The Mallard (Anas platyrhychos) has a broad range of habitat across the Northern Hemisphere. There are a couple of variations of the Mallard, the Common Mallard and the Greenland Mallard. Our focus is on the Common Mallard. The Common Mallard resides on the North American Continent. It ranges from Alaska to northern California and east to Ontario.and the Great Lakes, with recent breeding extensions into New England. The Mallard winter through much of its breeding grounds and as far south as northern Mexico (Johnsgard 221).
Distribution of Mallards in N. America.
Since the Mallard has such a broad range of habitat, it is difficult to separate preferred from acceptable breeding habitats. Some trends are evident that the Mallard accepts waters of almost any kind for breeding. The majority of Mallards prefer to nest within 50 meters of water on fairly dry ground (Johnsgard 225, 226). Habitat for the Mallard is classified into six categories:
1. Grassland: This includes all pastures and idle areas dominated by native or planted areas.
2. Hayland: This includes all areas planted to forage crops and areas mowed for hay.
3. Wetland: This includes all areas within wet meadow, shallow marsh, and deep marsh.
4. Cropland: This includes all areas that are annually tilled for production of grain and row crops.
5. Right-of-way: This includes the areas between fences, and the road surface along primary and secondary roads and railroads.
6. Odd areas: This includes an array of different features such as rock piles, haystacks, gravel pits, shelterbelts, and farmsteads (Cowardin 9).
Mallards feed upon a wide variety of foliage. Having the ability to utilize agricultural grain crops as well as natural foods depending upon availability. Mallards feed upon pondweeds, smart weeds, bulrushes, other emergent or submerges plants. Mallards also feed upon corn, sorghum, barley, oats and rye (Johnsgard 228).
There is a debate among wildlife managers that Mallard populations are declining. The purpose of this project was to find potential habitat for the Mallard in Cache County with the use of 30 meter Thematic Mapper imagery of the Cache County area.
Having found a general description of Mallard habitat, we could use this in locating the six categories classified by Cowardin. Upon knowing that Odd areas and Right-of-way would not show up, due to resolution of imagery, these two categories will be excluded from our study. Grassland will also be excluded from our study due to limited known sites of such vegetation and are lack of knowledge and identification of such grasses in the Cache County region. Wetland, Hayland, and Cropland will be the focus of our study. The study site is located between the Wellsville mountain range and the Cache Wasatch range on the valley floor.
Image of Study Area.
Finding suitable habitat for the Mallard duck is critical to wildlife managers, because once the habitat is found management programs can be implicated. Locating habitat over large landscapes is costly and time consuming using conventional methods (i.e air photos). It is our belief that with the use of satellite imagery habitat can be identified just as effective as using conventional methods. It is also our belief that this technology can be used to monitor the habitat overtime with less cost than conventional methods.
1: To locate all waterways and water bodies in the Cache county area. Knowing that the majority of the Mallards prefer to build nests within 50 meters of water.
2: Locating the cover types preferred for nesting using a supervised classification. All of the image processing and Gis analysis will be done using the ERDAS IMAGINE software.
We started out with a TM image of Cache County. The image contained the spectral values for bands 1-5, and band 7. The thermal band was not included in the data set. The image was taken in the early summer of 1988. Because we were only focusing on a portion of the county we felt that it was necessary to streamline the image and remove the areas that were not needed. The removal of these area was done using a inquire box tool with Imagine. The inquire box allowed us outline the area that was going to be used in the study. After this area was outlined, we made a subset of the image . This was preformed using the Interpreter/Utilities/Subset tool. Once this was completed we had an image of the study area. From this point the image of the study area will be referred to as cache.img.
As stated in the introduction the majority of nest were within 50 meters water it was necessary to build a buffer to isolate these areas. Instead of using a 50 meter buffer we decide that a 90 meter buffer would be used instead. The 90 meter buffer was used to insure that possible habitat was not left out of the study. In order to create the buffer we first had to identify the water bodies within the study area. This classification of water was done with the use of an unsupervised classification. The unsupervised classification used ISODATA to group the image pixels into 20 unique classes. ISODATA stands for " Iterative Self-Organizing Data Analysis Technique". It is "iterative" in that it repeatedly performs an entire classification and recalculates statistics. " Self -Organizing " refers to the way in which it locates clusters with minimum users input (ERDAS, Inc. 1990). Within Imagine the unsupervised classification was preformed on the cache.img using the Classification/Unsupervised Classification tool. With in this tool the users is able to set the maximum number of classes for the output file. We were advised by Dr. Ramsey to run this classification using different numbers of classes. The first classification was done setting the total number of classes at two. After reviewing the image it was found that this was not accurate enough for what was needed. The next three classification were run using 4, 10 and 20 classes in the output file. It was decided that the 20 class unsupervised classification would be used in the creation of the buffer. The buffer was created using the Interpreter/Gis Analysis/Search tool. Class one of the image was the only class specified with in this tool. Class one represented the water and shadow areas present with in the study areas. This buffer would be used to mask out the areas that did not fall with in the 90 meter buffer.
