Final Report

Introduction

There is increasing recognition that sensible use of finite and shrinking resources requires comprehensive planning to coordinate the placement of private and public facilities with amounts and locations of resources. Uncoordinated development can lead to inefficient use of these resources and undesirable environmental consequences. It also can lead to selection of sites of marginal suitability, or cause displacement of activities from their optimum sites. Coordinated development, on the other hand, must outline the kinds of land use patterns in specific sites, and favored locations for specific uses (Campbell,1987). Our research project was designed to compare the similarities and differences of two coverages of the same area from different sources. These Coverages deal primarily with land use. One source was the Thematic Mapper Data while the other was the Utah Department of Water Resources Land Cover database. We also set goals to learn to use the majority of the software of the computer system. We intended to also use the GPS in our research to help with the accuracy of the results.

Methodology and Discussion

In our research we used several types of software. For the most part, we used software developed by ESRI of Redlands, California. These included the following: Arc, Info, ArcEdit, ArcPlot, ArcTools, & ArcView2. Some additional software prototypes we used included XV and NCSA Mosaic for viewing what's on the World Wide Web as well as the .gif files. The software created by ESRI is designed specifically for map design and studies dealing with databases that are correlated and connected to a coverage or map.

Some of the coverages we chose to create were made inside of ArcPlot. We used data collected by the Utah Department of Water Resources to make this particular map. The data were collected by aerial photo's as well as by ground observations for the entire state of Utah; however, we simply clipped a section from this map to study the area in and around Cache Valley. This was a subset of the Bear River Hydrologic Basin. The Department of Water Resources Data is divided up into 11 Hydrologic Basins for the state with additional sub-sections. When we clipped the data we used the Clip command in Arc to cut from the Bear River Hydrologic Basin the polygons, arcs, and points found within Cache County. The county coverage was copied from Dr. Doug Ramsey's database.

Once we had the data clipped, we then categorized the data into general areas of interest. We found that the data would be best used if we categorized it into the following fields:

1) Riparian (Colored Purple on the Map)

2) Commercial (Grey)

3) Alfalfa (Dark Green)

4) Pasture (Brown)

5) Water Areas (Blue)

6) Residential (Red)

7) Crops (Light Green)

8) Public Areas (Light Beige)

After these categories had been programmed into Info, we used them to make the map. When the data was copied from the Department of Water Resources, there were codes already placed into the database referred to as Landcov codes (Land Cover). This code was classified into those areas which were Irrigated, Non-Irrigated, Woodlands (Empty), Water Areas, Barren Lands, & Altered Natural Environment (Built-Up Land), etc. One such code is Alfalfa (Surface Irrigated); its code is IA3a. We generalized these codes into the general categories listed above. We initially intended to create categories for some types of woodlands, but realized that the data was not present for this type of categorization. We made do with what was there.

Following the categorization process, we created an AML, or Arc Macro Language file. This is a file with the ".aml" extension on it. It is simply a file that contains a list of commands within it. After starting the ArcPlot software with display present, one issues the command in the following way if the file name is "george.aml". - &r george. The & is a signal to the computer that you're issuing a command that it will recognize and retrieve from another place in the database. The r means run. The file name is listed after stating that the thing to be run is the file "george.aml." It is automatically defaulted that the file must have the .aml extension following the name. If it does not have the .aml extension the file cannot be run as an Arc Macro Language file. Our AML file is entitled cache_lu.aml; Cache, represents the county of study. The underscore shows a differentiation between the next two letters lu and Cache. These two letters stand for land use. Our cache AML file is as follows:

This aml has been altered a number of times to make sure that it is going to work every time its run. In the aml file there is a reference to a file known as "cache_lu.aml". This file is a coverage of the roads in Cache County. We, however, did not want all the roads for the county, so we therefore edited this coverage inside of ArcTools by selecting large sections of the areas we did not want and then summarily deleted them. This left behind the area of Cache Valley in which we were most interested. We had to run the normal build and clean commands on the coverage so that it would have line topology afterwards. These also were executed inside of ArcTools under command tools.

