湖滨绿洲棕漠土有机碳含量高光谱估算  

Hyperspectral prediction of organic carbon content of brown desert soil in the lakeside oasis

作　　者：樊泳灼 李新国[1,2] FAN Yong-zhuo;LI Xin-guo(College of Geographical Sciences and Tourism,Xinjiang Normal University,Urumqi 830054,China;Xinjiang Key Laboratory of Lake Environment and Resources in Arid Regions,Urumqi 830054,China)

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：《江苏农业学报》2023年第6期1341-1348,共8页Jiangsu Journal of Agricultural Sciences

基　　金：新疆维吾尔自治区自然科学基金项目(2022D01A214);国家自然科学基金项目(41661047)。

摘　　要：以博斯腾湖湖滨绿洲为研究区,利用实测棕漠土有机碳含量与高光谱(350~2 500 nm)数据,应用竞争性自适应重加权采样算法(CARS)、连续投影算法(SPA)、竞争性自适应重加权采样-连续投影算法(CARS-SPA)筛选棕漠土有机碳含量响应的高光谱特征波段,分别采用全波段和特征波段结合随机森林(RF)模型构建棕漠土有机碳含量估算模型。结果表明:博斯腾湖湖滨绿洲棕漠土0~50.0 cm土层有机碳含量为1.40~40.92 g/kg,平均值为14.20 g/kg,变异系数为55.54%,呈中等变异水平。CARS、SPA、CARS-SPA等算法筛选出的棕漠土有机碳含量响应特征波段分别为122个、11个和10个。基于CARS-SPA算法筛选出的特征波段数据输入RF模型估算效果最好,验证集检验的决定系数(R^(2))、相对分析误差(RPD)、均方根误差(RMSE)分别为0.85、2.59和2.72 g/kg,该方法能有效减少光谱数据冗余、提高模型估算精度和运行效率。本研究结果为研究区棕漠土有机碳含量的估算提供参考。Using the measured organic carbon content of brown desert soil and hyperspectral(350-2500 nm)data acquired from the lakeside oasis of Bosten Lake,competitive adaptive reweighted sampling(CARS),successive projection algorithm(SPA)and competitive adaptive reweighted sampling-successive projection algorithm(CARS-SPA)were used to screen the characteristic bands of organic carbon content response in brown desert soil.The prediction model of organic carbon content in brown desert soil was constructed by using full band and characteristic band combined with random forest(RF)model.The results showed that the organic carbon content in the 0-50.0 cm soil layer of the brown desert soil in the lakeside oasis of Bosten Lake was 1.40-40.92 g/kg,with an average of 14.20 g/kg,and the coefficient of variation was 55.54%,showing a moderate variation level.The response characteristic bands of brown desert soil organic carbon content screened by CARS,SPA and CARS-SPA were 122,11 and 10,respectively.The best prediction effect was obtained when the characteristic band data selected by the CARS-SPA algorithm were input into the RF model.The determination coefficient(R^(2)),relative percentage difference(RPD)and root mean square error(RMSE)of the validation set test were 0.85,2.59 and 2.72 g/kg,respectively.This method could effectively reduce the redundancy of spectral data and improve the prediction accuracy and operation efficiency of the model.The results of this study provided a reference for the prediction of organic carbon content in brown desert soil in the lakeside oasis of Bosten Lake.

关 键 词：土壤有机碳含量 棕漠土 高光谱 竞争性自适应重加权采样-连续投影算法(CARS-SPA) 随机森林 

分 类 号：S127[农业科学—农业基础科学]