Machine learning benchmarking for land cover map production

Land cover map validation is a complex task. If you read French, you can check this post by Vincent Thierion which shows how the 2016 LC map of France produced by CESBIO stands with respect to data sources independent from those used for its production. But this is only one aspect of the validation. A land cover map is a map, and therefore, there are other issues than checking if individual points belong to the correct class. By the way, being sure that the correct class is known, is not so easy neither.

 

In this epoch of machine learning hype 1, it is easy to fall in the trap of thinking that optimising a single metric accounts for all issues in map validation. Typical approaches used in machine learning contests are far from enough for this complex task. Let's have a look at how we proceed at CESBIO when we assess the quality of a LC map produced by classification.
Continue reading

[MUSCATE News] A difficult start of 2018 for our production center

As you have probably noticed, our production rate has been very low these days and we are more than 10 days late in our delivery of L2A and snow cover products.

This seems to be due to an intervention on CNES cluster end of December to add new nodes and disk space. MUSCATE sometimes loses communication with the platform that handles the databases and crashes. As it also happened when CNES was closed for Christmas, we really lost a lot of time.  All the teams are on the deck to try to solve this issue and catch the delay up. We are very sorry for that inconvenience.

 

Revised spectral bands for Sentinel-2A

The Sentinel-2 mission status document, edited by ESA, is a very interesting reading. On its last edition of 2017, ESA announced very discretely that the spectral bands of Sentinel-2 had been revised, following a review of the pre-flight measurements. Very few details are provided on the nature of the error contained in the previous version, and on the validation of the new ones. But still, a new version of the spectral response function is available here,, since the 19th of December 2017. The site provides an excel file with the spectral response functions.

All the visible and near infrared bands have changed a little, even if only three bands have significant changes, B1, B2 and B8: B2 equivalent wavelength changes by 4 nm, B1 by 1 nm, and B8 by 2 nm. The SWIR bands did not change.

Old and new versions of five VNIR S2A spectral bands, together with that of S2B.

Most users should not use bands B1 and B2, as they are affected by atmospheric effects. So I do not think much of you will have to change the coefficients in your methods. But for us, who take charge of the atmospheric correction, and heavily rely on B1 and B2, it probably has an effect, and we are changing our look-up tables to account for that. Stay tuned for the results.

Best wishes for 2018 !

=>

May this new year bring you happiness, and not only related to image time series !

 

As usual, this beginning of year brings the opportunity to summarize 2017. Here is what I would record, in our field of interest :

  • the consecration of Copernicus program and of the Sentinel satellites. Since 2015, more than 110,000 people have registered to access the data since 2015! In my opinion, this success is due to the combination of several factors: the data are free and easy to access, the observations are repetitive, regular and frequent worldwide, and the data are of high quality. Congratulations to ESA and the EU, not to mention the contribution of CNES for the quality of Sentinel-2 images and the calibration of Sentinel-3. Continue reading

Tous nos voeux pour 2018 !

=>

Que cette nouvelle année vous apporte joie et bonheur, et pas seulement dans l'utilisation de séries temporelles !

 

Sans aucune originalité, ce début d'année est l'occasion de faire un petit bilan de l'année 2017. Voici, dans notre domaine, quelques uns des faits que je retiendrai :

  • la consécration du programme Copernicus, et des satellites Sentinel. Plus de 110 000 personnes se sont inscrites pour accéder aux données depuis 2015 !  A mon avis, ce succès est dû à la combinaison de plusieurs facteurs : les données sont gratuites et faciles d’accès, les observations sont répétitives, régulières et fréquentes sur le monde entier, et les données sont de grande qualité. Un grand bravo à l'ESA et à l'UE, sans oublier la contribution du CNES pour la qualité des images de Sentinel-2 et l'étalonnage de Sentinel-3. Continue reading

Venµs mission status, 5 months after launch

Hereafter some news on the status of the Venµs mission.

