The problem is you can't really validate the correctness of the data much except through ground-truthing or collecting spectra in-situ at the time of the overflight. One method some papers will use is comparison to another, well established, sensor/product. For NDVI and other products many researchers will compare their results to MODIS products which have been extensively calibrated. Another approach is to attempt to analyze the image as you intend on both uncorrected and corrected images, using your ground-truth data you can see which is more correct. Many more papers just state that the correction was done to solve some problem (such as BRDF) and have no way of proving that their correction was either applied appropriately or make any effort to show it made any difference to their results.

You can compare your results to other atmospheric correction products, such as ACORN, QUAC, DOS, or in the case of Landsat, LEDAPS.

(https://www.usgs.gov/land-resources/nli/landsat/landsat-surface-reflectance?qt-science_support_page_related_con=0#qt-science_support_page_related_con)

But you run into the same problem, no real validation that any are "correct".

The mid-case scenario is that you have reliable atmospheric data at/around the time of image acquisition and can reasonably apply those corrections. I think the coverage of the problem is fair in this link

(https://www.harrisgeospatial.com/Learn/Blogs/Blog-Details/ArtMID/10198/ArticleID/15452/When-Should-I-Correct-My-Imagery-for-Atmospheric-Effects).

Dark object subtraction is a basic approach that requires a valid dark object be present in your image. That you have negative values are to be expected. See (https://www.researchgate.net/post/What_is_the_correct_procedure_for_dark_object_subtraction) for additional information.

The summary is, if you should correct the imagery at all, it depends on what you're doing with the image(s). And if there are additional concerns depending on the type of analysis you'd like to achieve or any known confounding variables for your study area or land surfaces that you'll process.

Re-reading your question. It also appears that you might be seeing the impact of the histogram stretch on your image, which isn't able to both show detail in the brightest portion of the image (the snow) and the darkest (mtn shadow) at the same time. This will cause the histogram to peak at a lower value and what you get is a solid color above those values. You can play around with the histogram to see different aspects of the image, but its just what your eyes see, any computer analysis wouldn't care. I don't work with snow much, but those areas could also be oversaturated, recording a flat value of 255 for too bright surfaces:

(https://www.usgs.gov/land-resources/nli/landsat/oversaturation?qt-science_support_page_related_con=0#qt-science_support_page_related_con)

I think you can apply dark object substation in ENVI, you can find it in Basic Tools < Preprocessing < General purpose utilities < Dark subtract ( its also another atmospheric correction method ). Apply and check whether output is changing similar to FLASH. I think there is some error when you apply FLASH. 

What are the reflectance values after correction? 

Thank you Uday for the response. Some reflectance values are negative, I am wondering about those negatives values. I also tried the DOS but, when I compared with the reflectance value from DOS with FLAASH, the values are more better FLAASH than DOS, I guess.

The problem is you can't really validate the correctness of the data much except through ground-truthing or collecting spectra in-situ at the time of the overflight. One method some papers will use is comparison to another, well established, sensor/product. For NDVI and other products many researchers will compare their results to MODIS products which have been extensively calibrated. Another approach is to attempt to analyze the image as you intend on both uncorrected and corrected images, using your ground-truth data you can see which is more correct. Many more papers just state that the correction was done to solve some problem (such as BRDF) and have no way of proving that their correction was either applied appropriately or make any effort to show it made any difference to their results.

You can compare your results to other atmospheric correction products, such as ACORN, QUAC, DOS, or in the case of Landsat, LEDAPS.

(https://www.usgs.gov/land-resources/nli/landsat/landsat-surface-reflectance?qt-science_support_page_related_con=0#qt-science_support_page_related_con)

But you run into the same problem, no real validation that any are "correct".

The mid-case scenario is that you have reliable atmospheric data at/around the time of image acquisition and can reasonably apply those corrections. I think the coverage of the problem is fair in this link

(https://www.harrisgeospatial.com/Learn/Blogs/Blog-Details/ArtMID/10198/ArticleID/15452/When-Should-I-Correct-My-Imagery-for-Atmospheric-Effects).

Dark object subtraction is a basic approach that requires a valid dark object be present in your image. That you have negative values are to be expected. See (https://www.researchgate.net/post/What_is_the_correct_procedure_for_dark_object_subtraction) for additional information.

