Tuesday, August 25, 2015

Landsat image pre-processing in ArcGIS - tools for seamless mosaicking

Mosaicking adjacent Landsat tiles often produces visible seam lines at the boundary between the two scenes. The Landsat Toolbox for ArcGIS deals with this problem by selecting non-cloud, non-shadow, and non-snow pixels in the overlapping portions of the scenes and then performing linear regression on each band. To do this the user must first select a "master" or reference image and a "second" image. The values in the second image are adjusted to match the first image. Although this method is relatively simple it is not part of the standard mosaic tool in ArcGIS. As evidenced in the figure below this method is quite effective at removing seam lines producing a visually coherent mosaic.

The upper image contains portions of two overlapping scenes mosaicked using the standard mosaic tool. The lower image was mosaicked using the Landsat Toolbox for ArcGIS.




8 comments:

  1. Hi Tom,
    Is there any documentation for this software? Just looking at the radiometric normalisation and how to select output raster. It seems to be asking for a raster file that already exists is this correct?
    Thanks for your time,
    Andrew

    ReplyDelete
  2. Hi Andrew, Documentation is included in the form of a word document in the doc folder in the zip file. All the tools in this toolbox output new rasters. For the radiometric normalization you pick two images, two mask images, and a set of points that are supposed to be spectrally-invariant. Great care should be taken to make sure that the two types of points actually are spectrally invariant. Sometimes features we might expect to be invariant, such as desert playas, actually change quite a bit in response to soil moisture. I recommend topographic correction prior to radiometric normalization. Please feel free to post comments or question or e-mail me.

    ReplyDelete
  3. I'm working on improving the topographic normalization tool. The new version will stratify based on NDVI to help deal with heterogeneous landscape. I've found this approach to be a significant improvement.

    ReplyDelete
  4. Hi,
    Could you briefly explain what "Input points" file should contain (speaking about Topo correction)? Or what they do? Fmask returns .hdr file only and I couldn't find any description, also the senctence in your documentation is not finished. I would appreciate a hint on that.
    Thank you,
    Adam

    ReplyDelete
  5. Hi Adam,

    The input points can be random points or every grid cell converted to a point. They are just used in sampling the raster. I'm working on a version of the tool that eliminates the need for an input point shapefile and skips straight to a number of random points to be generated. The Fmask file will need to be converted to a geoTIFF using ENVI or some other software. The USGS now has reflectance products on Earth Explorer that include an mask band so that conversion step can be eliminated. I'm not sure which sentence you are referring to in the documentation, but if you point me to the right place I'll fix it. I'd be happy to answer any questions that you might have.

    ReplyDelete
  6. Hi, thank you for your answer, that helped me. I was referring to point 2) "Topographic correction ... ", section "Parameters" and point 3) "Input points". I downloaded your toolbox from ArcGIS website (in case you are looking at some other version) and the file is signed with 8/14/2015 date.
    Thanks a lot, I keep fingers crossed for your work to succeed.
    Adam

    ReplyDelete
  7. Hi,

    You are correct. The sentence should read "A point shapefile that could be random points or systematic points. The number of points represents a trade-off between accuracy and speed." Speed was accidentally omitted.

    ReplyDelete
  8. Thank you for the correction.

    ReplyDelete