Monthly cloud free syntheses merging Sentinel-2 and Landsat 8

To compute a cloud free synthesis of surface reflectances every month, a good repetitivity of observations is necessary. The weighted average method we developed at CESBIO, and which will be part of ESA's sen2agri system was coded by Cosmin Udroiu at CS Romania. It was meant to work with both Sentinel-2 sensors and an observation every fifth day. As we are still waiting for the launch of Sentinel-2B, the monthly syntheses obtained with Sentinel-2A alone really lack cloud free data.


On the left, the Sentinel-2A monthly synthesis, above Odessa (Ukraine) in May, and on the right its flag, with, in black, the areas flagged as cloud or cloud shadow. When a pixel is flagged as cloud or cloud shadow, the monthly synthesis provides the minimum blue reflectance, which tends to avoid clouds (if possible), but often selects cloud shadows.


Fortunately, the Sen2agri L3A processor is designed to work with LANDSAT 8 too, as both satellites have similar spectral bands, and as the MACCS atmospheric correction used to produce the L2A input products works for both sensors. Of course LANDSAT 8 geometric resolution is not that of Sentinel-2, so to avoid degrading Sentinel-2 imagery when LANDSAT8 data are available, we give Landsat 8 a very low weight in the weighted average. As a result, Landsat is really taken into account only when no cloud free Sentinel-2 was available during the synthesis period.


Same result as above, but including LANDSAT 8 data. A cloud free date at least is now found for every pixel. The water mask obtained from Level 2A product is a little wrong on the Landsat 8 image due to the presence of turbid waters and thin clouds. A solution for this problem will be implemented in next MACCS L2A version. Note that the monthly synthesis of both Sentinel-2 and LANDSAT-8 leaves nearly no visible artifacts on the lands.

For a better comparison of both versions, here is a little animation of composites with and without Landsat 8.

The Sen2Agri system is still in validation phase and should be released as open source next May, 6 months from now. The L3A synthesis processor will be also implemented within Theia and monthly L3A products will be distributed by Theia as it is already the case for L2A products.

Sentinel-2 MACCS L2A reflectance stability

As MACCS was used to produce a large amount of images (all the images acquired over France), we were able to make some more validation tests, such as checking the surface reflectance stability. To do that, the region of Arles in Provence (France), tile 31TFJ, is a good site, with a lot of clear skies, and the chance to be observed from two different orbits by Sentinel-2.


Here is an example for a coniferous forest (Pines probably): The reflectances of the 4 Sentinel-2 10m spectral bands are stable indeed, as it may be seen on the top plot, while the bottom plot shows the variation of the aerosol optical thickness measured by MACCS. You may have noticed that towards the summer, we can see a few couples of points, acquired 3 days apart, with small differences in surface reflectance, noticeable in the NIR infrared (black dots). These differences are probably due to directional effects as the observation from two different orbits are made with different viewing angles. These effects may be corrected. We do that for instance in our level 3 products.

Top : surface reflectances corrected with MACCS as a function of time (blue : B2 (blue), green B3 (green), red B4 (red), black B8 (Near infrared)). Bottom, Aerosol Optical Thickness measured by MACCS.

The overall stability of surface reflectances is clearly a good sign of the quality of MACCS processing.


First Sentinel-2 monthly cloud free syntheses from the Sen2Agri system



The Sen2Agri System, funded by ESA, is now on its pre-operational phase. It has just completed its first mass production and in this context, we just obtained our first Level 3A syntheses obtained with Sentinel-2A. Level 3A products are monthly syntheses of non cloudy pixels.


The monthly syntheses are produced by the Sen2Agri system. They use as input 2A level products processed by the MACCS processor, which provides surface reflectance along with cloud masks and shadow masks and snow and water masks. As their name suggests, the syntheses are produced once a month, but can be based on a bit longer time periods to increase chances to get cloud free observations.

Continue reading

Les premières synthèses mensuelles Sentinel-2 issues du système Sen2Agri



Le système Sen2Agri, financé par l'ESA, est en phase de validation pré-opérationnelle. Il vient donc de réaliser ses premières productions de masse. Dans ce cadre, nous venons d'obtenir, avec Sentinel-2A, nos premières synthèses de Niveau 3A, c'est à dire des synthèses mensuelles des réflectances de surface des pixels non nuageux.


