Classification of Land Use in the Salton Sea Area
Abstract Introduction Image Manipulation Analysis Summary References

Abstract

Located in both Imperial and Riverside Counties, the Salton Sea is California’s largest body of water. Created by a rupture in a Colorado River levee in 1905 the sea is located nearly 227 feet below sea level and has been recorded to be as deep as 51 feet in some areas. The Salton Sea is located within the Colorado Desert Region and receives little annual precipitation. Recharge occurs mainly through agricultural runoff from the surrounding Coachella, Imperial and Mexicali Valleys.

Using a dataset provided by Landsat.org the area was classified into five different land uses using Idrisi Andes. Composite images, vegetation indices, ISOCLUST, masks, and overlays were used in image compilation. Area analysis and histograms were created for calculation of land area for each cluster. The land uses in order from smallest land area to largest land area are as follows: non-active vegetation, dirt with no vegetation, active vegetation, water bodies, and urban areas. The following table shows the area of land in square kilometers for each land use.


Introduction

This project is based on a Landsat TM dataset acquired on May 3, 2000. The subscene is centered on the Salton Sea, located in both Imperial and Riverside Counties, California. Located in the Salton Trough The Salton sea was created by a Colorado River level break in 1905. The rupture was so great that it was not fixed in 1907. By that time the water had settled into the Salton Sink, or Salton Trough, creating what is now known as the Salton Sea. Today it is located nearly 227 feet below sea level, and is 51 feet deep in some areas.

The Salton Sea is located within the Colorado Desert Region. The climate in this area is characterized by low precipitation and high summertime temperatures. Despite the absence of recharge from percolation of rain water the Salton Sea is the still largest lake in California. Recharge to the is mainly from agricultural runoff from the surrounding Coachella, Imperial and Mexicali Valleys creating an overabundance of salts, nitrates and phosphates in the water.

The Coachella and Imperial Valleys surround the Salton Sea and are known for their agriculture. In 2000, over 580,000 acres of Imperial Valley land was farmed to produce over $900 million in field, vegetable and permanent crops. The largest crop produced is the date. The Coachella Valley produces 95 percent of all dates grown in the United States.

The Salton Sea is located in Southern California. It is located in both Riverside and Imperial Counties.

.

Spectral bands 1-5 and 7 of Landsat Thematic Mapper the were selected for this activity. The Landsat Thematic Mapper sensor uses multispectral scanning to achieve sharp spectral separation and resolution. The TM data are scanned simultaneously in seven spectral bands. Bands 1-5 and 7 have a spatial resolution of 30 m (100 feet). The thermal band (6) has a spatial resolution of 120 m. The spatial resolution for each of the bands is described in the table below.

Wavelength Resolution (micrometers)
Band 10.45 - 0.52
Band 20.52 - 0.60
Band 30.63 - 0.69
Band 40.76 - 0.90
Band 51.55 - 1.75
Band 610.40 - 12.50
Band 72.08 - 2.35

Using Idrisi Andes and the March 3, 2000 dataset I will delineate between water, agricultural, urban, and natural land uses. I will show different methods used for classification and comment about the effectiveness of each.

Back to Top


Image Manipulation

Obtaining Data

Datasets are widely available but can be expensive. The Landsat.org website provides many orthorectified datasets at no cost. Files are downloaded in zipped files which first must be extracted.

Once extracted, the images need to be imported in the correct format. To do this you go to file, Import, Government/Data provider formats, GEOTIFF/TIFF. A wizard will pop up and you indicate the bands that you want to be imported.

Composites

A composite is an image that combines three bands from the dataset to create an image. Composites can be made to look natural or to highlight a particular feature, such as vegetation. These images are called natural color composites and false color composites, respectively. Composites provide good reference points for data analysis.

A composite image using bands 1, 2, and 3 will provide a natural color image using the visible bands.

Note that the image appears much like an aerial photograph.

.

A composite image using bands 2, 3, and 4 will provide a false color image that is useful in delineating vegetation and other land uses.

Note how this image looks similar to infrared photography.

Another type of false color composite uses bands like before and one band that is a ratio of two additional bands. Band ratio images combine A composite image using bands 2, 4, and the 3/1 ratio provides a dramatic contrast in the various land types.

To create a band ratio false color composite you must first create a band ratio using the OVERLAY module using band 3 for the first image and band 1 for the second image and selecting “first/second” provides a satisfactory band ratio image to use in the composite.

