Physics-based models

Coupled climate fire models (CCFMs) are physics-based models that include dynamic global vegetation models (DGVMs) for simulating vegetation growth (Henderson-Sellers 1993) and some even include fire simulation modules (Goldammer and Price 1998, Lenihan 1998, Fosberg 1999, Arora 2005) including ignition and spread. Venevsky et al. (2019) provide a good example of the level of sophistication in such a model in terms of how fuel is grown, dried, ignited, and burned. A recent study by Li et al. (2019) provides a performance-based comparison of nine different DGVMs that participated in the Fire Model Inter-Comparison Project (Fire MIP) using satellite derived historical fire emissions from 2003-2008. At the global level (e.g., azimuthally average results by latitude) all models perform well but evaluation at regional levels (sub-continental) is challenging in part because of lack of observational data at the regional level.

CCFMs have the advantage of not only accounting for how and why vegetation grows and dries (through precipitation, sunshine, temperature, etc.), they also create burned area footprints using simulated weather for the region at the time the fire starts. Despite the complexity and advantages of CCFMs, performance – as determined by comparison of results with historical data – is still not great. The advantages can also become disadvantages because the output is only as good as the parameterization of the processes. Additionally, even if the resolution of a model is on the order of 50-250 km – and the fire models can have higher resolutions – the burned area results are not necessarily accurate at such a resolution. Output from Fire-MIP demonstrated significant differences in the results from different fire models even when using historical climate data as input (Rabin et al., 2017).

For forested regions, statistical models, especially when built on many years of observed fire (e.g., as opposed to remotely sensed fire) provide an advantage (Syphard et al. 2018).