Les synthèses mensuelles sont produites par le système Sen2Agri. Elles utilisent en entrée des produits de niveau 2A traités par la chaîne MACCS, qui fournit des réflectances de surface accompagnées de masques de nuages et d'ombres et masques de neige et d'eau. Comme leur nom l'indique, les syntheses sont produites une fois par mois, mais peuvent se baser sur des périodes de temps un peu plus longues pour accroître les chances d'avoir des observations claires.

Continue reading

Some news about Sentinel-2 from Living Planet Symposium


If I haven't posted for more than a week, it is because I have been participating to the Living Planet Symposium 2016 in Prague. With the preparation, travel, and participation to this crazy symposium. I say crazy because it is packed with about 3000 people, of which I know only about 300. When you want to go from a room to another, it takes as much as 20 minutes as you meet at least 3 or 4 of your colleagues and have a chat. But I do not need to tell you, as very likely, you were there too !

One of the 6 to 8 rooms, packed with people at the same time


The good news is that I have had access to a lot of information. I will start by some news about Sentinel-2 of course. Some of these news are not good.

  • The launch of Sentinel-2B has been postponed by to 2017, probably between March and June, due to a delay with the Rockot launcher. This is very annoying as accounting with the time needed for the commissioning phase, it means we will only rely on a 10 days repetitivity during modt of 2017.
  • The availability of Sentinel-2 ortho-rectified data with ground control points has also been postponed to the end of 2016, while initially it was planned in June. ESA says it is due to the fact that the global reference images (GRI) are not ready yet worldwide. We thus will have to cope with registration errors of about 1 pixel within the same orbit and 2 pixels when comparing data from 2 orbits. ESA had announced last year that they would introduce the reference data per continents starting by Europe, but it seems they changed their minds. Still I was told that the GRI for Europe and Australia are available or will be very soon, so why not starting a prodution of ortho-rectified data on those continents ?
  • ESA is also going to change and shorten the very long names of their products and start to distribute data tile wise. Of course, this is good news as the choices made before were not convenient, and it is better correcting it now, as the data backlog is short, but it means everyone will have to change his software. This might delay several productions downstream ESA ground segment.
  • ESA has published a new version of Sen2cor, which is said to seriously enhance the scene classification which was really bad in the previous version, especially for its cloud and cloud shadows mask. I will test it of course, as soon as I find some time.


This image background image is a monthly synthesis of Sentinel-2 images of august 2016, covering the whole Czech Republic. It was processed by MACCS to level 2A  and then to level 3A by the synthesis method we developed at CESBIO and implemented within Sen2Agri package. The overlayed landcover map itself was generated by GISAT in the Czech Republic based mainly on Sentinel-1 data, as Sentinel-2 data last summer were still quite scarce. This poster was shown on the 10x10m advert on the congress centre, alternating with a nice bikini (sorry, I only have the Sentinel image !)


I have been very positively surprised by how our user community has started using the time series, instead of using images only. It is clear we have entered a new chapter of remote sensing history on the application side, with much more robust results. The case studies based on one image have completely disappeared on the presentations, even if they are still present on the posters. Sentinel-1 examples were impressive and joint uses of both Sentinels 1 & 2 are rising. As the recently launched Sentinel-3A seems to be working well, it is clear ESA has set up a great system Europe can be proud of. And on top of that, ESA really know how to organise a symposium !


Slides from the second SPOT (Take 5) Workshop

Voici les présentations des Journées SPOT4 (Take5) du 18 et 19 novembre 2014.  Cette réunion a réuni une centaine de personnes au CNES pour faire un bilan des acquisitions et des produits distribués, prendre connaissance des travaux et résultats des utilisateurs et obtenir un retour d'expérience.
J'en profite pour remercier chaleureusement les deux organisatrices, Sylvia Sylvander et Danielle Barrère, du CNES (DCT/ME/OT), ainsi que les 23 orateurs qui nous ont proposé de brillantes présentations.


Here are the presentations from SPOT4 (Take5) users day on November 18th and 19th 2014. Around 100 people attended this meeting at CNES, which aimed at making an assessment of the acquisitions and products, show the users studies and results and obtain their feedback.
I would like to thank a lot the organisers, Sylvia Sylvander and Danielle Barrère from CNES, and the 23 speakers, who gave us amazing talks.

The table below gives access to all the slides.


