Potential Waterfowl Habitat of Culter Reservoir

Bonnie B. Banner, Wendy Goetz, Bob Hilderbrand and Steve Sheffey

Introduction

"The abundance and availability of suitable habitat is unquestionably the greatest limitation confronting waterfowl." - Baldessare and Bolen 1994.

The Great Basin:


The semi-arid intermountain regions of the U.S. are receiving increased demands on water resources of the area. These demands include the development of water resources for energy, industry, and domestic use, with agriculture accounting for 90% of the freshwater used in the 17 western states. Wetlands are in serious competition for sparse amounts of water, and wildlife habitat often receives last priority.

Our project focuses on determining the potential waterfowl habitat around Cutler Reservior. Cutler Reservior is located in the Great Basin. As a true basin it lacks an outlet to the sea. It receives an average of <15cm of rainfall per year, therefore most of the water comes from snowmelt. The typical vegetation of the area consists of Sagebrush (Artemisia spp.), Saltbush (Atriplex), and Greasewood (Sarcobatus vermiculatus). The Great Basin is used as as stopover by significant populations of migrating waterfowl in the Pacific Flyway. Although few ducks winter in the Great Basin, it is a crucial breeding habitat for a variety of waterfowl. Only Mallards (Anas playtrhynchos), Canadian Geese (Branta canadensis), and Trumpeter Swans (Cygnus buccinator) winter in the interior in significant numbers.

Cutler Reservior:

The study area is a privately owned reservior located in the Newton Quad in Cache Valley, Utah. It was built in the 1930's by Utah Power & Light (now owned by Pacific Corp.) for the purpose of generating power and irrigation. Today it is soley used for irrigation except in extremely wet years when it is used for peaking power.

Waterfowl Habitat:


Waterfowl:

There are three main categoies of waterfowl found in Cache Valley, each with its own habitat needs:
  1. GEESE - Grazing animals, strick vegetarians that feed on green grasses. Nest on upland islands.
  2. PUDDLE DUCKS - Feed mostly on invertebrates. Nest on upland areas.
  3. DIVING DUCKS - Feed mostly on invertebrates. Nest in bulrush.

Some of the marsh birds that can be found at Cutler Reservior are:
  1. Cinnamon Teal (Anas cyanoptera)
  2. Gadwall (Anas strepera)
  3. Green-winged Teal (Anas crecca)
  4. Mallard (Anas platyrhynchos)
  5. Northern Pintail (Anas acuta)
  6. Northern Shoveler (Anas clypeata)
  7. Redhead (Aythya americana)
  8. Ruddy Duck (Oxyura jamaicensis)
  9. Sandhill Crane (Grus pratensis)
  10. Whitefaced Iibes (Eudacimus albus)



Gathering information about the distribution and condition of wetlands is essential if we are to effectively manage for waterfowl. Our project focuses on using Remote Sensing and GIS technology techniques to assess the wetlands around Cutler Reservior. Through applying various approaches, we hope to gain new insights into wetland classification.

Objectives

Set within the parameters of a class project based on data sets supplied by Dr. R. Douglas Ramsey our group had three objectives.

The first objective was to compare the utility of two methods of identifying potential waterfowl habitat. One of which was to create a 100m buffer around the waterbodies within the Newton quadrangle and the other to run an unsupervised classification scheme on the same area.

We then wanted to determine if we could make a 10m resolution multispectral image by combining TM and SPOT images.

Assessing the feasibility of using 10m resolution multispectral data for classification purposes was the third objective of the study.

Methods
Study Area: Newton Quadrangle of Cutler Reservoir

Perspective view of the study area from the north.

For a prespective view looking toward Logan from the Oxbows click on icon.

This location was selected for the study, because of abundant water and wetland areas that waterfowl are known to use.

The fact that a quadrangle is a convenient area to work with in the confines of a class project limited to memory constraints was also a major factor in determing the study area.

Imagery and data:

We acquired LANDSAT Thematic Mapper (TM) 30m resolution imagery of Cache Valley from June 1988 processed by Dr. R. Douglas Ramsey.

Unrectified 10m resolution SPOT monochromatic imagery of the southern half of Cache Valley remotely sensed on April 14, 1994 was also used.

The Cache Valley (ca_dma.img) elevation data set provided by Dr. R. Douglas Ramsey established the base for developing perspective views.

Software:

ERDAS/Imagine 8.2 was used for all image processing and analyses.

Processes:

The Newton Quadrangle of Cache Valley comprising the study area was clipped from the Tm and SPOT images.

Rectification of the SPOT image was accomplished by using landmarks on the Tm image.

We then produced a 10m resolution image with multispectral values by performing a resolution merge of the rectified SPOT and Tm quad images.

A second version of the 10m merge image was created by running a 5x5 majority smoothing filter over the 10m merge image.

The smoothing filter was applied in an attempt to account for the splitting of each 30m Tm pixel into mine pixels in the 10m merge image and to reduce the number of mixed pixels.

Classification:

An unsupervised classification was done on Tm 30m, Tm-/SPOT 10m merged and smoothed Tm-SPOT 10m merged images. A supervised classification was also attemped, but failed.

Our classification goal stated identification of potential waterfowl habitats around Cutler Reservoir (not including all feeding sites in the surrounding area, mainly interested in potential breeding and nesting sites.)

Each image was originally classified into 30 classes. After evaluation the 30 classes were recoded to 4 classes. These consisted of:

Class 1 (water) which was denoted as water.

Class 2 (wetlands) that we considered preferred waterfowl habitat.

Class 3 (marginal habitat) was deemed potentially useable.

Class 4 (Uplands) was not considered to be potential waterfowl habitat.

