Ember transport for bushfire simulation - final report

A parameterisation is a simplified model of a complex process, intended to be used as a component of more complex models that are unable, usually for reasons of computational cost or capacity, to resolve that process more completely. The term is widely used in atmospheric simulation, where the complex and fine-scale processes in, say, cumulus clouds or the boundary layer, are typically represented by parameterisations. If they were not, predicting the evolution of a single cumulus cloud would be capable of consuming all the computer resources available for preparing tomorrow’s weather forecast. They thus represent a pragmatic solution to the problem of adequately modelling the evolution of a process that covers a very wide range of scales in time and space.

As simplified models, parameterisation development requires a careful balance between simplicity and fidelity. These two factors are more-or-less synonymous with reduced computational cost and accuracy, respectively. Parameterisation development requires that these conflicting aims be, so far as is possible, reconciled.

Firebrand transport in bushfire plumes is an example of a process that has a significant impact on the outcome of a prediction but cannot be fully resolved on present computational hardware in a sufficiently timely manner. For example, the explicit firebrand transport simulations reported by Thurston et al. (2017) consumed days of time on a supercomputer, and that was after some simplifying assumptions were made.

Firebrand transport, and the generation of spotfires, is important because it can increase the rate of bushfire spread, can cause bushfires to break containment lines, makes fires less predictable, and is often implicated in structure loss. The prediction of long-range ember transport is a particular problem for which there has been a lack of understanding and decision support tools.

Building on the work of Thurston et al. (2017) and similar results of Thomas et al. (2019), Kepert et al. (2022a,b) developed a simplified model of firebrand transport in a bushfire plume that agreed reasonably well with those explicit calculations. That model was designed to be useable as a stand-alone predictive tool, and as part of a fire-spread simulator. That parameterisation was incorporated into an earlier version of the Spark fire simulator, as described in Kepert et al. (2022a), and tested on one fire. That version of Spark was not particularly suitable for this purpose, however, with little of the necessary internal model data being readily available. However, the work yielded very encouraging results.

The aim of this project was to implement the firebrand transport parameterisation into an up-to-date version of Spark, specifically version 2. This later version is far more suitable for such tasks because the full state of the fire model is accessible to the developer, along with the ancillary data such as topography, fuels, and surface meteorology. The technical details of the implementation were described in the interim project report (Kepert 2022c). The algorithmic details are described in this report.

As already noted, the challenge with developing simplified models, such as this one, is achieving an appropriate balance between simplicity and fidelity. Almost inevitably, use of the parameterisation will reveal circumstances in which minor adjustment of the choices will improve performance. Systematic, continuous verification is crucial, both as an impetus to gradual refinement, and to enable users to interpret predictions from the model with an appropriate balance between confidence and caution. If one may make a comparison with parameterisations in numerical weather prediction systems, it is both hoped and expected that this initial version of the parameterisation will evolve and improve over time, as experience and scientific knowledge accumulate.

To begin this process of validation, the bulk of this report describes the performance of Spark with the firebrand transport parameterisation on three notable fires. 

This report is organised as follows. Section 2 gives a brief overview of the formulation of the firebrand transport parameterisation, and section 3 describes some aspects of its implementation into Spark. Section 4 presents case studies of three significant fires; the Kilmore East Fire on Black Saturday, 7 February 2009; the Reedy Swamp fire of 18 March 2018; and the Timbarra fire of 25 January 2019. Section 5 presents an overall discussion of results, and section 6 highlights some outstanding issue.