期刊文献+

基于机载L波段全极化无人机合成孔径雷达的森林地上生物量估测

Forest AGB Estimation Based on Airborne L-band Full-Polarization UAVSAR
下载PDF
导出
摘要 为探索L波段全极化合成孔径雷达(Synthetic Aperture Radar,SAR)数据估算森林地上生物量(Aboveground bio⁃mass,AGB)的潜力,基于非洲合成孔径雷达(AfricaSAR)项目无人机合成孔径雷达(Unmanned Aero Vehicle Synthetic Aperture Radar,UAVSAR)数据的冠层-地面散射分量构建5种极化散射比参数(R1、R2、R3、R4、R5),计算雷达植被指数(Radar Vegetation Index,RVI),采用六分量和七分量分解等4种基于模型的分解提取21个极化分解参数,最后合并所有特征并采用随机森林特征重要性筛选出最优特征组合,采用随机森林(Random forest,RF)、支持向量机回归(Support vector machine regression,SVR)、K最近邻回归(K-nearest neighbor regression,KNN)对不同特征组合估测非洲加蓬洛佩(Lope)区的森林地上生物量。结果表明,极化散射比参数、体散射(Vol)和雷达植物指数(Radar Vagetation Index,RVI)对森林AGB具有较高的敏感性,R2与AGB的相关性为0.823,最优特征组合为Vol、极化散射比参数和RVI。不同特征组合的机器学习模型均表现出较好的效果,基于极化分解参数机器学习模型的决定系数(R2)大于0.800,均方根误差(RMSE)小于88.000 Mg/hm^(2),效果最好的是基于最优特征组合的RF模型,对比单独使用极化分解参数,R2提高0.144,RMSE降低30.327 Mg/hm^(2)。极化散射比参数在森林AGB估计中具有一定的潜力,引入RVI提高模型精度,基于模型的分解适用于森林AGB估测,特征筛选的机器学习模型能较好地反演森林AGB,并在AGB达到400.000 Mg/hm^(2)未出现明显饱和点。 In order to explore the potential of L-band full-polarization SAR data to estimate forest aboveground biomass(AGB),five polarimetric scattering ratio parameters(R1,R2,R3,R4,R5)were constructed based on the canopy-ground scattering component of Unmmaned Aero Vehicle Synthetic Aperture Radar(UAVSAR)data of the AfricaSAR project.Calculating the Radar Vegetation Index(RVI),and 21 polarimetric decomposition parameters were extracted by four model-based decompositions,including six-component and seven-component decomposition.Finally,all features were merged and the random forest feature importance was used to screen out the optimal feature combination,and random forest(RF),support vector machine regression(SVR),K-nearest neighbor regression(KNN)were used to estimate forest AGB of Lope,The Gaboneses Repbulic,Africa,with different feature combinations.The results showed that the polarimetric scattering ratio parameters,bulk scattering(Vol)and RVI had high sensitivity to forest AGB,and the correlation between R2 and AGB was 0.823,and the optimal feature combination was Vol,polarimetric scattering ratio parameters and RVI.Machine learning models with different feature combinations had shown good performance,the coefficient of determination(R2)of the machine learning model based on the polarimetric decomposition parameters was bigger than 0.800,and the root mean square error(RMSE)was less than 88.000 Mg/hm^(2),and the best effect was the RF model based on the optimal feature combination,which increased R2 by 0.144 and decreased RMSE by 30.327 Mg/hm^(2) compared with the polarimetric decomposition parameters alone.The polarimetric scattering ratio parameter had certain potential in the estimation of forest AGB,the introduction of RVI improved the accuracy of the model,the model-based decomposition was suitable for forest AGB estimation,and the machine learning model based on feature screening can better invert forest AGB,and there was no obvious saturation point when the AGB reached 400 Mg/hm^(2).
作者 余琼芬 岳彩荣 罗洪斌 罗广飞 段云芳 孙妙琦 恒承志 徐天蜀 YU Qiongfen;YUE Cairong;LUO Hongbin;LUO Guangfei;DUAN Yunfang;SUN Miaoqi;NEHG Chengzhi;XU Tianshu(College of Forestry,Southwest Forestry University,Kunming 650224,China;Forestry 3S Technology Engineering Center of Yun-nan Province,Kunming 650224,China)
出处 《森林工程》 北大核心 2024年第5期17-29,共13页 Forest Engineering
基金 国家自然基金项目(42061072) 云南省科技厅重大科技专项项目(202002AA00007-015) 西南林业大学预科基金项目(18200139)。
关键词 UAVSAR 极化分解 极化散射比参数 特征筛选 机器学习 森林AGB UAVSAR polarimetric decomposition polarimetric scattering ratio parameter feature screening machine learn-ing forest AGB
  • 相关文献

参考文献4

二级参考文献35

共引文献105

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部