Nos premiers traitements de niveau 2A avec Sentinel-2

2015/07/06

=>

 

Maintenant que Sentinel-2A a commencé à acquérir des séries temporelles d'images sur l'Europe, nous avons pu tester notre chaîne de niveau 2A, MACCS, pour faire des corrections atmosphériques. Comme vous le savez probablement si vous avez suivi ce blog, MACCS travaille avec des séries temporelles d'images, pour détecter les nuages,les ombres, l'eau et pour estimer l'épaisseur optique des aérosols. Cette chaîne a donc besoin de quelques images pour s'initialiser correctement.

 

2015/07/16

2015/07/26

Beatrice Petrucci, du CNES, a lancé le premier traitement avec une série temporelle d'images de niveau 1C acquise par Sentinel-2A sur les Pyrénées au cours du mois de juillet. Les données N1C ont été produites au CNES sur le Ground Processing Prototype (GPP), qui a servi à mettre au point et valider les traitements et est utilisé maintenant pour produire les données nécessaire à la "recette en vol" du satellite.

 

Les résultats obtenus sont déjà chouettes, avec une très bonne détection des nuages et de leurs ombres, comme vous pouvez le voir sur les quicklooks ci-joints. Sur ces images, les nuages détectés sont entourés en rouge, les ombres en noir et l'eau en bleu. N'hésitez pas à cliquer sur les images pour les regarder à 100m de résolution (les fournir à 10m aurait accru les temps de transfert...). Les lecteurs observateurs pourraient se demander pourquoi, sur le Nord de l'image du 26/7, certains nuages sont entourés en noir : c'est un artefact de l'outil de calcul des quicklooks, qui fait disparaître le contour rouge quand un contour noir se superpose exactement.

 

La correction atmosphérique a l'air correcte, mais il faudra valider cela plus scientifiquement, quand nous aurons traité un plus grand nombre de données.

 

Ces bons résultats ont été obtenus avec un jeu de paramètres issus de notre expérience avec d'autres capteurs. Ce sont donc les paramètres adaptés à LANDSAT 8 qui ont été utilisés. On peut s'attendre à quelques améliorations encore quand nous aurons réglés ces paramètres avec un plus grand nombre d'images Sentinel-2.

Our First Sentinel-2 Level 2A processing

2015/07/06

=> 

Now that Sentinel-2A has started to acquire time series over Europe, we have been able to try our MACCS Level 2A processor to perform atmospheric corrections. As you probably know if you have been following this blog, MACCS works with time series to detect clouds, shadows, water and estimate aerosols optical thickness. Our processor therefore requires at least a couple of images to get a good initialization, .

 

2015/07/16

2015/07/26

Beatrice Petrucci, from CNES, launched the first processing with a time series of L1C images obtained near the Pyrenees. The L1C data were produced at CNES with the Ground Processing Prototype, which is used to define and validate the processing methods, and also produces all the data needed for the commissioning phase.

 

The results obtained on these L2A data are just great, with a very good detection of clouds and shadows, as you may see on the attached quicklooks. On the joined images, the detected clouds are circled in red, while the shadows are circled in black and water is in blue. You will need to click on the images to see them at 100 m resolution (we had to reduce it to let you download them quickly). You might wonder why on the top of the image of 26/7 some clouds are circled in black and not in red. There is no error in MACCS mask detection, but when cloud mask and shadow mask overlap exactly, the red contour is hidden by the black contour.

 

The atmospheric correction also looks fine, but of course needs to be validated scientifically.

 

These excellent results have been found with a first set of parameters coarsely tuned on the basis of our experience with other sensors; we expect to have even better results when the fine tuning of the parameters will be completed using much more images...

Landsat Science Team meeting slides

Twice a year, there is a  Landsat Science Team meeting, organised by USGS. One day is dedicated to LANDSAT news, and one or two days to the science team presentations. And what is good, is that the slides are made available just after the meeting. It is a good way to get news from LANDSAT, and even from Sentinel-2, and to see how LANDSAT data are used in the USA mainly, but also around the world.

 

Here is the address : http://landsat.usgs.gov/science_LST_Team_Meetings.php

 

The question is : will there be a Sentinel-2 Science Team ? There is a Sentinel-2 Mission Advisory Group, but the carefully selected members (I am one of them...) were only reunited twice during Sentinel-2 development phase, and the slides are not available. Even if there were also two Sentinel-2 symposiums (and two SPOT (Take5) workshops, with the slides here and there), the level of discussions and feedback with the project team is not the same as in a science team with a reduced size.

