Last modified on Jan 19, 2026 13:47

Contributors: Martin Wooster (EI), Weidong Xu (EI), Carsten Brockmann (BROCKMANN CONSULT GMBH), Grit Kirches (BROCKMANN CONSULT GMBH), Martin Boettcher (BROCKMANN CONSULT GMBH)

Issued by: Earth Insight (EI)/ M.Wooster & W. Xu

Date: 10/12/2025

Ref: C3S2_313e_BC_WP1-DDP-FIRE-AF&FRP-S3-SLSTR-v1.2-2024_202506_ATBD

Official reference number service contract: 2024/C3S2_313e_BC/SC1

History of modifications

List of datasets covered by this document

Acronyms

General definitions

Active Fire (AF)

A fire that was actively burning when the satellite observations were made. Satellite ‘Active Fire’ products are those that report information on these types of ‘hotspots’ using thermal remote sensing techniques. AF or ‘hotspot’ pixels are pixels classified as containing one or more fires when the observation was made. Generally, most AF or ‘hotspot’ pixels are related to landscape fires, but some are related to other phenomena such as gas flares, volcanos or even other warm industrial targets. A small proportion of AF pixels are also ‘false alarms’ caused for example by unmasked areas of sunglint, land-water boundaries, or solar heated warm ground. Removal of such false alarms is typically attempted prior to final generation of the dataset.

Fire Cluster

A group of fire pixels represents a set of spatially contiguous, or nearly contiguous, active fire pixels that, at the time of imaging, are considered to belong to a single landscape fire event.

Fire Radiative Energy (FRE)

The temporal integral of fire radiative power calculated over the fire’s lifetime, equating to the total amount of energy radiated by the fire. FRE is typically used to estimate how much material was burned in a fire and how much smoke was released.

Fire Radiative Power (FRP)

The rate of radiant heat output from a landscape fire, typically expressed in Watts x 10⁶ (MW). FRP is typically very well related to a fire´s combustion rate (how much material is being burned per unit time) and rate of smoke emission, and hence remotely-sensed FRP measures are commonly used to estimate these terms.  At the pixel scale, a satellite product typically is reporting the total FRP from all fires burning within that pixel at the time the observation was made.

Gas Flare

A controlled open flame used to burn off excess natural or other combustible gas, typically at oil and gas facilities, chemical plants and natural gas processing plants, or landfills. Sometimes known as a flare stack, flare boom, ground flare, or flare pit.

Hot Spot

A pixel within a satellite instrument dataset identified as containing a sub-pixel heat source (i.e. something hotter than its ambient temperature surroundings). This identification is based at least in part on it increasing the pixel signal (spectral radiance and or brightness temperature) in at least one of the instruments thermal infrared or shortwave infrared bands. Such hotspots may be caused by landscape fires, gas flares, volcanoes, industrial heat sources, or they may be “false alarms” generated by phenomena such as sunglint. Removal of such false alarms is typically attempted prior to final generation of the dataset.

Landscape Fire (LF)

A landscape fire that was actively burning when the satellite observations were made. The hotspot detections held within satellite ‘Active Fire’ products are typically mainly due to landscape fires.

Satellite Data Processing Levels

• Level 0 (L0) data are reconstructed, unprocessed instrument and payload data at full resolution, with any and all communications artefacts (e.g., synchronization frames, communications headers, duplicate data) removed.
• Level 1A (L1A) data are reconstructed, unprocessed instrument data at full resolution, time-referenced, and annotated with ancillary information, including radiometric and geometric calibration coefficients and georeferencing parameters (e.g., platform ephemeris) computed and appended but not applied to L0 data.
• Level 1B (L1B) data are L1A data that have been processed to sensor units (not all instruments have L1B source data).
• Level 1C (L1C) data are L1B data that include new variables to describe the spectra. These variables allow the user to identify which L1C channels have been copied directly from the L1B and which have been synthesized from L1B and why.
• Level 2 (L2) data are derived geophysical variables at the same resolution and location as L1 source data.
• Level 2A (L2A) data contains information derived from the geolocated sensor data, such as ground elevation, highest and lowest surface return elevations, energy quantile heights (“relative height” metrics), and other waveform-derived metrics describing the intercepted surface.
• Level 2B (L2B) data are L2A data that have been processed to sensor units (not all instruments will have a L2B equivalent).
• Level 3 (L3) are variables mapped on uniform space-time grid scales, usually with some completeness and consistency.
• Level 3A (L3A) data are generally periodic summaries (weekly, ten-day, monthly) of L2 products.
• Level 4 data are model output or results from analyses of lower-level data (e.g., variables derived from multiple measurements).

Descriptions of data processing levels ranging from Level 0 to Level 4 have been sourced from the following National Aeronautics and Space Administration (NASA) Earth Observation Data website: https://www.earthdata.nasa.gov/engage/open-data-services-and-software/data-information-policy/data-levels

A detailed description of the C3S FRP products, of their evaluation and verification method as well as the results, and of the application use cases is specified in the corresponding documents:

• Product User Guide and Specifications (PUGS) of the C3S FRP Products ( E.U. Copernicus Climate Change Service, 2021a & 2025a),
• Product Quality Assurance Document (PQAD) of the C3S FRP Products (E.U. Copernicus Climate Change Service, 2021b),
• Product Quality Assessment Report (PQAR) of the C3S FRP Products (E.U. Copernicus Climate Change Service, 2025b),
• Tutorial of the C3S FRP Products (E.U. Copernicus Climate Change Service, 2021c).

Executive summary

Landscape fires burn large amounts of vegetation and soil across the globe, releasing smoke containing many different chemical compounds. The Copernicus Climate Change Service (C3S) provides Climate Data Records for the Landscape Fire Essential Climate Variable, including active fire (AF) pixel (hotspot) detections with associated fire radiative power (FRP) estimates, the latter of which relates to rates of fuel-consumption and smoke-emission. The C3S Active Fire (AF) & Fire Radiative Power (FRP) products deliver global hotspot detection and FRP information from daily to monthly scales, with daytime and night-time Level-2 summaries in text format, and three Level-3 NetCDF gridded 'synthesis' products. Each C3S product file consolidates the key information contained within hundreds to thousands of Sentinel-3 SLSTR Level-2 Active Fire & FRP product granules into a single file covering the globe over a period of either one day, 27 days or one month depending on the product. The vast majority of the identified hotspots within the C3S AF & FRP products are related to actively burning landscape fires. Accompanying the C3S AF & FRP products are a set of similar C3S Gas Flare products that focus only on land and ocean flaring identified by spectral and temporal criteria, but only at night. Together these C3S products are designed to be simple to use and well suited to modelling, emissions estimation, trend analysis, and evaluation. The C3S products are designed to align with long-running MODIS fire records, that are expected to finish around 2026/27 when the Terra and Aqua satellites cease operation. Sentinel-3’s similar overpass time will help maintain a consistent long-term global active-fire and FRP record despite strong diurnal fire variability. The accompanying C3S PUGS document for these products provides detailed specifications, along with information required by users wishing to utilise the information contained in the product files. Quality assurance, verification and validation information is provided in separate PQAR documentation.

Each C3S AF and FRP product file, and each C3S Gas Flare product file, is generated from large numbers of Sentinel-3 SLSTR Level-2 Active Fire & FRP product granules. These Level 2 product files have been generated using a spatio-spectral contextual hotspot detection algorithm, and the accompanying FRP estimates derived using the MIR and the SWIR radiance methods. The algorithm applies spectral and temporal filtering to isolate genuine gas-flare pixels for incorporation into the C3S Gas Flare products. The C3S Summary Products summarise the key information extracted from these Level 2 product files at the location of identified hotspot pixels, whilst the C3S Level 3 synthesis product algorithm produces gridded statistical summaries of the Level 2 product file data at different spatio-temporal intervals. The gridded product files contain matching information on cloud cover, since this can mask surface hotspots from the spaceborne sensors´ view.

This ATBD describes the algorithms used to generate the C3S AF and FRP products, and the C3S Gas Flare products. Section 1 introduces the Sentinel-3 Mission and the SLSTR instrument from whose data the C3S AF Detection and FRP products and the C3S Gas Flare products are derived. It includes details of the forerunner Level 2 granule-based Active Fire Detection and FRP products that are generated directly from the Level 1 SLSTR data and which form the basis for each of the C3S product types.  Section 2 details aspects of the Level 2 AF Detection and FRP products which form the input files for each of the C3S product types. Section 3 details the algorithm used to generate each C3S Level 2 Summary and Level 3 Synthesis product from the many Level 2 product files used as input.  Finally, Section 4 details the output of the algorithm and the contents of each of the C3S product types.

