Image Mining within Meteosat Data: A Case of Modeling Forest Fire

Remote Sensing Images are being collected nowadays every 15 minutes from satellites such as Meteosat, covering large areas of land. These repositories of images can be used for a range of different purposes. For the human mind, it may be hard to consider each image individually, analyze it as well as their relationships with the previous images of varying time steps. In order to address that issue, this research attempts to develop a simple, time efficient and effective generic model to facilitate the process of pattern discovery from series of remote sensing imageries. The data used for the study were of Meteosat Second Generation. The focus of this study is on development of a model for monitoring and analyzing forest fires in space and time. As a case, a diurnal cycle of fire, which took place in Portugal, on 28th of July, 2004 was taken and analyzed. Kernel convolution method was used to characterize the hearth of the fire in space. The patterns of these fire objects in space were then extracted and tracked over time. These algorithms were automated to analyze for a series of imageries. The results thus obtained forms the knowledge gained about the fire in space and time. This knowledge was used for a further understanding of the behavior of fire with respect to vegetation and wind. This mining model allows one to better understand the behavior of the fire in space and time. Such a model may then be useful for making predictions of hazards at an almost real time basis. The research is promising for data mining, possibly allowing other spatio-temporal phenomena to be modeled as well.