Comparative study of basic and Bee Colony-optimized models cellular automation for prediction of wildfire spread

As one of the most widely used models for fire spread simulation, cellular automata models commonly require appropriate parameters and the optimized coefficients. The main aim of this study was to compare the basic models of cell automation set by previous studies and optimized models using the Bee Colony Algorithm (ABC) to predict the rate of fire spread in a reforestation area in northeastern Golestan province, Iran. Furthermore, a sub-objective was to test the ability of optical and radar sensors (Sentinel 1 and 2) for mapping the vegetation stand type and density required for fire spread modeling. Following pre-processing and extraction of vegetation indices, vegetation type and density were mapped by the Random Forest algorithm. The accuracy assessment showed that the best result was obtained by combining the optical and radar data (Kappa coefficient (KC) = 0.75 and Overall accuracy (OA) = 0.81). Moreover, the fire spread was modeled using previous and optimized coefficients from previous studies. Model coefficients were optimized based on environmental conditions using the BCA algorithm and were compared with the occurred fire to improve the results and comparative evaluation. The results showed that the optimized models were more consistent (Sorenson coefficient (SC) = 0.78; KC = 0.74 and OA = 0.93 for Alexandridis et al., 2011; and SC = 0.78; KC = 0.74 and OA = 0.92 for Alexandridis et al., 2008) with the observed fire than the other applied cellular automation models.