The resulting false color composite looks like the image to the left.

Notice the difference in coloration between vegetation and other land uses.

Creating a Vegetation Index

Agriculture is a key land use in the Imperial and Coachella Valleys. The vegetation index uses infrared and near infrared bands to highlight the agriculture use in an area. Using bands 3 (red) and 4 (near infrared) a in the VEGINDEX produces and image that provided different values for different land covers. NDVI (normalized difference vegetation index) is traditionally used for -----. For storage space reasons it is customary to convert raw NDVI values into a byte-binary scale.

The NDVI image to the left has been converted into a byte-binary scale.

Cluster

The CLUSTER module is a basic technique used to classify cells that have similar numerical values in all bands of the dataset. When this module is used Idrisi Andes recognizes natural groupings based on band values. Idrisi Andes may assign multiple values to a single dataset. This can be manually corrected by comparing the clustered image to previous composites and reclassifying the values to the desired number of classifications.

Clustered image before reclassification
Clustered image after reclassification

ISOCLUST

The ISOCLUST module is a hard classifier image processing technique. This module uses a “self organizing cluster analysis” to classify raw bands with the user indicating the number of clusters to process. After the bands are entered a histogram showing the frequency of clusters will appear. It is typical for most of the data in TM datasets to be located on the left side of the histogram in clusters 1-16. The processing time can be lengthy (in this particular case, two hours!). The ISOCLUST can then be reclassified to adjust for further analysis.

The ISOCLUST image as clustered by Idrisi Andes based on natural breaks is shown to the left.

.

The reclassified ISOCLUST image using five different land uses: water bodies, active vegetation, non-active vegetation, urban land, and dirt.
A close-up view of the reclassified ISOCLUST image of the Coachella Valley

Masks

To make a mask you must isolate one area of your image. Isolate your image by reclassifying your image into a boolean image. To do this you must reclassify the area you want isolated as 1, and everything else as 0. Do this by using the RECLASS or ASSIGN modules.

A mask of the Salton Sea is located to the left

Multiplying a mask and an NDVI image using the OVERLAY module will allow you to determine the status of vegetation within a certain area.

Back to Top


Analysis

The area of each cluster from the reclassified ISOCLUST image can be calculated with the AREA function. The following areas were calculated in square kilometers for the Salton Sea image.

Category Square Kilometers
Water bodies977
Active Vegetation8718
Non-active Vegetation12521
Urban Areas732
Dirt10764

Based on my classification of Active and Non-active vegetation you would expect to see higher NDVI values in the active vegetation than in the non-active vegetation class. To determine the accuracy of the classification schemes the status of vegetation within the active and non-active classification the NDVI image was multiplied with each of the masks to yield the results in the table below.

The active vegetation cluster had a mean NDVI value of 203.729 with a standard deviation of 4.34.

.

All values were below the minimum value for display of 200, therfore the minimum was set at the highest number available(176).

The non-active vegetation cluster had a mean NDVI value of 176 with a standard deviation of 0.

Back to Top


Summary

Based on close inspection of natural and infrared composites of the May 3, 2000 Landsat TM satellite imagery there were 6 major classification of land uses in the Salton Sea area. The classifications in order from largest to smallest are: non-active vegetation, dirt with no vegetation, active vegetation, water bodies, and urban areas. Comparing the vegetation index of the active vegetation with the non-active vegetation it appears that the types of vegetation were accurately classified.

Back to Top


References

Aber, James A.. "Advanced Image Processing Laboratory Exercises." ES 775 . Emporia State University, Spring 2008.http://academic.emporia.edu/aberjame/es775/syllabus.htm

"Agriculture and the Sea." Salton Sea Restoration: Ten Years of Progress. Salton Sea Authority. http://www.saltonsea.ca.gov/ltnav/why_agriculture.html.

"Historical Chronology" Salton Sea Restoration: Ten Years of Progress. Salton Sea Authority. http://www.saltonsea.ca.gov/about/history.htm.

"Landsat Thematic Mapper Data (TM)." Earth Resources Observation and Science (EROS). 22 Aug 2006. United States Geological Survey. http://edc.usgs.gov/guides/landsat_tm.html.

NASA Landsat Program, 2000, Landsat TM scene elp039r037_7t20000503, SLC-Off, USGS, May 5, 2008.

Back to Top


Created by Susan June for ES775

May 8, 2008

Emporia State University, Department of Earth Sciences