An effective proposal distribution for sequential Monte Carlo methods-based wildfire data assimilation

Sequential Monte Carlo (SMC) methods have shown their effectiveness in data assimilation for wildfire simulation; however, when errors of wildfire simulation models are extremely large or rare events happen, the current SMC methods have limited impacts on improving the simulation results. The major problem lies in the proposal distribution that is commonly chosen as the system transition prior in order to avoid difficulties in importance weight updating. In this article, we propose a more effective proposal distribution by taking advantage of information contained in sensor data, and also present a method to solve the problem in weight updating. Experimental results demonstrate that a SMC method with this proposal distribution significantly improves wildfire simulation results when the one with a system transition prior proposal fails.