M. Montesino, A. Fernández Militino, M. D. Ugarte Martínez
Cloud deletion, image reconstruction and outlier removal are key steps in pre-processing satellite images. Existing geostatistical methods are computationally demanding for data-intensive tasks. Hence, we propose the ‘Spatial Functional Prediction’ method (SFP) where; (1) images are spatially aggregated, (2) time-series of pixels are deconstructed into trend, seasonal and error components, (3) trends are defined as functional data and predicted with ordinary kriging, and (4) trends, seasonal and error terms are reassembled. To illustrate the application of SFP, we raise a simulation study using remote sensing data in northern Spain (Navarre), where clouds, noise and missing data are artificially introduced into the series of images. The performance of SFD is tested and compared against Thin-Plane Splines. Results show that SFP slightly outperforms Tps while the execution time is considerably lower. Therefore, SFP could be convenient for large data applications.
Keywords: Geostatistics, Remote Sensing, Spatial Functional Data, Thin-Plate Splines
Scheduled
EET-2 Spatial Statistics and Temporal Space
September 5, 2019 10:40 AM
I2L5. Georgina Blanes building