Forest fire occurs in the Russian Far East in summer of every year, and it makes many forest resources lost and accelerates global warming further. Because a large-scale forest fire occurs in this region at intervals of several years, it is necessary to develop the method for estimating the burnt down area accurately. In this paper, we present a time-series analysis method for detecting the forest fire from NOAA AVHRR images. This method is composed of two parts. One is hot spot detection, and another is a geometrical correction. The hot spot detection is based on two-dimensional histogram method. The geometrical correction is based on the GCP correction. A feature coastline in the Far Eastern Russian region was used to correct the gap between the image and the ground control point (GCP). It was corrected by the projective transformation because it was caused from the attitude of the satellite. This correction is automatically done and is processed within tens of seconds per scene. We used 1185 scenes from May to October from 1994 to 2003. To show the effect of the correction of the GCP, we compared the result of the correction by the GCP with that of no correction. We successfully processed all scenes and created fire maps. And we estimated the burnt down area of each year. As a result, we clarified that there were quite a lot of burnt down areas compared with other years in 1998 and 2003.