A fully-integrated, Hyperspectral optical and SAR (Synthetic Aperture Radar) constellation of small earth observation
satellites will be deployed over multiple launches from last December to next five years. The Constellation is expected to
comprise a minimum of 16 satellites (8 SAR and 8 optical ) flying in two orbital planes, with each plane consisting of
four satellite pairs, equally-spaced around the orbit plane. Each pair of satellites will consist of a
hyperspectral/mutispectral optical satellite and a high-resolution SAR satellite (X-band) flying in tandem. The
constellation is expected to offer a number of innovative capabilities for environment monitoring. As a pre-launch
experiment, two hyperspectral earth observation minisatellites, Spark 01 and 02 were launched as secondary payloads
together with Tansat in December 2016 on a CZ-2D rocket. The satellites feature a wide-range hyperspectral imager.
The ground resolution is 50 m, covering spectral range from visible to near infrared (420 nm - 1000 nm) and a swath
width of 100km. The imager has an average spectral resolution of 5 nm with 148 channels, and a single satellite could
obtain hyperspectral imagery with 2.5 million km2 per day, for global coverage every 16 days. This paper describes the
potential applications of constellation image in environment monitoring.
In the southwest of China, it is anticipated that synthetic aperture radar (SAR) will become an important tool for forest inventory because of its all-weather capabilities. The Zhazuo area in Guizhou Province of southwest China, with a typical Karst landform, was selected as the test site. Six RADARSAT-2 polarimetric images were acquired in order to analyze polarimetric backscattering behavior and temporal variation of forest and deforested area. Polarimetric decomposition was conducted, and Pauli and Freeman-Durden decomposition were demonstrated to be more suitable for identifying forest and deforestation respectively. Finally, a scheme for multitemporal polarimetric SAR data fusion was proposed, which could greatly improve image quality and make forest identification more efficient. Support vector machine classification showed that the overall accuracy for forest identification was 87.63%, and the accuracy could be enhanced to 91.49% after gamma filtering.