Buffer Created Using Unsupervised Classification.
Because the study area has a diversity of features within it , it was felt that we could not just identify the habitat types we were interested in. It was felt that the identification of other features would help increase the accuracy of the habitat types. Urban, bare ground, grasses, and mixed vegetation would have to be included in the identification.
In order to preform a supervised classification on the cache.img it was necessary to collect information concerning what cover types are present within the study area. This information is used to help train the computer to recognize features with in the study area. Training is the process of defining criteria by which patterns in image data are recognized of the purpose of classification (ERDAS, Inc. 1994). The collection of the training sites was done with the use of air photos, G.P.S., and personal knowledge of the study area. Because the image was taken in 1988 the agriculture features had to be identified with the use of the air photos. The photos were obtained from the USDA office in Logan UT.. The photos are of the Zollinger dairy farm located in Young Ward. The owners participated in the CRP program in 1988. Because of the participation of in the CRP program it was know what crop was in each field farmed during that year. This took out the guess work of identifying which crops were planted that year. Alfalfa, corn and spring barley were three crops identified with the use of the air photos. Because there were no coordinates available each field was visited with a G.P.S.. The urban, wetlands, areas with mixed vegetation along the rivers were also visited with the G.P.S.. At each point visited the coordinate was recorded along with type of cover that was present at that point on the ground. The points collected were later used to help train the computer to recognize those features.
One of Air Photos Used in Collection of Traing Sites
With the training sites collected the next step was to create a signature file. The signature file contains the mean and standard deviation of each sample (ERDAS, Inc 1994). Each training site had to be located on the cache.img. After reviewing the cache.img it was found that not all of the sites were readily visible in the image, specifically the wetlands. In order to be able to see the wetland sites two courses of actions were taken. The first was to change the histogram contrast of the cach.img. By turning off the blue and green and only displaying the red we were able to locate the wetlands much easier. By changing the contrast this only effects the way the image is displayed on the screen. The spectral values of each pixel remain the same. The second course of action taken was the creation of NDVI image. An NDVI (Normalized Difference Vegetation Index) indices was run on the cache.img. An indices is a process of creating an image mathematically combining the digital number values of different bands. The NDVI is created by combination of the addition, substraction and division of bands 3 and 4. The ratio looks the following band4-band3/band4+band3 (ERDAS, Inc 1994). The NDVI image produced highlighted the areas of vegetation in the study area. The NDVI not only helped identify the wetland sites but it also proved to be and invaluable tool with locating the agricultural fields with in the study area. With the combination of the NDVI and the cache.img with the histogram change we were able to locate the training sites to be used.
The above images are of the study area with the histogran change.
NDVI images used to help locate vegetation
Creation of the signature file was done with the aid of the AOI tool in the view. Using the inquire cursor we were able to enter in the coordinates of each training site. We then used the seed property tool in the AOI to build areas of interest. The spectral signatures of each AOI were then transferred to the signature editor to build the signature file.
Picture of signature file after training sites were entered.
Once all of the training sites were entered into the computer and the signature file was completed the next step was to run the classification on the cache.img. The supervised classification was preformed using Classification/Supervised Classification tool in Imagine. Within this tool the user has the ability to set the decision rules used in the classification process. The four rules that need to be specified are as follows: 1. Non-Parametric rule, 2. Overlap rule, 3. Unclassified rule, 4. Parametric Rule. The classification was run several different times with the parameters set up differently. It was found when the parametric rule was set on maximum likelihood that much of the image was classified as water. We felt this was attributed to the fact that we had a small number of training sites to preform the classification. The final image was produced with the following parameters: 1. Parallelepiped, 2. Parametric rule, 3.Parametric rule, 4.Minimum distance rule. Parellelepiped classification is describe by Campbell as being " Parallelepiped classification, sometimes known as box decision rule or level slice procedures, is base upon the ranges of values within the training data to define regions within multidimensional data space" (317). Minimum distance rule is defined in the ERDAS Field Guide as " a classification decision rule that calculates the spectral distance between the measurement vector for each candidate pixel and the mean vector for each signature" (593). It was felt that the using the following parameters gave us the most accurate representation of what was in the study area. With the completion of the supervised classification we now had an image that contained 39 classes. Each class represented a training site within the study area.
