Exploring Independent Spatial Controls on the Global Distribution of Burnt Area, Fire Size and Fire Intensity through Generalized Linear Modelling

Wildfires are fundamental for maintaining ecosystem structure and functioning, and thus it is important to know how projected climate and land use changes will affect wildfire regimes globally. Fire-enabled vegetation models can be used to predict changes in fire regime but are still far from perfect since many of the processes that control different aspects of the fire regime are still relatively poorly understood. In this work, we investigated the underlying relationships between potential controls of different fire properties, including fire size, intensity and burnt area, at a global scale. We fitted three generalized linear models (GLM) to monthly data from 2010 to 2015 for fractional burnt area from the Global Fire Emissions Database version 4.1 (GFEDv4), fire size from the Global Fire Atlas database and median fire radiative power divided by square root of median fire size (as a proxy for fire intensity) from the MODIS MCD14ML dataset. We used partial residual plots between each predictor and each response variable to show the underlying linear relationships fitted by each model. We show that there are different controls on burnt area, on fire size and on fire intensity. Specifically, whilst burnt area is driven mainly by fuel availability and dryness, fire size is driven primarily by wind speed and fire intensity by tree cover and road density. Land fragmentation was highly limiting for fire size and burnt area whereas dryness was limiting for fire intensity. These findings suggest that it is possible to develop empirical models of multiple aspects of fire regimes which could be used to predict how these will change in the future. Furthermore, they highlight the importance of including landscape fragmentation as a control on fire explicitly within process-based fire models. Additionally, the limiting nature of dryness on fire intensity could be due to antecedent vegetation conditions and highlights the need for better representation of these conditions and their effect on fuel load. Finally, these results also suggest that the current treatment of ignition sources as an important driver in these models is unnecessary.