Sentinel-3 active fire detection and FRP product performance - Impact of scan angle and SLSTR middle infrared channel selection

https://doi.org/10.1016/j.rse.2021.112460Get rights and content

Highlights

  • We study the impact of scan angle and selection of S7 or F1 MIR channel data use on SLSTR active fire detection and FRP product performance.

  • The F1 channel pixel footprint area is smaller and more consistent across the swath than that of S7.

  • SLSTR AF detection performance degrades with increasing pixel area, due to an increasing minimum FRP detection limit.

  • Using data from the F1 rather than S7 channel offers signifcant benefits during AF product generation and is recommended wherever possible.

Abstract

The Sentinel-3 satellites each carry the dual-Earth view Sea and Land Surface Temperature Radiometer (SLSTR). SLSTR data from the ‘near nadir’ scan are used to produce a set of global, daily active fire (AF) products similar to those produced from MODIS data. The Sentinel-3 AF products are generated in both Near Real Time (NRT) and Non Time Critical (NTC) mode and are expected ultimately to replace those from MODIS Terra when that mission starts to reach its end of life. Middle infrared (MIR) observations are critical for the AF application, and SLSTR has two MIR channels (S7 and F1) based on different detector designs and located at different positions in the instrument focal plane, delivering different dynamic ranges with the same spectral response function. The SLSTR AF detection and FRP retrieval algorithm uses both S7 and F1 channel data, but there are options on how these are combined to generate the final AF products. Past work has shown that MODIS' AF product characteristics are strongly view zenith angle (VZA) dependent, and since the SLSTR ‘near nadir’ scan actually extends out to a maximum VZA of 55° (similar to the maximum MODIS VZA), any such view angle dependency may also impact the choice of how the S7 and F1 channel data are combined to generate the SLSTR AF products. Here we use a combination of simulation modelling and analysis of more than 15,000 early SLSTR AF products to study how VZA affects the Sentinel-3 AF product characteristics, and how this varies when the S7 and F1 channel data are combined in different ways. Maximum use of S7 channel data provides an AF product whose statistical characteristics vary markedly with VZA around the near nadir scan. For example, the median FRP extracted from the more than 15,000 products studied varies by around an order of magnitude between the centre and edge of scan positions, a level of VZA-dependency similar to that seen with the MODIS AF products. By contrast, maximum use of data from the F1 channel provides a median FRP that varies only by a factor of ~3 over the same VZA range. Our simulation modelling demonstrates that this difference is a result of the area of the F1 channel ground pixel footprint remaining far more consistent around the near nadir scan compared to that of S7, which also leads to minimised F1 pixel oversampling even at high VZA observations. The result is that FRP retrievals for a fire derived using the F1 channel observations are far less sensitive to where in the scan the fire is observed. When generated in this way, the Sentinel-3 FRP product characteristics appear similar to those of the VIIRS AF products, since VIIRS also has a method of minimising pixel area growth away from nadir. We conclude that maximum use of the SLSTR F1 channel observations provides significant benefits to the Sentinel-3 AF product, and that unless there are other reasons not to do so any operational processing to generate the NRT and NTC product versions should consider this option as the default. Our findings also show the value of the novel detector design of the SLSTR F1 channel, whose advantageous pixel footprint characteristics mean that similar designs should be considered for future sensors aiming to support active fire applications.

Introduction

The Sea and Land Surface Temperature Radiometer (SLSTR) builds on the heritage of the Along Track Scanning Radiometer (ATSR, Delderfield et al., 1986) and operates on two concurrently orbiting European satellites - Sentinel-3A and -3B (S3A and S3B) launched in 2016 and 2018 respectively. SLSTR provides dual-viewing angle, highly accurate imaging radiometry in multiple channels spanning the visible to longwave infrared spectral regions (Llewellyn-Jones et al., 2001; Smith et al., 2020). The primary role of SLSTR is to provide the dual view, highly radiometrically accurate data required to generate very high quality, daily mapping of Earth's sea and land surface temperature (SST and LST; Coppo et al., 2010, Coppo et al., 2015).

