Docstoc

Kevin

Document Sample
Kevin Powered By Docstoc
					Identification of Kleingrass in Gonzales, Texas
                Kevin Hankinson
                    ES5053
                   Fall 2004
OBJECTIVE:

To determine the amount and relative distribution of Kleingrass (Panicum coloratum) in
and around Gonzales, Texas.

VEGETATION:




Panicum coloratum, also known as Kleingrass is not native to Texas. It was
recommended for import from Africa in the 1950’s by the Texas Agricultural
Experimentation Station because of its ability to resist drought, survivability in variable
soil conditions, and its tolerance to salt. Kleingrass makes excellent high quality hay and
forage for cattle. Kleingrass is also used as a conservation tool to stabilize soils and
promote revegitation of depleted range land. Kleingrass does however have several
drawbacks. Saponins, glycosides with a distinctive foaming characteristic, in the grass
have been found to cause liver damage in horses, sheep, and goats; cattle are not affected.
All things considered Kleingrass is still a very popular feed for cattle.

LOCATION:

The study area is located approximately 75 statute miles East of San Antonio, Texas on
Interstate 10 and approximately 12 statue miles south on U.S. Highway 97. This area is
located in the Gulf Coastal plains region of Texas. The soil in this area is usually sandy
with a high concentration of iron, overall the soil is nutrient deficient. The study area
encompasses approximately 144 square kilometers.
DATA SOURCE:

The images for the study were downloaded from http://www.texasview.org/. which
provided Landsat ETM+ images with 30 meter resolution for the study area for
12/16/1999, 4/25/2001, 11/6/2002, 11/22/2002, and 3/30/2003. Unfortunately Texasview
could not provide images for consecutive months or years. Each file contained
approximately 48MB of information. Header files ranged in size from 7KB to 12KB.

METHOD:


Each landsat ETM+ 742 composition image was first resized to a more manageable area.
This resized image intentionally contained a known Kleingrass field of approximately 30
acres. To accomplish this resizing upper left and lower right points were determined and
applied to the basic tools-resize-spatial subset-map feature which then performed the
resizing operation. This new image was assigned to memory for later analysis. This
process was repeated for each of the five images. Next, each image was viewed to
determine visually that a difference existed between the Kleingrass field and adjacent
fields of other types of vegetation. Only one of he five resized images raised any doubt
that a distinction could not be made between the Kleingrass and other vegetation. This
was the image from 12/16/1999, and the known (ground truth) Kleingrass field was
strikingly similar to the adjacent field of Coastal Bermuda, therefore it was not used in
the study.
Resized 742 composition of study area, note the 30 acre Kleingrass field with stock tank centered within
the red box. Gonzalez Texas is just out of view in the lower left.



Within the 144 square kilometer study area only 28 of the 159600 pixels were defined as
the region of interest (ROI), indicated in red, and used for the study. The ROI was
defined by selecting the overlay-region of interest-zoom function in ENVI. Then, by use
of the cross hair, the region was defined and saved to be applied to the other images. The
ROI does not encompass the entire 30 acre Kleingrass field that was identified by field
observation. The reason for this is that several trees located in the southern portion of the
field contaminated the otherwise homogenous Kleingrass field.




After the ROI was defined it was used as a classification tool or training area in order to
determine the location of other areas of Kleingrass in the image. To determine the
location and coverage of Kleingrass the supervised-classification-spectral angle mapper-
import ROI function was used. The spectral angle mapper (SAM) compares all available
bands (n-bands), in this case there are six bands, in each pixel with the ROI classification.
SAM measures the spectral angle, which is the angle between each endmember spectrum
and each pixel vector. So smaller angles represent a closer match to the training sited, in
this case the ROI. An angular difference measure of .1 (radians) was initially used (0
being no difference or the same as the training area and 1 being totally opposite). The
results with the .1 angular differences were stored in memory and used to open a new
display that depicted any pixels that were similar as red and pixels that were the different
as black. To verify that the results were valid the 742 composition resized image was
linked to the newly generated spectral angle mapper image. When this technique is used
on calibrated data it is insensitive to illumination and albedo effects. By toggling between
the two images it was possible to see if other pixels outside the ROI but within the 30
acre Kleingrass field would be displayed as red, indicating Kleingrass. If the red colored
pixels roughly resembled the shape of the Kleingrass field the angular difference was
considered to be valid. It was determined that the optimum angular difference to use was
.03 radians. The process was repeated for each of the remaining images using the same
ROI and angular difference.




Above is the image created using the ROI to classify the 742 composition image with an spectral angular of
.03. Note that the red “blob” located in the center of the red square acceptably matches the 30 acre known
Kleingrass field.




To determine how much of the study area was populated with Kleingrass statistical
calculations were performed on the newly created spectral angle mapper image (red or
black pixels) which yielded the total number of pixels that had a digital number (DN) of 0
(black) or 1 (red) and the respective percentages.
Above is an example of the statistical report that is generated using the spectral angle mapper image.

RESULTS:

The statistical analysis for the spectral angular difference images indicated the percentage
of Kleingrass within the study area ranged from a low of 1.95% (3117 pixels) from the
data gathered on November 6th 2002 to a high of 10.77% (17184 pixels) from the data
gathered on March 30th 2003.

Date               # pixel with DN of 1 (red)                  percent coverage
4/25/2001                 11627                                7.29
11/6/2002                 3117                                 1.95
11/22/2002                8395                                 5.26
3/30/2003                 17184                                10.77

As of now no conclusive answer has been determined to be the cause of the substantial
difference in the percent coverage of Kleingrass. However, some theories do come to
mind. For instance the spectral signature of the Kleingrass does change from week to
week and month to month. In order to compensate for this change a new spectral angle
value must be chosen for each new image. Another theory that might explain this is that
as the Kleingrass becomes active or dormant the spectral signature might change at
varying rates and may at some point be too similar to other vegetation to be
distinguished. To compensate for this the study would need to look out of the visible
spectrum and into the IR bands for more refined results.

CONCLUSION:

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:7
posted:10/31/2011
language:English
pages:6