基于GF-5高光谱影像的滇中高原灌区土壤有机碳含量反演研究  

Inversion of soil organic carbon content in irrigation area of Central Yunnan Plateau based on GF-5 hyperspectral images

作　　者：严正飞 杨明龙[1] 唐秀娟 夏永华[1] 杨赈 李万涛 YAN Zheng-fei;YANG Ming-long;TANG Xiu-juan;XIA Yong-hua;YANG Zhen;LI Wang-tao(Faculty of Land Resources Engineering,Kunming University of Science and Technology/Yunnan University Spatial Information Mapping Technology Application Engineering Research Center,Kunming,Yunnan 650093,China;Kunming Institute of Surveying and Mapping,Kunming,Yunnan 650091,China)

机构地区：[1]昆明理工大学国土资源工程学院/云南省高校高原山区空间信息测绘技术应用工程研究中心,云南昆明650093 [2]昆明市测绘研究院,云南昆明650091

出　　处：《南方农业学报》2025年第1期124-134,共11页Journal of Southern Agriculture

基　　金：国家自然科学基金项目(62266026)。

摘　　要：【目的】基于GF-5高光谱影像构建针对滇中高原灌区土壤有机碳(SOC)含量反演模型,为后续开展滇中高原灌区SOC含量反演研究提供参考依据。【方法】选取云南省楚雄州姚安县为研究区,以GF-5高光谱影像为基础数据源,筛选出与SOC含量相关性较高的预处理方法并构建光谱指数,基于连续投影算法(SPA)和竞争性自适应重加权算法(CARS)筛选特征波段组合,以筛选的特征波段、光谱指数、地形因子及Sentinel-1后向散射系数为辅助变量进行组合,结合实地采样的SOC含量数据,运用XGBoost模型进行SOC含量反演。【结果】在21种数据预处理方法中以AM-Normalize的预处理效果最优,与实测SOC含量的相关系数为0.7544;其次是SG-FD、SD和FD的预处理效果,与实测SOC含量的相关系数分别为0.6791、0.6671和0.6202。SPA筛选的波段反演效果最优,其决定系数(R^(2))较CARS和全波段数据分别提升了0.0739和0.1524,均方根误差(RMSE)分别降低了0.9279和1.2793。引入地形因子的变量模型G2,其R^(2)较变量模型G1(特征波段+光谱指数)提升了0.0398,RMSE降低了0.1685;进一步加入Sentinel-1后向散射系数,变量模型G3的R^(2)较变量模型G2提升了0.0255,RMSE降低了0.1385。基于GF-5高光谱影像的SOC含量反演结果显示,滇中高原姚安灌区的SOC含量范围为9.8443~29.2514 g/kg,平均为19.4447 g/kg,与土壤样本SOC含量实测值的范围(10.47~30.11 g/kg)及平均值(20.6307 g/kg)较接近。【结论】基于GF-5高光谱影像构建的XGBoost模型,经AM-Normalize预处理降低噪声干扰、SPA筛选特征波段及引入光谱指数、地形因子和Sentinel-1后向散射系数后,能有效提升SOC含量反演的精度和适用性,为滇中高原地区SOC含量预测提供技术支撑。【Objective】Based on GF-5 hyperspectral images,a model for inverting soil organic carbon(SOC)content in the irrigation area of Central Yunnan Plateau was constructed,which could provide reference basis for subsequent research on SOC content inversion in the irrigation area of Central Yunnan Plateau.【Method】Yao’an County,Chuxiong Prefecture,Yunnan Province was selected as the research area,and GF-5 hyperspectral image was used as the basic data source to screen out preprocessing methods with high correlation with SOC content and spectral index.The feature band combination was screened based on continuous projection algorithm(SPA)and competitive adaptive reweighting algorithm(CARS).The selected feature band,spectral index,topographic factor and Sentinel-1 backscattering coefficient were combined as auxiliary variables,combined with the SOC content data collected in the field,XGBoost model was used to invert SOC content.【Result】Among the 21 data preprocessing methods,AM-Normalize had the best preprocessing effect,with a correlation coefficient of 0.7544 with the measured SOC content;followed by SG-FD,SD and FD,with correlation coefficients with the measured SOC content of 0.6791,0.6671 and 0.6202 respectively.The band inversion effect of SPA screening was the best,with coefficient of determination(R^(2))increasing by 0.0739 and 0.1524 compared to CARS and full-band data respectively,while root mean square error(RMSE)decreased by 0.9279 and 1.2793 respectively.The variable model G2,which introduced topographic factors,had an R^(2) increase of 0.0398 compared to the variable model G1(characteristic bands+spectral indexes),and RMSE decreased by 0.1685;further adding the Sentinel-1 backscatter coefficient,the R^(2) of the variable model G3 increased by 0.0255 compared to the variable model G2,and RMSE decreased by 0.1385.The SOC content inversion results based on GF-5 hyperspectral images showed that the SOC content range in the Yao’an irrigation district of the Central Yunnan Plateau was 9.8443-29.2514 g

关 键 词：土壤有机碳(SOC) GF-5高光谱影像 光谱指数 XGBoost模型 滇中高原 

分 类 号：S127[农业科学—农业基础科学]