The widely used ESA World Fire Atlas (WFA; Arino and Rosaz, 1999) was generated from the ATSR and Advanced ATSR (AATSR) data records, forerunner instruments to SLSTR. SLSTR was equipped with certain new design features that would further improve the detection and characterisation of active fires beyond that provided by (A)ATSR. Specifically, the SLSTRs wider swath and two middle infrared (MIR) bands (the ‘standard’ S7 channel and the high dynamic range ‘F1’ fire channel; Coppo et al., 2010, Coppo et al., 2015; Wooster et al., 2012) make SLSTR able to provide greatly enhanced active fire (AF) information compared to (A)ATSR. This capability is more akin to that of the MODIS sensors deployed on the Terra and Aqua satellites, whose data are used to respectively generate the very widely used MODIS MOD14 and MYD14 AF products (Giglio et al., 2003, Giglio et al., 2016). Beyond AF detection, which was already included in the ESA WFA (Arino and Rosaz, 1999) but which is conducted using a far more comprehensive algorithm in the case of SLSTR (Wooster et al., 2012), the data from S7 and F1 allow routine calculation of the fire radiative power (FRP) of each detected AF pixel (Xu et al., 2020). FRP equates to the rate of the radiant heat output from a fire, and acts as a means of estimating rates of fuel consumption and smoke emission (Wooster et al., 2005). The pre-launch SLSTR AF product algorithm (Wooster et al., 2012) was recently significantly updated now that a number of years of real SLSTR data are available (Xu et al., 2020). Based on an operational version of this algorithm, Level 2 night-time AF products are now produced from SLSTR near nadir-view scan Level 1 data. Since the Sentinel-3 satellites (currently S3A and S3B) have a local equatorial crossing time (10:00 h and 22:00 h) that is very close to that of Terra (10:30 h and 22:30 h), the Sentinel-3 AF products are expected to ultimately replace those generated from the MODIS sensor onboard Terra as that mission reaches its end of life (expected early 2020s). Near real-time (NRT) versions of the night-time Sentinel-3 AF Detection and FRP Products have been publicly available from EUMETSAT since March 2020 (https://www.eumetsat.int/S3-NRT-FRP) and the non-time critical (NTC) product versions generated using the most up-to date ancillary information (e.g. the atmospheric profiles used for FRP atmospheric correction; Xu et al., 2020) are made operationally available by ESA with a short delay at the Copernicus Open Access Hub (https://scihub.copernicus.eu/dhus/#/home). Full daytime versions of both Level 2 NRT and NTC products are expected to be released in 2021. The purpose of the current work is to detail some key aspects of the SLSTR Level 1 data that influence generation of the Level 2 AF products, and specifically to evaluate how choices between how data collected in the two different dynamic range middle infrared bands (S7 and F1) are combined can affect the characteristics of the final AF product data record. In work akin to that conducted by Freeborn et al. (2011) for the MODIS AF products, we also evaluate how the SLSTR AF products characteristics are affected by the wide variations in view zenith angle (VZA; out to 55°) found around the instruments near-nadir scan, and how this varies depending on how the data from the S7 and F1 bands are combined. This work is designed to inform choices made to maximise the performance of the operational Sentinel-3 SLSTR AF products and their future ability to be merged with those from MODIS Terra to provide the type of long-term data record suitable for climate related investigations.

Section snippets

Sentinel-3 SLSTR instrument scans

As introduced in Section 1, a key aspect of SLSTR is the collection of data in two scans of the Earth within a few minutes of one another but made under very different view zenith angle (VZA) conditions. These two near contemporaneous scans deliver the data required to very accurately account for atmospheric effects on the measured infrared brightness temperatures, facilitating the SST retrieval process in particular (Coppo et al., 2015). The first scan made of an Earth location by SLSTR is the

SLSTR observation modelling methodology

The SLSTR S7 and F1 pixel footprints shown in Fig. 3, Fig. 4 were generated using an observation simulator from STFC Rutherford Appleton Laboratory which is akin to that used to simulate ATSR-2 data by Godsalve (1995). The ground sampling distance of the footprint maps used within the observation simulator is 40 m, and an input surface-leaving radiance map representing the ground target whose SLSTR view is to be simulated is generated at the same 40 m resolution. Two different ground targets

