Understanding the drivers of wildfires using JULES model simulations and machine learning emulators

Fires play a critical role in shaping ecosystems, driving biogeochemical cycles, and influencing atmospheric composition. In many regions historically affected by fire, the frequency, intensity, and size of fires have undergone rapid change in recent decades, especially in high-latitude forests. Meanwhile, wildfire extremes are now emerging across many of the world’s forests and fire-sensitive ecosystems including regions such as the Amazon, Congo, Indonesia, and the Pantanal. Many of these ecosystems have evolved with little or no fire, increasing the impacts of these fires’ potential risk of climate-driven tipping points. It is therefore essential to accurately represent wildfire dynamics within Earth system models to quantify their influence on carbon–climate feedbacks and predict ecosystem responses, including potentially rapid and irreversible ones, to environmental change.

Modelling and understanding wildfires processes remain challenging due to complex interactions among climate, vegetation, human activity, and land-use change. The Joint UK Land Environment Simulator (JULES) provides a robust framework for simulating the dynamics of terrestrial hydrology, vegetation, carbon storage, and the surface exchange of water, energy, and carbon. Complementary Machine Learning (ML) techniques allow development of model emulators, enabling large-scale data processing and quantification of model uncertainty for a comprehensive analysis of potential wildfire driving factors.

Here, we will present an ML-based emulator for the JULES-INFERNO model to: (1) Analyse and understand the key climatic drivers for wildfire, characterising recent trends (such as the size, frequency and intensity of wildfires) across JULES model simulations; and (2) Evaluate and identify the potential for monitoring early warning signals for tipping points by combining model simulations, remote sensing data and Artificial Intelligence. The analysis and evaluation will contribute to a better understanding for wildfire processes and provide comprehensive information for policy makers.