Open greenspin opened 8 years ago
I have the same problem. How can I use fmask as a windows stand-alone version? What's the command I have to enter in the commandline? Or which folder do I have to use as input in a Sentinel-2 folder? When using Version 3.2 for Windows, downloaded from http://ftp-earth.bu.edu/public/zhuzhe/Fmask_Windows_3.2v/
I get the error message
Output argument "clr_pct" (and maybe others) not assigned during call to "C:\Users...\AppData\Local\Temp...\mcrCache8.1\Fmask_0\Users\zhuzhe\Documents\FmaskStandAloneWindows\autoFmask.m >autoFmask"
MATLAB: unassignedOutputs
Error in mclFreeStackTrace
When I use it with Landsat images, it works fine. What am I doing wrong? I used the following commands:
E:\Users\S2_Arles_Test\S2A_OPER_PRD_MSIL1C_PDMC_20151202T224525_R008_V20151130T105641_20151130T105641.SAFE\GRANULE\S2A_OPER_MSI_L1C_TL_EPA__20151202T202847_A002293_T31TCJ_N02.00 C:\fmask\fmask
Even if I use
cd /D E:\Users\S2_Arles_Test\S2A_OPER_PRD_MSIL1C_PDMC_20151202T224525_R008_V20151130T105641_20151130T105641.SAFE\GRANULE\S2A_OPER_MSI_L1C_TL_EPA__20151202T202847_A002293_T31TCJ_N02.00 C:\fmask\fmask
it doesn't work.
I run several tests and all currently available stand-alone versions of the fmask does not work for S-2 data (but they run perfectly for all Landsat scene). So I guess you compile and run processors correctly, just the input data are 'wrong'.
Based on my experience, I concluded that for a life being S-2 data can be processed only with the MATLab version of the code, which is a pity for all not-MATLab folks. Unfortunately Octave is not able to handle the code.
An update has been made!
'The 3.2 version of Windows stand alone software for Sentinel 2 images can be downloaded at this link (provided by Sean Griffin segriffin@gmail.com)' [please look on the main fmask page]
It runs on Sentinel-2 data! [;
Thank you! Yes, it runs on Sentinel-2 data. However, if I use the default parameter settings (3 3 0 22.5), then I discovered some issues with urban areas and shady areas in the mountains. For example, urban areas are classified as clouds and cloud shadows (even if there aren't any clouds) and shady regions are classified as water.
Has anyone already tested different parameter? Is it scene-dependent? What do I have to change?
Thanks!
Accuracy if the fmask greatly depends on a scene. Unfortunately Sentinel-2 has no thermal band which makes it much more difficult to capture clouds correctly. From my experience fmask outperforms Sen2Cor classification, but, as you pointed out, some over/underestimation might still a big problem. You can play with parameters, but since thermal information is missing cloud is defined as a very bright object, which corresponds with a spectral response of bright built-up class (rooftops). I am afraid you will need to go either for over or underestimation...
As I understood, there is only a matlab version for sentinel-2 data cloud masking. The other versions show the "No L*MTL.txt" file error when startin on S-2 data. Is there a windows stand-alone version for cloud masking S-2 data? Or do I have to compile the matlab code somehow? Thanks