Calculating NDVI using Landsat 8- Cloud/Atmospheric Correction

Question

System: Windows 7

Software: Arcmap 10.4.1

Data: Landsat 8, NDVI calculations

I am attempting to calculate the NDVI values over my state over a period of 3 months. Naturally, there are going to be clouds in my image. I am aware that there are ways to correct for cloud interference (using the QA band, Fmask) but I'm not sure exactly how I should approach it. My plan was to use Landsat 8 bands 5 and 4 on their own but I'm unsure if I should apply corrections before or after calculating the NDVI. Would it be better to run corrections then the NDVI or vice versa? Does the order matter?

Answer 1

I would like to suggest Radiometric Normalization as atmospheric correction.

Check Relative Radiometric Normalization and Atmospheric Correction of a SPOT 5 Time Series. This can be applied in Landsat and is the best way to compare NDVI from different sensors (L5, L7 and L8).

You have to take account radiometric resolution from each Landsat sensor, not all sensor has the same width in NIR (for example, L7 senses 0.77-0.90 wavelenght range; L8, 0.851 - 0.879). Because of this, the result can't be compared even if you make an absolute atmospheric correction.

You need to take one reference image (in surface reflectance) and compute invariant pixels among the time series to create linear models to make correction for each date. If you only need NDVI, you can do this only for NDVI values.

Clouds need to be removed, check in detail the process described in the paper that I linked in this post. CFMask do a great job with clouds and cloud shadows.

Answer 2

In order to check how to interpret and use QA band of Landsat images you can use this blog entry. On the other hand while you are performing your analysis you can give weights to the corresponding cloud and/or cloud shadow pixels.

It would be better to use surface reflectance values which you can download directly from USGS site; so you don't need to perform atmospheric corrections. You can also download and use the NDVI band that is calculated with SR bands directly from USGS too.