摘要
为探究植被覆盖条件下GF-1卫星反演农田土壤含水率的可行性,以河套灌区解放闸灌域沙壕渠为研究区,采用GF-1卫星遥感影像作为数据源,通过全子集筛选法确定不同土壤深度下光谱指数的最优自变量组合,并分别采用多元线性回归(MLR)、BP神经网络(BPNN)、支持向量机(SVM)3种算法,构建不同深度下土壤含水率反演模型。结果表明,全子集筛选后模型反演精度有较大提升,且过拟合现象减弱;植被覆盖条件下各深度土壤含水率敏感程度从大到小依次为0~40 cm、0~60 cm、20~40 cm、0~20 cm、40~60 cm;植被覆盖条件下各模型对土壤含水率反演能力由强到弱依次为BPNN、SVM、MLR;筛选后BPNN在深度0~40 cm下的建模集和验证集R~2均能达到0.50以上,RMSE在0.02%以内。研究结果可为植被覆盖条件下利用GF-1卫星监测农田土壤含水率提供参考。
In order to explore the feasibility of GF-1 satellite inversion of farmland soil moisture content(SMC)under the condition of vegetation coverage,taking Shahaoqu District of Hetao Irrigation Area as study area,and GF-1 satellite remote sensing images as the data source.Simultaneously,the soil moisture content data were collected with various depths at 0~20 cm,20~40 cm,40~60 cm,0~40 cm,and 0~60 cm.Then a set of independent variables,including four bands and 15 spectral indices were obtained based on the GF-1 data,and the full subset selection was used to select the optimal combination of independent variables at five depths.Based on these,the combinations before and after full subset selection were used to build soil moisture content inversion models(multiple linear regression,MLR;back propagation neural network,BPNN;support vector machines,SVM)at five depths in the vegetated area,and evaluate the sensitivity of GF-1 to SMC at different depths and the inversion capability of the models.The model performance was assessed by using adjusted coefficient of determinationR~2and root mean square error(RMSE).The results showed that the model inversion accuracy was greatly improved after the full subset selection,and the over-fitting phenomenon can be reduced.The sensitivity of GF-1 to the SMC at different depths under vegetation coverage was ordered from the largest to the smallest as follows:0~40 cm,0~60 cm,20~40 cm,0~20 cm,and 40~60 cm.The SMC inversion capabilities of all the three models under vegetation coverage ordered from the largest to the smallest were as follows:BPNN,SVM,and MLR.After the full subset selection,the R~2of the modeling set and verification set of BPNN at depth of 0~40 cm can reach more than 0.50,and the RMSE was within 0.02%.The research result can provide a reference for using GF-1 satellite to monitor SMC of farmland under vegetation coverage.
作者
姚一飞
王爽
张珺锐
黄小鱼
陈策
张智韬
YAO Yifei;WANG Shuang;ZHANG Junrui;HUANG Xiaoyu;CHEN Ce;ZHANG Zhitao(College of Water Resources and Architectural Engineering,Northwest A&F University,Yangling,Shaanxi 712100,China;The Key Laboratory of Agricultural Soil and Water Engineering in Arid Areas,Ministry of Education,Northwest A&F University,Yangling,Shaanxi 712100,China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2022年第9期239-251,共13页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家重点研发计划项目(2017YFC0403302)
国家自然科学基金项目(41804029、51979232、51979234)
关键词
土壤含水率
遥感
反演
GF-1卫星
全子集筛选
光谱指数
soil moisture content
remote sensing
inversion
GF-1 satellite
full subset selection
spectral index