Another command of importance in the aml is the reference to the Legend. We referenced a file which told the AML to order our legend box categories in a certain order. This file is quite simple. We typed the number of the color we wanted in the box like : ".5". It is mandatory that the number be preceded by a decimal so that the computer will recognize it as a color number and simply not text that is to be placed in the legend box. After hitting return, the text that is to be place next to the colored box is entered. So the file should look something like this:

Two final command that are essential to the aml are quite interesting. The first is the shadeset command. This command tells the computer which shadeset we are going to use to color the polygons in the coverage. We decided to create our own shadeset inside of ArcPlot by typing Shadeedit and the ArcPlot prompt. This command brings up a window that has a custom button on it. By hitting this button, one can create the specific colors and layers of every box/polygon type found in the legend that is to be colored. This is a good way to get exactly what you want. We decided to create our own shadeset. Similar to the AML file, this file must have the ".shd" extension on it for the computer to recognize it for what it is, a shadeset description. We called our shadeset cache_lu.shd, for the same reasons stated above. The command that follows this command actually does the coloring of the polygons of the map once all the general parameters are set in Info. This command, polygonshade tells the computer to color the polygons of any given item in the Info database according to the shadeset description. Generally, a color item is created in Info with numerical data to tell the computer which type is to be colored what. We created two additional items in the Info database after receiving the coverage and supplemental database.

GPS System

The GPS part of our research didn't work out exactly as we planned due to some unexpected troubles that cropped up in the middle of our research. We used a GPS Pathfinder, by Trimble in Sunnyvale, California. Dr. Ramsey let us borrow it on April 26th, 1995. We decided to start at a place that would be easy to recognize with the Thematic Mapper classifications as well as the Land Use GIS coverage as well. We started at the crossroads of the airport runway just north of Logan. In this way we would know if our points were really accurate or if they were way off. Our second point was the grassy fields right next to the airport runway to the west. These two GPS points would have different reflectance values. The third site we chose was the pasture lands just north of the Logan Settling Ponds. This was a field in the lowland areas of Cache Valley where the grass was pretty low do to intensive grazing. The fourth point we selected was the Logan cemetery up near the northeast end of Utah State University. We figured we could get values showing the coniferous and deciduous trees in the cemetery to help classify the data in the program Imagine 8.2. Following this, we did our fifth point in the gravel pits just north of the cemetery. There has been quite an amount of growth in this area recently. This could have some affect on the values that we get in the data from the TM as opposed to what exists there now. Our final point was a grain field in North Logan. We took photo's of all the sites where we collected GPS points and made it a point to make sure that the GPS had found 4 "SV's", or satellites, before we recorded the point and moved on to the next site.

On the way back to the university that night, we monitored our progression along the road to make sure that the GPS was actually doing what we thought it was doing. It looked like everything was correct. As we headed south towards the university, the values of the Y coordinate were decreasing as we went. We placed the sensor on top of my sunroof and held the GPS computer system in the car as we went along. We also made a mental note that the X coordinate remained about the same as we went along. This would seem to make sense as our course down the road was pretty much due south. We were going to use the GPS coordinates to help improve accuracy levels while analyzing the data for comparison between the TM image and the Land Use data from the Utah Department of Water Resources. After the coordinates were collected we down-loaded them with the PFinder Program in the GIS Student Laboratory on the IBM. From here, we could look at the respective points and there relative positions in relation to one another without having the TM image or Land Use coverage as a background coverage. From this point, the points seemed to be accurate and good. We then recorded the points manually from the statistics tables generated on PFinder and used them to find coordinates inside the TM Image. The unearthly thing about this was that when we plugged the points into the program Imagine, the points were consistently to the west of where they were supposed to be. Additionally, they were more than 30 meters off, the general error range caused by Selective Availability is around 30 meters. We had averages the points for all the GPS collection sites. When the mean was taken, the point was supposed to be relatively accurate, within 6 - 10 meters. We're not sure as to the reason for this. We have a feeling it could be caused from the wrong coordinate system being used when recording the GPS points. Dr. Ramsey did get the Selective Availability for us from Hill Air Force Base, Utah. Yet, these data files were only given for certain days, one of which was not April 26th, 1995. The nearest one was April 27th, 1995.

Arcview

We have been creating a map of the Cache Land Use data in all of the software programs available. We created one in Imagine; one in ArcPLot; and now we've also created one in ArcView2 entitled "pasture.apr". In ArcView, we pulled up all the coverages that were made available to us through Dr. Ramsey's database. We decided we were going to use a number of these coverages to help with the analysis of our data. In the end, we actually only ended up using the roads coverage, river coverage, and the data that we copied from the Utah Department of Water Resources for the Land Use Coverage.

Imagine 8.2

We used the software package Imagine quite extensively to view our various remotely sensed images. For our project, we were hoping to use a black and white Spot image with ten meter resolution and combine this with a TM color image with thirty meter resolution to create a ten meter resolution color image to use in our land cover analysis. Unfortunately, every attempt we made at this endeavor crashed. We began this process by using the AOI (Area of Interest) feature contained inside Imagine. We used the AOI to cut a small section of Cache Valley from the Cache County TM and Spot images. The reason we did this was to make our data we were trying to combine smaller so that it would be more manageable later when we would be classifying the data. For some reason which we do not know, we were unable to combine the two AOI images using Imagine functions. Because of the time factor, we chose to eliminate the black and white Spot image and AOI from our project. We continued our project of comparing land cover data with the TM data and the land cover data from the Division of Water resources. To do this, we imported are vector ARC coverage into Imagine as a raster coverage.

