尺度效应对多光谱遥感监测土壤水分影响研究

Study on the Effect of Scale Effect on Soil Moisture Monitoring by Multispectral Remote Sensing

尺度效应是指在无人机遥感观测中,随着遥感分辨率的变化,各尺度遥感反演得到的数据表现不一致的现象,影响多光谱遥感监测土壤水分的精度。为探究尺度效应对于无人机遥感监测土壤水分的影响,以冬小麦为研究对象,通过无人机搭载多光谱相机分别在19、37、55、74、92 m的不同高度(对应分辨率为10、20、30、40、50 mm)拍摄多光谱图像,并同时采集土壤含水率数据。使用ENVI5.3(64 bit)对多光谱图像进行掩膜处理,包括无掩膜处理(no mask,NM)、去土壤掩膜处理(Soil background removal by masking,SRM)、去土壤去阴影掩膜处理(Soil and shadow background removal by masking,SSRM),获取各个高度下冬小麦的纹理特征。使用灰色关联法(Grey Relational Analysis,GCA)优选数据后,采用多元线性回归(Multiple Linear Regression,MLR)、BP神经网络(Back PropagationNeural Network,BPNN)和随机森林(Random Forest,RF)3种方法反演土壤含水率,使用R2和RMSE评价反演效果。研究结果表明:在采取的各种掩膜处理中,无掩膜方法反演效果最好。在各掩膜处理方法下表现最好反演模型是BPNN模型,在大部分情况下表现稳定有较好的反演效果,证明机器学习在遥感监测领域应用的可行性。冬小麦土壤含水率30 mm-NM-RF模型为最佳反演模型。在分辨率为40 mm,即无人机飞行高度74 m时,不同模型综合反演效果最佳,研究成果可应用于今后无人机监测土壤水分确定飞行高度。

The scale effect refers to the phenomenon that the data obtained from the remote sensing inversion of each scale are not consistent with the change of remote sensing resolution in the UAV remote sensing observation,which affects the accuracy of soil moisture monitoring by multispectral remote sensing.In order to investigate the influence of scale effect on soil moisture monitoring by UAV remote sensing,winter wheat was taken as the research object,and multispectral images were taken at different heights of 19 m,37 m,55 m,74 m and 92 m(corresponding to resolutions of 10 mm,20 mm,30 mm,40 mm and 50 mm)by UAV with multispectral cameras,and soil moisture content data were also collected.The multispectral images were masked by using ENVI 5.3(64 bit),including no mask(NM),soil background removal by masking(SRM),soil and shadow background removal by masking(SSRM),and soil and shadow background removal by masking(SSRM),to obtain the texture features of winter wheat at each height.After optimizing the data by using Grey Relational Analysis(GCA),the three methods of multiple linear regression(MLR),Back PropagationNeural Network(BPNN)and random forest(RF)were used to invert the data.The results of the study showed that among the various mask treatments adopted in this paper,the unmasked method had the best inversion results.The best performing inversion model among the mask treatments is the BPNN model,which performs consistently well in most cases,demonstrating the feasibility of machine learning applications in remote sensing monitoring.The best inversion model was the 30 mm-NMRF model for winter wheat soil moisture content.At a resolution of 40 mm,i.e.a UAV flight altitude of 74 m,the inversion of different models was the best,and the research results can be applied to determine the flight altitude for future UAV monitoring of soil moisture.