S. Sylvander CNES Introduction
S. Sylvander CNES Preparation and progress of SPOT4 (Take5) experiment
O.Hagolle CESBIO/CNES SPOT4 (Take5) validation, users feedback and lessons learned
M.Kadiri CESBIO Definition, test evaluation of a monthly composite product for Sentinel-2 based on SPOT4 (Take5) data
M.Claverie U. Maryland / NASA An update on the LANDSAT8/Sentinel-2 merged reflectance product project
M.Le Page CESBIO SAT-IRR : Satellite for irrigation scheduling
V.Simonneaux CESBIO Water budget monitoring of irrigated perimeters in semi-arid areas using high resolution NDVI image time series
O.Arino ESA ESA Studies: Agriculture, Forest, Wetland, Costal Water. Preparing for Sentinel-2
M.El-Hajj TETIS/IRSTEA Estimation of soil moisture using radar and optical images over Grassland areas
J.Inglada CESBIO Automatic land-cover map production of
agricultural areas using supervised classification of SPOT4(Take5) and Landsat-8 image time series
S.Valero CESBIO Real-time crop mask production using high-spatial-temporal resolution image times series
D.Morin CESBIO Estimation of biophysical variables and cartography of irrigated surfaces with high temporal and spatial resolution images
W.Li INRA Avignon Deriving ECVs GAI and FAPAR from SPOT4 and LANDSAT8 sensors: evaluation of the consistency and comparison with ground measurements
A.Roumiguié Dynafor/EI-Purpan Forage production monitoring
C.Jacqueminet EVS-ISTHME, U. Saint-Etienne
Discrimination of herbaceous habitats using multitemporal Spot and Landsat images (Massif central – France)
V.Gond CIRAD Phenological monitoring of tropical forest ecosystems in the North of Congo
A.Verhegghen JRC Assessing Forest Degradation in the tropics using Time Series of Fine Spatial Resolution Imagery
M.Szulkin CEFE-CNRS Inferring blue tit reproductive phenology using SPOT4 imagery
S.Sylvander / O.Arino ESA SPOT5(Take5) Operations in 2015
J.P.Dedieu LTHE / CNRS Snow cover monitoring in the French Alps physical properties of surface snow, snow cover dynamics impact on vegetation
A.Facello IRPI-CNR (Italy) Snow Water Equivalent and Slope Movements from Satellite Data: potential of space-borne observations with high spatial and temporal sampling. Case study: Tena Valley (Central Pyrenees, Spain)
S.Cerisier GIP Loire-Estuaire Remote sensing data with high repetitivity : a contribution to coastal and estuarine processes knowledge
V.Lafon Geo-Transfert / EPOC, U. Bordeaux
SPOT4 (Take5) Experiment: simulation of Sentinel-2 time-series to monitor the maximum turbidity zone of tidal estuaries

Directional effect correction for Sentinel-2 composites.


Sentinel-2 orbits

Swaths observed by Sentinel-2A, for day 1 (green), 4 (Blue), 7 (grey), 8 (Pink). For Sentinel-2B, we will have to shift that by 5 days. Distance between swaths was computed so that a little overlap is available at the equator.

The Sentinel-2 orbit was set so that the swaths observed by the satellite have a little overlap at the equator. The width of the overlap increases quickly at higher latitudes. For instance, at France latitude (45 degrees), about half of the surface will be observed twice per satellite cycle, from two adjacent swaths.

It is not very fair, since it will always be the same places that will be observed twice and the rest of the world will only be observed once (The Cesbio site is well located !)

Same as above, with a zoom over France. Here, half of the land will be observed twice per cycle (red segments), and the other half (yellow segments) once per cycle.




Directional correction for composites.


Well, the issue is that each point within the overlap zone will be observed twice, but under two different viewing angles, and therefore will have different reflectances in each swath, due to the directional effects. The users of our data often ask for monthly syntheses as cloud free as possible, that merge the data observed from different orbits.To obtain such products, a directional correction is therefore necessary.

Monthly syntheses in Toulouse region, without directional correction on the left,  with a directional correction on the right.

N.B.. The scattered green points you may see are invalid points due to saturated pixels (saturations are often observed with SPOT? which will not be the case for Sentinel-2).


