Interpolation framework to speed up near-surface wind simulations for data-driven wildfire applications

Abstract

Local wind fields that account for topographic interaction are a key element for any wildfire spread simulator. Currently available tools to generate near-surface winds with acceptable accuracy do not meet the tight time constraints required for data-driven applications. This article presents the specific problem of data-driven wildfire spread simulation (with a strategy based on using observed data to improve results), for which wind diagnostic models must be run iteratively during an optimisation loop. An interpolation framework is proposed as a feasible alternative to keep a positive lead time while minimising the loss of accuracy. The proposed methodology was compared with the WindNinja solver in eight different topographic scenarios with multiple resolutions and reference – pre-run– wind map sets. Results showed a major reduction in computation time (~100 times once the reference fields are available) with average deviations of 3% in wind speed and 3° in direction. This indicates that high-resolution wind fields can be interpolated from a finite set of base maps previously computed. Finally, wildfire spread simulations using original and interpolated maps were compared showing minimal deviations in the fire shape evolution. This methodology may have an important effect on data assimilation frameworks and probabilistic risk assessment where high-resolution wind fields must be computed for multiple weather scenarios.