Detection Method Of Spatio-temporal Temperature Anomalies And Application In Coal Fire Area

Coal fire disasters are widely distributed worldwide,and China is one of the countries most severely affected by coal fire disasters.Although efforts have been made by the national and local governments,early detection of coal fires to reduce mitigation costs remains a global challenge.Traditional temperature anomaly detection methods are unable to meet the requirements of "homogeneity"(various manifestations of anomalies from the same fire source),"multi-phenomena"(multiple surface anomalies),and "multi-image"(image containing surface anomalies from multiple data sources).In this study,we propose using multiple data sources for downscaling surface temperatures to improve the accuracy of fire area extraction and develop a coal fire area anomaly detection algorithm based on Isolation Forest.This algorithm calculates the anomaly degree of images using multiple data classes(LST,NDVI)and performs anomaly detection from temporal and spatial perspectives using a temporal overlay method.The proposed methods are applied to the Jiangjun gobi coal fire area in Xinjiang,and experiments for fire area extraction are conducted comparing Adaptive Gradient Method,Spatiotemporal Temperature Threshold Method,and Empirical Temperature Threshold Method.Finally,long-term monitoring of the fire area is performed using Seasonal-Trend Decomposition Using LOESS(STL)decomposition on the Google Earth Engine(GEE)platform.The main research content and achievements of this study are as follows:(1)Research on downscaling surface temperature based on ensemble learning.To address the problem of low resolution in Landsat 8 thermal infrared images for highprecision detection,we introduce downscaling surface temperature based on ensemble learning into the field of coal fire detection.This method uses the Stacking method to combine multiple factors for downsizing surface temperature data,using highresolution Sentinel-2 data to downscale surface temperature images to a resolution of10 m.Compared with the traditional Ts HARP method,the Stacking method improves the average absolute error,absolute median deviation,and mean square error by 0.44,1.3,and 0.39,respectively.The proposed method is then applied to coal fire detection,and experimental results show that it can effectively reduce jagged edges in the extracted fire area and improve the boundary extraction performance.(2)Research on spatiotemporal temperature anomaly detection method in coal fire areas based on Isolation Forest.In response to the limited consideration of temporal correlation and poor scalability in traditional temperature anomaly detection algorithms,the Isolation Forest anomaly detection method is introduced into the field of coal fire detection to construct a temperature anomaly detection method based on Isolation Forest.The results of fire area detection are obtained using a temporal overlay method,where the image results from different time periods are overlaid on an annual basis,yielding the spatiotemporal temperature anomaly regions of the Jiangjun Gobi fire area for each year from 2016 to 2020.(3)Analysis of temporal variations of fire area indices based on STL.Temporal data of various indices in the fire area are obtained through the GEE platform.Combined with the STL method,the fire area is analyzed and compared with the surrounding normal areas,revealing differences in various indices between the fire area and the normal areas.The significant differences obtained through differential methods combined with STL decomposition are used to analyze the trend of coal fire changes.The results of STL decomposition show that the various remote sensing indices of the Jiangjun Gobi fire area have returned to normal by early 2022,demonstrating the effectiveness and speed of the mitigation actions.(4)Development of coal fire anomaly detection software.To meet the requirements of the "Multi-source Remote Sensing Soil Moisture Monitoring and Anomaly Detection in Coal Fire Areas" National Natural Science Foundation project,we have developed a temperature anomaly detection software based on Landsat 8 data using Python and QT.The software’s overall functions include data reading,radiometric calibration,surface temperature inversion and downscaling,anomaly detection,overlay analysis,and calculation of commonly used indices,meeting the needs...