Sources of uncertainty in the SPITFIRE global fire model: development of LPJmL-SPITFIRE1.9 and directions for future improvements

Abstract. Since its development in 2010, the SPITFIRE global fire model has had a substantial impact on the field of fire modelling using dynamic global vegetation models. It includes process-based representations of fire dynamics, including ignitions, fire spread, and fire effects, resulting in a holistic representation of fire on a global scale. Previously, work had been undertaken to understand the strengths and weaknesses of SPITFIRE and similar models by comparing their outputs against remotely sensed data. We seek to augment this work with new validation methods and extend it by completing a thorough review of the theory underlying the SPITFIRE model to better identify and understand sources of modelling uncertainty. We find several points of improvement in the model, the most impactful being an incorrect implementation of the Rothermel fire spread model that results in large positive biases in fire rate of spread and a live grass moisture parametrization that results in unrealistically dry grasses. The combination of these issues leads to excessively large and intense fires, particularly on the dry modelled grasslands. Because of the tall flames present in these intense fires, which can cause substantial damage to tree crowns, these issues bias SPITFIRE toward high tree mortality. We resolve these issues by correcting the implementation of the Rothermel model and implementing a new live grass moisture parametrization, in addition to several other improvements, including a multi-day fire spread algorithm, and evaluate these changes in the European domain. Our model developments allow SPITFIRE to incorporate more realistic live grass moisture content and result in more accurate burnt area on grasslands and reduced tree mortality. This work provides a crucial improvement to the theoretical basis of the SPITFIRE model and a foundation upon which future model improvements may be built. In addition, this work further supports these model developments by highlighting areas in the model where high amounts of uncertainty remain, based on new analysis and existing knowledge about the SPITFIRE model, and by identifying potential means of mitigating them to a greater extent.