The summary is, if you should correct the imagery at all, it depends on what you're doing with the image(s). And if there are additional concerns depending on the type of analysis you'd like to achieve or any known confounding variables for your study area or land surfaces that you'll process.

Re-reading your question. It also appears that you might be seeing the impact of the histogram stretch on your image, which isn't able to both show detail in the brightest portion of the image (the snow) and the darkest (mtn shadow) at the same time. This will cause the histogram to peak at a lower value and what you get is a solid color above those values. You can play around with the histogram to see different aspects of the image, but its just what your eyes see, any computer analysis wouldn't care. I don't work with snow much, but those areas could also be oversaturated, recording a flat value of 255 for too bright surfaces:

(https://www.usgs.gov/land-resources/nli/landsat/oversaturation?qt-science_support_page_related_con=0#qt-science_support_page_related_con)

Thank you Srikanta for the information. 

Yes, Kenneth, I was also thinking the same, it is very difficult to validate. Of course I will try to compare the result with other sources. Again, the value reflected by the water bodies and mountain shadow looks similar, how can I mask out the shadow in this case. I should either apply the some slope threshold. Do you have any idea on making the certain part of the images that have certain range of pixel value.

Thank you again, thank you so much.

Mohan

Hi Mohan,

(1) You can use band math in ENVI to eliminate the negative values:

-This is the post-FLAASH  equation:

(B1 le 0)*0+(B1 ge 10000)*1+(B1 gt 0 and B1 lt 10000)*float (b1)/10000

- The post-DOS equation is similar but use 1 instead of 10.000 like this

(B1 le 0)*0+(B1 ge 1)*1+(B1 gt 0 and B1 lt 1)*float (b1)/1

(2) You need to collect spectra profiles in-situ at the same time that the pass of Landsat to make comparisons...

good luck!

Santiago 

Hello, Sanitago, Thank you very much for the information. I applied the same equations and negative values are eliminated. When I tried to remove some noise using MNF transform function in ENVI, it again shows the negative value. Can I apply the same equation once again to get the result without negative values.

Thank you,

Mohan

Mohan,

It is not possible, MNF don't have the same units!

Are two things completely different. (MNF vs Reflectance values)

Use MNF Rotation transforms to determine the inherent dimensionality of image data, to segregate noise in the data, and to reduce the computational requirements for subsequent processing (Boardman and Kruse, 1994).

You can also use the MNF transform to remove noise from data by performing a forward transform, determining which bands contain the coherent images (by examining the images and eigenvalues), and running an inverse transform using a spectral subset to include only the good bands, or smoothing the noisy bands before the inverse.

You can apply these equations to eliminate the negative values, if you apply the inverse MNF to Reflectance values.

The post-Flaash/DOS equations is only for reflectance values!

Regards,

Santiago 

Dear Santiago, Thank you very much for more clarity. What is the threshold of eigenvalues that we can observe. Can we apply the certain values so that we can minimize the noise. I think, during processing we just observe the eigenvalues, if values are less than 2 (I guess), we consider there is noise otherwise no noise, am I right.

Thank you,

Regards,

Mohan

Mohan,

Maybe this document (attached) can help you to define the best bands in the MNF analysis (free of noise). Please go to p. 3, items 8 to 12. 

What kind of Landsat images did you work, Landsat-8 OLI / L-7 ETM+/5?

Regards, 

Santiago

Dear Santiago,

Thanks a lot for documents and suggestions. Definitely it will be helpful for me. I am thinking about the find out the evolution of some lakes those are increasing in their size and volume at higher altitudes since. Though these lakes very small in size it is very difficult to map them with Landsat Images but still useful. If Possible to get some high resolution images worldview, geoeye etc will be great for me.  Right Now I am thinking of use of Landsat for long term evolution and Sentinel 2.

Thanking you,

Warm regards,

Mohan

Again, I was trying to improve my result with different atmospheric correction methods available in ENVI. I had tried Quick Atmospheric correction, which gives result without those green parts that covered with snow. I think this method normalized the reflectance values for different objects. Hopefully, it can corrects the images.

Dear Flaash correction is related to reality