+/- inf values in OutputPCA_8_PCs raster

[nklaudy]

Hello,

I am having an issue where the output raster of the PCA (file named OutputPCA_8_PCs) contains many +/- inf values in each band. I am using the package sen2r to download Sentinel-2 data, which is in ENVI format. What may cause the PCA to produce such values and how can I solve it? I can supply more precise information if it is helpful.

Thank you very much in advance!

Best, Nikolas

[jbferet]

Dear Nikolas, I would need more information to provide an answer and solve the problem.

• did you use L2A reflectance? ESA recently added an offset to S2 data, so you may need to account for this offset when you get L2A reflectance.
• did you perform radiometric filtering to discard clouds, shaded pixels and non-vegetated pixels? if you did not correct the offset, then the default values of the thresholds for radiometric filtering should be adjusted.
• Is it possible that you share your scripts with me, so I can reproduce these results and understand the problems ?
• continuum removal may result in infinite values if reflectance data include 0 values. The radiometric filtering, if properly applied, is supposed to solve such problems

thank you, Jean-Baptiste

[nklaudy]

Dear Jean-Baptiste,

Thank you for your answer and sorry for the late reply! I have now set continuum removal to "False" which has resulted in normal-looking values in the PCA rasters and gives me spectral species maps and alpha/ beta diversity maps that also contain realistic values.

Thank you again for your help!

Best, Nikolas

[jbferet]

OK great! You may need to check your data in order to identify more precisely the reason for such infinite values. They may evidence improper data correction, or may be cause by very little amount of pixels which could be masked with no effect on final products, and would allow you to apply continuum removal (which proved to be very important to strengthen the relationship between spectral diversity and taxonomic diversity). --> I suggest that you produce a mask discarding all pixels with B2 or B12 reaching 0 value. Then you can judge if the amount of pixels discarded from your image is acceptable or not. If acceptable, combine this mask with the cloud mask and give it another try with continuum removal.

[nklaudy]

Hello Jean-Baptiste,

I checked and in my raster B6, B7 and B9 contain pixels reaching 0 value. Does it make sense to create a mask for these bands as well?

Again thank you very much for your help!

[jbferet]

0 value for B6, B7 and B9 should not lead to infinite values after continuum removal, unless the full spectrum shows 0 or negative values.

if you are working with vegetation, it seems pretty unlikely that B6 and B7 have reflectance value of 0, unless completely shaded. So yes, I would recommend masking these pixels as the signal will not provide any relevant information.

I am usually not using B9 as it is used for atmospheric corrections, and 60 m resolution which leads to questions regarding interpolation. Some colleagues do use B9, so I may be overly cautious.

Are you working with Sentinel-2 L2A images? I strongly recommend using L2A, not L1C. It may be easier to switch to emails to further discuss your case. my email is on my gitlab personal webpage. you can contact me directly for further assistance

[nklaudy]

Perfect thank you, I will mask the pixels and if I still have any questions/ issues I would contact you directly.