Publications

Xu, WD; Wooster, MJ; Polehampton, E; Yemelyanova, R; Zhang, TR (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.

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 similar to 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.

DOI:
10.1016/j.rse.2021.112460

ISSN:
0034-4257