Abstract: The urban thermal anomaly extraction (U-TAE) algorithm was proposed for the study of urban heat island, which was evolved from the TAE algorithm for thermal anomalies extraction of underground coal fires. The U-TAE algorithm detects the entire thermal infrared image by moving windows with different sizes, and the key of this robust algorithm was to set up thresholds for pixels with land surface temperatures (LST) higher than mean plus standard deviation. Mean and standard deviation were calculated from all pixels and those pixels in the moving windows (scope from 5×5 to 151×151 pixels). This algorithm can distinguish UHI areas from background areas according to the cumulative times of pixels, and determine UHI intensity according to the cumulative times in UHI areas. Cumulative times of zero are background areas and the others are UHI areas. This U-TAE algorithm has dynamic thresholds and unbias characteristics. Results showed that the heat island areas were between 55.31 km² and 68.35 km² in Changsha in 2001. In UHI intensity analysis, the UHI studies at macro-scale should use larger moving windows to reflect the relationship between urban and rural areas. Micro-scale studies should use smaller moving windows to analyze specific distribution of UHI, underlying surface and driving factors. Compared with traditional methods, U-TAE algorithm makes the results comparable in different time periods.

Key words: remote sensing, urban heat island, TAE algorithm, moving window