多尺度土壤质地与光谱空间非平稳性关系探究

Exploration of the spatial non-stationary relationship between soil texture and spectra from a multiscale perspective

基于210个杞麓湖流域实测土壤质地样本数据,对比了偏最小二乘回归(PLSR)、地理加权回归(GWR)、多尺度地理加权回归(MGWR)3种不同方法反演土壤质地的效果,并探究差异化作用尺度下土壤质地与光谱关系的空间异质性。结果表明,MGWR的R2、MAE、RMSE指标优于GWR和PLSR,其拟合效果最好,原因在于MGWR考虑了系数间差异化的空间尺度,能更好拟合土壤质地(黏粒、粉粒和沙粒)与光谱之间的关系;MGWR回归结果揭示了不同光谱潜变量对黏粒、粉粒和沙粒含量的多尺度差异化的空间响应规律。其中,光谱潜变量LV1对黏粒、粉粒、沙粒响应程度最大,其MGWR标准化回归系数分别为0.373、0.426、0.422。

A total of 210 samples collected from the Qilu lake basin were used.The performances of partial least squares regression(PLSR),geographically weighted regression(GWR)and multi-scale geographically weighted regression(MGWR)in measuring soil texture were evaluated,and the spatial heterogeneity of the relationship between soil texture and spectra under different scales were ex⁃plored.The results showed that the R2,mean absolute error and root mean square error of MGWR were better than those of GWR and PLSR,which indicated that MGWR outperformed other models.This was because MGWR can better fit the relationship between soil texture(clay,silt and sand contents)and spectra by considering different scales of different variables coefficients.MGWR results re⁃vealed that the spatial scale of spatial response of different spectral latent variables to clay,silt and sand content was different.Specifi⁃cally,the spectral latent variable LV1 had the largest response to soil texture.Its standardized coefficients of clay,silt and sand con⁃tents were 0.373,0.426 and 0.422,respectively.