A Data Mining Framework for Forest Fire Mapping

Abstract

Forest fires constitute the major reasons for the loss of biodiversity and degradation of ecosystems. Locally, forest fires are one of the major natural risks in the Kroumirie mountains, northwestern Tunisia. In these massifs, fires occur frequently, and this requires understanding the complex biophysical parameters of this phenomenon. The special attention of the research is paid to the spatial forecasting of forest fires. Different types of classical frequent itemset algorithms have been tested and employed to reveal forest fire patterns that relate the spatial parameters with the probability of fire occurrence. Extracted frequent patterns are then being aggregated through a defined measurement of pertinence. The forest fire risk zone maps are then generated, resulting in extracted spatial patterns. The experiments showed that, the integration of these patterns into GIS could be advantageous to determine risky places and able to produce good prediction accuracy.