1. Missions and Instruments

1.1. Active Fire and Gas Flare Background

Landscape fire is commonly studied and quantified via satellite Earth Observation - since it perturbs a greater area over a wider variety of biomes than any other natural disturbance agent. The trace gas and aerosol emissions released significantly impact Earth's atmospheric composition and chemistry, radiation balance, climate and air quality. Landscape fires are often large-scale, sporadic and highly dynamic, making satellite Earth Observation (EO) vital for their quantification. The need for data on landscape fires extends not only to scientific studies, but also to actionable information with which to support real-time monitoring and decision making. 'Active fire' (AF) products are one of the most widely used satellite EO datasets related to landscape fires. Satellite AF products originally recorded the location and timing of fires burning at the time the satellite observation was made (Giglio et al., 2003; 2016), and more recently extended this to often include data on the fires radiative power (FRP) output. FRP has been shown to be well related to rates of fuel consumption and smoke emission (e.g. Freeborn et al., 2008; Nguyen and Wooster, 2020), and thus satellite AF products containing FRP information provide a route to estimating these important Earth system parameters from spaceborne data - see review by Wooster et al. (2021). With respect to gas flares, flaring releases pollutants including carbon dioxide, methane, and black carbon - potentially contributing to climate impacts and negatively affecting local environments as well as burning off a potentially useable resource.  Night-time satellite datasets in particular are routinely used to keep track of changes in global gas flaring (Fisher et al., 2019; Elvidge et al., 2023).

1.2. Mission and Instruments used within the C3S Products

The C3S AF & FRP and the C3S Gas Flare products are designed to summarise and make easy to access and use active fire pixel detection, FRP and gas flare information generated from observations collected by the Sea and Land Surface Temperature Radiometer (SLSTR), a 1 km spatial resolution (at nadir) scanning radiometer carried onboard the Sentinel-3 series of polar-orbiting satellite platforms. SLSTR delivers large scale, accurate and timely multi-channel radiometric data covering Earth's land, ocean and atmosphere. SLSTR operates concurrently onboard two polar-orbiting Sentinel-3 satellites, S3A and S3B at the time of writing. SLSTR builds on the heritage of the (Advanced) Along Track Scanning Radiometer ((A)ATSR) series of instruments, carried by ERS-1, ERS-2 and ENVISAT. As with these prior instruments, SLSTR provides dual-view, highly accurate imaging radiometry across the visible to longwave infrared spectral range (Table 1). SLSTR data (primarily in the thermal infrared) from the instruments near nadir view scan are used to generate an operational Level-2 Active Fire Detection and FRP product (FRP product) based on the night-time algorithm of Xu et al. (2020) and the (slightly different) daytime algorithm of Xu et al. (2023), with day and night products showing some performance differences. Developments by Fisher et al. (2018, 2019) for SWIR hotspot detection and FRP retrieval are used at night to generate further information of particular relevance to gas flares.  These Level-2 Active Fire Detection and FRP products record in a single 3-minute SLSTR granule the timing and FRP of hotspot pixels detected in the LWIR and the SWIR. These hotspots are mostly active landscape fires burning at the time the SLSTR observations were made. A relatively small proportion (<<1%) of these active fire pixels are in fact due to other hotspot types such as active volcanoes, industrial activity and gas flares for example.

1.3. Algorithmic Approach

To derive the Level 2 Active Fire Detection and FRP products that form the input to the C3S product algorithms, an active fire detection and FRP retrieval algorithm is used. This exploits some of the significant advances that SLSTR possesses over those of (A)ATSR. These include an extended near-nadir view swath width (~ 1700 km), a 0.5 km spatial resolution for the visible to shortwave infrared (VIS to SWIR) channels (retaining 1 km for thermal infrared channels), and most importantly the addition of further spectral channels beyond those in the (A)ATSR design. The additional channels include a low-gain 'fire channel' (F1) located in the same middle infrared (MIR) spectral region (centred on 3.74 µm) as the 'standard' S7 channel, but able to measure unsaturated signals up to far higher brightness temperatures (though with a higher NEdT than S7). Wooster et al. (2012) provides more details of the overall SLSTR instrument specification including the S7 and F1 channels and their relevance to active fire detection and FRP retrieval, along with an early description of the L2 FRP product algorithm some years pre-launch. Coppo et al. (2010) provide comprehensive detail about all aspects of the SLSTR instrument, whilst Xu et al. (2020; 2021; 2023) detail some of the Level 2 FRP product algorithms evolution on the way to the current status. The algorithms are based on 1 km (nadir) spatial resolution hotspot detections made in the thermal infrared (TIR).  Fisher et al. (2018, 2019) detail the background behind the addition of the L2 gas flare specific algorithm components used in the Level 2 Active Fire Detection and FRP products that are based on 500 m (nadir) spatial resolution hotspot detections in the shortwave infrared (SWIR). Each of the C3S products combines global data from vast numbers of Level 2 Active Fire Detection and FRP products. Each SLSTR instrument generates around 15,000 of these Level 2 products per month both day and night and from each S3 satellite.

The S7 and F1 channels that sense electromagnetic radiation in the MIR in particular are central to the SLSTR FRP products, since actively burning fires emit electromagnetic radiation very strongly in this spectral region. For this reason, thermal infrared channels operating in the MIR are very sensitive to the presence of even highly sub-pixel heat sources - such as actively burning fires (Wooster et al., 2021). However, the two SLSTR MIR channels (S7 and F1) possess certain characteristics that make their combined use non-trivial – for example their data shows a small spatial offset, has a different pixel size and shape, and F1 records anomalously low brightness temperatures (BTs) down-scan of the type of high BT pixels characteristic of active fire (AF) pixels (See Xu et al., 2023 for example). However, F1 can record MIR brightness temperatures in excess of 450 K, whereas S7 starts to saturate beyond 311 K. Thus F1 can be used to retrieve the FRP of AF pixels that are saturated in S7 and where accurate retrieval of FRP using S7 channel data is impossible. Subsequent to the work described in Xu et al. (2023), the S3 FRP products use MIR observations in S7 to perform the initial detection of AF pixels where possible (along with matching observations in the S8 longwave infrared (LWIR) channel and some of the VIS-SWIR channels), but F1 to retrieve all FRP values. F1 is also used in place of S7 to provide the initial AF pixel detections where necessary (generally this is only by day - e.g. when the ambient land is saturated in S7), but when this is done the AF detection process is a little degraded compared to when S7 is used. For this reason, night-time and daytime L2 FRP products have somewhat different performance characteristics, and this is one reason why the Level-2 Active Fire Detection and FRP product files derived from daytime and night-time observations are kept separate, something that is continued with the C3S FRP products that are derived from them. There are also typically more fires present by day than by night in fire-affected regions. The C3S product files containing only gas flare information are issued only from night-time observations and in their own separate files, being based on the SWIR hotspot detections held within the Level 2 Active Fire Detection and FRP products . The performance of the L2 products issued from each S3 satellite appear extremely similar, but still the L2 and the derived C3S products coming from each S3 satellite are also delivered in separate files.

Table 1: Sentinel-3 SLSTR spectral channels. The MIR 'fire channel' (F1) has a greatly increased saturation temperature compared to the matching standard ('S7') MIR channel, though at the expense of increased noise (NEdT) characteristics. The S7 MIR channel starts to saturate at brightness temperatures exceeding 311 K, whereas F1 can record signals up to 450 K and beyond. Note that the F2 LWIR 'fire' channel is currently not used in either the L2 nor C3S FRP product algorithms since the S8 channel is sufficient for the purpose.

2. Input and auxiliary data

2.1. Level 2 Input Data Characteristics

Each C3S product is generated using multiple Level-2 Active Fire Detection and FRP products, each issued in non-time Critical (NTC) mode by the European Space Agency (ESA) as part of the EU Copernicus Programme. Each Level 2 product file is generated from a 3-minute SLSTR Level 1 granule dataset captured during an ~ 90 min orbit. Around 15,000 such granules are generated every month from each operating S3 satellite (Table 2).  A full ATBD for the all the Level-2 Active Fire Detection and FRP products can be found at https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-3-slstr/. Different C3S AF and FRP products are generated from these L2 products globally over land both day and night, whilst the C3S Gas Flare products are generated globally over land and ocean but only at night. The C3S AF and Fire Radiative Power (FRP) products are based primarily on the SLSTR thermal channel hotspot detections present in the L2 product granules, whilst the C3S Gas Flare products are based primarily on a filtered version of the SWIR channel detections.

Each Level 2 Active Fire Detection and FRP Product granule that feeds into the C3S product processor consists of a set of eleven measurement and annotation data files corresponding to information on active fire hotspots (mostly landscape fires; but also other hotspot types) that were active at the time the SLSTR observation was made. Each Level 2 Product is generated from a SLSTR Level 1 Product granule collected over a period of 3-minutes, with many such files generated per 90-minute orbit

Table 2:  SLSTR Sentinel‑3 (ESA/EUMETSAT)

Information related to the thermal infrared (1km) AF pixel detections and their FRP is stored in the Level 2 product as two different datasets: a "LIST" of the characteristics of each confirmed active fire (AF) pixel and a 2-d array "SUMMARY FLAG" dataset. The "LIST" dataset stores the characteristics of each confirmed hotspot pixel present in the Level 1 granule and detected with the 1 km thermal infrared channels of SLSTSR. Data stored here include an FRP estimate for each confirmed AF pixel derived from the SLSTR MIR channel data, together with its FRP uncertainty, the time and position of the measurement, and various other auxiliary data. During the night-time observations, a SWIR channel FRP value is also derived for the 1km detected hotspot locations. The full list of parameters held as 1D arrays within the List dataset of the Level 2 Active Fire Detection and FRP Product can be found in Table 3 , with the classification hotspot legend shown in Table 4. Note that the L2 Active Fire Detection and FRP product files report data derived from any 1 km thermal-infrared hotspot pixel. While these hotspots predominantly correspond to actively burning landscape fires, other phenomena—such as volcanic activity, industrial processes, and gas flaring—can also produce AF pixel detections and FRP retrievals. However, these non-fire sources are far less common than active landscape fires.

