Abstract
Spatial downscaling of remotely sensed products is one of the main ways to obtain earth observations at fine resolution. Area-to-point (ATP) geostatistical techniques, in which regular fine grids of remote sensing products are regarded as points, have been applied widely for spatial downscaling. In spatial downscaling, it is common to use auxiliary information to explain some of the unknown spatial variation of the target geographic variable. Because of the ubiquitously spatial heterogeneities, the observed variables always exhibit uncontrolled variance. To overcome problems caused by local heterogeneity that cannot meet the stationarity requirement in ATP regression kriging, this paper proposes a hybrid spatial statistical method which incorporates geographically weighted regression and ATP kriging for spatial downscaling. The proposed geographically weighted ATP regression kriging (GWATPRK) combines fine spatial resolution auxiliary information and allows for non-stationarity in a downscaling model. The approach was verified using eight groups of four different 25 km-resolution surface soil moisture (SSM) remote sensing products to obtain 1 km SSM predictions in two experimental regions, in conjunction with the implementation of three benchmark methods. Analyses and comparisons of the different downscaled results showed GWATPRK obtained downscaled fine spatial resolution images with greater quality and an average loss with a root mean square error value of 17.5%. The analysis indicated the proposed method has high potential for spatial downscaling in remote sensing applications.
Highlights
Remote sensing techniques, with their capacities for large-scale, real-time, simultaneous, and dynamic high-frequency monitoring at low cost, have become fundamental to the provision of earth observations at regional and global scales
The four downscaling methods that consisted of the trend model and the residual model were implemented on each coarse surface soil moisture (SSM) product by incorporating three auxiliary variables (LST, normalized difference vegetation index (NDVI), and blue-sky albedo (BSA))
The downscaled predictions were compared with ground-based observations using several statistical parameters, including the root mean square error (RMSE) (m3·m−3), mean error (ME) (m3·m−3), correlation coefficient (R), and slope (SLOP) of linear regression between ground observations and downscaled predictions
Summary
With their capacities for large-scale, real-time, simultaneous, and dynamic high-frequency monitoring at low cost, have become fundamental to the provision of earth observations at regional and global scales. There are techniques for downscaling continua and downscaling categories based on whether a continuous or a categorical variable is being predicted [6] The latter, referred to as super-resolution mapping, has its own research system, and it is often used to represent land cover or land use [7]. The method of ATP regression kriging (ATPRK) has emerged [25,26] It incorporates regression kriging [27] and ATPK, and it can consider both the correlated variables and the change-of-support problem during the downscaling process. Based on ATPRK, this paper proposes a hybrid spatial statistical method that integrates GWR and ATPK for spatial downscaling with high-resolution auxiliary information.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
