Application of Bispectral Approach in FY-4B/AGRI Sea Surface Temperature
Article
Figures
Metrics
Preview PDF
Reference
Related
Cited by
Materials
Abstract:
Fengyun-4B is the first operational satellite of the second-generation FengYun geostationary meteorological satellite. The Advanced Geostationary Radiation Imager (AGRI), a multiple channel radiation imager, which adds a low-level water vapour detection channel and an adjusted spectrum range of four channels to improve the quality of observation, is one of the primary payloads onboard FY-4B. As one of the basic quantitative remote sensing products of FY-4B/AGRI, the operational sea surface temperature (SST) derives from the split-window nonlinear SST (NLSST) algorithm in real time. The stripe noise is a common issue in sea surface temperatures (SSTs) retrieved from thermal infrared data obtained by satellite-based multidetector radiometers. It degrades not only image quality but also the accuracy of retrieved SSTs. The stripe noise is observed in FY-4B/AGRI SSTs. It is more obvious in the brightness temperature difference (BTD) of the split window data, but the stripe noise is invisible in brightness temperature (BT) images. The stripe noise originates from the relative noise in the BTD. It propagates into SSTs by degrading the atmospheric correction. The bispectral filter approach for removing the stripe noise is applied to FY-4B/AGRI data. The bispectral filter is a Gaussian filter and an optimal estimation method for the differences between the data obtained at the split window. A kernel function based on the physical processes of radiative transfer has made it possible to reduce stripe and random noise in retrieved SSTs without degrading the spatial resolution or generating bias. For the assessment of the bispectral filter approach, the retrieved FY-4B/AGRI SST is validated against in-situ SST measurements available from in-situ SST Quality Monitor (iQUAM). Robust statistics are used to assess the impacts of the bispectral filter on SST accuracy. The accuracy and precision of the bispectral filter approach are assessed by determining the robust standard deviation and median bias between FY-4B/AGRI SST and quality-controlled in-situ SST from May to Norember 2023. The matchup space-time window is 4 km and 30 mins from the buoys’ location to the centre of the SST pixel. The validation results demonstrate the effectiveness of the bispectral filter, which reduces stripe noise in the retrieved FY-4B/AGRI SSTs. The image of a bispectral-filtered BTD is clearer than that of an unfiltered BTD. It also improves the accuracy of the SSTs by about 0.04 K to 0.06 K in the robust standard deviation. Furthermore, the bispectral filter approach is based on a simple Gaussian filter and is easy to implement. However, the bispectral filter cannot remove the stripe noise in BT. Such noise should be removed before applying the bispectral filter.