Table 3: Information held within the LIST component of the Level-2 SLSTR Level 2 Active Fire Detection and FRP Product files related to the thermal infrared hotspot detections made at 1 km (nadir) resolution. Note prior to March 2022, SWIR-derived information was produced only at locations of thermal-infrared-detected hotspots during night-time observations. Since March 2022, SWIR-derived fields have been suppressed, and SWIR-based FRP retrievals are not generated during daytime because they are generally not feasible.

Table 4: FRP classification byte values for the List product parameter "Classification", related to the thermal infrared hotspot detections made at 1 km (nadir) resolution. Classification is based on fixed geographic coordinates of zones known to contain gas flares, volcanic and industrial features. Note that due to large areas of water being masked in the Level 2 product algorithm in relation to the thermal infrared detected hotspots, the "offshore gas flares" is not much used for the thermal infrared detected hotspots. The vast majority of the thermal infrared detected hotspots are classed as '0' (vegetation fires) since they are not located in a known gas flare, volcanic or industrial zone.

Currently within the L2 Active Fire Detection and FRP product files, the classification of an AF pixel into the classes shown in Table 4 is based on fixed geographic coordinates - including as “onshore gas flares” and “offshore gas flares”. However, there is another way to identify hotspot pixels at night - based on the fact that many of them emit significantly in the shortwave infrared (SWIR). This approach is implemented within the Level-2 Active Fire Detection and FRP product algorithm to generate a dedicated set of night-time SWIR-detected hotspots at 500 m spatial resolution over land and ocean. Information related to these SWIR (500 m) hotspot pixel detections and their SWIR-derived FRP is stored in the Level 2 product as two further datasets: a "LIST" of the characteristics of each confirmed SWIR-detected hotspot pixel and a 2-d array "SUMMARY FLAG" dataset. The full list of parameters held as 1D arrays within this second List dataset of the Level 2 Active Fire Detection and FRP Product can be found in Table 6. Note that these hotspot pixels are detected with the SWIR band data of SLSTR and so are not the same set of hotspots present in the 1 km thermal infrared detected LIST product, though there are certainly many hotspots common to both.

The importance of these SWIR-detected hotspots is in particular in relation to containing gas flares, since the SWIR radiance method of FRP derivation performs better for hotter (gas flare) targets than the MIR radiance method (Fisher et al., 2019) and using the spectral radiance ratio in the SLSTR S5 and S6 shortwave infrared (SWIR) channels of Table 5 gas flare hotspots can be discriminated from landscape fires (see Fisher et al., 2019).  The C3S Gas Flare products this spectral radiance ratio approach is combined with filtering based on temporal persistence to confirm a SWIR-detected hotspot present in the L2 Active Fire Detection and FRP product files as a gas flare. Only these confirmed gas flare pixels are included in the C3S Gas Flare products. Table 5 details the full list of parameters held as 1D arrays for SWIR-detected hotspots present within the Level 2 Active Fire Detection and FRP Product files, which then form the input to the C3S Gas Flare products.

Table 5: Information held within the LIST component of the Level-2 Active Fire Detection and FRP Product files related to the SWIR hotspot detections made at 500 m (nadir) resolution.

As stated above, a second category of dataset held within each Level 2 Active Fire Detection and FRP Product granules is generated on the SLSTR image grid. This consists of a 2D "SUMMARY FLAG" dataset providing the results of the various hotspot pixel detection tests and associated processing steps. The SUMMARY FLAG dataset stores information on every pixel in the SLSTR Level-1 granule, whether or not it was detected as an AF pixel. There are two SUMMARY FLAG datasets in every Level 2 Active Fire Detection and FRP Product granule, one for the thermal infrared detected hotspots (1km grid) and one for the SWIR detected hotspots (500 m grid). Those for the former thermal infrared channel data are shown in Table 6, and those of the corresponding SWIR channel data are shown in Table 7.

Table 6: FRP Summary Flag byte values held in the SUMMARY FLAG dataset of the Level 2 Active Fire Detection and FRP Product granules related to the thermal infrared hotspot detections made at 1 km (nadir) resolution.

Table 7: FRP Summary Flag byte values held in the SUMMARY FLAG dataset of the Level 2 Active Fire Detection and FRP Product granules related to the shortwave infrared hotspot detections made at 500 m (nadir) resolution..

2.2. Level 2 Input Data Performance

The data contained within the Level 2 Active Fire Detection and FRP Products are similar in nature to that provided within the MODIS Active Fire and Thermal Anomaly (MOD14) products.  The latest version (Collection 6) of the Level 2 MODIS Active Fire products (Giglio et al., 2016) were first issued in 2015. They are similar to the Sentinel-3 Level 2 Active Fire Detection and FRP Products in that in each case these Level 2 products contain information on the characteristics of each confirmed active fire pixel present in the Level 1 granule (FRP estimate, AF pixel detection time and position etc), along with a 2D array providing a classification of every pixel in the Level 1 granule, i.e. regardless of whether it was identified as an AF pixel (e.g. water, cloud, cloud-free land, AF pixel).  The algorithm used to generate the Level 2 products from the Level 1 granules is a spatio-spectral contextual hotspot detection algorithm used to identify pixels that are spectrally and spatially dissimilar to their neighbours in ways that indicate they contain a (sub-pixel) high temperature heat source, with the accompanying FRP estimates of this heat source derived using the MIR and the SWIR radiance methods. Ultimately the Sentinel-3 FRP products will take over from the MODIS products in order to continue a globally consistent data record for the morning and evening overpass times (S3 equatorial crossing time of 10:00 hrs and 22:00 hrs respectively). Figure 1 shows an example of the AF pixel counts and FRP detected at night globally for one month from a single SLSTR instrument.  Performance from S3A and S3B is almost identical, but evidence indicates that when the Sentinel-3 Level 2 Active Fire Detection and FRP Product and the MODIS Terra AF products are produced for the same area near simultaneously at night, the SLSTR product tends to detect significantly more active fire pixels. However, the FRP total from the near-simultaneously observed regions appear to be very similar from SLSTR and from MODIS Terra since the additional AF pixels detected by SLSTR tend mostly to be associated with low FRP fires (see examples in Xu et al., 2020). By day however, due to the requirement to use F1 more often in the AF pixel detection stage, the performance of the S3 L2 FRP product is somewhat different than at night. The daytime S3 product still tends to detect more AF pixels that make up a detected fire than does the matching MODIS data, but more single pixel fires or very low FRP fires can remain undetected by the S3 product than the MODIS product. Despite this lower daytime performance than night-time performance however, the FRP detected across a region by the S3 and MODIS FRP products is still very similar by day (as it is at night), as the differences are almost all at rather low FRP fires and fire pixels (see Xu et al. 2020 and Xu and Wooster, 2023). 

Figure 1: Monthly global map of night-time (a) active fire pixel count and (b) total FRP, both derived from thousands of Sentinel-3A Level 2 FRP product files generated that month from SLSTR observations whose data are combined within a single C3S Level 3 Monthly Gridded Active Fire and FRP Product.  Grid cell size is 0.25°.

Atmospheric phenomena, particularly cloud cover, can interfere with active fire detection. For example, active fires cannot be detected through thick cloud, and clouds can in some cases cause false AF detections. However, AF pixels can usually be detected through smoke. Thus, to generate the Sentinel-3 Level 2 Active Fire Detection and FRP Products, SLSTR pixels in the L1b near-nadir scan data are first masked for interfering atmospheric phenomena such as thick cloud cover (whilst ideally masking areas of smoke) using a set of simple thresholding tests developed specifically for the active fire application (Wooster et al., 2012; Xu et al.,2020). There also exists the possibility of to use some of the Level 1b cloud masks instead of the internal FRP product cloud mask, and this option is controlled by a switch in the processing chain used to deliver the Level 2 FRP product. Each pixel passing those tests unmasked is then tested to see if it seems likely to contain actively burning fires at the time of observation. The contextual AF detection algorithm used to identify AF pixels via their radiometric contrast with their non-fire neighbours is based on the kinds of principles applied when generating the MODIS AF products (Giglio et al., 2003; 2016), but the method has been adapted for SLSTR pre-launch by Wooster et al. (2012), and then substantially modified and optimized by Xu et al. (2020; 2021) and Xu and Wooster (2023) - including with procedures to attempt to avoid false alarms from such phenomena as cloud edges, homogeneously warm surfaces, and (by day) sunglint. The SLSTR MIR channel (S7 and F1) data (Table 1), along with data from various other SLSTR channels such as the SWIR S5 and S6 bands at night, provide extreme sensitivity to locations of combustion and other high temperature phenomena present at the time of the Sentinel-3 overpass. Due to this sensitivity, SLSTR pixels containing areas of combustion covering down to only around 10^-3 to 10^-4 of a pixel can typically be successfully identified. Once identified the FRP of each confirmed AF pixel is then retrieved using the MIR radiance method of Wooster et al. (2003; 2005), with the SWIR radiance method of Fisher and Wooster (2019) also provided and the most appropriate for gas flares. The full algorithm is detailed in Xu et al. (2020; 2021) and Xu and Wooster (2023) and the ATBD for the family of NTC Level-2 FRP Products can be found at S3-L2-SD-03-T04-KCL-ATBD_FIREPRODUCT - Sentinel-3 SLSTR Level 2 Active Fire Detection and FRP2022 - 5.1.pdf. The following Section details the algorithms used to generate each of the C3S product types from these Level 2 datasets.

