一种改进的地理加权回归模型水深反演方法

A water depth inversion method based on improved geographically weighted regression model

利用多光谱卫星遥感影像反演浅海水深是水深测量的一种重要方式。提出一种基于主成分分析的地理加权回归模型(PCA-GWR),采用WorldView-2多光谱卫星遥感影像数据,对经过数学变换后的波段反射率数据先进行主成分分析,将得到第一主成分量进行地理加权回归分析,并与双波段比值模型、多波段线性模型和地理加权回归模型(GWR)的水深反演结果进行比较。结果显示,各个反演模型反演水深值与实测水深值的相关系数r均大于0.75,其中PCA-GWR模型水深反演结果最好,r为0.96、RMSE为1.56 m、MAE为1.06 m。研究表明,PCA-GWR模型可有效去除数据变换后的冗余信息,降低数据空间非平稳性,具有较高的反演精度与可靠性,适用于浅海水深反演。

Inversion of shallow water depth using multispectral satellite remote sensing images is an important way of water depth measurement.In this paper, a geographically weighted regression model based on principal component analysis(PCA-GWR) is proposed.Using WorldView-2 multispectral satellite remote sensing image data, the band reflectivity data after mathematical transformation is firstly analyzed by principal component analysis, and the obtained first principal component quantity is analyzed by geographically weighted regression, and compared with the water depth inversion results of dual-band ratio model, multi-band linear model and geographically weighted regression model(GWR).The results show that the correlation coefficients between the inverted water depth values of each inversion model and the measured water depth are all greater than 0.75,and the PCA-GWR model has the best water depth inversion results, with r of 0.963,RMSE of 1.56 m, and MAE of 1.06 m.The research shows that the PCA-GWR model can effectively remove redundant information after data transformation, reduce data spatial non-stationarity, and has high inversion accuracy and reliability, which is suitable for shallow water depth inversion.