1. Classify existing riparian vegetation types into accurate and functional classes of vegetation.
Although riparian areas make up only about 2 percent of the world's terrestrial ecosystems, arguably they may be one of the most important systems in the biosphere. These areas have some of the highest rates of biological production and biodiversity, often taking on the role of sediment and nutrient sinks. They also play a minor role in flood control, and provide habitat for a wide range of species. Despite their importance and the wide range of roles they play within the environment these areas are being degraded and destroyed under the increasing pressures of human uses. (Hawkins, 1994) With growing concern about the condition of riparian areas throughout the world it is important that functional information concerning riparian vegetation types is available to researchers. Such information could be used to quantify remaining riparian areas as well as their loss and perhaps help control the degradation of these important ecosystems.
On a more specific regional level the riparian areas throughout Cache county are also threatened. With only a preliminary study of the vegetation types that make up these riparian zones it became apparent that little original vegetation still exists along the waterways in the lower lands of Cache county. Today there is little known about the vegetation that historically occured along these streams in the valley because of the changes brought about by irrigation, grazing, high rates of erosion, and urban development. With these changes taking place in the valley it is only reasonable to assume that riparian areas throughout the county are threatened to some degree and in need of analysis and possible future protection. With these concerns in mind we decided that a complete vegetation classification of all of the riparian areas in Cache county could be a useful asset to local land managers.
This paper describes the methods used to create a vegetation classification with these specific objectives in mind:
1. Classify existing riparian vegetation types into accurate and functional classes of vegetation.
2. Determine which vegetation types are dominant in riparian communities at varying elevations.
3. Become familiar with Erdas imagine software
Rectification
To begin our project it was necessary to see the overall distribution of water courses in Cache county. We accomplished this by overlaying the hydrological vector file "cache_hydro" over the Landsat raster image of Cache county "cache_tm.img". For a geographic reference tool, we also overlayed the vector file "utahrd_ca" on the cache_tm image which displayed the roads and trails of Cache county. With both roads and water courses displayed over the cache_tm image we noticed that the vectors did not coincide with the hydrological and road reflectance features of the cache_tm image.
To alleviate the problem we first attempted to move the vector layer by pixels(30 meters) from the upper left 'X' and upper left 'Y' values in the map information section to allign with the hydrological features of the cache_tm image. In doing so, we discovered that this was not a linear error. We therefore used the rectification process to geometrically correct the cache_tm image so that it would correlate to the vector layers. The particular type of rectification that we used involved image-to-image transformation (Registration). A source image (to be rectified) and a destination image (the vector layers) were used to initiate this process. We found that it was necessary to have a raster layer on top of the vector to begin using the GCP editor. We then overlayed the raster image as an invisible layer so we could view the vectors to help us locate GCPs.
We used the GCP editor to locate points on the destination image and match these points to corresponding locations on the source image. The transform editor was used in conjunction with the GCP Editor to quantify the statistical errors for registration. The GCP editor quantified the data values by forming a polynomial equation. Because the image was not linear, we set the Transformation Editor to 2nd order to compensate for geographical skewness. When using a second transformation order, we needed to assign six points to each image so the Transform Editor could formulate the polynomial equation. The GCP editor then predicted the location of designated GCPs on our preselected "source" image by using the formulated equation. When the predicted points were accurately located by the computer, we determined that we had enough points designated on the image to initiate rectification. The root mean square (RMS) error was also useful in the rectification process to determine if we had located a sufficient number of points on the image. The RMS error described the distance between the desired output coordinate and the actual output coordinate for the same point. Our final RMS was approximately 33.2. meters.
Unrectified image
Rectified image
The final procedure was Resampling which extrapolated the data values from the destination image and converted them to the source image. We used the Nearest Neighbor method in Resampling to assign the transformed data to the closest pixel values of the source image. Because the image and the vector were the same size, there was no overlap "stairstep" effects. Our final image was successfully rectified.
Buffering
Because riparian vegetation occurs within a limited area to either side of a waterway we were primarily interested in only that information on the Cache_ tm image which occured along the streams and rivers throughout the image. We determined that a 90 meter swath along each bank would include all appropriate vegetation. Our goal, then, was to discard all parts of the image that did not fit this 90 meter criteria. The purpose of this step was to isolate our study area, including only the relevant spectral and spatial information in order to more efficiently assess the vegetation in those areas.