3. Algorithms

3.1. Sentinel-3 FRP Climate Data Record (CDR) Product Types

The C3S product suite is fully described in the accompanying PUGS documentation. The suite incorporates both global Active Fire & FRP products (day and night, land only) and Gas Flare products (night-time only, land and water). In total, twelve different product types are generated, each using information from hundreds to thousands of Sentinel-3 NTC Level-2 Active Fire Detection and FRP product files that each cover a small portion of the globe over a period of 3-minutes. Figure 2 shows the workflow for the generation of the product suite. Eight of the C3S products relate to hotspots detected over land at 1 km (nadir) spatial resolution in the thermal infrared (TIR) - with separate products for day and night and most of these hotspots being associated with landscape fires, whilst the other four of the C3S products relate to gas flares detected in the shortwave infrared (SWIR) at 500 m (nadir) spatial resolution over land or ocean (night only).

Two Level-2 Summary Active Fire & FRP products (day and night) provide text (CSV) summaries of TIR-detected hotspots extracted from multiple Level-2 Active Fire & FRP List Product files. Local solar time is included for user reference. A hotspot-type classification from the Level-2 files is included only in the Level-2 Monthly Global Active Fire & FRP Summary Product.  FRP from both MIR and SWIR methods is included, with MIR preferred for landscape fires. Six Level-3 synthesis products supply gridded statistics of the same variables at daily, 27-day and monthly intervals, again split into day and night and stored separately for each S3 satellite. Grid cell cloud-fraction information is included in the gridded products so hotspot pixel counts can be cloud-cover adjusted if desired. These Level-2 and Level-3 fire products include only land-based TIR hotspots, and the vast majority of these are landscape fires.  

The four C3S Gas Flare product types focus on SWIR-detected hotspots related specifically to gas flaring at 500 m resolution. The C3S algorithm applies spectral and temporal filtering to isolate genuine gas-flare pixels, which may differ from the Level-2 geographic-based classifications. Because SWIR FRP retrievals work best at night, all Gas Flare products are night-time only. Only Level 2 hotspot data with clear spectral signal characteristic of flares, and temporal persistence - also typical of flares - are retained in the C3S Gas Flare products.  As with the fire products, the Gas Flare suite includes a Level-2 Monthly Global Gas Flare Summary Product (CSV) and three Level-3 Gas Flare synthesis products (daily, 27-day, monthly), all in NetCDF format.

Figure 2: Data processing workflow for the C3S Sentinel-3 SLSTR Active Fire and FRP products. Multiple individual Level-2 granules covering the globe for different time periods are aggregated into monthly text-based summary (Level-2) products, and three different gridded (Level-3) products (daily, 27-day and monthly), with separate files for daytime, night-time, and gas flare data types in each case, as well as separate files for each S3 satellite. The approach used to generate the Level-2 products from the Level-1 granules is introduced in Section 2.2, whilst the approach used to generate the C3S products from multiple Level 2 product files is detailed in Section 3.2. 

3.2. C3S Product Algorithms

3.2.1. Level 2 Monthly Global Active Fire and FRP Summary Product (Daytime and Night-Time) & Level 2 Monthly Global Gas Flare Summary Product (Night-Time Only)

The Level 2 Monthly Global FRP Summary Products provide a monthly global AF location and FRP summary to users in an easily accessible (e.g. simple ACSII) CSV format, collating the information from many of the L2 List product files detailed in Table 2. Each Sentinel-3 satellite collects around 15,000 Level 1 3-minute granules per month globally, each of which results in a Level 2 Active Fire Detection and FRP Product. Thus, each Level 2 Monthly Global AF location and FRP Summary Product is based on information from thousands of Level 2 Active Fire Detection and FRP Product files and contains for each detected active fire pixel information of the type listed in Table 2 - including the equivalent Level 1 granule sample number, image line number, latitude, longitude, etc. This summary is placed into a single CSV file for ease of use by users, with separate files for the different satellites and for daytime and night-time data. It is important to realise that the separation of the night-time and daytime AF pixel detections included in the Level 2 Monthly Global FRP Summary Product (and all three Level 3 product types) is based on pixel solar zenith angle NOT time of day NOR whether the observations were from the orbital descending or ascending node. The L2 processing chain defines night-time AF pixels as those having a solar zenith angle exceeding 85°. Whilst it is the case that the vast majority of AF pixels defined as being “night-time” using this criteria will have been detected during the evening (ascending-node) S3 overpass, at high latitudes some AF pixels detected during the ascending node pass can have a solar zenith angle < 85°. Any hotspot pixels identified here will be classed as daytime detections via the solar zenith angle criteria, and be included in the Daytime Level 2 Monthly Global FRP Summary Product alongside those detected in the descending node passes. Therefore, in order for users to be able to include their own time-based discrimination of daytime and night-time hotspots if they so wish, the ‘local solar time (LST)’ of each AF pixel is provided in the Level 2 Summary Product file based on the following equations

Where LST is the local solar time in decimal hour, LT is the local time and TC is the Time Correction calculated with equation 2

Where LSTM is the Local Standard Time Meridian. LSTM is a reference meridian used for a particular time zone and is similar to the Prime Meridian, which is used for Greenwich Mean Time. EoT is the equation of time in minutes (see eq. 3 below).

where dT is the time difference between the Local Time and Universal Coordinated Time (UTC) in hours.

EoT is an empirical equation that corrects for the eccentricity of the Earth's orbit and the Earth's axial tilt. An approximation of EoT with accuracy within ½ minute is

where

in degrees and d is the number of days since the start of the year.

Whilst the original Level 2 FRP product algorithm uses a geographic classification process to classify hotspot AF pixels into the classes listed in Table 3, since this relies on static flare locations reported in other datasets and databases it is not very rapidly adapting. Therefore, whilst the C3S Level 2 Active Fire Detection and FRP Summary Product hotspot classification also uses this static classification, the C3S Level 2 Gas Flare Summary Product derived from the SLSTR SWIR band measurements uses the S5-to-S6 (shortwave infrared band) spectral radiance ratio to classify pixels as likely gas flares. Higher values of this ratio indicate hotter combustion sources - though the method is only really effective at night and so the C3S Gas Flare products are night-time only. The output C3S Level 2 Gas Flare Summary Product summarises information from the L2 Gas Flare product files listed in Table 4, both over land and ocean and only at night. Since gas flares change less rapidly than active fires, night-time only data is not deemed to be such a limitation - and follows other night-time gas flare focused products (e.g. from VIIRS; Elvidge, et al, 2023).

Since gas flares are highly sub-pixel targets at the scale of the SLSTR observations, any difference in the point spread function of the S5 and S6 spectral bands may affect the value of this ratio, as will any spatial mis-registration between the bands. To minimise such impacts, the algorithm follows the strategy adopted by Zhukov et al. (2006) for sub-pixel hotspot analysis, and calculates the spectral radiance ratio metric on a ‘hotspot cluster’ basis rather than at the level of individually detected hotspot pixels. Following use of this spectral radiance ratio metric to classify each cluster, and thus each pixel in a cluster, as a gas flare or not, a temporal persistence approach is then used to remove hotspot pixels identified as probable gas flares but which appear to be insufficiently persistent over time to justify them be classed as such. Details on these two approaches used to identify the hotspot pixels to include in the C3S Level 2 Monthly Night-time Gas Flare Summary Product file (and also the Level 3 Gas Flare Gridded Product files) is provided below.

Clustering and Spectral Radiance Ratio Thresholding: The clustering method used in the C3S algorithm is based on connected-component labelling, a standard technique in image processing and computer vision (He et al., 2017). Within the C3S workflow, this method groups together spatially neighbouring hotspot pixels detected in the S6 shortwave infrared band into single hotspot 'clusters'. This clustering approach can be broken down into the following steps:

1. Create a binary image:
A binary mask is created matching the (x,y) size of the Level 2 Active Fire product granule. Pixels identified as hotspots in the Level-2 product file are set to 1, while all other pixels are set to 0.

2. Define the connectivity rule:
A 3×3 square structural element array is used to determine which pixels count as “neighbours.” Two hotspot pixels identified as “1” in the binary image and located within the same 3×3 pixel array are to be defined as connected and belonging to the same 'cluster'.