To do that, directional models have been developed, such as the ones of Roujean or Ross-Li, that model the directional variations as a function of viewing angles and solar angles, with a rather good accuracy for most types of surfaces. Here is how they look like :

 \rho= \rho_0 (1 + K_1. F_1(angles), + K_2. F_2 (angles))


 \rho is the reflectance for the actual viewing and solar angles  \rho_0 is the reflectance for a given angular condition chosen to standardise the data (for instance viewing at nadir and solar angle at 45 degrees), F1 and F2 are the directional functions that depend on the angles, and  K_1 and K_2 are the coefficients of the directional model, that depend on the observe pixel type of surface.

Fortunately, in the case of S2, the angle differences are low, no more than 20 degrees. We have tried, as a first test; to find mean coefficient that could work more or less for all surfaces. Tu compute these coefficients, we used the SPOT4 (Take5) sites which have been observed under two viewing directions. These are Maricopa (In the USA), and Midi-Pyrénées, Bretagne and Provence in France. They show very different landscapes, with desert and irrigated crops in Maricopa, a very diverse agricultural landscape in Bretagne and Midi-Pyrénées, and Mediterranean forests and vineyards in Provence. We have used all the available couples of clear images separated by less than 5 days and we searched for the coefficients K_1 and K_2 that allow to minimise differences.


Finally, these coefficients were used to correct the data and produce composites. The monthly syntheses are finally obtained by computing  a weighted mean value of the reflectance of cloud free pixels obtained during a period of 42 days. The images above or below show the results obtained by M. Kadiri on the French sites (Maricopa is still running), with on the left the synthesis without directional correction, and on the right the one with directional correction. The shading observed from right to left on the image without correction almost disappears on the images with correction. It is the same for all 3 sites and the chosen images are the ones which show the highest differences. Knowing that the angle difference is greater for SPOT4 (Take5) than for Sentinel-2, we have good hopes that this simple method could work for Sentinel-2.

However, our sampling of 4 sites is not sufficient, we will have to prove that theses results still hold for other types of surfaces. We could do that with SPOT5 (Take5) or with the first Sentinel-2 data (which should come soon !).


Same as above, for Provence-Languedoc.


Same as above, for Bretagne

Fauchée d'un instrument : c'est la surface observée par un satellite au cours d'un passage.

Correction des effets directionnels pour les synthèses mensuelles de Sentinel-2


Les orbites de Sentinel-2

Fauchées observées par Sentinel-2A, le jour 1 du cycle (vert), le 4 (Bleu), le 7 (gris), le 8 (rose). Pour Sentinel-2B, il faut décaler le tout de 5 jours. L'espacement entre les fauchées est déterminée par l'orbite, qui est calculé pour permettre un léger chevauchement des fauchées adjacentes à l'équateur.

L'orbite de Sentinel-2 a été calculée pour que les fauchées observées par le satellite aient une petite intersection à l'équateur. La largeur de cette intersection augmente rapidement lorsqu'on s'éloigne de l'équateur et qu'on se rapproche des pôles. A la latitude de la France (45 degrés), c'est quasiment la moitié des surfaces qui pourront être observées à deux reprises, à partir de deux fauchées adjacentes.

C'est d'ailleurs un peu injuste car ce seront toujours les mêmes endroits qui seront observés deux fois tous les 5 jours, alors que d'autres endroits ne seront observés qu'une fois, mais à la verticale. Le site Sudmipy du CESBIO semble faire partie des endroits observés deux fois, mais, je ne suis pas sûr de disposer des orbites définitives de Sentinel-2).

Zoom sur la France de l'image ci-contre. On constate qu'à la latitude de 45 degrés, la moitié des terres (trait jaune) est observée une fois par cycle, et l'autre moitié (trait rouge) deux fois par cycle (donc deux fois tous les 5 jours avec les deux satellites).




La correction directionnelle.


Bref, le problème, c'est qu'un point à l'intersection de deux fauchées adjacentes sera observé sous deux angles différents et n'aura pas les mêmes réflectances sur les deux images, en raison des effets directionnels. Or les utilisateurs de nos données (oui, vous) nous demandent souvent des images de synthèses mensuelles (si possible sans nuages), et assemblant les données acquises depuis plusieurs orbites, de préférence sans que les coutures entre orbites soient visibles. Pour obtenir de tels produits, il faut donc pratiquer une correction des effets directionnels.

Synthèses mensuelles calculées avec la méthode de la moyenne pondérée, sans correction directionnelle à gauche, avec correction directionnelle à droite.