Venµs was successfully launched from the Kourou space port on August 1st by a VEGA launcher:

 

https://vega.cnes.fr/en/live-vega-launch-venus-august-first-2017

 

First images were acquired by mid-August (see below for examples). The commissioning phase is still running. This phase consists in checking the whole system, including the satellite, the camera, the download of the images and data to the Kiruna receiving station, the ground processing chains, as well as the geometric and radiometric calibrations. All these components are in a good health and working well. However, given the very demanding requirements in terms of multitemporal registration, more work than anticipated is needed to fine tune the AOCS (Attitude and Orbital Control Subsystem) and the processing algorithms. In addition, the first part of the in-flight demonstration of the electric engine developed by RAFAEL (called IHET) will take place from mid-december to mid-january.

 

For these reasons, CNES and ISA plan to resume the systematic acquisitions of the scientific sites early 2018. Preparing the reference images for every site and checking the quality of time series will also take some weeks. We anticipate delivering the Level 1 (top of the atmosphere reflectances, orthorectified) products by april 2018, but all the data acquired from the beginning of systematic acquisitions will be processed and made available on the Theia web site:

http://www.theia-land.fr/en/products/venus

 

Level 2 data (top of the canopy reflectances) might be available slightly later since the method we use requires a time series of data.

You will find some examples of images following the links below :

 

https://presse.cnes.fr/en/france-israel-space-cooperation-venus-vegetation-monitoring-satellite-sends-back-first-images

and here:

 

http://www.cesbio.ups-tlse.fr/multitemp/?p=11215

http://www.cesbio.ups-tlse.fr/multitemp/?p=11096

http://www.cesbio.ups-tlse.fr/multitemp/?p=11344

 

The Moon is also acquired for calibration monitoring purposes:

http://www.cesbio.ups-tlse.fr/multitemp/?p=11344

 

We thank you for your patience. We are doing our best to provide you with quality products.

Happy New Year to you all

Our blog's audience in 2018

=>

A sixth year begins for the "Séries temporelles" blog, and as usual, it is an opportunity to review its audience, and to get a little self-satisfaction.

The blog is always receiving more visits. even if the annual growth rate is much lower than the previous years, but still 20%... French visitors constitute only 35% of visits. The United States rank second, followed by Morocco, which is probably an effect of CESBIO's long presence in Marrakesh. Then come France neighbouring countries.

 

2013 2014 2015 2016 2017
Number of visits 13985 22928 34723 47773 57692
Number of viewed pages 30922 46940 66947 89555 105846

Continue reading

Fréquentation du blog en 2017

=>

Une sixième année commence pour le blog Séries Temporelles, et comme d'habitude, c'est l'occasion de revenir sur sa fréquentation, et de faire un peu d'autosatisfaction.

 

Le blog reçoit toujours plus de visites. même si la croissance devient inférieure à 20%..en un an. Les visiteurs Français ne constituent plus que 35% des visites. Les Etats-Unis viennent en second, suivis par le Maroc, ce qui est probablement un effet de la longue implantation du CESBIO à Marrakech.

2013 2014 2015 2016 2017
Nombre de visites 13985 22928 34723 47773 57692
Nombre de pages lues 30922 46940 66947 89555 105846

Continue reading

Data upload on MUSCATE distribution server is slow

It seems that the upload of data to MUSCATE distribution server is quite slow these days. This results in a delay in the provision of data. As CNES is closed between Christmas and the new year, we will have to wait a little to have it repaired. We are sorry for that inconvenience. But you can use this opportunity to rest and spend time with your families without having to process our scenes :) . We wish you all a happy Christmas week.

Il semble que la mise à jour du serveur de distribution de MUSCATE très lente depuis quelques jours,c e qui cause d'importants retards de distribution. Comme le CNES est fermé entre Noel et le jour de l'an, nous devrons attendre pour réparer ce problème. Nous vous prions e nous en excuser. Mais vous pouvez en profiter pour passer des vacances en famille sans avoir à traiter nos données ;) . Passez tous de bonnes fêtes !