Tracking extreme fire behaviour through assimilated fuel variables: A lens on climate change impact

Fire-prone ecosystems — including savannas, boreal forests, and Mediterranean regions — are experiencing increasingly intense wildfires, with climate change identified as a key driver. Changes in temperature, precipitation, and vegetation dynamics are altering fuel availability and dryness, leading to shifts in fire behaviour. However, as fuel remains poorly represented and constrained in global models, our ability to predict these shifts is still limited.

In this study, we present a long-term record of key fuel state variables — specifically fuel load and moisture content — derived from remote sensing. Creating long-term records of these essential climate variables can serve as indicators of climate variability and ecosystem vulnerability. Funded by ESA-FUELITY project, this work integrates multiple satellite datasets into the SPARKY-Fuel Characteristics dataset, covering the period from 2019 to 2021. It combines L-band vegetation optical depth (VOD) from SMOS and solar-induced fluorescence (SIF) from Sentinel-5P. A hybrid approach is employed, leveraging supervised machine learning to improve the connection between satellite observations and modelled fuel variables, enabling more accurate updates to vegetation conditions.

This dataset can be used in studies related to vegetation productivity, water stress, and carbon cycling, providing a valuable resource for detecting environmental change. Constraining these variables with satellite observations offers significant potential to capture climate-driven shifts in ecosystem flammability and drought response, particularly in fire-prone regions where traditional vegetation indices often lack sensitivity.

Here, we analyse the capability of this dataset to detect fire activity globally and identify regions of shifting fire behaviour. We will demonstrate how this dataset can help refine our understanding of fuel–fire interactions, providing a powerful lens for identifying possible tipping points in vulnerable environments.

Our work reflects a growing need for cross-disciplinary approaches that combine physical modelling, remote sensing, and artificial intelligence to track biosphere responses in a changing climate.