3. Apply connected-component labelling:
Using the algorithm described by He et al. (2017), each group of 'spatially connected' hotspot pixels identified via the above means is assigned a unique label. Each label corresponds to a distinct cluster of hotspot pixels.

4. Compute cluster-level spectral ratios:
For every hotspot cluster, the S5 and S6 spectral radiances of all active-fire pixels are summed. The total S5 radiance is then divided by the total S6 radiance to give a cluster-level S5/S6 spectral radiance ratio. Each cluster therefore has its own ratio value.

5. Assign ratios back to individual pixels:
The cluster’s S5/S6 ratio is then copied to each individual hotspot pixel belonging to that cluster.

6. Gas-flare classification:
Finally, each pixels S5/S6 ratio (R) is used to classify that pixel as a gas flare or not, based on a version of the the criteria defined by Fisher et al. (2019). Gas flares exhibit much higher combustion temperatures than vegetation fires, so their R ratio must exceed a temperature-dependent threshold (e.g., corresponding to ~1400 K). A second upper spectral radiance ratio threshold is also applied to help avoid false gas flare classifications caused primarily by residual reflected sunlight in night-time high-latitude scenes.

The threshold tests classify hotspot pixels detected in the SWIR as gas flare pixels if:

Note that since all AF pixels within a particular cluster have the same R value, then the entire cluster will either be classed as containing all gas flare pixels, or all non-gas flare pixels. Also note that the two thresholds used in equation (eq. 6) maybe subject to revision once more comprehensive data and analyses become available.

Persistence Filtering: To ensure that only persistent thermal anomalies are treated as true gas flares—rather than short-lived events such as misclassified vegetation fires or occasional false alarms—a final persistence filter is applied. This filter keeps the information on gas flares only in those grid cells that show sufficiently consistent gas flare detections over time.

From each S3 satellites data, a grid cell must register at least one hotspot classified as a gas-flare (based on the aforementioned S5/S6 spectral radiance ratio threshold) during three consecutive 27-day S3 satellite orbital cycle. Using the full 27-day cycle ensures that observations are comparable from one cycle to the next.

To implement this, for every 27-day orbital cycle, a global 0.1° × 0.1° grid is generated in which each cell stores the number of nights on which a gas-flare hotspot detection occurred. A cell is then set to zero unless it contains at least one detection in any of the following three-cycle combinations:

• the current cycle and the cycle immediately before and after it, or

• the current cycle plus the two preceding cycles, or

• the current cycle plus the two following cycles.

This temporal filter removes grid cells showing only occasional or very intermittent hotspot detections that may seem to be gas flares based on the spectral radiance ratio threshold, but which seem unlikely to be real gas flares as they are too transient.

Example: If over an 18-month period a potential gas flare is detected in the 27-day orbital cycles corresponding to months 1, 2, 3, 4, 6, 7, 9, 11, 12, 1, 2, 3, 4 and 6, then the persistence criteria is satisfied only for months 1, 2, 3, 4, 9, 11, 12, 1, 2, 3 and 4—i.e., only where three consecutive orbital cycles contain gas flare detections.

The final set of gas-flare pixels included in the C3S Gas Flare products are those located within the non-zero grid cells after this filtering step. By combining the spectral-radiance-ratio classification with this temporal persistence criteria, the C3S algorithm identifies gas flares without relying on fixed geographic flare lists, which may become outdated. These spectrally and temporally filtered detections are then stored in both the Level-2 Monthly Night-time Gas Flare Summary Product and the Level-3 Gas Flare Gridded Products.

The contents of each C3S Gridded Product type are described below.

3.2.2. Level 3a Daily Gridded Global Active Fire and FRP Product (Daytime and Night-Time) & Level 3a Daily Global Gas Flare Product (Night-Time Only)

All Level 3a Daily Gridded products - both active fire and gas flare - are generated on a global 0.1 degree resolution grid (G×G), with data stored in NetCDF format.  Data coming from observations made by the different S3 satellites are stored separately, and only observations at thermal channel detected hotspot pixels classed as being land are included in the Level 3a Daily Gridded FRP Product (since landscape fires are limited to land), with daytime and night-time detection data stored separately, but with the Level 3a Daily Gridded Gas Flare product containing data over land and water, but only at night. The latter also contains only those Level 2 product file SWIR-detected hotspots classed as gas flares using the spectral radiance ratio test and the persistence test detailed above (and only at night).

Level 3a Daily Gridded Global Active Fire and FRP Product stores eight layers: (i) total active fire (hotspot) pixel count, (ii) mean FRP from the MIR observation, (iii) uncertainty on the mean FRP, (iv) total number of pixels on land identified as containing atmospheric phenomena that might interfere with AF detection (e.g. thick cloud cover), (v) total number of pixels covering the grid cell, (vi) total number of pixels whose surface conditions might impact AF detection (e.g. which contain too great an area of surface water), (vii) the mean “cloud fraction” (i.e., atmospheric conditions affecting AF detection) of the non-water (i.e. land) pixels in a 1.1° grid cell centred on the 0.1° cell, and (viii) total ‘cloud adjusted’ active fire pixel count in the 0.1° cell.

There is a separate product file for each S3 satellite, and for daytime and night-time. 

Details on how each metric is derived from the data contained within the original Level 2 FRP product files is shown below:

Total AF pixel count on land is derived from the NTC Level 2 Active Fire Detection and FRP Product files generated from the SLSTSR Level 1 data acquired at time t, summed over a regular grid of resolution G×G grid box (0.1 degree in this case). For each grid cell (x_G , y_G ):

Where N_fl is the total AF pixel count on land and i_f,j_f is the set of SLSTR Level 1 granule pixels contained within the grid cell on the land.

The FRP values acquired at time t at the set of land based active fire pixels are also summed over the 0.1 degree resolution grid box. For each grid cell the total FRP is calculated from the MIR-radiance derived FRP values as:

This summation is performed over all the FRP values in the 0.1 degree grid cell, provided the pixel at which the FRP is being assessed is classed as land. The mean land based FRP ( \( \overline{FRP_l} \) ) is the total land FRP ( \( FRP_l \) ) divided by total land AF pixel number ( \( N_{f)} \) ).

Uncertainty on the MWIR-derived mean FRP ( \( U_{\overline{FRP_l}} \) ) is calculated as:

Where U_FRP is the FRP uncertainty of an individual AF pixel.
Similarly, for each grid cell the number of pixels over land identified as containing atmospheric phenomena that might interfere with AF detection (e.g. thick cloud cover), N_cl is calculated as:

And the total number of observed pixels in the grid cell N_o as:

And the total number of pixels classed having surface conditions that might impact AF detection (e.g. which contain too great an area of surface water) N_w as:

Finally, the total number of "cloud-adjusted" active fire pixels on land N'_fl is estimated as the number of active fire pixels present if fires were burning with the same frequency under the pixels identified as containing atmospheric phenomena that might interfere with AF detection (e.g. thick cloud cover as they do under cloud free conditions. For this calculation, the mean "cloud" fraction ( \( \overline{F_{cl}}(t,i_{G2},j_{G2}) \)  of the "non-water" (i.e. land) pixels in a 1.1 degree grid cell (G2) centred on the 0.1 degree grid cell to be adjusted is calculated using:

Then in each 0.1 degree grid cell, the number of "cloud-adjusted" active fire pixels burning on the land surface (N'_fl) is calculated as:

The fraction of the land surface that is cloud free – i.e. the denominator of Equation 15 - is calculated at a 1.1 degree grid resolution instead of 0.1 degree. Calculating the cloud cover at 0.1 degree would result in many grid cells having complete 100% cloud cover – posing a problem for the subsequent adjustment of the active fire counts in Equation 15. The use of a 1.1 degree cell is designed to try to ensure that the cell size for this calculation is large enough to make a reasonable assessment of the relevant atmospheric conditions – such as thick cloud cover. The 1.1° × 1.1° grid cell within which the "cloud cover" fraction is calculated has the 0.1° grid cell under consideration at its centre. If they require it, the total "cloud-adjusted" FRP can be calculated by users simply by multiplying at each grid cell the total number of "cloud-adjusted" active fire pixels over land ( N'_fl) (generated by Equation (eq. 15)) by the mean FRP ( \( \overline{FRP_l} \) ) (generated by Equation (eq. 9)). Note that when the 1.1 degree grid cell is considered too affected by conditions such as thick cloud to provide a "cloud adjusted" AF pixel count, i.e.,  \( \overline{F_{cl}} \) > Threshold (currently envisaged as 90%), Equation 15 is not applied. A value of N'_fl = -1 is provided in Level 3 product file to indicate this, and in place of N'_fl users can instead use the "non-cloud adjusted" value AF pixel count (N_fl) should they wish.

For the Level 3a Daily Gridded Gas Flare Product, there are six data layers derived from the 500 m SWIR-derived information held within each relevant Level 2 product file. The following information is stored at each grid cell: (i) total gas flare pixel count, (ii) mean FRP from the SWIR observations, (iii) uncertainty on this mean FRP, (iv) total gas flare pixel count where the grid cell was fully observed cloud-free, (v) mean SWIR-derived FRP for these fully observed measures, and (vi) uncertainty on this mean FRP. Data are from night-time SLSTR observations only.

