Abstract: Since the fireline state and input parameters of wildfire spread prediction models inevitably contain errors and uncertainties, there are usually large deviations between wildfire spread predictions and actual spread results. The Kalman filter method based on data assimilation has the advantage of dynamic estimation in reducing model state errors. In this paper, we propose a dual estimation method for fireline position and fuel parameters based on Multi-Layer Perceptron (MLP) neural network, combined with the Ensemble Transform Kalman Filter (ETKF) algorithm. This method utilizes fireline position data from two consecutive observations and employs ETKF and MLP algorithms to estimate fireline position state and fuel parameters respectively under different prediction periods and fuel adjustment factor resolutions, aiming to reduce the uncertainties in both model state and model parameters. We validated the dual estimation method using a 10 000 m2 prescribed burning experiment and historical wildfire cases. The results demonstrate that the dual estimation algorithm can effectively reduce errors in both fireline state and model parameters, with the predicted fireline positions being to colser actual observations compared to single estimation methods. The prediction period significantly influences the performance of the dual estimation method in reducing fire spread model errors. The longer the prediction period, the greater the improvement in fireline position prediction accuracy compared to single estimation methods. In historical wildfire cases, the dual estimation algorithm shows good overall performance in fireline position prediction, validating the predictive capability of our algorithm in practical scenarios.

Key words: wildfire spread prediction, fuel adjustment factor, parameter estimation, ensemble transform Kalman filter algorithm, FARSITE