Application of logistic models to predict human-caused forest fires
Georg Buchholz
Fire Ecology Research Group, Max Planck Institute for Chemistry, Biogeochemistry Department,
c/o Freiburg University, P.O. Box, D-79085 Freiburg, Germany.
Tel.
: +49-761-808011,
Fax: +49-761-808012,
E-mail: buchholg@ruf.uni-freiburg.de
Abstract:
The logistic model was used to predict human-caused forest fires in the Krasnoyarsk Region, Siberia. The database included fire and weather observations for fire season 1987 to 1991. Four models were developed using the Canadian Fire Weather Index Variable FFMC. The models were tested on an independent dataset with quadratic scoring rules. Average score was for all four models over 0.75. Furthermore, a two by two classification table was used to test the performance of the models. In this case the overall percentage of correctly predicted values was between 75% and 84%. These models can be used on an operational basis in assisting fire management to assess the fire danger on a daily base in the Krasnoyarsk region.