N.B.. Les points verts que l'on voit par-ci par là sont des pixels invalides, car tout le temps nuageux ou saturés (sur SPOT, les saturations sont fréquentes, ce qui ne sera pas le cas sur Sentinel-2)


Pour cela, il existe des modèles directionnels, comme celui de Roujean, ou ceux de Ross-Li, qui permettent de modéliser l'évolution des réflectances en fonction des angles de prise de vue et des angles solaires, avec une précision correcte pour la plupart des surfaces. Ils se présentent sous la forme suivante :

 \rho= \rho_0 (1 + K_1. F_1(angles), + K_2. F_2 (angles))


 \rho est la réflectance dans les conditions de la prise de vue,  \rho_0 est la réflectance pour une direction donnée (par exemple, observation à la verticale et élévation solaire à 45 degrés), F1 et F2 sont des fonctions directionnelles qui dépendent des angles de prise de vue et des angles solaires, et  K_1 et K_2 sont les coefficients du modèle directionnel, qui vont en général dépendre de la nature du pixel observé.


Dans le cas de Sentinel-2, nous avons la chance que les différences d'angles de prise de vue entre deux orbites adjacentes soient faibles, tout au plus 20 degrés. Nous avons donc tenté de trouver des coefficients moyens qui fonctionneraient à peu près pour tous les paysages. Pour trouver ces coefficients, nous avons utilisé les sites de l'expérience SPOT4 (Take5) qui ont été observés sous deux angles différents. Il s'agit de Maricopa (aux USA), Midi-Pyrénées, Bretagne et Provence en France. Il s'agit de paysages très différents, avec du désert et de l'agriculture irriguée à Maricopa, un paysage agricole varié en Bretagne et en Midi-Pyrénées, et un paysage de forêts méditerranéennes et de vignes en Provence. Nous avons utilisé tous les couples d'images claires séparées par moins de 5 jours et cherché les coefficients  K_1 et K_2 qui permettent de minimiser les différences.


Enfin, ces coefficients ont été utilisés pour corriger les données et produire les composites. Les produits de synthèses mensuelles, sont finalement obtenus en calculant la moyenne des pixels non nuageux pendant une période de 42 jours. Les images ci-dessus présentent les derniers résultats obtenus par Mohamed Kadiri sur le site de CESBIO près de Toulouse. à gauche, sans correction directionnelle, à droite avec correction directionnelle. Le dégradé de couleurs qui apparaît sur la partie droite de l'image de gauche, sans correction, disparaît presque complètement sur l'image de droite (avec correction). Il en va de même pour toutes les dates et pour les 3 autres sites, et j'ai choisi ici l'image qui comportait les effets les plus prononcés. Sachant que la différence angulaire entre les images SPOT acquises depuis des orbites adjacentes est plus grande que pour Sentinel-2, je pense qu'on peut espérer obtenir de bons résultats avec Sentinel-2 avec cette méthode simple.

Ceci dit, notre échantillon statistique, composé de 4 sites est largement insuffisant, ces résultats devront donc être confirmés, par exemple avec SPOT5 (Take5), ou avec les premières données de Sentinel-2 (c'est bientôt !)


Même figure que ci-dessus pour le site Provence-Languedoc.


Même figure que ci-dessus pour le site Bretagne

Fauchée d'un instrument : c'est la surface observée par un satellite au cours d'un passage.

Comparison of Level 3A compositing methods


As said in a previous post, we are testing various methods of level 3A production, using SPOT4 (Take 5). The Theia Land Data Center will the use these methods to process Sentinel 2 data. In case you did not click on the link above, let's recall that the level 3A products are monthly composite products of cloud free reflectances. For each pixel, our method computes the weighted average of the reflectances of the dates when the pixel is cloud free. For more details, you will need to follow this link.


The work of Mohamed Kadiri at CESBIO, which is funded by the CNES budget for Theia, adressed first the definition of quality indexes for composite products (for more details, may I suggest that you follow this link ?). This work showed that our product has nice performance, but we knew some one would ask us to compare them to the classical methods for level 3A products.


Therefore, we compared our product with the famous NDVI Maximum Value Composite (NDVI MVC), developped by our remote sensing ancestors, and used since the most remote antiquity to process AVHRR time series. This method consists in using for each pixel of the level 3A, the reflectances of the date which has the greatest NDVI.  Why ? Mostly because the NDVI of a cloud is very low, often negative, and therefore this method will rather select cloud free pixels. The NDVI MVC comes from a time when the cloud masks were not very accurate.


