The Blue Lake wetlands area is a relative oasis in the Utah West Desert, surrounded by salt flats and desert vegetation. Owned by the US Air Force and managed by the state, this wetlands wildlife management area is fed by geothermal hot springs and maintains a warm temperature year round, making it popular with several species of fish, waterfowl and divers. This project involves the utilization of Clark Lab’s Idrisi Kilimanjaro geographic analysis and image processing software to process and analyze Ikonos multispectral imagery from October of 2000. These image processing techniques have enabled detailed analysis of the wetlands area as well as the water bodies within the wetlands.
Importing Ikonos Data
Isolating The Study Area
Unsupervised Classification - Cluster Analysis
Water Body Analysis
Blue Lake is located in western Utah, just south of Wendover, Nevada. The property is owned by the United States Air Force and is part of the Utah Test and Training Range. However, the Blue Lake wetland is managed by the state of Utah and is designated as a wildlife management area with open access to the public. It is a unique wetland area surrounded on one side by high desert vegetation and on the other by barren salt flats. (The Bonneville Salt Flats, where land speed records are set is roughly 25 miles to the North.) Blue Lake is frequently visited by scuba divers who enjoy the year round warm water temperatures provided by the geothermal springs which feed Blue Lake from the lake bottom. Blue Lake is also home to small mouth bass, bluegill and waterfowl, attracting fishermen and waterfowl hunters.
|Blue Lake Shoreline||Blue Lake Wetlands Area|
|Bonneville Salt Flats||Utah High Desert Vegetation|
The objective of this project will be to utilize Ikonos 2 multispectral data to isolate and analyze the Blue Lake wetlands area for vegetation and water body analysis. The Ikonos 2 dataset was provided by Mr. Sanford Moss of the Hill Air Force Base Environmental Management Division and was acquired on October 19th, 2000. Mr. Moss also helped frame the scope of this project through discussion indicating that wetland management on Air Force range property is an important aspect of the Air Force’s effort to preserve the environment in which it operates.
This Ikonos 2 multi-spectral dataset was acquired on October 19th, 2000. Included are Ikonos bands 1-4 which are Blue (444.7nm-516.0nm), Green (506.4nm-595.0nm), Red (631.9nm-697.7nm) and Very Near Infrared (757.3nm-852.7nm) respectively with 4 meter resolution. The original dataset file was in GeoTIFF format. Mr. Sanford Moss of the Hill AFB Environmental Management office exported a single GeoTiff file utilizing ESRI ArcGIS software. From the Idrisi import menu functions, I was able to import this dataset as a raster group file with bands 1-4. Bands 1-3 were utilized for a natural color overview of the image seen below.
In order to eliminate the black area image borders as well as further isolate the Blue Lake wetland area for analysis, the window operation was performed. Experimenting with the Idrisi Zoom Box feature, I was able to crop a much more suitable image on the natural color composite. This was saved as a new Idrisi image. Then, utilizing the Window feature under the REFORMAT menu, I selected the Blue Lake raster group file as the files to be reformatted and utilized the cropped Idrisi natural color composite as a reference file for the new rows and columns.
In the natural color composite, much of the terrain and vegetation cover blends together in shades of brown and red. Vegetation features in the wetlands area are not as easily distinguished from the surrounding areas. The combination of various Ikonos multispectral bands in RBG color composites, especially the VNIR band (4), can prove useful to further distinguish the study area as well as accentuate the wetlands vegetation and water bodies. I decided to make several false color composites utilizing Ikonos Band 4 (VNIR). First I made a standard false color composite combination of bands 2, 3 and 4 assigned to blue, green and red respectively. I also made a special false color composite by combining bands 1, 4 and 2 assigned to blue, green and red.
The standard false color composite does an excellent job of accentuating the vegetation in a nearly natural color because the vast majority of this vegetation exists naturally in varying shades of brown and reddish brown. Water bodies are more apparent as dark blue to black with shallower water areas appearing in shades of cyan and light blue. Blue Lake itself stands out well in this image. However, vegetation discrimination is only slightly more enhanced over the natural color image.
The special false color composite where band 4, the VNIR band is assigned to green, does a slightly better job of pulling out the actual vegetation from the surrounding landscape and gives similar distinction capabilities to water bodies as the standard false color composite.
Infrared bands are useful in accentuating vegetation in an image because of the relatively higher reflectance values of vegetation due to chlorophyll reflectance. Conversely, bands in the red area of the electromagnetic spectrum exhibit higher absorption qualities in vegetation. However, in order to further break down the wetland area as well as water bodies for analysis, additional image processing is required.
Unsupervised classification is a function whereby the reflectance values across the bands of an image dataset are analyzed and grouped according to spectral similarities. Algorithms are utilized by Idrisi Kilimanjaro to break down image band pixels by relevance to group into like categories. Several iterations occur within this analysis of grouping and regrouping until a final image is produced with maximum likelihood classifications of the clusters.
I first performed the CLUSTER function, under Image Processing / Hard Classifiers, and selected to retain all clusters. The resultant image had 42 clusters identified. Utilizing HISTO under GIS Analysis / Database Query, I produced a histogram for this cluster analysis and determined the point on the slope where there is the biggest drop off occurs is after 15 clusters.
I then re-performed the CLUSTER on the dataset. By cross referencing my false color composites with the resultant 15 cluster image, I was able to identify the following clusters as being primarily wetlands vegetation areas: 2,5,7,and 12. Utilizing the RECLASS function under GIS Analysis / Database Query, I was able to produce a Boolean image which approximately isolates the Blue Lake wetlands area of interest. Using the AREA function under GIS Analysis / Database Query, I calculated the Blue Lake wetland area to encompass approximately 18.49 km2.
My next objective in this project was to isolate the water bodies in this image for a surface area analysis. This information could be useful for recreational activity and wildlife management as Blue Lake is frequented by scuba divers, fishermen and waterfowl. Unfortunately, the cluster analysis managed to lump together several of the surface water areas with other land classifications, so I needed another method.
I attempted several different overlays with mixed result before I settled on the Overlay function under GIS Analysis / Database Query represented by the formula (First-Second)/(First+Second). Since I was using Bands 4 and 3, this overlay also represented the NDVI function. The Normalized Difference Vegetation Index is used to depict the relative amount of vegetation biomass in an image by establishing a ratio of Infrared and Red radiant flux. This ratio can also be used to identify water bodies as the resultant image, which depicts water bodies with lower values than vegetation and surrounding land. This is due to the fact that water absorbs high amounts of both infrared and red radiant flux.
In order to isolate the water bodies for a surface area analysis, I created a Boolean image by using the RECLASS function under GIS Analysis / Database Query. I established a cutoff value for water bodies thereby grouping them together and then grouped the remaining non-water values together. Using the AREA function again, I was able to estimate the water body surface area at .98 km2.
In both the unsupervised cluster analysis and NDVI reclassification operations, the targeted areas were well isolated with only a minor amount of outlaying pixels that did not meet exactly the land classification parameters I had established. By cross referencing false color composite images, personal observations and ground photos my situational awareness of the study area was greatly enhanced. This awareness was essential in both the cluster analysis and NDVI overlay as it enabled me to refine input parameters to the analysis process until I was able to produce a product that I believed was an accurate representation of the wetlands area.