Data assimilation for wildland fires

Abstract

Two wildland fire models and methods for assimilating data in those models are presented. The EnKF is implemented ina distributed-memory high-performance computing environment. Data assimilation methods are developed combining EnKF with Tikhonov regularization to avoid nonphysical states and with the ideas of registration and morphing from image processing to allow large position corrections. The data assimilation methods can track the data even in the presence of large corrections, while avoiding divergence. The methods can assimilate gridded data, but the assimilation of station data and steering of data acquisition is left to future developments. A semi-empirical fire spread model is implemented by the level-set method and coupled with the WRF model.