Details on how each metric is derived from the data contained within the original Level 2 FRP product files is shown below.

Total gas flare pixel count (N_gas) is derived from the NTC Level 2 Active Fire Detection and FRP Product files generated from the SLSTSR Level-1 500 m pixel resolution SWIR data acquired at time t, summed over a regular grid of resolution  grid box (0.1° in this case). Firstly, within the Level 2 products the night-time observation and spectral radiance ratio tests described above, along with the subsequent 5 x 5 pixel window test, are used to identify only the night-time ‘gas flare’ pixels (N_gas) that should be included in the Level 3 products.

Then for each grid cell (x_G , y_G ) the total AF pixel count on land is derived from the NTC Level 2 Active Fire Detection and FRP Product files generated from the SLSTSR Level 1 data acquired at time t, summed over a regular grid of resolution G×G grid box (0.1 degree in this case).

For each grid cell (x_G , y_G ):

Where Ngas is the total gas flare pixel count and i_f, j_f is the set of night-ime SLSTR Level 2 granule gas flare pixels contained within the grid cell.

The SWIR-based FRP values acquired at time t at the set of gas flare pixels are also summed over the 0.1° resolution grid box. For each grid cell, the total FRP is calculated from the SWIR-radiance derived FRP values as:

This summation is performed over all the FRP values in the 0.1° grid cell, whether or not the pixel at which the FRP is being assessed is classed as land or water. The mean gas flare FRP is the total FRP divided by total gas flare pixel number:

Uncertainty on the SWIR-derived mean FRP is calculated as:

For each grid cell, the Level 2 SWIR-derived observations where the grid cell is fully observed (i.e. the Level 1b file covered the grid cell completely) and without any pixels classified as cloud are then used again with eq. 17 to 19 to calculate the total gas flare pixel count where the grid cell was fully observed cloud-free, the mean SWIR-derived FRP for these fully observed measures, and the uncertainty on this mean gas flare FRP respectively, i.e. the terms:

3.2.3. Level 3a 27-Day Gridded Active Fire and FRP Product  (Daytime and Night-Time) & Level 3a 27-Day Gridded Gas Flare Product (Night-Time Only)

These are multi-layer global files, also derived at 0.1 degree grid cell resolution as with the daily data, but now summarising data from the S3 satellites collected over 27 days. This time period matches the Sentinel-3 satellites standard orbital repeat cycle. The reason for this selection is that the characteristics of the retrieved FRP values are somewhat dependent on view zenith angle even when derived from the (less dependent) F1 channel observations (Xu et al., 2021), and this angle changes markedly around the near-nadir view scan (from close to 0° to close to 55°). Each 27 day repeat cycle sees the same view angle conditions repeated however – so summarising the FRP data over a 27-day period is designed with an approach aimed at providing as consistent a dataset with respect to observation geometry variability as possible.

For each S3 satellite, and for daytime and night-time separately, there are the same 8 data layers stored in the Level 3a 27-Day Gridded FRP Product as for the daily product, but now calculated over the 27-day period over land. Each layer is calculated using the relevant equations detailed in the Daily Gridded product, but now with a calculation time t of 27 days. Similarly, for the Level 3a 27-Day Gridded Gas Flare Product (which is only issued with night-time data) there are there are the same 6 data layers as for the Level 3a Daily Gridded FRP Product, but now calculated over the 27-day period over land and ocean. In each case, each layer is derived using the relevant equations detailed in the Daily Gridded product information above, but now with a calculation time t of 27 days.

3.2.4. Level 3 Monthly Gridded Active Fire and FRP Product  (Daytime and Night-Time) & Level 3 Monthly Gridded Gas Flare Product (Night-Time Only)

The C3S Monthly Gridded FRP product provides monthly global AF detection and FRP data at a grid cell size of 0.25 degrees both day and night, with the cloud cover now assessed at 1.25 degrees. The 0.25 degree resolution matches that of the MODIS Climate Modelling Grid (MOD14CMQ ) monthly summary active fire product. As with the C3S Level 3 27-Day Gridded FRP Products and the 27-Day Gridded Gas Flare Products, the C3S Monthly Gridded products contain the same data layers as the respective Daily Gridded Products, but now but now with a calculation time t equivalent to one month. Also as before, for the Monthly Gridded FRP product the total "cloud-adjusted" FRP can be calculated at each grid cell by users if required, simply by multiplying at each grid cell the total number of "cloud-adjusted" active fire pixels over land ( N'_fl ) by the mean grid cell FRP ( \( \overline{FRP_{fl}} \) ). Note that when the 1.25 degree cell is considered too affected by atmospheric phenomena such as thick cloud, i.e.,  \( \overline{FRP_{lc}} >Threshold \) (currently envisaged as 90%) , a value of -1 is provided instead for (N_f). Users should then use the "non-cloud adjusted" value of active fire pixel count over land (N_fl) should they wish a value for that grid cell.
Note the "cloud fraction" are applied here at a reduced (1.25 degree) grid cell resolution, for the reasons outlined earlier. The 1.25 x 1.25 degree area over which the "cloud fraction" is calculated has the 0.25 degree grid cell under consideration at its centre.

Similarly the matching C3S Monthly Gridded Gas Flare product stores the same information as the Daily and 27-Day Gridded Gas Flare products, but now similarly at 0.25 degree monthly intervals and for night-time only.

3.3. Additional Information regarding the S7- and F1-Derived FRP Values

Since the S7 and F1 channels of SLSTR detailed in Table 1 both measure in the same MIR spectral band, FRP values can be derived from either measurement (Xu et al., 2021). In fact, data from both S7 and F1 must be combined for any FRP production algorithm to be most effective, because brightness temperatures in excess of 311 K can only be accurately provided by F1, whereas cooler BTs are more precisely measured by the lower-noise S7 channel (Xu et al., 2020; 2021). However, the instantaneous field of view (IFOV) of the S7 and F1 channels are very different due to their differing detector shapes (Xu et al., 2021), with that of F1 being far narrower than that of S7, and there is a slight spatial offset between the two channels as well. At any point around the near nadir scan, the matching S7 and F1 channels thus have quite different ground pixel footprint shapes, and are smaller for F1. Around the scan, F1 also shows far more limited growth in the area of its pixel footprint compared to S7. For lower FRP clusters whose S7 brightness temperature observations remain unsaturated across all AF pixels, the FRP can be retrieved in all AF pixels via either the S7 (termed the “F1_OFF” option) or F1 (“F1_ON” option) data. Before switching to the latter “F1_ON” option after August 2020, the NTC Level-2 FRP product used the former “F1_OFF” option, and this meant that fire clusters whose S7 channel observations remained unsaturated across all its AF pixels had their FRP and constituent pixel positions derived from the S7 observations. Conversely, fire clusters containing one or more saturated S7 pixels had their FRP and pixel positions recorded from F1. The C3S Level-2 Summary Product contains an “F1_flag” to record this, with 1 indicating an FRP derived from the F1 channel measurement and 0 indicating it came from S7. However, for all three C3S Level-3 FRP Synthesis Products, the active fire count and FRP metrics are gridded from the Level-2 data regardless of which of S7 or F1 was used in the FRP retrieval. After August 2020, the Level-2 NTC night-time product algorithm was set to use the “F1_ON” option for all fires, so all AF position and FRP data in the C3S products come from the F1 channel data after this time.

Note also that the Sentinel-3 L2 FRP products upon which the C3S products are based utilise pixel coordinates from both the SLSTR F1 and S7 channels, not only for the fire pixels but also with regard to certain flags used in Sentinel-3 L2 FRP products such as "cloud" and "water". No matter what pixel coordinates are used for the AF pixel locations, these latter flags are always based on the S7 channel grid since the relevant classifications of pixel type are based on observations from the SLSTR ‘S’ channels. In the case of the C3S FRP L2 Summary product, the source of the original L2 AF pixel location coordinates is specified by the "F1_flag" introduced above, which indicates whether these come from the S7 or F1 pixel grid. After August 2020 these always come from the F1 pixel grid for the night-time data (and indeed day-time data as well). However, generating the C3S FRP L3 products necessitates aggregating the SLSTR L2 FRP products, including the corresponding "cloud" and "water" flags that always come from the S7 channel grid. This difference in the source of the grid can occasionally result in AF and FRP being placed in one grid cell, whilst the corresponding "cloud" and "water" flags are counted in the neighbouring cell, thereby causing a small inconsistency.

4. Output data

4.1. Level 2 Monthly Global Active Fire and FRP Summary Product (Daytime and Night-Time) & Level 2 Monthly Global Gas Flare Summary Product (Night-Time Only)

Each C3S Monthly Level 2 Active Fire & FRP Global Summary Product is a comma delimited ASCII text CSV file containing all the information of daytime or night-time detected land-based hotspot pixels from the Level 2 products generated that month, and with the Global Gas Flare product specifically only those identified at night as being gas flares either on land or ocean based on the S5/S6 spectral radiance ratio and persistence tests.

The information held within the C3S Level 2 Monthly Level 2 Active Fire & FRP Global Summary Product CSV file is provided in Table 8.

