Optimized Spatial Ensemble Approaches for Approximating the Fire Weather Index With Interpolation

Wildland fire science and management rely on modeling indicators of wildland fire risk, such as the Fire Weather Index (FWI). Spatial interpolation methods can estimate FWI across a region, but accuracy can vary with observation density and daily spatial dependencies. Our new algorithm approximates ideal spatial ensemble mixing weights to combine multiple interpolation methods to improve overall accuracy and reliability across the region of Ontario, Canada. We also develop a new Adjusted‐Adaptive Inverse Distance Weighting (AAIDW) method combining features from two generalized IDW methods, Adaptive IDW, which changes its power parameter based on local observation density, and Adjusted IDW, which accounts for nearby observations shielding farther observations in the same general direction. Through a cross‐validation study of 559,993 station‐day combinations between 1991 and 2025, along with a simulation study, we find that both features are beneficial, but the marginal benefit of an adaptive power parameter is lessened when shielding is already accounted for. We demonstrate that our spatial ensemble methods are comparable on average and more reliable than the best individual interpolation methods considered when FWI is in its Extreme risk class, reducing interpolation (extrapolation) errors by approximately 3.5% (7%), as quantified by the Score metric, while also improving Extreme‐class classification rates.