Satellite-informed fuel estimation using hybrid data assimilation
| Title | Satellite-informed fuel estimation using hybrid data assimilation
|
Report
|
|
| Date Published |
12/2025
|
| Series/Collection |
ESA Contract Report
|
| Document Number |
4000144712/24/I-DT-bgh
|
| Author | |
| Abstract | This deliverable presents the final outcomes of the Fuelity project, which aims to improve the representation of fuel variables in the SPARKY-fuel model through the physically constrained use of satellite information. To achieve this, we combine satellite observations with land surface modelling using a hybrid data assimilation framework implemented in the European Centre for Medium-Range Weather Forecasts (ECMWF) ecLand-integrated forecasting system (IFS). The system ingests three complementary satellite data streams: L-band vegetation optical depth (VODL) -SMOS- and solar-induced fluorescence (SIF) -TROPOMI- to update a monthly leaf area index (LAI) climatology, and advanced scatterometer (ASCAT) backscatter to constrain upper-layer soil moisture. These observations are integrated using a simplified extended Kalman filter (SEKF), which updates the model background while accounting for uncertainties in both the satellite signals and the physical model, ensuring that analysis increments remain physically consistent. Two machine-learning models, an extreme gradient boosting (XGBoost) model and a multilayer perceptron (MLP) neural network, were evaluated as observation operators to map satellite observations to their model-equivalent counterparts. The models were trained on 8-daily data over 2019–2020 to predict SIF and VOD-L from meteorological conditions, vegetation and soil states. XGBoost was retained as it provided the most effective balance of simplicity, flexibility, interpretability, and predictive skill. The operator reproduced the satellite-driven seasonal cycle with high fidelity: Validation against an independent 2021 dataset showed that LAI explained 67% of the variance in SIF and 72% in VOD-L, with corresponding RMSE values of 0.12 and 0.16 m2 m−2. On the basis of this validation, the data assimilation framework was used to produce a 25-year global satellite-informed reanalysis of live fuel moisture content (LFMC) using the SPARKY model definition. Compared with a climatology-based LFMC benchmark, the reanalysis substantially improves the depiction of seasonal variability, drought response, and interannual fluctuations, with the greatest benefits observed in semi-arid and fire-prone ecosystems. The system was further evaluated across three main application domains: • Vegetation dynamics: Improved representation of phenology, drought-induced canopy decline, and fuel desiccation during major fire events. • Fire forecasting: Earlier detection of fuel drying and improved spatial localisation of high-risk conditions, supporting the potential for near-real-time (NRT) integration into operational fire danger systems. • Numerical weather prediction: Improved short-range forecasts of lower-tropospheric and nearsurface temperatures (1000 hPa, 850 hPa, and 2 m). These developments provide a scalable, observation-driven framework for global fuel estimation. The framework enabled the generation of satellite-constrained LFMC fields over more than 20 years. Its modular structure also creates clear pathways for future extensions to dead fuel moisture and fuel load as additional satellite missions become available, allowing progressively more complete and physically consistent fuel characterisation. |
| URL | https://www.ecmwf.int/en/elibrary/81701-satellite-informed-fuel-estimation-using-hybrid-data-assimilation |
| DOI | 10.21957/b732109ed9 |
| Download citation |