基于无人机多光谱遥感的土壤含水率反演研究

摘要

To get the soil moisture of the large scale rapidly and the best monitoring depth in bare soil by UAV muhispectral remote sensing technology,the clay loam soil was prepared into two different depths (5 cm and 10 cm) and the soil moisture ranged from 3％ to 30％ of the different samples.The UAV was equipped with a Micro-MCA muhispectral camera to monitor the soil samples at 3 p.m.for three consecutive days.The soil spectral reflectance values of six bands (490 nm,550 nm,680 nm,720 nm,800 nm and 900 nm) were collected.The surface moisture content (about 1 cm) and overall moisture content of soil samples of two different depths were also measured.The regression models between soil moisture and the reflectance of different bands were established by the regression methods of partial least squares regression,stepwise regression and ridge regression.Quantitative relationship was analyzed of the regression modes and the methods.The results showed that the three regression models had statistical significance (P ＜ 0.001) for predicting soil moisture content.The accuracy evaluation of the model through the validation set showed that the stepwise regression model had good prediction ability (R2 were 0.775,0.764,0.798 and 0.694,RMSE were 0.028,0.042,0.037 and 0.038 and RPD were 2.22,2.04,2.20 and 1.75),followed by ridge regression method and partial least squares method.The regression models of the surface soil had good inversion effect in monitoring depth.The inversion effect was decreased as the increase of monitoring depth.The relationship between the soil moisture and the wavelength of 720 nm,680 nm and 550 nm band was better among the six bands.The results showed that the best regression method was stepwise regression method,and the best monitoring depth was the surface layer (about 1 cm) of the soil samples.The research result can provide reference for the rapid monitoring of soil moisture in the area by using multispectral remote sensing of UAVs,and promote the further development of precision agriculture.%为研究无人机多光谱遥感技术对裸土土壤含水率的大范围快速测定和最佳监测深度的确定,以杨凌地区粘壤土为试验材料,分别配制成2种不同深度(5 cm和10 cm)、含水率为3％ ～ 30％的土壤样本.用无人机搭载多光谱相机对土样连续监测3d,监测时刻均为15:00.采集6个波段(490、550、680、720、800、900 nm)处的土壤光谱反射率,同时对2种不同深度的土壤样本表层(约1 cm)含水率和整体含水率进行测定.分别采用偏最小二乘回归法、逐步回归法和岭回归法,建立不同波段光谱反射率因素反演土壤含水率的回归模型,并分析其定量关系.试验结果表明,逐步回归预测精度最佳,决定系数(R2)分别为0.775、0.764、0.798、0.694,而预测均方根误差(RMSE)分别为0.028、0.042、0.037、0.038;其次为岭回归法;偏最小二乘法的预测精度最低.综合比较得最佳回归方法为逐步回归法,最佳监测深度为土壤表层(约1 cm),其次为5 cm深度,最后为10 cm深度.