)The correlation between the fractional-order differential hyperspectral data and soil conductivity showed that as the fractional-order differential order increased,the number of characteristic bands gradually increased.The second order was the order with the largest number of characteristic bands,the number of characteristic bands is 335(P=0.01),and the maximum absolute value of the correlation coefficient was 0.64.(2)The hyperspectral data of the fractional differential optimtzation spectral index shows that with the increase of the fractional-order differential order,the spectral matrix chart showed that the correlation coefficient fluctuated greatly between positive and negative values,and the absolute value of the correlation coefficient of the 0.8 order the spectral index DSI was the maximum 0.75.The maximum absolute value of the correlation coefficient of the square root,the logarithm,and the reciprocal was 0.64.(3)The soil conductivity estimation model was constructed based on PLSR and SVM.The estimation model based on characteristic bands selected by the 0.8 order differential and spectral index DSI had a good estimation effect.The estimation model constructed by SVM was the best,and the model accuracy was RSVMc 2=0.89,RMSESVMc=0.03,RSVMv 2=0.80,RMSESVMv=1.12.The SVM estimation model could be used to quantitatively estimate the soil conductivity in the study area.

关 键 词：土壤电导率 分数阶微分 光谱指数 特征波段 湖滨绿洲 

分 类 号：S156.4[农业科学—土壤学]