Coal fire detection and evolution of trend analysis based on CBERS-04 thermal infrared imagery

Abstract

Coal fire hazards not only causes loss of coal resources but also induces serious geological hazards and environmental degeneration. Wuda Coalfield (China), known for its widespread coal fires, was used as the study area for this study. Owing to an inherent noise problem, CBERS (China & Brazil Earth Resource Satellite)-04 thermal infrared images have not been explored and applied in the field of coal fire detection. An adaptive-edge-threshold (AET) method was proposed to delineate coal fire maps based on nighttime CBERS-04 data from 2015 to 2020. Iterative Self-Organizing Data Analysis Technique Algorithm (ISODATA) was employed to cluster coal fire spots into different classes, then coal fire propagation directions were generated by connecting the clustering centers of the same class in chronological order. The results of quantitative analysis of coal fires showed that AET method achieved the highest accuracy among four coal fire recognition algorithms; the accuracy of AET in field verification reached 81.25%. The total area of coal fires increased by 137%, while the total intensity of coal fires decreased by 18% from 2015 to 2020. Overall accuracies of ISODATA clustering coal fire spots were between 90.25% and 100.00%, the Kappa coefficients were between 0.82 and 1.00. The evolution trend of coal fires obtained based on this clustering accuracy is reliable. The evolution trend of coal fires is as follows: coal fires of Class I spread to the northwest; coal fires of Class II first propagated to the northwest, then to the south and then to the west; coal fires of Class III essentially stayed in the same place; coal fires of Class IV spread to the northwest; coal fires of Class V spread to the southeast and then to the northeast.