Top of Atmosphere reflectance in the 4 high resolution channels of Sentinel-2

The above plot shows the TOA reflectance time series gathered by Sentinel-2 over a pixel chosen randomly in a tile in the centre of France (tile 31TDK, pixel 3000-7000), from L1C products.  Looking at the time series, it is rather difficult to tell what kind of surface was observed, even if a vegetation cycle seems to be present.  As we will see below, most of the observed noise is due to the presence of clouds or cloud shadows.

The plot below shows that after removing all clouds and cloud shadows occurrences, the top of atmosphere reflectance is already quite smooth, and it is much easier to understand the type of surface observed. It is even easier with the surface reflectance bottom plot : the pixel is a little green all year, but gets much greener at the beginning of summer. It is a mid altitude meadow.

You have probably noted that the atmospheric correction removes some noise, as well as the directional effect correction, but the main part of the noise reduction is the result of cloud detection obtained with the MACCS processor.  Such kinds of time series are used in the iota2 processor to automatically determine the land cover. A less accurate cloud masking would probably leave more errors. Other colleagues, in Sen2Agri project, use such vegetation cycles to determine the start and end of the vegetation cycle, A less accurate cloud masking would result in leaving some noise in the time series, providing much less accurate results. And we also produce monthly cloud free syntheses of surface reflectance which are also quite sensitive to the presence of a forgotten cloud or shadow. We checked what happened with Sen2Cor cloud mask (version 2.3.0) for the same time series. This is not that easy because Sen2Cor provides three cloud masks : Low probability (LP), Medium Probability (MP) and High Probability (HP). Low probability can be ruled out as it provides really too many false cloud detections, each bright pixel being classified as a cloud. The amount of false detection is still high for Medium Probability but further reduced with High Probability, but in that case, some clouds are missed. Here are three figures for 3 different cases already shown in this post (a meadow, a coniferous forest, and a building roof). Each figure shows the results for (top to bottom), MACCS, SEN2COR using the medium probability cloud mask, and SEN2COR using the the high probability cloud mask.

Comparison for a mid altitude meadow, top MACCS, Middle, Sen2cor Medium probability, bottom, sen2cor High probability
Comparison for a pine forest, top MACCS, Middle, Sen2cor Medium probability, bottom, sen2cor High probability
Comparison for buiding roof, top MACCS, Middle, Sen2cor Medium probability, bottom, sen2cor High probability. The Medium probability mask always flag that building as cloudy, while the high probability mask doesn't, but some times forgets a few clouds.
 This means that to use time series from Sen2cor operationally, one will have to add an outlier detection (with HP mask), or accept to have some pixels always flagged as cloudy with MP mask. With MACCS, even if the time series are not always perfect, the consistency of time series is much better and can be used automatically without outlier detection, at least in these cases chosen randomly.     

Laisser un commentaire

Votre adresse e-mail ne sera pas publiée. Les champs obligatoires sont indiqués avec *

Ce site utilise Akismet pour réduire les indésirables. En savoir plus sur comment les données de vos commentaires sont utilisées.