On the Application of Modelling Forest Fire in the Environment: A Bayesian Model Averaging Approach

One important environmental concern that has the potential to inflict serious ecological harm and also affect human life is a forest fire. Several factors can lead to forest fires, but uncertainties on the correct model specification can be a serious issue in the model selection processes. The variable selection approach Bayesian Model Averaging (BMA) accommodates for variability in model by aggregating the quantities by their Posterior Model Probabilities (PMP). This study used a model averaging strategy to model the major factors contributing to forest fires using a model prior that is uniform in choice as recommended by most in literature and uniform information parameter prior. The covariates Duff Moisture Code (DMC), and temperature with a Posterior Inclusion Probability (PIP) of 100% and a fairly large coefficient band appear to be the most relevant of the ten predictors investigated. Other relevant covariates in modeling burned area in the forest are y axis geographical coordinate with PIP of 91.8% and wind with PIP of 84.5%. With a Posterior Model Probability, the best model, 53.4%, found involves the y axis spatial coordinate, DMC, temperature and wind. Having a PIP of more than 50%, the four variables are important in modeling forest fires.