Another validation of CESBIO's 2016 France land-cover map

In this post, a validation of the land-cover map of France produced by CESBIO for the 2016 period was presented. This validation used independent data (that is data collected by different teams and using different procedures than the data used for the classifier training), but the validation procedure consisted in applying classical machine learning metrics which, as described in this other post, have some limitations.

A fully independent validation following a sound protocol is costly and needs skills and expertise that are very specific. SIRS is a company which is specialised in the production of geographic data from satellite or aerial images. Among other things, they are the producers of Corine Land Cover for France and they are also responsible for quality control and validation of other Copernicus Land products.

SIRS has recently performed a validation of the 2016 France land-cover map. The executive summary of the report reads as follows:

This report provides the evaluation results of the CESBIO OSO 2016 10m layer and the CESBIO OSO 2016 20m layer.

The thematic accuracy assessment was conducted in a two-stage process:

  1. An initial blind interpretation in which the validation team did not have knowledge of the product’s thematic classes.
  2. A plausibility analysis was performed on all sample units in disagreement with the production data to consider the following cases:
  • Uncertain code, both producer and operator codes are plausible. Final validation code used is producer code.
  • Error from first validation interpretation. Final validation used is producer code
  • Error from producer. Final validation code used is from first validation interpretation
  • Producer and operator are both wrong. Final Validation code used is a new code from this second interpretation.

Resulting to this two-stage approach, it should be noticed that the plausibility analysis exhibit better results than the blind analysis.

The thematic accuracy assessment was carried out over 1,428 sample units covering France and Corsica.
The final results show that the CESBIO OSO product meet the usually accepted thematic validation requirement, i.e. 85 % in both blind interpretation and plausibility analysis. Indeed, the overall accuracies obtained are 81.4 +/- 3.68% for the blind analysis and 91.7 +/- 1.25% for the plausibility analysis on the CESBIO OSO 10m layer. The analysis on the 20m layer shows us that the overall accuracy for the blind approach is 81.1 +/-3.65% and 88.2 +/-3.15% for the plausibility approach.
Quality checks of the validation points have been made by French experts. It should be noticed that for the blind analysis, the methodology of control was based mostly on Google Earth imagery, no additional thematic source of information that could provide further context was used such as forest stand maps, peatland maps, etc.

These results are very good news for us and for our users. The report also contains interesting recommendations that will help us to improve our algorithms. The full report is available for download.

Theia produit au niveau 2A toute la bande côtière du Maghreb

Les tuiles en bleu viennent d'être ajoutées aux zones où Theia fournit des produits de niveau 2A.

=>

Depuis la découverte des récents bugs et la mise en place de moyens de les contourner, le rythme de  production de MUSCATE s'est accru, et nous avons donc pu étendre un peu les zones où nous fournissons des données Sentinel-2 (A&B) corrigées des effets atmosphériques, avec un bon masque de nuages: il s'agit de produits de niveau 2A fournis par la chaîne MAJA.

 

Nous venons de mettre en ligne toutes les données acquises sur la zone côtière du Maghreb, du Maroc à la frontière entre Algérie et Tunisie. Quelques tuiles manquantes ont également été rajoutées au sud du Maroc, et sur le Cap Bon, en Tunisie. Avec cet ajout de 50 tuiles, toutes les terres qui bordent la méditerranée occidentale sont donc suivies pat les produits de Theia (enfin, il manque les Baléares).

Continue reading

[MUSCATE] Theia releases Sentinel-2 L2A products on the whole Maghreb coastal region

The blue tiles were just added to the zones where Theia provides level 2A products with MAJA.

=>

Since we understood the bugs which were slowing MUSCATE, and we found ways to mitigate them, the production rhythm of MUSCATE improved and we were able to extend the zones where we provide Sentinel-2 (A&B) L2A products. L2A products provide surface reflectances after correction of atmospheric effects and with a high quality cloud mask. The products we deliver are provided by MAJA processor.

 

We just released all the data acquired on the Maghreb coastal zone, from Morocco to Tunisia. A few missing tiles have also been added in South Morocco, and on Cap Bon in Tunisia. With these new tiles, we now monitor all the lands that surround the occidental part of Mediterranean sea, adding 50 tiles to those already processed.

Continue reading

[MUSCATE news] Distribution server in maintenance

New update: Muscate distribution server is back but you should use the following address until Monday: https://theia.cnes.fr/atdistrib/rocket/#/home . New products will be uploaded starting from Monday too. And since the week-end is there, it is not compulsary to downloard our L2A images until then (even if you can...)

Update : the server will also be down of Friday . Sorry for that

In order to solve the issue described in this post, MUSCATE distribution server is in maintenance. The distribution of new products should resume  tomorrow (Wednesday) or Thursday. Meanwhile, the production is going on. Sorry for the inconvenience.

Le serveur de distribution de MUSCATE est en maintenance pour résoudre le problème décrit dans cet article. La distribution de nouveaux produits devrait reprendre demain (mercredi) ou jeudi. Pendant les travaux, la production continue. Veuillez nous excuser pour ces perturbations.

