Downscaling remotely sensed imagery using area-to-point cokriging and multiple-point geostatistical simulation
Downscaling remotely sensed imagery using area-to-point cokriging and multiple-point geostatistical simulation
复制标题
DOI：
10.1016/j.isprsjprs.2014.12.016
复制
发表时间：
2015-03
影响因子：
12.7
通讯作者：
Yunwei Tang;P. Atkinson;Jingxiong Zhang
中科院分区：
文献类型：
--
作者：
Yunwei Tang;P. Atkinson;Jingxiong Zhang
A cross-scale data integration method was developed and tested based on the theory of geostatistics and multiple-point geostatistics (MPG). The goal was to downscale remotely sensed images while retaining spatial structure by integrating images at different spatial resolutions. During the process of downscaling, a rich spatial correlation model in the form of a training image was incorporated to facilitate reproduction of similar local patterns in the simulated images. Area-to-point cokriging (ATPCK) was used as locally varying mean (LVM) (i.e., soft data) to deal with the change of support problem (COSP) for cross-scale integration, which MPG cannot achieve alone. Several pairs of spectral bands of remotely sensed images were tested for integration within different cross-scale case studies. The experiment shows that MPG can restore the spatial structure of the image at a fine spatial resolution given the training image and conditioning data. The super-resolution image can be predicted using the proposed method, which cannot be realised using most data integration methods. The results show that ATPCK-MPG approach can achieve greater accuracy than methods which do not account for the change of support issue.
