I'm working in RStudio environment for those approaches. There are time-efficient packages like RStoolbox which is designed to easily remove clouds from Landsat. As for Sentinel I found this: https://gis.stackexchange.com/questions/267260/remove-clouds-in-sentinel-2a-with-r

However, for both satellites, mosaicing works pretty straightforward in R with "raster" package- function "mosaic" which is basically a one-line code.

Other than working in R you might find useful tools in SAGA, I am not sure but I think I glanced some cloud-correction tools as well.

Hope this helps.

Usually you would not mosaic them as each scene has its unique atmospheric conditions, sun-angle, etc. You would process each scene as a independent entity. If you want a mosaic like air photos then GDAL merge will do it for free.

Thanks for the answers! I am aware of the options to simply stitch images together and I think you both pointed out the best options available to do this. However, as Stuart wrote - the problem is to create seamless mosaics (maybe I should rather talk about artificial scenes) where effects from atmosphere, sun-angles, clouds, cloud-shadows etc. are all minimized or in ideal case completely removed. So that it becomes possible to conduct analyses over large areas without the need to process each image individually (as in the later case you will also require reference/training data for all scenes). I for example talked to some people who used time-series of data to interpolate missing pixels. Algorithms like STARFM which use MODIS data to interpolate missing Landsat data may also be one option. But I assume there is more currently being developed?

The Remote Sensing and GIS Library (RSGISLib) python module and ARCSI (Atmospheric and Radiometric Correction of Satellite Imagery) can do this. You can atmospherically correct and cloud mask the data using ARCSI and then mosaic them using the maximum NDVI value per pixel (as per a quality mosaic in Google Earth engine). This is currently available for Landsat only, but if you're comfortable with python it can be easily modified to work with Sentinel-2 also

Thanks Nathan! This sounds like an interesting option. I will have a look into this. Will be exiting to see whether there are more alternatives out there. I am sure a lot is being developed right now.

Might be helpful:

https://rus-copernicus.eu/portal/the-rus-offer/ict-offer/

https://developers.google.com/earth-engine/landsat

https://github.com/fitoprincipe/geebap

Thanks Mustafa! This also looks really helpful. Particularly the links with respect to the Google Earth Engine processing options.

You could take a look at FORCE https://www.uni-trier.de/?id=63673, developed by David Frantz

HenninJust had a brief look! Looks really great. Thanks! I will try this

Hi,

I think we rather succeeded i doing so with Sentinel2 data, using a good atmospheric correction and cloud detection software (MAJA), and a monthly synthesis software based on a weighted average method (WASP). It works well over land at least in summer (a few issues to solve over water). WASP and MAJA are available.

More info :

http://www.cesbio.ups-tlse.fr/multitemp/?p=14192

Hi Fabian,

That's a good question for the satellite information we are traversing. We have developed a pre-processing workflow within Google Earth Engine that mosaics together tens to thousands of Sentinel-2 tiles masking out the highest reflectance pixels(cloud, atmosphere, Sentinel-2 granule neighbouring lines, sunglint, waves-the two latter over the marine part). If you have an area of interest/application, I could send you one pre-processed composite for comparison with other approaches perhaps. May I ask which is your field/area of interest?

Dear Olivier, dear Dimos,

thanks for adding your comments. Both approaches sound very interesting. Maybe there is an opportunity to make some comparison - including also some of the aforementioned approaches. It is really great to know that so many options are available. I hope to find some time soon to have a closer look!

Dear Fabian,

I designed and implemented a processing chain for composite vegetation indices and derive time-series from GEE as part of a publication I did this year. For the moment it is just for Landsat 4-8:

https://code.earthengine.google.com/4509a6a9e5d041a8f36e04b0edb80644

as the another one I did for Sentinel 2, require further development. Let me know if it you require some explanation for use this script.

Best regards.

Dear Fabian,

thanks for sharing your code! We will definitely have a look at it!

Thanks for very interesting discussion

Good afternoon,

I ran some tests using USDA' implementation of StarFM [1] to fill in the clouds of Landast 8 Surface Reflectance over the Amazon forest. The results are good for visual interpretation, but not so much for automatic supervised classification. We also tried linear interpolation and the results were slightly better.

On the other hand, we think spectral linear mixing models could harmonize satellite images across platforms and sensors. There is already a paper on on this subject for Landsat 7 & 8 [2].

Best regards,

[1] STARFM https://www.ars.usda.gov/research/software/download/?softwareid=432&modecode=80-42-05-10

[2] Global cross-calibration of Landsat spectral mixture models Article Global cross-calibration of Landsat spectral mixture models

Hi Alber Ipia

Related to StarFM.

I aim to achieve the fusion of Landsat and Modis time series data using StarFM. However, I cannot run it. Please technically guide me.

Thanks

Hi Trong Can Nguyen. I'm not part of StarFM's development team but I can tell you what I did to run it. Can you send me an e-mail with the details of your problem? Cheers!

A new approach?

See the problem as the human eye would.

Use the derivatives of the radiation spectrum and the total energy detected.

Good Luck

Preprint A Model of Color Vision for Remote Sensing 1

…