Remote Sensing Winter Wheat of 1988, 1992, & 1996

by

Jon Vopata

ES 771 Remote Sensing, Emporia State University


Abstract
Introduction
Importing Imagery
234 Composites
NDVI Composites
Using Isocluster
Calculating Area
Conclusion


Abstract
Remote sensing is the process of obtaining information about an object or feature from a distance. Satellite images are often used for remote sensing of the earth’s surface. Satellite images of the earth can be used to solve a variety of problems relating to an array of disciplines. Software such as Idrisi32 is a tool that helps manipulate satellite images for better analysis. This is a project of remote sensing satellite images to study Kansas winter wheat production in 1988, 1992, and 1996.

Return to the top


Introduction

This project is an attempt to compare the winter wheat production in the spring of 1988, 1992, and 1996 in a small area of Wabaunsee County, Kansas. The study area includes the communities of McFarland, Newbury, and Paxico. The analysis is based on satellite imagery from path 28 and row 33, taken by Landsat 5 and processed with Idrisi32 software. The images used were taken on April 27, 1988, April 22, 1992, and April 17, 1996.

Return to the top


Importing Imagery

The first step in this project was to use the BILIDRISI general conversion tool to import the imagery into Idrisi. The imagery consisted of seven bands and covered most of east central Kansas. These images occupied a vast amount of computer memory and needed resized to my individual study area. I used Idrisi’s ‘window’ to resize the images to my specified study area. ‘Resample’ was also used to insure that the seven image bands from all three years could overlap perfectly with the same number of rows and columns and the same x-min, x-max, y-min, and y-max values. One way of checking the accuracy of the resampled images was to create a multi-temporal image using one single band from the three different years. For example, I used band 4 from 88, 92, and 96 to create this composite image.

click on the image to enlarge (354kb)

Although the colors of the image may appear unusual, there are no signs of a double exposure effect. The colors of this image give a lot of information about the areas of winter wheat for 1988, 1992, and 1996. In this multi-temporal composite 1988 is assigned to the blue image band, 1992 is assigned to the green image band, and 1996 is assigned to the red image band. Thus, the color blue indicates fields where winter wheat was only grown in 1988. Green indicates fields where winter wheat was grown only in 1992. Red indicates fields where winter wheat was grown only in 1996 and white or light gray areas indicate where winter wheat was grown in 1988, 1992, and 1996.

Return to the top


234 Composites

After ensuring that all the imagery was projected correctly and all bands were windowed and resampled, I could begin creating false-color composite images of the study area. Here is one of the first composites created. It is a false-color composite of April 27, 1988. Band 2 is the blue image band, band 3 is the green image band, and band 4 is the red image band. Using these band designations, the winter wheat is easily discernible by its red color.

click on the image to enlarge (378kb)

Click here for a larger 1988 (234) composite image overlaid by a digital raster graphic to give a better perception of the topography and location of the study area.

Here are the 234 composites for each year side by side. One can easily see the differences in active winter wheat from 1988 to 1992 to 1996.

1988                                  1992                                  1996
click on the images to enlarge

Return to the top


NDVI Composites

Next I wanted to make more composite images consisting of the haze corrected bands 2 and 5 and a NDVI (normalized difference vegetation index) image. Then I could compare the new haze corrected NDVI composite with the 234 composites. The first step in doing this was to perform haze correction for all bands of each year. This was done by using the image calculator to make the minimum value of each band equal to one. Next, I used the image calculator to create the NDVI image with this equation: (band 4-band 3)/(band 4+band 3) = NDVI. Then I rescaled the values of my NDVI images and converted them into byte-binary format. Finally I constructed the false color composite with band 2 as the blue image band, the NDVI image as the green image band, and band 5 as the red image band. Here are the 2NDVI5 composites for each year side by side.

1988                                  1992                                  1996
click on the images to enlarge

As one can see, these composite images are similar to the 234 composites. The only major difference is the color of the wheat and the surrounding land. In the new NDVI composites the wheat appears as bright green while the surrounding landscape is purple.

Return to the top


Using Isocluster

More than just visual analysis I wanted to utilize Idrisi’s software capabilities to find quantitative changes in the winter wheat crops of 1988, 1992, and 1996. I used the haze-corrected bands 3, 4, and 5 to create 8-bit composites of 1988, 1992, and 1996. I then used the 345 composites to run Idrisi’s cluster and isocluster. For cluster I accepted the default settings and assigned the 345 composite as the composite file. For isocluster, I assigned bands 1-5 & 7 as the files to be classified, the 345 composite as the image for seeding, 3 as the number of iterations, and ten as the number of clusters. I found that isocluster produced far better results than cluster. So, I then discarded the cluster results and decided to focus on the isocluster images.

Here is the isocluster image for 1988.
click on the image to enlarge (732kb)

Comparing this image to the 1988 (234) composite it is apparent that cluster 10 correlates with active winter wheat growth. I also found this to be true for the 1992 isocluster image and the 1996 isocluster image.

Return to the top


Calculating Area

Now with the isocluster images, I used reclass to separate the 10 cluster from all other clusters. This was done in order to designate the active wheat fields from the surrounding landscape. Doing this for each year, I created the following images.

1988                                  1992                                  1996
click on the images to enlarge

As one can see, the isoclustered and reclassed images are not exactly accurate at distinguishing wheat from all other vegetation, especially the 1988 image, but this is good enough for a general interpretation.

Next, with these reclassed images I used Idrisi’s “area” function to determine how many square kilometers were covered by winter wheat in each year. Here is a graph depicting the results from the tabular area calculation.

Return to the top


Conclusion

Using satellite images with the aid of Idrisi software I was able to create numerous composite images of my study area. These images enhanced the winter wheat fields and allowed me to do some analysis of the wheat production of 1988, 1992, and 1996. I found that there seems to be a slight decrease in winter wheat production from 1988 to 1992 to 1996. I was also able to distinguish which fields contained winter wheat and in what years the wheat was present. I also realized that few fields contained winter wheat in all three years.

Return to the top


References

Aber, J.S. Remote Sensing. ES771 Remote Sensing. Emporia State Universiy. 2001
http://academic.emporia.edu/aberjame/remote/remote.htm

Clark Labs. Geographic Analysis and Image Processing Software. http://www.clarklabs.org/


This web page was created to fulfill the requirements for Remote Sensing Class at Emporia State University.
Return to student webpages.
This page was created on 12/13/02.