 

Please remember that the next SPOT (Take5) workshop will be held within the Living Planet Symposium in Prague, next May.

 

Number of LANDSAT scenes processed and delivered by USGS. More than 700 000 Landsat scenes were distributed in March 2015, which makes quite a difference with the period when LANDSAT data were not free, until 2008 (source USGS)

Feedback on Sentinel-2A first image

Sentinel-2A first image

=> 

Several colleagues asked me about the apparent defect on the maritime part of Sentinel-2 first image.  I would like to reassure you, it is not a defect, only a directional effect.

 

Let's explain it :
- For each Sentinel-2 spectral band, 30 000 elementary detectors are necessary to obtain a field of view of 300km with a resolution of 10m. But the current technology does not allow to provide 30000 detectors in a row. What is used are 12 linear arrays of detectors, which are put together to cover the whole field of view. But as each linear array is surrounded by an edge, it is not possible to stitch the pieces together with no space between the linear arrays. It was decided to shift the 10 pieces within the focal plane, as in the drawing below. The odd chips are looking forward, and the even chips are looking backward, with angle differences that may reach 3 degrees for some bands.

 

Sentinel-2 VNIR focal plane

 

On Sentinel-2A first image, we observe the see surface with the specular reflection of the sun on the sea surface (also called sunglint).  Such a phenomenon tends to change quickly as a function of the viewing angle, as may be seen on the picture below. Bertrand Fougnie, at CNES, computed the amplitude of this effect as a function of the detector, for different wind speeds. It is provided on the plot below, on which you can see the line breaks between the odd and even detectors. The plot is provided for different wind speeds. When the wind is low, the sunglint makes a small spot with a large variation with angle, which increases the effect.

 

The "defect" we observe is therefore only linked to a difference of observation angle. This phenomenon is not at all visible above lands, because there, the directional effects are much lower except in some special cases (the "hot spot"), which should be avoided by Sentinel-2..

 

A few concerned users asked if that would prevent the use of data above ocean, but I do not think so. If one is able to remove the sunglint effect on the data, then one is able to account for the angle variations.

Retour sur la première image de Sentinel-2

=>

Vous êtes plusieurs à m'avoir parlé du défaut apparent, visible sur la partie maritime de la première image de Sentinel-2A. Je voudrais vous rassurer, ce n'est pas un défaut, mais un effet directionnel.

 

Je vous explique :
- pour chaque bande spectrale de Sentinel-2, il faut 30000 détecteurs pour couvrir le champ de 300 km à 10m de résolution. Ces 30000 détecteurs ne peuvent pas être construit en une seule barrette. Ce sont 10 ensembles de barrettes qui sont assemblés pour constituer la ligne de 30000 détecteurs. Pour des raisons d'encombrement, il n'est pas possible de juxtaposer ces barrettes, qui doivent être décalées dans le champ. Les barrettes paires visent légèrement vers l'avant, les barrettes impaires vers l'arrière, avec des différences d'angle de vue pouvant atteindre 3 dégrés dans certaines bandes.

 

Plan focal de Sentinel-2

 

Or sur la première image de Sentinel-2, nous observons sur la mer la réflexion spéculaire du soleil sur la mer, qui est une cible qui varie très rapidement en fonction de l'angle de vue, comme on peut le voir sur la photo-ci dessous. Bertrand Fougnie, du CNES, a calculé la réflectance observée au dessus de la tache de réflexion spéculaire, dans les conditions géométriques de la première image Sentinel-2. On observe bien d'assez fortes différences entre les détecteurs pairs et impairs. Ces différences dépendent de la vitesse du vent. Quand le vent est faible, la tache spéculaire couvre une plus petite surface et ses variations angulaires sont plus fortes.

 

Le "défaut" que nous observons est donc tout simplement lié à une différence d'angle d'observation. Il n'est d'ailleurs pas visible au dessus des terres, car les variations directionnelles des surfaces terrestres sont beaucoup plus lentes, excepté pour certaines conditions très particulières, qui ne devraient pas être visibles avec Sentinel-2.

