A spatial-temporal-spectral blending model using satellite images

Abstract

Due to the budget and technical limitations, remote sensing sensor designs trade spatial resolution, swath width and spectral resolution. Consequently, no sensor can provide high spatial resolution, high temporal resolution and high spectral resolution simultaneously. However, the ability of Earth observation at fine resolution is urgently needed for global change science. One possible solution is to “blend” the reflectance from a variety of satellite data sources, including those providing high spatial resolution and less frequent coverage (e.g., Landsat Thematic Mapper, TM), daily global data (e.g., Moderate Resolution Imaging Spectroradiometer, MODIS), and high spectral resolution and infrequent revisit cycle (e.g., Hyperion). However, the previous algorithms for blending multi-source remotely sensed data have some shortcomings, especially with regard to hyperspectral information. This study has developed a SPAtial-Temporal-Spectral blending model (SPATS) that can simulate surface reflectance with high spatial-temporal-spectral resolution. SPATS is based on an existing spatial-temporal image blending model and a spatial-spectral image blending model. The performance of SPATS was tested with both simulated and observed satellite data, using Landsat TM, Hyperion and MODIS data, as well as heterogeneous landscapes as examples. The results show that the high spatial-temporal-spectral resolution reflectance data can be applied to investigations of global landscapes that are changing at different temporal scales.