We first explored the possibilities of creating these buffer zones in Imagine but determined that no viable method for this type of project was within its capabilities. We therefore, had to use ARC/INFO in the first steps of the process, later importing our work back into Imagine to complete the extraction of the buffered zones from the TM image. Our lack of experience with ARC/INFO led us to utilize Thad Tilton as consultant and ARC expert extraordinaire.
Within ARC, we utilized the buffer command on the vector image containing all waterways in Cache county. The buffer command allowed us to create buffered zones 180 meters wide following the pathes of the Cache_hydro riverways. For the next task of transposing the buffer image into a grid coverage we used the Polygrid command. Polygrid transformed the image to various polygons, assigning each a value. Our first attempt was unsuccessful because the polygons created by a circle of buffers were assigned values equal to the buffer zone polygons. ARC/INFO categorized the buffered polygons as '100' in the "INSIDE" attribute table characteristic, while the remainder of the image was '0'. We were able to isolate the incorrectly assigned polygons and rectify the problem by making sure all of the polygons outside the buffer zones were assigned a '0'. This finally resulted in a grid image of a 90 meter buffered zone on either side of the Cache_hydro vectors.
Buffer from grid
Buffered TM image
The next step was to import this grid coverage into Imagine using the Import command, which converted the grid into a usable image within Imagine. We then began the Mask process to cut out the buffered zones from our rectified Cache_TM image. The trick with this process was to make sure the projections of both the buffered image and the Landsat image were equal. Success in this process resulted in an image of 180 meter swaths along the county's waterways containing spectral information from the TM image.
Classification
After adding buffers to a GIS overlay of the river systems in Cache county and converting this back to an image created from Landsat TM data, our next step was classification. We decided to use an unsupervised classification to automatically categorize the riparian zone into separate spectral classes. This method proved to accurately denote the specific vegetation habitats after groundtruthing the area. Seven classes were chosen as an appropriate division of spectral values for the different vegetation classes found. This was determined by trial and error. For example, when using only four classes as a basis for unsupervised classification, the spectral value breakdown was too broad.
Five Classes
Seven Classes
During the classification process we came to the realization that there would be a problem with agricultural fields in the valley having similar spectral signatures to those of forest vegetation types in the mountains. Had we decided to classify the entire county with only one set of vegetative classes this would give the false impression that there were large tracks of coniferous and deciduous trees growing in the lowlands of Cache county. In order to overcome this obstacle we decided to reclassify the buffered water systems in the valley separate from the buffered drainages in the mountains using AOI(Area of Interest) commands. After applying unsupervised classification again to the valley AOI, we realized that many of the buffered areas in the valley were created from agricultural ditches or canals and were not representative of riparian vegetation. To alleviate this problem supervised classification was necessary to isolate the areas of true riparian vegetation.
The spectral areas we focused on included cottonwood/willow zones, as well as water zones and wet grasses. Training sites were located in the image for each of these classes and selected using the seed properties magnifier and a euclidean distance of 30. This value was selected by trial and error of the range of spectral values that were grouped. When satisfied with the categorization of each area we added them to a signature editor. In the signature editor we looked at the statistical information and spectral signatures of each class and compared them with the spectral signatures from similar classes from the unsupervised classified image. Because they had similar signatures, we concluded they were the same vegetation type.
Euclidean Distance 15 Euclidean Distance 30
For another comparison of spectral groupings we created an AOI of the mountains in Cache county excluding the valley vegetation. We reclassified this area using unsupervised classification to observe the changes in the spectral signatures. As suspected, very noticeable changes occurred. For example, the water/shadow class dramatically increased in shadow spectral values. We determined that the reason for this was because there was a lot of water in the valley and very little water in the mountains. Therefore, after cutting out the valley section, the computer needed to assign a larger range of spectral values to the water/shadow class to compensate for this lack of water. We concluded that our first classification for the entire image was more representative of true vegetative types than the mountain classification alone. Therefore, we overlayed the classified (supervised) AOI image of the valley onto the classified (unsupervised) image of Cache county using the Mask utility. In the Setting Recode option in Mask we were able to code similar valley and mountain classes as the same class, e.g. water, cottonwood/willow, riparian grasses. The resulting image was made up of eight classes, the seven original and the additional "other valley vegetation."
