Jan Mandel, University of Colorado Denver
Assimilation of MODIS and VIIRS satellite active fires detection in a coupled atmosphere-fire spread model
Coauthors
Aimé Fournier, James D. Haley, Mary Ann Jenkins, Adam K. Kochanski, Sher Schranz, Martin Vejmelka, Tian Yu YenAbstract:
We present a system for the assimilation of active fires detection from polar-orbiting satellites into a wildland fire spread model coupled with the Weather Research Forecasting (WRF) model. The state of the fire model is encoded as the fire arrival time on a spatial domain and modeled as a Gaussian random field with the covariance a sufficiently large negative fractional power of the Laplace operator to have differentiable realizations, which controls the magnitude of the corrections to the spread rate. The data likelihood is based on the probability of detection in the MODIS/VIIRS Level 2 product pixels, estimated from the fire state by logistic regression in literature. Confidence of detection and cloud mask are used to weigh the contributions of the fire detection pixels to the log likelihood. A preconditioned steepest descent method is developed, which can find a sufficient approximation of a maximum aposteriori probability estimate in a single iteration and avoids local maxima in practice. Adjustment of the atmospheric state for consistency with the analysis fire state is achieved by rerunning the atmospheric model with the modified heat forcing from the analysis fire arrival time. Ignition time and location are estimated by fitting to an early fire detection. Computational results are shown using data from the 2013 Patch Springs fire. This work was partially supported by NASA grant NNX13AH59G. Development of data assimilation methodology was partially supported by NSF grant DMS-1216481 and Czech Science Foundation grant GA13-34856S.