Modelled impact of scan position on retrieved FRP

Building on the simulations made at the near-nadir view centre and edge of scan positions shown in Fig. 6, Fig. 8, results from all scan positions across the full SLSTR near nadir swath for both the 800 K sub-pixel (37.2 MW) artificial fire and the real (184 MW) boreal forest fire are shown in Fig. 9. At each scan position, the graphs show the ratio of the retrieved FRP compared to that retrieved at the scan centre, based on statistics derived for 28 neighbouring pixel sampling locations around

Summary and conclusions

The Sea and Land Surface Temperature Radiometer (SLSTR) operating onboard the Sentinel-3A and -3B satellites provides dual-view, global observations of the Earth every day in multiple VIS, MIR and -LWIR spectral channels. Observations from the SLSTR near nadir view scan are being used to generate a set of Level 2 Active Fire Detection and FRP Products based on an operational version of the algorithm presented in Xu et al. (2020). The S3 satellites equatorial crossing time of ~10:00 h and

Credit author statement

W.Xu: Data curation, Investigation, Methodology, Formal Analysis, Writing - original draft, Visualisation. M.J. Wooster: Conceptulisation, Funding Aquistion, Methodology, Project Administration, Resources, Supervision, Validation, Writing - review & Editing; Visualisation; E. Polehampton: Methodology, Software. R. Yemelyanova: Methodology, Software, Investigation. T. Zhang: Methodology, Software, Investigation.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

This work was supported by NERC National Capability funding though the UK's National Centre for Earth Observation (NE/R016518/1) and by the European Space Agency FIDEX project (ESA Contract 4000122813/17/I-BG). The NERC Airborne Research Facility (NARF) and British Antarctic Survey are thanked for support in use of the twin otter aircraft. ACRI ST are thanked for processing of the Level 2 SLSTR data products used herein.

References (26)

  • O. Arino et al.

    1997 and 1998 world ATSR fire atlas using ERS-2 ATSR-2 data

  • P. Coppo et al.

    SLSTR: a high accuracy dual scan temperature radiometer for sea and land surface monitoring from space

    J. Mod. Opt.

    (2010)
  • Peter Coppo et al.

    Sea and land surface temperature radiometer on Sentinel-3

  • Cited by (11)

    • Global burned area mapping from Sentinel-3 Synergy and VIIRS active fires

      2022, Remote Sensing of Environment
      Citation Excerpt :

      A full daytime version was released in 2021, although for now only the night-time version is available as standard global, monthly summarised files (https://cds.climate.copernicus.eu/cdsapp#!/dataset/satellite-fire-radiative-power?tab=overview, last accessed July 2022). The SLSTR sensor can complement MODIS or VIIRS active fire information (Xu et al., 2021). This paper presents the development of a global burned area algorithm based on SYN reflectance and VIIRS 375 m active fire products.

    • Satellite remote sensing of active fires: History and current status, applications and future requirements

      2021, Remote Sensing of Environment
      Citation Excerpt :

      Contextual algorithms still remain the most commonly used AF detection method. Since their initial development (Flasse and Ceccato, 1996; Giglio et al., 1999; Kaufman et al., 1998), they have been applied to data from numerous LEO sensors, including VIIRS (Schroeder et al., 2014; Csiszar et al., 2014; Zhang et al., 2017), MODIS (Kaufman et al.,1998; Giglio et al., 2016), the BIRD Hot Spot Recognition System (HSRS, Zhukov et al., 2006), the TRMM Visible and Infrared Scanner (VIRS, Giglio et al., 2000), SLSTR (Wooster et al., 2012b; Xu et al., 2020), and Landsat (Schroeder et al., 2016; Kumar and Roy, 2018), and also GEO sensors such as those carried by Meteosat (e.g. Wooster et al., 2015; Amraoui et al., 2010; Di Biase and Laneve, 2018), GOES (Prins et al., 1998; Xu et al., 2010; Schmit et al., 2017; Xu et al., 2021a), Himawari (Xu et al., 2017; Wickramasinghe et al., 2016), and FengYun (Xu et al., 2010). The majority of AF detection algorithms are applied to single date imagery, with some adding basic temporal constraints to remove possible false alarms (e.g., Prins et al., 1998; Xu et al., 2010; Kumar and Roy, 2018).

    View all citing articles on Scopus
    View full text