Elevation data
Our last procedure was for the pupose of quantifying the area of various riparian vegetation in different elevation zones. We aquired an elevation image from Steve Creek which consisted of 34 classes of 50 meter intervals beginning at 1339 meters and ending at 3001 meters, the elevation range of Cache County.
We then used the Mask utility to combine our classification data with this elevation data. The result was an image with 272 combinations merging the 34 elevation zones and the 8 vegetation classes. As you might imagine this image was very difficult to decipher. We attempted to color code it to create a more useful image but were unsuccessful with this process. Instead we utilized the Summary tool under the GIS Analysis option in the Interpreter menu to generate the statistical data of this merge. Using this summary report, we were able to analyze elevation characteristics that were associated with the different elevation classes more easily. To simplify the data even further, we grouped the elevation classes into four categories consisting of the following:
-Valley Riparian 1339-1500m
-Low Elevation Riparian 1500-1800m
-Mid Elevation Riparian 1800-2550m
-Alpine Riparian 2550-3001m
We choose to tabulate data for the conifers, cottonwoods/willows/aspen, and riparian grass classes only since our these are primarily the classes of true riparian vegetation and this was our focus.
As expected we found that there was a strong correlation between elevation zones impacted the composition and quantity of riparian vegetation
Valley riparian (1339-1550m)
-Conifers 2.61 ha
-Cottonwood/Willow 2557.98 ha
-Riparian Grasses 282.69 ha
Low Elevation Riparian (1500-1800m)
-Conifers 843.31 ha
-Cottonwood/Willow/Aspen 1218.87 ha
-Riparian Grasses 1310.49 ha
Mid Elevation Riparian (1800-2500m)
-Conifers 5653.89 ha
-Cottonwood/Willow/Aspen 1820.88 ha
-Grasses 4019.58 ha
Alpine Riparian (2550-3001m)
-Conifers 267.93 ha
-Cottonwood/Willow/Aspen 66.62 ha
-Grasses 203.76 ha
With the close of our project we have gained valuable information pertianing to the process of vegetation classification using remotely sensed data. As with every project of this type there were inital problems with the use of unfamiliar processes and software. Much of the time spent with our project involved trial and error as we tried to use specific processes to overcome obstacles associated with classification or riparian vegetation. Thad Tilton and the other teaching assistants rapidly became priceless sources of information as our procedures became more complicated.
There were some problems associated with our project that are typical when dealing with vegetation classification. One of these was the genralization required in order to seperate riparian vegetation into a relatively small number of classes. In such a variable habitat it would be hard to accurately place vegetation types into eight classes without lumping some together. At the same time to divide vegetation into an infinite number of classes would make that classification impossible to accurately ground truth.
Another minor inconsistancy in our classification is a result of the 90 meter buffer zone along the streams that we used as our study area. Along some of the larger drainage areas in the county a ninety meter classification on either side of the stream is accurate. However on some ephemeral or annual first order streams a buffer area this size would include vegetation that is not necessarily associated with riparian areas. Here again was a neccessary generalization. Not all streams in the county would have this wide of riparian area in association with them. In order to be inclusive, in classifying all riparian areas regardless of size, this large of an area had to be used.
When it came time to overlay our seperate moutain and valley classes, that would display our entire classification of the county, we noticed another problem. The transition from the mountain classes of vegetation to the valley classes was very abrubt. This is a result of cutting the valley out as an AOI and classifying it seperately. When you cut an area out as an AOI that has different colors representing each vegetation class these colors do not depict the accurate inferface of these seperate types of vegetation. This gives the viewer the incorrect impression that at a certain area between the valley and the mountains the vegetation suddenly become very different. This is a necessary side effect of classifying these two associated areas as if they were seperate regions.
Despite these minor setbacks our project ended with successful classification of riparian vegetation. From this we were also able to provide information on vegetative class dominance in association with varying elevations. Our analysis demonstrates a general evaluation of the kind of information that can be attained from remotely sensed imagery. With this classification complete it will be possible for researchers to use this information for habitat, grazing impact, or erosion studies throughout Cache county. From such studies, necessary information will be available for effective management of these fragile ecosystems. With newly informed land managers local riparian areas could be spared further degradation.