In order to make a valid comparison between the coverage created from the Pasture project and the TM data, it was necessary to format the TM data to the same specifications of the Pasture data. We did this by using functions found in Imagine. We classified the data in three different ways in an attempt to discover one classification that closely resembled the coverage already created. We first pulled up our TM image in the viewer in Imagine. From here, we used the Spectral Editor to establish the spectral values of the various land cover types which we were interested in. We found the spectral values of land features such as wetlands, water, alfalfa, pasture, crops, urban areas, and barren land areas. We selected these features because they closely resembled the data used in our coverage. In the Spectral Editor, we found the red, green and blue values for the selected features. We used this data to run a Supervised Classification for the aforementioned features.

After running the Supervised Classification, we decided to run two unsupervised classifications because we knew the computer would not make classification mistakes like we had. The first Unsupervised Classification of the thematic mapper AOI utilized seven feature classifications. To name the seven classes we found the points on our image that we had taken using the GPS , one for each class , and using the identifier cursor in the Imagine tools we were able to determine with reasonable accuracy to which class each of our GPS points belonged. This means that the computer would classify all of the data on the image into one of seven classes. We hoped these seven classifications would closely resemble the seven land use categories in the coverage. This would make comparing statistics between the unsupervised classification and the cache valley land use coverage we had created easier. After viewing our unsupervised classification with seven categories, we decided that the spectral signatures of several of the classes were so similar that it was impossible to identify accurately between classes.

We then decided to merge from seven to five categories in order to reduce the number of classes and hopefully improve the accuracy of the classification. The two new combination classes we created were riparian/pasture and bare ground/land use. Comparing the two unsupervised classifications we decided that reducing the number of classes was not any more accurate than our previous attempt in estimating land-use in Cache Valley. To do this comparison we used the GIS summary under the Interpreter feature of imagine to generate a statistics file which included total area for each class. We also did a visual comparison by using two windows one in Arcview the other a viewer in Imagine to simultaneously view both the land use coverage and the two unsupervised classifications. Without even viewing the summary statistics we realized something was wrong when the mountains east of Logan were shaded blue the class we had designated for water bodies. We hypothesize that this was so because of shadow created by the aspect of the mountain range and the position of the sun at the time that the TM image was taken. Had time permitted us, we would have liked to run some kind of filter to allow us to compensate for the affect of shadow.

One tool in Imagine that we found helpful to our project was the perspective viewer which allowed us to view our classified images with respect to elevation. We created several 3-Dimensional views by selecting both points to look from and to. We also chose an elevation height, with AWG, from which we wanted to view the image. The results of this technique were mainly visual in aiding us in improving our ability to determine Cache Valley land use from DEM data. Noticable was the general fact that built up areas were toward the east boundary, the lowlands between urban and crops where Logan River runs, was mostly pasture. Large areas of crop planting also occur in west cache valley on the high ground toward the Wellsvilles. The idea that elevation plays a role in determining land use in the Cache Valley should be further studied in light of the fact that general patterns can be seen.

Conclusions

When we originally began this experiment, we intended to compare the two maps; the Land Use Map & the TM Map. We nascently believed that we could use the TM Image to check the accuracy of the Land Use Coverage, however, it now seems that the opposite has taken place. When ended up using the Land Use Coverage to more accurately define the colors and categories for the Unsupervised Classification of the TM Image. If we were to take a point from every polygon in the Land Use Coverage and test its reflectance value on the TM Image, we would be able to find errors in the Land Use Coverage if any do exist. But, due to the time and money constraints, we could not do research of this caliber. Additionally, we noticed that the generalized nature of the Land Use Coverage was in stark contrast with the gradual change with literally 256 raised to the third. Image classification error may be partly attributable to the assignment of class boundaries to an image where, in fact, a gradient of change exists (Hunnicutt, 1992). This is an interesting project to do, and it does provide insight into the possibilities of checking map accuracy. The accuracy checking, however could go either way depending on the amount of field research that needs to be done.

We found that with proper knowledge of the software, these tools can be very helpful in the management and development of land use. Both methods have their individual peaks and valleys as far as data analyzation goes. In using both techniques, we have gained insight into the strengths and weaknesses of each.

Bibliography