How to preprocess and remove noisy data from Landsat-8 image Bands for ndvi time-series analysis

Question

I'm working on satellite image processing, I'm calculating NDVI for time series analysis and vegetation prediction. I'm having a tough time removing the noisy data like,

1. Cloud coverage
2. Haze
3. Cloud shadow

on Landsat-8 satellite image bands.

Is there any way to do this in Python?

I came across fmask package in python to mask cloud coverage but wasn't clear how to use it.

Then used pymakser but was only able to create a separate cloud mask file using BQA band of Landsat-8 but couldn't apply the mask on the image bands.

Finally tried using rio-cloud mask package but wasn't able to create the pre-requisite top of atmosphere reflectance for landsat-8 bands.

Also it'll be great if there's a method to just use the BQA band of Landsat-8 to remove noisy data from the TIF files using python.

Answer 1

Hi for anyone landing here in 2021 or later. Handling the QA masks in Landsat and other remote sensing data is indeed a pain, so I made a small package just for that.

unpackqa: A python package for QA/QC bit unpacking and labeling in earth science data products

Full Documentation

Here is a short example, given a numpy array, img_qa_data which is the data from a Landsat 8 image QA band, you can create a cloud mask like so:

cloud_only_mask = unpackqa.unpack_to_array(img_qa_data, 
                                           product='LANDSAT_8_C2_L2_QAPixel', 
                                           flags=['Cloud'])

plt.imshow(cloud_only_mask, interpolation='none')