6

I am working with Landsat 7 ETM+ SRC-off imagery. I'm doing some land-cover classification for a place in Kenya. My problem is the no-data stripes in the image.

Is there a way in which I can interpolate data for such 'empty' areas using the neighbouring pixels?

5
  • 2
    What program do you use ? Commented Oct 14, 2013 at 12:23
  • 3
    Hi Ken - I would be very careful interpolating pixel values for these areas. In addition to the NoData values, adjacent pixels in the banded areas of the SLC-off imagery are often corrupted or have inaccurate values. This means that interpolating off of these pixels will simply fill in your NoData with...BadData. Your nearest actual data may be 10-20 pixels away, meaning that you will be interpolating a fairly significant area off of data that may be 600m removed. Something to keep in mind depending on the desired resolution of your output.
    – JWallace
    Commented Oct 14, 2013 at 15:21
  • Check these answers gis.stackexchange.com/questions/30250/… and gis.stackexchange.com/questions/60822/…
    – nadya
    Commented Oct 14, 2013 at 16:06
  • 1
    The procedure of filling the Landsat7 Null stripes is called 'Destriping' . There's vivid bibliography and many techniques (that are beyond my unterstanding atm) on how to accomplish the task. Just a comment to know how its called so you can search for it by name.
    – nickves
    Commented Nov 14, 2013 at 14:19
  • 1
    If you want to avoid the problem entirely you can try acquiring new Landsat-8 imagery from glovis.usgs.gov
    – Radar
    Commented Nov 14, 2013 at 16:46

4 Answers 4

5

There are multiple "gap fill" techniques which essentially take multiple L7-SLC off scenes and combines them to create a gap free image. This may or may not work for you application. For land classification it should be acceptable, assuming you can find other L7 scenes of a similar vintage.

You should be very aware of what these procedures are doing and be sure they are acceptable for your use case. Generally, this is better than any interpolation method since the "gap fill" pixels would simply be older, not entirely made up.

If you are working with crop classification this may not be acceptable, since even a few days can significantly alter the signature of vegetation.

The USGS has a great starting point for "gap fill" procedures http://landsat.usgs.gov/using_Landsat_7_data.php.

Tools such as ERDAS Imagine and ENVI have native tools to tackle this problem.

-Tim

1

I work with LiDAR data daily. I have frequently come across a DEM where there are No Data pixels, and you just have to go with it. Interpolating a surface area will only give a false analysis later on. What if these areas are extreme peaks or valleys? Interpolating pixels will not accurately represent the land surface. Search for a different DEM available on the web... I know its difficult finding data with the Fed Gov Shutdown... But see what's out there! If you find a few different DEMs, you can always create a mosaic dataset within a file geodatabase and "stitch" the images together by adding multiple rasters to the mosaic dataset.

I know this doesn't exactly answer your question on how to interpolate empty values, but figured I'd include my input being that I work with this data daily and even the data I receive is through the State Government... Just have to analyze what's available and leave the rest for what is it... Part of working in the GIS field is accepting error and distortion, but you always want to be as accurate as possible, when possible.

0
0

As an alternative to interpolating the SRC errors have you considered using Landsat 5 data? There should be coverage for the same time period.

2
  • Landsat 8 imagery is now available with good coverage.
    – Radar
    Commented Nov 14, 2013 at 16:45
  • That's true, I was assuming the OP wanted older data that is not available under Landsat 8.
    – dblanchett
    Commented Nov 14, 2013 at 21:27
-1

You first need to understand why they are null. Is it a specific area/feature type that presents issue ie water area or steep mountain area. How prevalent are these null areas across your data?

If you only want to fill those null areas within you data out reinterpolating the entire data set, you have a few options.

Install QGIS & use the GRASS plugin with r.fillnulls.

Install SAGA and use the Grid -- Tools:

  • Close gaps (there are several to select from)
  • Close One cell gaps
1
  • 2
    They're NULL because the SRC (Scan Line Corrector) failed at May 31, 2003, resulting stripes of false data. The only areas that are unaffected by the failure are those are are at the nadir of the sensor
    – nickves
    Commented Nov 14, 2013 at 14:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.