Improvements of MAJA cloud masks in production

We are continuously working on improving the Sentinel-2 L2A products delivered by Theia. Since the 2nd of February, a new parametrization was put in production in order to improve two points of the cloud masks. There is a page which summarizes the variations of Theia product versions. Because of some error within Muscate production centre, we are not able to change the product version when we change a parameter. You have to look at the production date to know which version was used. This inconvenience will be corrected soon.

Clouds above water

Although MAJA is optimized for cloud detection and atmospheric correction over land, it is important to detect clouds well over water. For instance to detect the shadows of these clouds that are cast over land. But until now, the clouds mask was missing quite a lot of clouds over water. This was not a surprise as we had passed only a few hours determining the thresholds to apply. We recently found some more time to obtain a better tuning of the detection thresholds above water.

Thresholds in the SWIR have been halved (from 0.08 to 0.04), in the absence of sunglint. When sunglint is likely, due to the geometry of acquisition, the threshold is still higher (0.016, but it was 0.25 before).This means that when the sunglint flag is raised, the accuracy of cloud detection is reduced.

Impact of new cloud detection threshold over water : left, old threshold, right, new threshold.

Cloud dilation

 

 

 

 

The plane contrail illustrates the difference in observation angles of the spectral bands of Sentinel-2. The cumulus clouds above has the same effect, although lower because the cloud is at a lower altitude.

As it may be seen on the image above, the cloud limits are fuzzy, and except for the big and nice cumulus clouds, the pixel next to the cloud mask is also affected by clouds. Even if the cloud has sharp limits, it also changes the illumination of neighbouring pixels, and, in the case of Sentinel-2, as all the bands do not observe exactly in the same direction, there is a "parallax effect" which results in different cloud positions depending on the spectral band. For all these reasons, we need to dilate the cloud mask (by the way, this is an identified drawback of Sen2cor, which does not dilates its clouds).

 

Our dilation buffer was originally 480m. Following complaints from some users working in very cloudy countries, we had reduced the buffer to 240m in may 2017. But we recently figured out that the reduced dilation was degrading our estimates of  aerosols, as undetected clouds are considered as aerosols.  Due to this, the new version which runs since the2nd of February has once again a dilation of 480m.

Left, aerosol validation results with a dilation of 240m, right with a dilation of 480m. The new result brings a significant improvement (Merci à Bastien Rouquié pour ce résultat)

 

La neige de Pyeongchang

Connaissiez-vous ce dicton coréen "On n'est jamais trop prudent" ?

 

Voici une série d'images Sentinel-2 près du site olympique de Pyeongchang. On voit de la neige artificielle apparaître dès le mois de novembre !

 

neige Pyeongchang

 

Les images suivantes montrent qu'au cours du mois de janvier de la neige naturelle a finalement recouvert le site...

Using NDVI with atmospherically corrected data

NDVI

 

NDVI is by far the most commonly used vegetation index. NDVI was developed in the early seventies (Rouse 1973, Tucker 1979), and widely used with remote sensing in the nineties until now. It is computed from the surface reflectance in the red and near infra-red channels on each side of the red-edge.

 

 NDVI=\frac{\rho(NIR)-\rho(RED)}{\rho(NIR)+\rho(RED)}

 

where \rho(NIR) and  \rho(RED) are reflectances in the NIR and RED. Although several users still use top-of atmosphere reflectances (TOA), surface reflectances should be used to reduce sensitivity to variations of aerosol atmospheric content.

 

A time profile of surface reflectance from Sentinel2 satellite for the blue, green, red and NIR spectral bands for a summer crop in South East France. The observation under constant viewing angles minimizes directional effects.One can also notice that reflectance variations related to vegetation status are greater in the near infra-red, while the noise is usually lower. As a result, a vegetation index should rely more on the NIR than on the red.

I think NDVI is mainly used for the following reasons (but feel free to comment and add your reasons) :

  • it has the large advantage of qualifying the vegetation status with only one dimension, instead of N dimensions if we consider the reflectances of each channel. Of course, by replacing N dimensions by only one, a lot of information is lost.
  • it  enables to reduce the temporal noise due to directional effects. But with the Landsat, Sentinel-2 or Venµs satellites, which observe under constant viewing angles, the directional effects have been considerably reduced.

 

I therefore tend to tell students that if NDVI is convenient, it is not the only way to monitor vegetation.

Continue reading

Users of THEIA Sentinel2 products are welcome in Toulouse, 13-14th June 2018

This is Toulouse in mid-June last year. Isn't that a nice place and time to gather the users of Sentinel-2 products delivered by Theia ?

CNES and THEIA are very happy to invite you to provide your feedback on the Sentinel-2 products we have been delivering for more than one year, thanks to the MUSCATE ground segment and the MAJA processor.

 

All users of MACCS/MAJA procesor and of THEIA products are welcome to tell us their findings, suggestions, and share experiences on methods, applications, and results.  We will also provide an update  about the processing perspectives, validation results, description of new versions, and you will have the opportunity to influence us on the related choices.

 

Please register and send your abstract to the workshop site. Here are the important dates to remember :

  • abstract submission deadline : March the 8th.
  • registration deadline : June the 3rd (we only have 100 seats, so please register quickly)
  • workshop dates 13-14th June 2018