A Hybrid Image Fusion Model for Generating High Spatial-Temporal-Spectral Resolution Data Using Operational Land Imager-Moderate Resolution Imaging Spectroradiometer-Hyperion Satellite Imagery
Nowadays, a large number of Earth observation satellites have been launching, with various Spatial, Temporal and Spectral Resolution (STSR), which contribute significantly to Earth surface monitoring ability. It seems that there is a boom in Earth observation field, nevertheless, because of the limitations of satellite sensor’s technology and budget constraints, there exist compromises between STSR of satellite data. That is to say, even though so many satellites have been launched, none of them can obtain high STSR data simultaneously. These compromises limit the application of existing remotely sensed data significantly, especially for the remote sensing applications that requires fine spatial detail, long-term and frequent coverage, and hyper-spectral (HS) satellite imagery, such as precise global or regional change detection, time series analysis, urban dynamic monitoring, natural disaster monitoring, real-time air quality monitoring, etc. Image fusion provides a feasible means to overcome the limitations of the current Earth observation data, mainly including spatial and spectral fusion, and spatial and temporal fusion. However, these fusion technologies can only solve part of the resolution enhancement problems, which cannot generate synthetic images with high STSR simultaneously. This study proposed a Hybrid Spatial-Temporal-Spectral image Fusion Model (HSTSFM) to generate synthetic satellite data with high STSR simultaneously, which adopts the Landsat-8 Operational Land Imager (OLI, including its panchromatic (PAN) and multi-spectral (MS) bands), Moderate Resolution Imaging Spectroradiometer (MODIS) and Hyperion images as data sources. HSTSFM blends the high spatial resolution from OLI-PAN image, i.e., 15 m, the high temporal resolution from MODIS images, i.e., daily observations, and the high spectral resolution from Hyperion image, i.e., 146 good-quality bands out of 242 bands to produce daily HS images with a spatial resolution of 15 m, which aims to satisfy the growing demand of high STSR satellite images. The proposed HSTSFM model contains three fusion steps: (1) PAN and MS image fusion with an Enhanced Synthetic Variable Ratio (ESVR) algorithm to get high spatial resolution OLI-MS image at the base date; (2) Spatial and temporal image fusion with One-Pair Image Dictionary Learning method to obtain high spatial resolution OLI-MS image at the prediction date; (3) Temporal and spectral image fusion with a Modified Color Resolution Improvement Software Package (MCRISP) algorithm to obtain the high spatial resolution Hyperion image at the prediction date, i.e., the final predicted high STSR image. The proposed method adopted the Beijing area, China as a test region, and used the actual OLI-PAN and Hyperion images at the prediction date to evaluate the precision of the fusion result from spatial detail information and spectral properties similarities aspects. The experimental results indicate that HSTSFM can capture the land cover changes accurately and has good spatial and spectral fidelity to the reference OLI-PAN and Hyperion images. The proposed HSTSFM can fuse spatial, temporal, and spectral information from different types of satellite sensors, which has the potential to generate synthetic data to support the studies that require high STSR satellite imagery.