标题：Automated volcanic hot-spot detection based on FY-4A/AGRI infrared data
作者：Chu, S. S.; Zhu, L.; Sun, H. F.; Li, Q. W.; Zhang, X. R.; Chen, T. T.; Qiao, L.; Zhu, W. R.; Zhao, D. X.; Zhang, Y. H.
作者机构：[Chu, S. S.; Sun, H. F.; Li, Q. W.; Zhang, X. R.; Qiao, L.; Zhu, W. R.; Zhao, D. X.; Zhang, Y. H.] China Univ Min & Technol Beijing, Coll Geosci & Sur 更多
通讯作者：Sun, HF;Sun, HF
通讯作者地址：[Sun, HF]China Univ Min & Technol Beijing, Coll Geosci & Surveying Engn, Beijing, Peoples R China.
来源：INTERNATIONAL JOURNAL OF REMOTE SENSING
摘要：A new algorithm (FYVOLC) for the automated detection of volcanic hot spots using data obtained from China's new-generation satellite FY-4A is developed and tested. FYVOLC improves the detection ability of the Volcanic Anomaly SofTware (VAST) algorithm by incorporating a Normalized Brightness Temperature Difference Index (NBTDI) to reduce the influence of 'cold' cloud. In addition, FYVOLC introduces a mid-infrared brightness temperature criterion to identify volcanic hot spots by making calculations based on the image itself without artificially determining any parameters. To test the volcanic hot-spot detection performance of FYVOLC, FY-4A Advanced Geostationary Radiation Imager data were used for eruptions from four volcanoes: Mayon Volcano in the Philippines (25-26 January 2018), and Bromo (1-2 September 2018), Lawu (3-4 September 2018), and Soputan volcanoes (3-4 October 2018) in Indonesia. A total of 147 images of the above four volcanoes were used, and the results obtained using the FYVOLC algorithm were compared with those from three existing volcanic hot-spot detection algorithms: the simplified contextual, VAST, and HOTSAT algorithms. It is shown that the simplified contextual and VAST algorithms are prone to generating false alerts (with a maximum false alert rate of up to 41% and 48%, respectively), whereas the simplified contextual and HOTSAT algorithms are prone to missing hot-spot pixels (with a maximum miss rate of up to 71% and 54%, respectively). The FYVOLC algorithm has the best detection accuracy owing to the adopted NBTDI and image-based mid-infrared brightness temperature criterion. The maximum false alert rate of FYVOLC is 12%, and the maximum miss rate is 11%. By analysing the thermal anomaly time-series of the 147 images, it was found that the detection results of FYVOLC are basically consistent with the actual hot spots, except for some images that were strongly affected by cloud cover. This study is the first to realize the automated detection of volcanic hot spots and monitor temporally dynamic thermal phenomena based on FY-4A satellite data. The results have significance for the continuing development of global volcanic early-warning systems and for the dynamic monitoring of volcanoes after eruptions.