Estimation of urban vegetation fraction by image fusion and spectral unmixing
基于图像融合与混合像元分解的城市植被盖度提取

Estimation of urban vegetation fraction is helpful for urban green space protection and urban land use planning. With the development of remote sensing technologies, the spectral unmixing method has been widely used in estimating urban vegetation fraction based on middle-resolution multispectral imagery. However, the spectral unmixing method largely depends on the spatial resolution of the images used, limiting its extensive applications in practice. Taking Hangzhou as a case study, we proposed the Gram-Schmidt (GS) algorithm to fuse the Landsat Enhanced Thematic Mapper plus (ETM+) PAN band with the ETM+ multispectral bands. A linear model of spectral unmixing was then applied in the estimation of vegetation fraction based on the fused ETM+ image. Finally, the accuracy of vegetation fraction derived from the fused ETM+ image was assessed using high-resolution SPOT imagery. The results show that the fused image had a higher standard deviation, information entropy and average gradient than the original image. The relative deviation between the images was less than 0.07, indicating advantages of increasing spatial resolution while maintaining spectral consistency with the original image by GS method. Based on random sampling, we found the estimated results of vegetation fraction from the fused ETM+ and SPOT images were comparable, which was reflected by more than 75% of samples having similar values of vegetation fraction from the two data sources, except for a few unmatched pixels with very high or low vegetation fraction. Furthermore, the root-mean-square error and systematic error of the fused image decreased by 0.01 compared with those of the original image. These results suggest that the new method holds potential for improving the estimation accuracy of urban vegetation fraction without the substantial cost of acquiring high spatial resolution images.