 

Certains m'ont dit que celà pourrait donc poser des problèmes pour l'utilisation des données au dessus de l'eau. Je ne le pense pas, parce que si l'on veut pouvoir utiliser ces données sur l'eau, en présence de réflexion spéculaire, il faut être capable de corriger ce phénomène, et on peut donc tout à fait prendre en compte les différences d'angle.

Champagne et gourmandises (à 5h00 AM)

C'est fait, Sentinel-2A est en orbite. Il a commencé à dialoguer avec ses stations de contrôle et a déployé ses panneaux solaires. Il faudra quelques jours pour vérifier et configurer la plate forme avant d'allumer l'instrument et de réaliser les premières acquisitions.

 

Félicitations aux équipes de l'ESA, d'Airbus et des industriels associés à la fabrication du satellite, et à celles d'Arianespace et du CNES à Kourou pour ce lancement. A nous de nous préparer à l'utilisation des données, notamment en utilisant les séries temporelles de SPOT (Take5) !

 

Its done, Sentinel-2A is in orbit. It already started its dialog with the control stations, and the solar panels were deployed successfully. A few days will be needed to check the bus before opening the instrument to take the first pictures.

 

Congratulations to ESA, Airbus and all the teams that built the satellite, and to Arianespace and CNES in Kourou for this launch. It is now our turn to finish our preparation for using the data, for instance using SPOT (Take5) time series !

Sentinel-2 is launched tonight
Sentinel-2 est lancé ce soir

Patience is not my best quality. For instance, regarding the availability of Sentinel-2 data,  I even had the orbit of two satellites changed to obtain time series quicker :) . So you cannot imagine how happy I will be to watch Sentinel-2's launch  tonight, and how even my alarm at 2:30 am will sound as a nice music. Here are two nice links from ESA's Website, and from THEIA's :

 

 

La patience n'est pas mon point fort. Par exemple, plutôt que d' attendre la disponibilité des données Sentinel-2, j'ai fait détourner deux satellites de leur orbite ;) . Aussi vous ne pouvez pas imaginer comme je serai heureux, ce soir, d'assister au lancement de Sentinel-2A. Même la sonnerie stridente de mon réveil à 2h30 la nuit prochaine me fera l'effet d'une douce musique. Si vous voulez participer vous aussi, voici quelques liens issus des sites de l'ESA et de THEIA :

 

 

First Level2 products from SPOT5 (Take5)

 

 

=>

As you already know if you have been following this blog, our L2A method for cloud detection and atmospheric correction, named MACCS, makes a large use of multi-temporal information. It assumes that over lands, the surface reflectances in the blue do not change quickly with time, except on some exceptions that have to be detected. Based on this, a sudden and large increase in surface reflectance will be detected as a cloud, a more subtle change will be detected as a variation of Aerosol Optical Thickness (AOT). This is the basis of our MACCS method, which of course is a little more complex than that.

 

So, to do that, we need to process time series of data, and we need several dates to get a nice convergence of results. That's why we had not shown any level 2 product from SPOT5 (Take5) yet. Now that 3 cycles of data have been processed, we have processed our first level 2A time series, selecting only regions with a nice weather. The processor worked well nearly at once (except for a "something.dat" file which had to be renamed "SOMETHING.DAT").

 

 

Lake Balaton Hungary

 

Pornic, France

 

The results are good (see the images of Lake Balaton, Hungary and Pornic, France, above). If you click on the images, you will see that the very thin plane contrails, circled in green, have been detected. The plane contrail shadows (circled in balck) were missed, but it is very difficult to detect such a faint signal. On the Pornic image the very thin cloud is well detected as well as its shadow. In both cases. A very thin cloud is missed in the south east corner, because this region was only seen on one image, and therefore the multi-temporal detection does not work. Not only the cloud detection works, but the atmospheric correction also looks fine.

So this is good news, we should be able to start operational production soon !

On Lake Balaton site, It is also amazing, to see how the vegetation changed in 10 days : on the first image, most of the forests were dark with no leaves, while they become bright... red on the second image. In the Pornic images, you may see two dark zones corresponding to two wetlands, and we should see the vegetation grow when the water retreats during summer. The forests are much more advanced than in Hungary, and highlight the difference between an Atlantic and a continental climate, and all this shows that Sentinel-2 will be a great tool to study vegetation phenology at a high resolution.

 

Good observers will also notice that the footprints of the SPOT5 images may change from one image to the other. This is the main issue we have had to deal with so far, but we think we have found a solution, and the footprints should be more stable from now on.