Table 8: Information held within each C3S Level 2 Monthly Level 2 Active Fire & FRP Global Summary Product File at the location of each detected land-based active fire (AF) pixel. Daytime and night-time data stored in separate files from each S3 satellite.

Figure 3 presents a C3S Level‑2 monthly summary product for January 2021 from Sentinel-3A, displayed in Excel. The first row lists variable names, and the data records begin in the second row. Only a small portion of the data is shown because the full record is extremely long.

Figure 3: An example of C3S Sentinel-3A Level-2 monthly summary product in Jan.2021, displayed in excel.

Details of all the information held within the C3S Level 2 Monthly Global Gas Flare Summary Product CSV file is provided in  Table 9.

 Table 9: Information held within each Level 2 Monthly Global Gas Flare Summary Product File at the location of each detected active fire (AF) pixel. Night-time data only. Data stored in separate files from each S3 satellite.

4.2. Level 3a Daily Gridded FRP Product & Level 3a Daily Gridded Gas Flare Product

Each Level 3a Daily Gridded FRP Product file is a tiled, multi-layer NetCDF file covering the globe at a 0.1 degree grid cell resolution. Each file stores information contained in one day of Level 2 Sentinel-3 AF detection and FRP Product files. Only data in grid cells containing land are provided in the Level 3a Daily Gridded FRP Product files in order to make the product most relevant to only vegetation fires and organic soil burning (avoiding oceanic gas flares). Separate files are provided for the different S3 satellites, and separate for day and night observations. In each grid-cell location eight values as stored:

1. Number of active fire pixels detected in each land grid cell (grid-cells containing only water are given zero AF detections, since any fires would likely be due to oceanic gas flares rather than biomass burning)
2. Mean FRP of the detected land-based AF pixels
3. Uncertainty of the Mean FRP of the detected land-based AF pixels
4. Total number of pixel observations made within the grid cell
5. Total number of pixels in the grid cell whose surface conditions might impact AF detection (e.g. which contain too great an area of surface water)
6. Total number of pixels on land identified as containing atmospheric phenomena that might interfere with AF detection (e.g. thick cloud cover)
7. The mean "cloud fraction" of the land pixels in a 1.1° x 1.1° degree grid cell
8. Total active fire pixel count adjusted for atmospheric conditions that impede AF detection (e.g. thick cloud cover).

Figure 4 shows an example of data extracted from the Level3a Daily Gridded FRP product for 15 September 2024, in this case a night-time S3A product file. These data are reported at a 0.1° × 0.1° grid cell resolution, and panel (a) reports the number of land-based active fire pixel detections per grid cell (1. in the above list), while panel (b) reports the total Fire Radiative Power (FRP) of these fires in MW as calculated from these active fire pixel counts multiplied by the mean FRP of the detected land-based AF pixels (2. in the above list). The spatial patterns shown in these datasets clearly highlight the extensive biomass burning occurring in the Southern Hemisphere at this time, particularly across the Amazon Basin and southern Africa.

Figure 4: Data extracted from the C3S Level3a Daily Gridded FRP product for 15 September 2024, in this case a night-time S3A product file; (a) land-based Active Fire Pixel Count within each grid cell, and (b) total FRP within each grid cell as derived from the product of active fire pixel count and mean per-pixel FRP.

Each Level 3a Daily Gridded Gas Flare Product file is similar to the above format, but only exists for night-time data and contains six rather than eight values for each grid cell location:

1. Total number of detected gas flare pixels
2. Mean FRP (derived from SWIR channel observations) of all detected gas flare pixels
3. Uncertainty on this SWIR-derived mean FRP
4. Total number of detected gas flare pixels when the cell was fully observed cloud-free
5. Mean SWIR-derived FRP derived from the flare pixels detected when the cell was fully observed cloud-free
6. Uncertainty on this SWIR-derived mean FRP

4.3. Level 3a 27-Day Gridded FRP Product & Level 3a 27-Day Gridded Gas Flare Product

The Level 3a 27-Day Gridded FRP Product time interval is selected to match the standard Sentinel-3 orbital repeat cycle. Data from each S3 satellite are maintained separately, as are results based on daytime and night-time observations, and each file stores eight layers each calculated over the 27-day period in question:

1. Number of active fire pixels detected in each land grid cell
2. Mean FRP of the detected land-based AF pixels
3. Uncertainty of this mean MIR-derived FRP of the detected land-based AF pixels
4. Total number of pixel observations made within the grid cell
5. Total number of pixels in the grid cell whose surface conditions might impact AF detection (e.g., which contain too great an area of surface water)
6. Total number of pixels on land identified as containing atmospheric phenomena that might interfere with AF detection (e.g., thick cloud cover)
7. The mean “cloud fraction” of the land pixels in a 1.1° x 1.1° grid cell
8. Total active fire pixel count adjusted for atmospheric conditions that impede AF detection (e.g., thick cloud cover)

Figure 5 shows an example of data extracted from the Level 3a 27-day Gridded FRP product for the period 11 September to 7 October 2024, in this case a night-time S3A product file. These data are reported at a 0.1° × 0.1° grid cell resolution, and panel (a) reports the number of land-based active fire pixel detections per grid cell (1. in the above list), while panel (b) reports the total Fire Radiative Power (FRP) of these fires in MW as calculated from these active fire pixel counts multiplied by the mean FRP of the detected land-based AF pixels (2. in the above list). Compositing over the 27-day orbital repeat cycle of the Sentinel-3 satellite effectively fills orbital gaps, lessens the effect of cloud obstruction inherent in the daily products, and ensures all locations are viewed with the same viewing geometry range in each product file, thus providing a comprehensive view of fire activity - in this case at night during the peak of the southern hemisphere burning season.

Figure 5: Data extracted from the C3S level 3a 27-Day gridded product for the period 11 September to 7 October 2024, in this case a night-time S3A product file; (a) land-based Active Fire Pixel Count within each grid cell, and (b) total FRP within each grid cell as derived from the product of active fire pixel count and mean per-pixel FRP.

Each Level 3a 27-Day Gridded Gas Flare Product file is similar to the above format, but only exists for night-time data and contains six rather than eight values for each grid cell location:

1. Total number of detected gas flare pixels
2. Mean FRP (derived from SWIR channel observations) of all detected gas flare pixels
3. Uncertainty on this SWIR-derived mean FRP
4. Total number of detected gas flare pixels when the cell was fully observed cloud-free
5. Mean SWIR-derived FRP derived from the flare pixels detected when the cell was fully observed cloud-free
6. Uncertainty on this SWIR-derived mean FRP

4.4. Level 3 Monthly Gridded FRP Product & Level 3 Monthly Gridded Gas Flare Product

The Level 3 Monthly Summary FRP Product provides global data at a grid cell size of 0.25 degrees, matching that of the Terra MODIS Climate Modelling Grid (MOD14CMQ) active fire products. Data from each S3 satellite are maintained separately, as are results based on daytime and night-time observations, and each file stores eight layers each calculated over the month in question:

1. Number of active fire pixels detected in each land grid cell
2. Mean FRP of the detected land-based AF pixels
3. Uncertainty of this mean MIR-derived FRP of the detected land-based AF pixels
4. Total number of pixel observations made within the grid cell
5. Total number of pixels in the grid cell whose surface conditions might impact AF detection (e.g., which contain too great an area of surface water)
6. Total number of pixels on land identified as containing atmospheric phenomena that might interfere with AF detection (e.g., thick cloud cover)
7. The mean “cloud fraction” of the land pixels in a 1.25° x 1.25° grid cell
8. Total active fire pixel count adjusted for atmospheric conditions that impede AF detection (e.g., thick cloud cover).

Figure 6 shows an example of data extracted from the Level‑3 Monthly Gridded FRP product for September 2024, in this case a night-time S3A product file. These data are reported at a 0.25° × 0.25° grid cell resolution, and panel (a) reports the number of land-based active fire pixel detections per grid cell, while panel (b) reports the total Fire Radiative Power (FRP) of these fires in MW (1. in the above list), as calculated from these active fire pixel counts multiplied by the mean FRP of the detected land-based AF pixels (2. in the above list). This monthly composite provides an easy to use, comprehensive view of night-time fire activity during the peak Southern Hemisphere burning season, and is designed to be broadly compatible with the long-standing Terra MODIS Climate Modelling Grid (MOD14CMQ) active fire product that has the same monthly interval and 0.25° grid cell size.

Figure 6: Data extracted from the C3S Level‑3 Monthly Gridded FRP product for September 2024, in this case a night-time S3A product file; (a) land-based Active Fire Pixel Count within each grid cell, and (b) total FRP within each grid cell as derived from the product of active fire pixel count and mean per-pixel FRP.

