Reducing FRP uncertainty: early findings of University of Southampton sensor intercomparison and pixel-level modelling

At the FRM4Fire mid-point project meeting (PM2 October 2025), Gareth Roberts (University of Southampton) presented results of his modelling work for the project to that point. These covered the sub-tasks Work Package (WP) 1.2 – Sensor Intercomparison and WP 1.3.2 – Pixel-Level Modelling. Here we share some results of that modelling work, which is now contributing to project goals of establishing a metrological framework for Fire Radiative Power (FRP) retrievals, interoperability across sensors and platforms, and more reliable fire monitoring from space.

The University of Southampton also leads WP1.3 titled ‘characterisation of FRP uncertainty effects through modelling exercises’.  Other activities include a sensitivity and comparison analysis of different FRP datasets and laboratory analysis of thermal instrumentation used in airborne campaigns. Together, these activities are contributing to a comprehensive characterisation of all the main uncertainty sources impacting FRP measurements.

Context of Southampton modelling activities for the objectives of FRM4Fire

FRM4Fire aims to establish traceable, SI-consistent methods for FRP retrieval from satellite sensors. The University of Southampton contributions support this process by helping to reduce measurement uncertainty of FRP retrievals, improving the teams’ understanding of sensor geometries and optics as sources of bias, and supporting future modelling efforts, such as integrations with the Discrete Anisotropic Radiative Transfer Model (DART) model WP 1.3.3–1.3.4, another part of FRM4Fire led by Southampton. 

University of Southampton expertise

These contributions are led by Dr. Gareth Roberts, a specialist in environmental remote sensing with research interests focused on measurement and monitoring of land surface dynamics using optical and thermal remote sensing methods. His previous research has improved the understanding of fire in the natural environment and advanced the use of satellite observations to address questions about fire effects and their impacts. 

Based in Southampton’s Environmental Change and Sustainability research group (School of Geography and Environmental Science), he is experienced in developing active fire detection and fire characterisation algorithms. This includes intercomparing active fire datasets acquired from various satellite sensors including Spinning Enhanced Visible and Infrared Imager,(SEVIRI), Moderate Resolution Imaging Spectroradiometer (MODIS), and Sentinel-2 and in validating satellite FRP retrievals – encompassing the use of helicopter-mounted thermal imaging cameras to monitor controlled (prescribed) burns in Africa. 

Some of the algorithms he helped develop for detecting landscape fires are implemented operationally by the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) 

EUMETSAT Land Surface Analysis Satellite Application Facility (LSA SAF). Active fire data have been used to parameterise atmospheric transport models for air quality forecasting which have subsequently been used to quantify the impact of landscape fire emissions on air quality and human health (Roberts and Wooster, 2021). 

Gareth is a past co-chair of Committee on Earth Observation Satellites (CEOS) Land Product Validation ‘Fire Disturbance’ theme and is a member of the Global Observation of Forest Cover and Land Dynamics (GOFC‑GOLD) Fire Implementation Team. 

Inter-comparison of FRP from Meteosat-11 and Meteosat-08 

For this task, Southampton completed a detailed analysis that compared FRP retrievals from SEVIRI instruments onboard Meteosat-11 and Meteosat-08 (IODC), focusing on thermal anomalies observed simultaneously between 2017 and 2020. This built on earlier community efforts to develop and validate FRP products, including the FRP-PIXEL product (Wooster et al., 2015). 

The preliminary findings suggest: 

• A consistent bias of ~32% in per-pixel FRP retrievals, with Meteosat-08 typically reporting higher values. 
• Improved agreement (~21% difference) when comparing fire clusters. 
• Variability linked to view zenith angle (VZA) and pixel area differences, which affects radiance measurements, background characterisations, and identification of fire location within a sensor’s instantaneous field of view (IFOV). 

These results provide evidence that an improved understanding of sensor-specific biases and geometric influences will be needed. The results could also inform future efforts to harmonise FRP products across geostationary platforms. 

Figure 1: (Left) Mean (2017-2020) per-pixel FRP from Meteosat-11; (centre) mean (2017-2020) per-pixel FRP from Meteosat-08 and pixel area difference (km2) in overlapping areas between Meteosat-11 and Meteosat-08. (Southampton/FRM4Fire) 

Modelling impacts of fire location within image pixels

In parallel, Southampton developed a simulation framework that explored how the location of a fire within SEVIRI’s Instantaneous Field of View (IFOV) affects FRP retrieval accuracy.  

Key observations included: 

• FRP underestimation can exceed 80% depending on fire location and temperature. 
• When a fire is centrally located in the Point Spread Function (PSF), the average error is -12% across a range of temperatures. 
• This error increases to -37% (illustrative SEVIRI data error = 32%) on average when accounting for variation of a fire location in the PSF due to differences in the number of detected and saturated pixels. 
• Radiance distributed into neighbouring pixels can influence both fire detection and background characterisations. 
• The least error (~ -1.8%) is found where pixel area is similar between sensors and greatest (+/- 50%) when pixel area difference exceeds 5 km2 

Figure 2 Simulated MWIR brightness temperature, FRP and the number of detected fire and saturated pixels from simulations where a fire at four temperatures (500, 700, 900 and 1100 K) is centrally located in the PSF. Also shown by way of example are actual SEVIRI MWIR images (far right) which display similar spatial patterns in brightness temperature as found in the simulations. Note that the fire characteristics (e.g. size and temperature) in the real data are unknown. (Southampton/FRM4Fire) 

While these simulations were idealised, they likely offer a useful reference for interpreting retrieval variability and could support future algorithm development or uncertainty quantification. 

Next steps 

Both sub-tasks are being integrated into the deliverable D1 ‘Uncertainty budget document in support of Copernicus Sentinel-3 calibration and validation’.  

Southampton continues work on fire scene simulations using the Fire Dynamic Simulator (FDS) and DART models, focusing on environmental influences such as vegetation structure and smoke plume dynamics. 

• Fire Dynamics Simulator (FDS) is a large-eddy simulation (LES) code for low-speed flows, with an emphasis on smoke and heat transport from fires. It is free and open-source software provided by the National Institute of Standards and Technology (NIST) of the United States Department of Commerce.  
• The Discrete Anisotropic Radiative Transfer (DART) model is a comprehensive physically-based 3D model that simulates Earth-atmosphere radiation interactions from visible to thermal infrared wavelengths. It models optical signals at the entrance of imaging radiometers and laser scanners on board of satellites and aircraft, as well as the 3D radiative budget of urban and natural landscapes for experimental configurations and instrument specifications. DART was developed at Centre d’Etudes Spatiales de la Biosphère (CESBIO) from 1992 and patented in 2003. Paul Sabatier University (now Université de Toulouse) distributes licenses which are free for research and teaching activities.  

Further updates will be shared as the FRM4Fire team continues to build the scientific and technical foundations for trusted, interoperable fire monitoring from space.