Crosta Techniques

CROSTA TECHNIQUE
Using Principal Component Analysis

By iwaL and Adopted from ER Mapper Application Help
The following image processing technique provides a simple, robust way to map
alteration zones for mineral exploration using Landsat Thematic Mapper (TM) satellite
imagery. It is based on the use of Principal Components Analysis (PCA) and is
sometimes called the 'Crosta technique' after the researcher who carried out the initial
studies (Crosta and McM.Moore, 1989).
The following steps describe how to build the PCA alteration mapping algorithm in
ER Mapper. In order create the algorithm, you need to create and examine dataset
statistics to account for variation in the spectral properties of different areas and TM
images. After choosing the appropriate PCs for your image, you can display the result as
an RGB algorithm for analysis. For complete details on the concepts and theory behind
the PCA alteration mapping algorithm, please refer to the references listed at the end (this
particular technique is described in the Loughlin paper).
1. Create two Virtual Datasets, one that contains TM bands 1, 3, 4 and 5 (this is dataset
1), and another that contains bands 1, 4, 5 and 7 (this is dataset 2). Add a label to each
layer indicating the band.
For example: ETM_1345_113062_20040521.ers (Dataset 1)
2. Be sure to delete the transforms from each layer to avoid rescaling the data by click
button in the algorithm window and then in the transform window click Edit
button choose delete this transform.
3. Calculate statistics for each of the Virtual Datasets by: go to main menu and click
process menu and then choose Calculate Statistic..
4. Display the statistics for dataset 1 and examine the Covariance Eigenvectors as the
following picture:
Identify which PC has the greatest loadings (values) for TM bands 1 and 3 but that also
has opposite signs (+ or -). Typically this is either PC3 or PC4. For example, in this
dataset you would choose PC4 (you are looking at VDS bands 1 and 2 below as they
refer to actual TM bands 1 and 3):
For example:
ETM_1345 (Dataset 1)
Band 3
This PC represents the "iron oxide" (F) component.
Note: When examining the statistics, remember that the "Band" number above refers to
the order of bands in the VDS, *not* to the actual TM band number. For example, TM
band 3 may actually be Band 2 in your VDS. In this case, you must first determine which
VDS band numbers correspond to TM bands 1 and 3, then examine the PC3 and PC4
statistics for those two bands.
5. Display the statistics for dataset 2 and examine the Covariance Eigenvectors. Identify
which PC has the greatest loadings (values) for TM bands 5 and 7 but that also has
opposite signs (+ or -).
For example:
ETM_1457 (Dataset 2)
Band 7
Typically this is also either PC3 or PC4. This PC represents the "hydroxyl" (H)
component.
6. Create a 2-band Virtual Dataset that calculates PC4 (or PC3 as appropriate) from
bands 1,3,4,5 in one layer (the F image), and PC4 from bands 1,4,5,7 in a second
layer (the H image).
7. Delete the transforms to avoid rescaling the data and add a label for each.
8. After saving the VDS, calculate statistics for it. You will use this VDS later to
generate PC1 of the H and F images.
9. Create PC1 as follow:

10. Create an RGB algorithm with the following contents (remember to use the
appropriate PC for your scene; PC4 is shown below):
Red = PC4 of bands 1,4,5,7 (the "H" image)
Green = PC1 of the "H" and "F" images (using the Virtual Dataset created in step 5)
Blue = PC4 of bands 1,3,4,5 (the "F" image)
On each layer, apply a linear contrast stretch that clips the darkest portion of the
histogram - you are interested in highlighting the frequently occurring values in the
center of the histogram. You might start by applying a 99% clip and then moving the
bottom node of the transform line in closer to the histogram center.
The resulting image is usually a dark bluish color composite image on which alteration
zones are unusually bright. White pixels are both iron-stained and argillyzed, bright
reds and oranges are more argillyzed than iron-stained, and bright cyan to bluish tones
are more iron-stained than argillyzed.
Optional: To add the major structural features of the image back into your RGB
algorithm, you can add an Intensity layer that generates PC 1 of the TM image and
contrast stretch it. This may make it easier to interpret the image because the overall
scene brightness or albedo information is restored to the scene. In some cases, however, it
may detract from interpretation so you need to experiment.
In addition, you may wish to add low pass (averaging) filters to the RGB layers to
smooth noise in the higher order PCs, and/or a high pass (sharpening) filter to enhance
structural edge features in the PC 1 Intensity layer if you decide to use it.

Crosta Techniques

Uploaded by

Copyright:

Available Formats

Crosta Techniques

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Crosta Techniques

Uploaded by

Copyright:

Available Formats

CROSTA TECHNIQUE

Using Principal Component Analysis

For example: ETM_1345_113062_20040521.ers (Dataset 1)

This PC represents the "iron oxide" (F) component.

9. Create PC1 as follow:

Red = PC4 of bands 1,4,5,7 (the "H" image)

Blue = PC4 of bands 1,3,4,5 (the "F" image)

You might also like