Binary Regresion Model for Automated Wildfire Early Prediction and Prevention

This paper presents a model for wildfire early prediction and prevention. The model is derived by binary logistic regression using temperature, humidity, solar radiation, rain and fuel moisture conditions as initial input parameters. The input data is registered on 24/7 bases and includes 20 of the biggest fires in a 10-year period from 2007 to 2017 in a specific region. The variables in the model are presented as well as the classification table. The most significance input parameters are the fuel moisture conditions, rain precipitation as well as the maximum temperature, measured per day. The model shows very high result for true negatives and lower results for its sensitivity. The false positives and negatives are also determined. The total success rate of the model is calculated to be 84,4%. The model is tested with 3 real fires out of 15 events. The results of probability of observing fire in these 3 cases are P=78%, P=90% and P=76%.