Example of a monthly synthesis obtained with the NDVI MVC methods Example of a monthly synthesis obtained with the weighted average method

This post uses the SPOT4-Take5 data to show a comparison of the performances obtained on the Versailles site, with the NDVI MVC method on the left, and the weighted average on the right. One can clearly see, on the left, the presence artefacts made of whiter and darker dots which are not seen on the image on the right. These artefacts appear when the selected date changes from one pixel to the other. These artefacts are much less visible on the vegetation covered plots, as, for this composite obtained in spring, the vegetation increases quickly, and all the pixels come from the last cloud free date of the synthesis.


If we have a look at our quality indicators, which were described in our previous post about composite products , it is obvious that the performances obtained by the weighted average method are much better than those of the NDVI MVC method, either as regards the similarity to the central date image of the Level 3A (in yellow, for the 70 % best pixels and in green for the 95% best pixels), and moreover as regards as the amplitude of artefacts (in blue). The abscissa of the plot is the half of the number of days used in the synthesis, and our recommended value is 21.



NDVI Maximum Value Composite Weighted Average Composite

Produits de Niveau 3A : comparaison avec la méthode classique du maximum de NDVI


Comme nous le disions dans un précédent article, nous utilisons les données SPOT4 (Take 5) pour tester différentes méthodes de création de produits de niveau 3A. Le pôle Thématique appliquera par la suite ces méthodes aux données Sentinel-2.   Pour rappel, si vous n'avez pas cliqué sur le lien ci-dessus, les produits de niveau 3A sont des synthèses périodiques (a priori mensuelles) de réflectances de surface sans nuages. Notre méthode est basée sur des moyennes pondérées de réflectances des pixels non nuageux obtenus pendant une certaine période de temps. Pour plus de détails, il faudra vraiment aller voir ce lien...


Les travaux de Mohamed Kadiri au CESBIO, financés par le budget CNES de THEIA, ont d'abord porté sur la mise au point d'indices de qualité (décrits dans le lien ... bon, j'arrête), et ont permis de montrer que notre méthode produit de bonnes performances. Nous avons voulu comparer ces résultats avec la méthode du "Maximum de NDVI", développée par nos ancêtres télédétecteurs, et appliquée depuis des temps immémoriaux aux données de moyenne résolution comme celles des instruments AVHRR. Cette méthode consiste, pour chaque pixel, à utiliser dans le produit de Niveau 3A, la date dont le NDVI est le plus grand. Pourquoi ce choix ? Principalement parce que le NDVI d'un nuage est très faible, et donc que cette méthode permettra de choisir préférentiellement les pixels non nuageux. Cette méthode date d'un temps où les masques de nuages n'étaient pas très précis.


Extrait de synthèse mensuelle obtenue avec la méthode du maximum de NDVI Extrait de synthèse mensuelle obtenue avec la méthode de la moyenne pondérée.

Nous vous présentons ici, avec les données SPOT4-(Take5), une comparaison des résultats obtenus sur le site de Versailles, avec  la méthode du maximum de NDVI à gauche et la méthode de moyenne pondérée à droite.  On note, sur l'image de gauche, la présence de nombreux artefacts sous la forme de points brillants ou sombres que l'on ne voit pas sur l'image de droite. Ces points de niveau différents sont dûs au fait que d'un pixel à l'autre, une date différente a été utilisée, en fonction de la valeur du maximum de NDVI. Ces artefacts sont moins présents sur les zones couvertes de végétation (en rouge), car pour cette synthèse obtenue au printemps, la croissance de la végétation fait que le maximum de NDVI correspond à la date la plus tardive de la synthèse..


Si l'on regarde les valeurs de nos critères de qualité, décrits dans l'article précédent (cela faisait longtemps ;) ), on note que les performances de la méthode par moyenne pondérée sont bien meilleures que celles de la méthode du maximum de NDVI, vis à vis de la fidélité à l'image de la date centrale de la synthèse mensuelle (en Jaune, pour les 70% de pixels les meilleurs, et Vert, pour les 95% de pixels les meilleurs), et surtout, vis à vis de la présence ou non d'artefacts (en bleu). L'abscisse des courbes correspond à la moitié du nombre de jours utilisés pour chaque synthèse, nous recommandons la valeur 21.


Maximum de NDVI Moyenne Pondérée