Overview

The chosen study area for this assignment was near Memphis Tennessee including three scenes running North to South along the Mississippi River. The Landsat imagery was downloaded both from libra.developmentseed.org and explorer.usgs.gov/. The images downloaded were on path 23, rows 35 (August 8, 2014), 36 (August 22, 2014) and 37 (August 22, 2014). The listed images were selected for low snow and cloud cover. Row 35 and 37 were downloaded from libra and row 36 was downloaded from the usgs explorer site as the download from libra would not work. The mississippi river was chosen for it’s variety of land cover types and relatively flat terrain.

The downloaded scenes are in WGS 1984 and projected in UTM Zone 20. Data could have been reprojected in either ERDAS or PCI to NAD83, but this step was not deemed necessary for this project.

 

Mosaicking : Comparing Results from ERDAS Imagine and PCI Geomatica

Figure 1 shows the study area near Memphis Tennessee

Mosaicking : Comparing Results from ERDAS Imagine and PCI Geomatica

Figure 2 shows the location of the landsat imagery before any corrections or mosaicking.

Mosaicking : Comparing Results from ERDAS Imagine and PCI Geomatica

Figure 3 shows an overview of the mosaicked landsat 8 imagery in PCI.

Mosaicking : Comparing Results from ERDAS Imagine and PCI Geomatica

Figure 3 shows an overview of the results of mosaicking in ERDAS using feathering and histogram balancing.

Mosaicking in PCI

In order to mosaick in the PCI Geomatica software first the focus interface must be loaded so that the landsat images can be imported for atmospheric correction. To import the landsat imagery the *.tar.gz file (or *.tar.bz file in libra) must be extracted. The extracted bands include a *.mtl file which can be dragged into the PCI focus environment to open either Pan, Multispectral or Thermal imagery.

The atcor correction was then run on these files with default settings. The results were better than the originals but only marginally (See right).

Next a project was created in orthoengine, the images to be mosaicked were defined and the automatic mosaic tool was used with defaults and the overlap method to produce a new imagery. The results were nearly seamless with good colour balancing (shown in Figure 3).

 

  • Raw Imagery

    Raw Imagery

  • ATCOR Corrected

    ATCOR Corrected

Mosaicking in ERDAS

In order to mosaick in ERDAS the landsat imagery must be imported using the manage data –> import data tool. Once complete the multispectral imagery is available for manipulation. Due to extension limitations and the determination in PCI that the atmospheric correction was not absolutely required, the erdas data was not corrected. To accomplish this using available licenses would have required dark pixel subtraction or TOA correction using self built models in spatial model.

The 2D Mosaic Pro tool was used for the mosaicking of the three images. The results were not seamless and despite trying many settings good results were not obtained. Default settings vs averaging and feathered seamline methods are shown to the right. The best results were obtained using default settings. Attempting colour corrections seemed to result in worse results.

Non default settings attempted  include seamline settings such as feathered edges, overlapping edges and averaging as well as colour corrections including colour balancing and histogram balancing and illumination equalization.

Manual radiometric adjustments may aid in better results (i.e adjusting the most northern image manually) but this was not done for this project.

  • Default Mosaic

    Default Mosaic

  • Feathered Edge with HistogramMatching

    Feathered Edge with HistogramMatching

Comparison of ERDAS to PCI

Overall the better result was achieved through PCI Geomatica. The colour balancing was more complete and the result was more seamless. In ERDAS there was a clear difficulty with making the images radiometrically similar using automatic settings.

The ERDAS workflow is slightly more intuitive with the use of Mosaic Pro but the difficulty with showing demos is a clear set back. ERDAS has a functionality for this but I had difficulty making it work.

PCI creates a responsive edge between the two images instead of a straight line such as ERDAS which may be related to the better results. The downside with PCI is that the resultant image is darker than in ERDAS, but this is easy to fix by changing very simple contrast and brightness settings.

It is possible that atmospheric correction may have provided an improved ERDAS result and with proper extensions this could be better tested.

Additionally the import workflow in PCI is very user friendly and does not require exporting within the software. In using downloads from the Libra website there were no compatibility issues in PCI while ERDAS required file conversion from .tar.bz to .tar.gz which resulted in corruption of one file. PCI prevents this by having the user extract the data before importing it into the software.

Visual comparisons of some attempted mosaicking settings are shown below. PCI obtained great results on the first try so the settings were not changed resulting in only one PCI mosaic image.

Mosaicking : Comparing Results from ERDAS Imagine and PCI Geomatica
Mosaicking : Comparing Results from ERDAS Imagine and PCI Geomatica
Mosaicking : Comparing Results from ERDAS Imagine and PCI Geomatica
Mosaicking : Comparing Results from ERDAS Imagine and PCI Geomatica