论文部分内容阅读
为实现干旱地区土壤水分含量(soil moisture content,SMC)的快速监测,该文以渭干河-库车河绿洲为靶区,采用小波变换(wavelet transform,WT)对反射光谱进行1~8层小波分解,通过相关性分析确定最大分解层数,再通过竞争性自适应重加权(competitive adaptive reweighted sampling,CARS)滤除冗余变量,筛选出与SMC相关性较好的波长变量,并叠加各层特征光谱的优选波长变量作为最优变量集,用偏最小二乘回归(partial least squares regression,PLSR)构建土壤水分含量预测模型并进行分析。结果显示:1)小波分解过程中,土壤反射率与SMC的相关性不断增强,到小波变换第6层分解(L6)处达到最高,因此小波变换最大分解层数为6层分解;2)通过对土样进行WT-CARS耦合算法筛选出变量,得出的最优变量集包括400~500、1 320~1 461、1 851~1 961、2 125~2 268 nm区域之间共131个波长变量;3)相对于全波段预测模型,各层特征光谱的CARS优选变量预测模型的精度均高,并且基于最优变量集的预测模型的精度最高,该模型的建模集均方根误差0.021、建模集决定系数0.721、预测集均方根误差0.028、预测集决定系数0.924、相对分析误差2.607。说明WT-CARS耦合算法使其在建立模型时尽可能少地损失光谱细节、较为彻底的去除噪声,同时还能对无信息变量进行有效去除,为该研究区SMC的预测提供新的思路。
In order to rapidly monitor soil moisture content (SMC) in arid regions, this paper takes the Weigan-Kuche River oasis as the target area and carries out wavelet transform (WT) Wavelet decomposition was used to determine the maximum decomposition level through correlation analysis. Then, the redundant variables were filtered out by competitive adaptive reweighted sampling (CARS), and the wavelength variables with good correlation with SMC were screened. The optimal wavelength of layer spectra was used as the optimal set of variables, and the soil moisture content prediction model was constructed by partial least squares regression (PLSR). The results show that: 1) During the wavelet decomposition, the correlation between soil reflectivity and SMC is increasing, reaching the maximum at the sixth layer decomposition (L6) of the wavelet transform, so the maximum number of decomposition layers in the wavelet transform is 6-layer decomposition; 2) The WT-CARS coupling algorithm was used to screen out the variables. The optimal set of variables was obtained from 131 samples with wavelengths ranging from 400 to 500,1 320 to 1 461, 1 851 to 1 961 and 2 125 to 2 268 nm 3) Compared with the whole band prediction model, the accuracy of the CARS preferred variable prediction model for each layer of the characteristic spectrum is high, and the prediction model based on the optimal variable set has the highest accuracy. The model set has a root mean square error of 0.021 , Modeling set decision coefficient 0.721, prediction set root mean square error 0.028, prediction set decision coefficient 0.924, and relative analysis error 2.607. It shows that the WT-CARS coupling algorithm makes it possible to minimize the spectral details, minimize the noise and eliminate the no-information variables effectively. It provides a new idea for the SMC prediction in the study area.