Improving the LPJmL4-SPITFIRE vegetation–fire model for South America using satellite data

Vegetation fires influence global vegetation distribution, ecosystem functioning, and global carbon cycling. Specifically in South America, changes in fire occurrence together with land-use change accelerate ecosystem fragmentation and increase the vulnerability of tropical forests and savannas to climate change. Dynamic global vegetation models (DGVMs) are valuable tools to estimate the effects of fire on ecosystem functioning and carbon cycling under future climate changes. However, most fire-enabled DGVMs have problems in capturing the magnitude, spatial patterns, and temporal dynamics of burned area as observed by satellites. As fire is controlled by the interplay of weather conditions, vegetation properties, and human activities, fire modules in DGVMs can be improved in various aspects. In this study we focus on improving the controls of climate and hence fuel moisture content on fire danger in the LPJmL4-SPITFIRE DGVM in South America, especially for the Brazilian fire-prone biomes of Caatinga and Cerrado. We therefore test two alternative model formulations (standard Nesterov Index and a newly implemented water vapor pressure deficit) for climate effects on fire danger within a formal model–data integration setup where we estimate model parameters against satellite datasets of burned area (GFED4) and aboveground biomass of trees. Our results show that the optimized model improves the representation of spatial patterns and the seasonal to interannual dynamics of burned area especially in the Cerrado and Caatinga regions. In addition, the model improves the simulation of aboveground biomass and the spatial distribution of plant functional types (PFTs). We obtained the best results by using the water vapor pressure deficit (VPD) for the calculation of fire danger. The VPD includes, in comparison to the Nesterov Index, a representation of the air humidity and the vegetation density. This work shows the successful application of a systematic model–data integration setup, as well as the integration of a new fire danger formulation, in order to optimize a process-based fire-enabled DGVM. It further highlights the potential of this approach to achieve a new level of accuracy in comprehensive global fire modeling and prediction.