Accounting for temporal information from dense time-series coarse-scale satellite data for spatio-temporal downscaling

초록

A spatio-temporal downscaling model is proposed in this paper to fuse complementary information in spatial and temporal resolution of remote sensing data. The proposed model is based on decomposition of a target attribute of interest into trend and residual components, which is named as a decomposition-based spatiotemporal downscaling (DSPAT). Based on the component decomposition, DSPAT is composed of (1) estimation of trend components and (2) correction of residual components. Unlike conventional spatio-temporal downscaling models, DSPAT fully considers temporal information of dense time-series coarse-scale (DC) data during trends estimation. The residuals are predicted by a geostatistical interpolation at a dense time-series coarse-scale and then incorporated with the trends. To evaluate the applicability of DSPAT, an experiment was conducted using simulated reflectance data. In addition, its performance was compared with conventional models including a spatial and temporal adaptive reflectance fusion model (STARFM) and an enhanced STARFM (ESTARFM). From the experimental results, DSPAT showed better performance than STARFM and ESTARFM even when the difference between the prediction date and the acquisition time of both DC and SF increased. These results demonstrate the potential of DSPAT that can fully account for the temporal information of DC data. © 2020 40th Asian Conference on Remote Sensing, ACRS 2019: Progress of Remote Sensing Technology for Smart Future. All rights reserved.

키워드