Each Level 3 Monthly Gridded Gas Flare Product file is similar to the above format, but only exists for night-time data and contains six rather than eight values for each grid cell location:

1. Total number of detected gas flare pixels
2. Mean FRP (derived from SWIR channel observations) of all detected gas flare pixels
3. Uncertainty on this SWIR-derived mean FRP
4. Total number of detected gas flare pixels when the cell was fully observed cloud-free
5. Mean SWIR-derived FRP derived from the flare pixels detected when the cell was fully observed cloud-free
6. Uncertainty on this SWIR-derived mean FRP

References

Coppo, P., Ricciarelli, B., Brandani, F., Delderfield, J., Ferlet, M., Mutlow, C., Munro, G., Nightingale, T., Smith, D., Bianchi, S., Nicol, P., Kirschstein, S., Hennig, T., Engel, W., Frerick, J. and Nieken J. (2010) SLSTR: a high accuracy dual scan temperature radiometer for sea and land surface monitoring from space, Journal of Modern Optics, 57:18, 1815-1830, DOI: 10.1080/09500340.2010.503010.

Elvidge, C.D., Zhizhin, M., Sparks, T., Ghosh, T., Pon, S., Bazilian, M., Sutton, P.C. and Miller, S.D., 2023. Global satellite monitoring of exothermic industrial activity via infrared emissions. Remote Sensing, 15(19), p.4760.

E.U. Copernicus Climate Change Service, 2021a. Product User Guide and Specification ICDR - SLSTR Fire Radiative Power Products, D3.3.16-v1.0_PUGS_CDR-ICDR_FRP_SENTINEL3_v1.0.x_PRODUCTS_v1.0.1

E.U. Copernicus Climate Change Service 2021b. Product Quality Assurance Document ICDR SLSTR Fire Radiative Power Product: D2.2.16-v1.0_PQAD_CDR-ICDR_FRP_SENTINEL3_v1.0_PRODUCTS_v1.0

E.U. Copernicus Climate Change Service 2021c. Tutorial Fire Radiative Power - SLSTR Fire Radiative Power Products, D4.3.5_Tutorial_CDR-ICDR_FRP_v1.0

E.U. Copernicus Climate Change Service, 2025. Product User Guide and Specification (PUGS) v1.0 - CDR  and ICDR Sentinel-3 Active Fire & Fire Radiative Power Day-time, Night-time Products & Gas Flare Products (v1.2), C3S2_313e_BC_WP1-DDP-FIRE-AF&FRP-S3-SLSTR-v1.2-2024_202506_PUGS_v1.0

E.U. Copernicus Climate Change Service, 2025., Product Quality Assessment Report (PQAR) v1.0 - CDR  and ICDR Sentinel-3 Active Fire & Fire Radiative Power Day-time, Night-time Products & Gas Flare Products (v1.2), C3S2_313e_BC_WP1-DDP-FIRE-AF&FRP-S3-SLSTR-v1.2-2024_202506_PQAR_v1.0

Fisher, D. and Wooster, M.J., 2018. Shortwave IR adaption of the mid-infrared radiance method of fire radiative power (FRP) retrieval for assessing industrial gas flaring output. Remote Sensing, 10(2), p.305.

Fisher, D. and Wooster, M.J., 2019. Multi-decade global gas flaring change inventoried using the ATSR-1, ATSR-2, AATSR and SLSTR data records. Remote Sensing of Environment, 232, p.111298.

Freeborn, P.H., Wooster, M.J., Hao, W.M., Ryan, C.A., Nordgren, B.L., Baker, S.P. and Ichoku, C., 2008. Relationships between energy release, fuel mass loss, and trace gas and aerosol emissions during laboratory biomass fires. Journal of Geophysical Research: Atmospheres, 113(D1).

Freeborn, P.H., Wooster, M.J., Roberts, G., & Xu, W.D. (2014). Evaluating the SEVIRI Fire Thermal Anomaly Detection Algorithm across the Central African Republic Using the MODIS Active Fire Product. Remote Sensing, 6, 1890-1917.

Giglio L., Descloitres J., Justice C.O., Kaufman Y.J (2003). An Enhanced Contextual Fire Detection Algorithm for MODI. Remote Sensing of Environment, 87, 273-282.

Giglio, L. (2007). Characterization of the tropical diurnal fire cycle using VIRS and MODIS observations. Remote Sensing of Environment, 108, 407-421.
Giglio, L., Schroeder, W. and Justice, C.O., 2016. The collection 6 MODIS active fire detection algorithm and fire products. Remote Sensing of Environment, 178, pp.31-41.

Godsalve, C. (1995). Bi-directional reflectance sampling by ATSR-2: a combined orbit and scan model. International Journal of Remote Sensing, 16(2), 269-300.

He, L., Ren, X., Gao, Q., Zhao, X., Yao, B. and Chao, Y. (2017). The connected-component labeling problem: A review of state-of-the-art algorithms. Pattern Recognition, 70, pp.25-43.

Nguyen, H. M., & Wooster, M. J. (2020). Advances in the estimation of high Spatio-temporal resolution pan-African top-down biomass burning emissions made using geostationary fire radiative power (FRP) and MAIAC aerosol optical depth (AOD) data. Remote Sensing of Environment, 248, 111971.

Roberts, G., Wooster, M.J., Perry, G.L., Drake, N., Rebelo, L.M. and Dipotso, F., 2005. Retrieval of biomass combustion rates and totals from fire radiative power observations: Application to southern Africa using geostationary SEVIRI imagery. Journal of Geophysical Research: Atmospheres, 110(D21).

Roberts, G., Wooster, M. J., & Lagoudakis, E. (2009). Annual and diurnal african biomass burning temporal dynamics. Biogeosciences, 6(5), 849-866.

Sessa, R. (Ed.). (2008). Terrestrial Essential Climate Variables: For Climate Change Assessment, Mitigation and Adaptation. GTOS-Secr., Food and Agriculture Organization of the United Nations.

Wooster, M. J., Roberts, G., Perry, G. L. W., & Kaufman, Y. J. (2005). Retrieval of biomass combustion rates and totals from fire radiative power observations: FRP derivation and calibration relationships between biomass consumption and fire radiative energy release. Journal of Geophysical Research: Atmospheres, 110(D24).

Wooster, M. J., Xu, W., & Nightingale, T. (2012). Sentinel-3 SLSTR active fire detection and FRP product: Pre-launch algorithm development and performance evaluation using MODIS and ASTER datasets. Remote Sensing of Environment, 120(0), 236–254.

Wooster, M.J., Roberts, G., Freeborn, P.H., Xu, W., Govaerts, Y., Beeby, R., He, J., Lattanzio, A., Fisher, D. and Mullen, R., 2015. LSA SAF Meteosat FRP products–Part 1: Algorithms, product. Atmospheric Chemistry and Physics. Chem. Phys, 15(22), pp.13217-13239.

Wooster, M.J., Zhukov, B., & Oertel, D. (2003). Fire radiative energy for quantitative study of biomass burning: derivation from the BIRD experimental satellite and comparison to MODIS fire products. Remote Sensing of Environment, 86, 83-107.

Wooster, M.J., Roberts, G.J., Giglio, L., Roy, D.P., Freeborn, P.H., Boschetti, L., Justice, C., Ichoku, C., Schroeder, W., Davies, D. and Smith, A.M. (2021). Satellite remote sensing of active fires: History and current status, applications and future requirements. Remote Sensing of Environment, 267, p.112694.

Xu, W., Wooster, M. J., He, J., & Zhang, T. (2020). First study of Sentinel-3 SLSTR active fire detection and FRP retrieval: Night-time algorithm enhancements and global intercomparison to MODIS and VIIRS AF products. Remote Sensing of Environment, 248, 111947.

Xu, W., Wooster, M.J., Polehampton, E., Yemelyanova, R. and Zhang, T., 2021. Sentinel-3 active fire detection and FRP product performance-Impact of scan angle and SLSTR middle infrared channel selection. Remote Sensing of Environment, 261, p.112460.

Xu, W. and Wooster, M.J., 2023. Sentinel-3 SLSTR active fire (AF) detection and FRP daytime product-Algorithm description and global intercomparison to MODIS, VIIRS and landsat AF data. Science of Remote Sensing, 7, p.100087.

Xu, W., Wooster, M. J., Polehampton, E., Yemelyanova, R., & Zhang, T. (2021). Sentinel-3 active fire detection and FRP product performance-Impact of scan angle and SLSTR middle infrared channel selection. Remote Sensing of Environment, 261, 112460.

Zhukov, B., Lorenz, E., Oertel, D., Wooster, M. and Roberts, G. (2006). Spaceborne detection and characterization of fires during the bi-spectral infrared detection (BIRD) experimental small satellite mission (2001–2004). Remote sensing of Environment, 100, pp.29-51.

Related articles

• Page:
  Active Fire & Fire Radiative Power (AF & FRP) & Gas Flare Products V1.2: Product Quality Assessment Report (PQAR) (Copernicus Knowledge Base)

  • c3s
  • ecv
  • af
  • frp
  • pqar
• Page:
  Active Fire & Fire Radiative Power (AF & FRP) & Gas Flare Products V1.2: Product User Guide and Specifications (PUGS) (Copernicus Knowledge Base)

  • c3s
  • ecv
  • af
  • frp
  • pugs
• Page:
  Active Fire & Fire Radiative Power (AF & FRP) & Gas Flare Products V1.2: Algorithm Theoretical Basis Document (ATBD) (Copernicus Knowledge Base)

  • c3s
  • ecv
  • af
  • frp
  • atbd