When the signature file was created the training sites were entered in order. This was done to help make the recoding of the classes faster. The recoding was done using the Interpretation/Gis/Recode tool with Imagine. The image was recoded from 39 classes to 8 classes. Those classes are as follows: 1.Wetlands, 2.Water, 3.Alfalfa, 4.Croplands, 5.Urban, 6.Bare ground, 7.Mixed vegetation, and 8.Grasses.
Image of Study Area After Being Recoded to the Final Eight Classes.
With the image now recoded the next step was to mask out the areas that did not fall with in the buffer. Using the 90 meter buffer and the Interpreter/Gis Analysis/Mask tool these areas outside the 90 meters were masked out. We were now left with an image containing the possible areas of for the mallard habitat. This image will be referred to as the final.img.
Final Image of Study Area After Masking. Using 90 Meter Buffer.
The final.img identified 2817 hectares ( 46% of the area) within the 90 meter buffer as potential mallard habitat. Because the study area contains several large bodies of water it was felt that amount (hectares) of habitat should be looked at for those areas as well. Those areas looked at include Cutler reservoir, Hyrum dam, Little Bear River, Newton dam, and the Logan sewage treatment facility. Areas that do not fall with in the above bodies of water will be called miscellaneous areas. Miscellaneous areas include the Bear River and any body of water not listed above. Refer to graph #1 for the results. Miscellaneous areas contain the majority of the habitat identified with 59.93%. The Little Bear River was next with 23.09% of the habitat. Cutler Reservoir contained 12.5%, Sewer treatment facility 2.14%, Newton dam 1.27% , and Hyrum dam 1.07% of the total habitat identified in the study area.
We found that the potential habitat was not directly associated with the larger bodies of water within the study area. The majority of habitat was found within the miscellaneous areas of the study area.
Graph 1
Image of Newton. 
Image of Hyrum. 
Image of Little Bear.
Image of Cutler. 
Image of Sewer. The use of GIS and Remote Sensing has the potential of becoming valuable tools in wildlife management. The identification of the habitat for the common mallard can be identified with the use of the current technology and the software available.
The accuracy of the final image is not know. We were unable to preform an accuracy assessment for the final image. After reviewing the image and comparing it to the real world we feel the accuracy would not be acceptable for making management decision. This can be directly attributed to the fact that we had a limited number of training sites to work with. This was especially true dealing with the collection of the alfalfa points and the agriculture points. The use of the 30mx30m imagery was only able to be used for the identification of the wetlands,haylands, and croplands. The identification of the odd areas and right-of -ways we felt could not be identified with the image at 30mx30m resolution. The identification of these habitat types would be necessary and would probably have to be done with the use of ground truthing and air photos.
We recommend if these tools are to be used to keep an inventory of habitat for any species that the researcher only work with the imagery taken from that year. The reason we recommend using only current imagery is that it is a lot easier to identify the areas that change on a year of biyearly basis (i.e agricultural fields). We also recommend that the researches collect as many training sites as they possible can. This should help increase the accuracy of the final image.
It is our belief that the image produced during the course of this project does not have the accuracy needed for wildlife management. We also feel that if more training sites had been used in the project the image would represent the features on the ground better. Although our project was not as successful as we had hoped it to be we did learn that when these tools are used correctly they can be used very successfully.
Campbell, James B. Introduction to Remote Sensing. New York:The Guildford Press, 1987.
Cowardin, Lewis M., et al. Mallard Recruitment in the Agricultural Environment of North Dakota . Wildlife Monographs. 92: Pp. 1-37.
ERDAS, Inc. 1990. ERDAS Field Guide, Version 7.4, Jan. 1990.
ERDAS, Inc. 1994. ERDAS Field Guide, Version 8.1, Feb. 1994.
Johnsgard, Paul A. Waterfowl of North America. Indiana: Uni Press, 1975.
Mallard Picture . Photo. Ducks Unlimited. Jan/Feb 1987. Pp.21
Air Photos. Photos. USDA. 1988