Forest fire detection based on temporal and spatial correction of background brightness temperature using GF-4 PMI data

Abstract. Forest fires threaten human life and property, making timely and accurate fire monitoring essential for fire prevention and control efforts. Satellite remote sensing meets the requirements of large-scale, high-frequency observations for forest fire monitoring and has been widely applied in this field. Chinese Gaofen-4 (GF-4) satellite, a geostationary satellite equipped with a mid-infrared sensor, holds significant potential for forest fire monitoring. However, existing fire detection methods for GF-4 data, which use fixed initial thresholds and insufficiently account for the influence of fires on background brightness temperatures, often result in high rates of false positives and missed detections. To maximize the application potential of GF-4 data in forest fire monitoring and improve the accuracy of fire detection, this study proposes a novel fire detection method based on spatiotemporal correction of background brightness temperature, tailored to the characteristics of GF-4 PMI data and incorporating a contextual fire detection approach within the infrared spectrum. In this method, dynamic thresholds based on brightness temperature distributions are employed to extract potential fire points, and the background brightness temperature is corrected by utilizing imagery from the same time on the previous day and the brightness temperature from the outer edges of the background window, thereby reducing fire effects on background temperatures. Final fire detection is achieved by distinguishing potential fire points based on the difference between the brightness temperatures of potential fire points and the corrected background, effectively filtering false positives. In case studies of two fires in Ganzi Tibetan Autonomous Prefecture, Sichuan Province, and Chongqing, China, visually interpreted fire detection results were used as references. The proposed method significantly reduced false and missed detections compared to traditional contextual threshold methods. It achieved an overall evaluation index exceeding 0.81, demonstrating high reliability and applicability for forest fire detection and extraction using GF-4 PMI imagery.