doc - The University of Texas at San Antonio

advertisement
Title: Urban Growth Pattern of the San Antonio area
Name: Alecia C. George
Abstract
The project that I worked on was the urban growth pattern of the San Antonio
area. I worked with Landsat images of San Antonio from the years 1986 and 2003. I
received the 1986 image from Dr. Xie and the 2003 image from the Remote Sensing class
projects. I inputted the images in to ENVI and tried to statistically determine what the
change was between the two images.
Background
The Landsat Program is the United States oldest land-surface observation satellite
system. It has been obtaining data for the U.S. since 1972. The first Landsat Satellite to
be launched in July 1972 and the last one (Landsat 7) was launched in 1999 (Jensen,
2000). The Landsat satellites cover the Visible, the Near Infared and the Thermal Infared
bands. The swath that is covered by the satellite is 185 km wide and covers the entire
surface of the land every 16 days (Xie, 2003).
Study Area
The Study Area that was used was all of San Antonio and the immediate
surrounding area.
Data and Methods
When starting my project the first thing I did was to email the U.S. Census bureau
to find out what the population of San Antonio was for the years 1986 and 2003. I did
this so that I see what the population growth was between those years. The census
bureau did not have a count for those specific years and so I asked for the years that were
closest to them. When they responded to me they gave me the population for the years
1980, 1990 and 2000. In 1980 the population was 785,880, in 1990 the population was
935,933 and in 2000 the population was 1,144,646. Between the years 1980 and 2000
there was a growth of 46% in San Antonio, Texas.
To begin my project on ENVI I had to find two images of San Antonio that were
at least 10 years apart so that I could look at them and show what the differences and
growth had been. The oldest image that I used was a Landsat image from 1986 (Figure
Ia) and was given to me by Dr. Xie and the newest image that I used was a Landsat image
from 2003 (Figure Ib) that I got from our Remote Sensing class projects. The image that
I received for 1986 was a larger than the image for 2003. Dr. Xie had to shrink the 1986
image so that the two images had the same coordinate points. Dr. Xie did this by taking
the pixel point from the top left hand corner and the pixel point from the bottom right
corner of the 2003 image. Dr. Xie then inputted these coordinates into the 1986 image to
in order to shrink it to the same size as the 2003 image.
I then brought up the two images in ENVI and then linked them geographically.
After doing this I went to specific areas in San Antonio to show what the changes looked
like from 1986 to 2003. The specific areas that I looked at were Braunig Lake, the I-10
and Loop 1604 interchange, Kelly Air Force Base, Hwy 281 and Loop 1604 and the San
Antonio International Airport. In my presentation I pointed out the major changes for
each area.
I decided to run a classification to so that I could determine statistically what the
differences were between the images. The first classification that I did in ENVI was the
ISODATA unsupervised classification. The Isodata calculates the class means that is
evenly distributed in the data space and then clusters the remaining pixels using a
minimum distance technique. I ran the classification first for the 1986 image and then
again for the 2003 image using a maximum iteration of one and then again with a
maximum iteration of five. I then ran the K-Means classification, also with a maximum
iteration of one and five. The K-Means classification calculates the class means that is
evenly distributed in the data space and then clusters the pixels into the nearest class
using a minimum distance technique. After running all four classifications I decided to
use only the images with the maximum iteration of five for both the Isodata and the KMeans classification.
In order to determine the statistics I went under the Basic Tools and clicked on
Change Detection and then Change Detection Statistics. I ran the statistics first for the
Isodata classification and then for the K-Means classification. For the initial stage I
chose the classified 1986 image and for the final stage I chose the classified 2003 image.
Results
When the tables appeared giving the information for the changes it showed that
there were no changes between the two images for both the Isodata classification and the
K-Means classification. Tables I and II show the information for the Isodata
classification and Tables III and IV show the information for the K-Means classification.
However, when looking at the images there is clearly a difference between the 1986
images and the 2003 images. Figures IIa and IIb show the differences between the
Isodata images and Figures IIIa and IIIb show the differences between K-Means images.
I then decided to go into ENVI to see if there was another way to determine what
the differences where between the two images. I went under Classification and chose the
option for Post Classification and then Class Statistics. I chose the Isodata image for
1986 for the classification input file and the Isodata image for 1986 for the Statistics
Input file. I asked for the text report and then clicked OK. I then repeated the steps for
the 2003 Isodata image. This gave me two tables (Table V and VI) that listed the number
of pixels for each number for both the year 1986 and 2003. I then calculated the percent
difference between the two years (Table VII). This is not very accurate because the 1986
image only had nine classifications and the 2003 image had 10 classifications. However,
this did show that there was a difference between the two images.
Conclusion
I determined that the possible reason that there may have been problems in
finding the differences is that the 1986 image was taken from a larger image and the 2003
image was already smaller. This may have caused the images to not be geographically
linked correctly. If the images were not linked up exactly by even a few pixels this might
have caused the problem with calculating the statistics. If I had had more time to work
on the project I would have liked to have resized the 1986 image to see if that would have
made any difference when calculating the statistics.
FIGURES
Figure Ia
Figure Ib
San Antonio in 1986
San Antonio in 2003
Figure IIa
Figure IIb
Isodata image for 1986
Isodata image for 2003
Figure IIIa
Figure IIIb
K-Means image for 1986
K-Means image for 2003
TABLES
Table I
Pixel Count – Change Detection Statistics for Isodata with Maximum Iteration of 5
Table II
Percentage – Change Detection Statistics for Isodata with Maximum Iteration of 5
Table III
Pixel Count – Change Detection Statistics for K-Means with Maximum Iteration of 5
Table IV
Percentage – Change Detection Statistics for K-Means with Maximum Iteration of 5
Table V
Table VI
Table VII
% Difference For:
class 1 = 188%
class 2 = -54%
class 3 = -46%
class 4 = -3%
class 5 = 103%
class 6 = 473%
class 7 = 767%
class 8 = -75%
class 9 = -93%
REFERENCES
Remote Sensing of the Environment: An Earth Resource Perspective, John R. Jensen,
2000, Prentice Hall press
Lecture I, Dr. Hongjie Xie, 2003
Download