A large-region fog detection algorithm at dawn and dusk for high-frequency Himawari-8 satellite data

Abstract

ABSTRACT Satellite remote sensing provides effective methods for fog detection at large scales. At dawn or dusk, however, the separation of fog from the surface remains a great challenge because of the minor radiation difference between fog and the surface. This study develops a novel algorithm for fog detection suitable for dawn and dusk from high-frequency data of the Himawari-8 satellite. Methodologically, the area was divided into three subregions according to the distance of the terminator line. In regions at daytime that are far from the terminator line, a fog detection index of the low solar altitude angle (FDI LSAA) was built to isolate the fog through the relationship between the mid-infrared (MIR) brightness temperature (I MIR ) and its difference with the thermal infrared (TIR) brightness temperature (BTD = I MIR – I TIR , where I MIR and I TIR are the brightness temperatures at the MIR and TIR bands). In regions near the terminator line, the fog was extracted from the surface by a Gaussian mixture model (FD-GMM NTL) by the differences in MIR and TIR radiation changes. After that, the fog was further extracted from the low and mid-high clouds by the enhanced low cloud detection index (ELCDI) and by the difference in BTD and TIR characteristics. In regions at night, a BTD threshold value was adopted to detect fog. Validation results demonstrated that the algorithm in this study could precisely detect large-region radiation fog including dawn and dusk, with the overall probability of detection (POD) accuracies range from 76.3% to 89.2% at dawn and 67.3% to 77.0% at dusk. The relatively low accuracy at dusk was mainly attributed to desert in the study area, which has a similar BTD with fog after a day of solar radiation. Results of this study provide new insight for fog detection at dawn and dusk, which enhances the application of remote sensing in meteorology forecasting.