The evaluation process for recode was based on spectral signature proximity to water bodies, and field visits to the study area.

Global Positioning System (GPS):


Thad Tilton: (GPS) point collection team leader

Field observations and locations recorded and marked with Trimble Basic+ field unit global positioning system (GPS) points within Newton quadrangle of Cache Valley helped to verify homogeneous plant communities within the classified images. Photo points were marked as GPS points. The GPS points were downloaded and processed, and corrected. GPS points were also used in creating perspective views of the study site.

Buffer:

In addition to the unsupervised classification, we attempted to delineate potential waterfowl habitat by constructiong a 100m buffer around water bodies.

The 100m buffer was used based on the assumption that this area around all water bodies would be suitable waterfowl habitat.

Mask:

The buffer image was applied as a mask on the Tm, Tm-SPOT, and smoothed Tm-SPOT to assure assessment on the same total area within the study site.

Statistics:

Summary matrices comparing the smoothed Tm-SPOT and the Tm-SPOT merged images developed an accuracy assessment of the utility in using a 10m resolution image in a classification scheme.

Total area of the potential waterfowl habitat identified by the methods applied to the Tm, Tm-SPOT, smoothed Tm-SPOT and Buffer images were compared and assessed.

Results and Discussion

After traveling to the study area in order to collect GPS points we were pleased to find from field observations that the area did in fact have abundant water, wetlands, waterfowl, and habitat known to be suitable and used by waterfowl. Therefore, the team felt that our study was on the right track and that we were ready to continue with our objectives.

We processed our GPS points with Pfinder in order to ready them for use in our classification verification and perspective views. On April 21, 1995 the GPS team visited the study area for the purpose of collecting GPS points with the Trimble Basic+ unit. Field observations and locations recorded and marked with the global positioning system (GPS) helped to verify homogeneous plant communities within the classified images. Photo points were marked as GPS points. The GPS points were downloaded, processed, and corrected with recorded field observations instead of a differential correction, because we did not have access to base station points for April 21, 1995. GPS points were also used in creating perspective views of the study site. The photo points marked with the GPS were used as eye locations and target locations in our perspective views.

In that Dr. R. Douglas Ramsey had supplied us with the data sets needed to accomplish our project we took the first step by clipping out the Newton Quadrangle of Cache Valley that contained the study area from the TM 30m image amd the SPOT 10m image.This was accomplished using aoi and subset in Imagine 8.2.

An unsupervised classification was then applied to the TM 30m data. 30 classes were used in the classification, after which they were recoded into 4 classes of habitat. Class 1 was determined to be water. Class 2 was set as wetlands which in our estimation should be considered preferred waterfowl habitat. Marginal habitat was denoted as class 3 and was also deemed to be potentially useable by waterfowl. The Uplands of class 4, even though they are used for feeding were not included as potential waterfowl habitat. The evaluation of the 30 classes was done based on spectral signatures, proximity to water bodies and information collected at each GPS collection point during the field visit to the study area.

Now that the classification of the TM 30m Newton quadrangle was complete we were able to create a 100m buffer around the waterbodies that had been delineated in the image. The buffer was created by using search in Imagine 8.2 under GIS ANALYSIS in Interpreter.

The process of creating an image with 10m multispectral values with the bands included in TM 30m data was begun by rectification of the SPOT 10m image using the GCP editor and landmarks on the TM 30m image. A resolution merge using principle component was run with the the clipped quad of the TM 30m data and the clipped rectified quad of the SPOT 10m data. After successfully completing the resolution merge with the principle component model we ran the process with the multiplicative model. The multiplicative model was found to be the best of the two algorithms.

The multiplicative merged 10m resolution multispectral image was used in the unsupervised classification, because it seemed to have the best spectral clarity based on a visual acessment. The same guidelines for classification were applied to the merged image as were used for classification of the TM 30m image. Because of the discrepancies in the classification of the merged image and the TM image we made a second version of the 10m merged image by running a 5x5 smoothing filter over it to see if the problem was due to going from 1 TM pixel in the original image to 9 pixels in the merged image and to reduce the number of mixed pixels. We then performed the same classification on the smoothed merged image which had been run on the unsmoothed merged image.

Having completed the image preperation and classifications the buffer image was used as a mask on all of the images in order to assure that the same total area in all images would be accessed. The results of the accessment were acquired in report and summary of Imagine 8.2. The statistics were based on the summary matrices comparing the smoothed TM-SPOT and the TM-SPOT merged images that developed an accuracy assessment of the utility in using a 10m resolution image in a classification scheme. The total area of the potential waterfowl habitat identified by the methods applied to the TM, TM-SPOT, smoothed TM-SPOT and the buffer images were compared and assessed. The results of these processes are clearly shown below with the appropriate images.

Potential waterfowl habitat was determined by constructing a 100 m buffer around water bodies and assuming all area within the buffer was suitable habitat. Total habitat is 1640 hectares or 14% of the total image.

The second approach was accomplished in comparing the TM and the TM-SPOT images.

Original Tm 30m Newton quadrangle image

Tm-SPOT merged image

Unsupervised classification of the original TM image. Habitat types are: = water; = wetland; = marginal habitats; = upland (non-useable habitats).

Unsupervised classification of the TM-SPOT merged image. Habitat types are: = water; = wetland; = marginal habitats; = upland (non-useable habitats).

A comparison of area in hectares and percentage of each habitat for the unsupervised classifications of the original TM image and the TM-SPOT merged image.

This comparison did not show very good agreement between methods.

In order to test for multiplication errors in going from one TM pixel to nine pixels in the TM-SPOT merged image, we ran the same classification procedure on a merged image that was smoothed with a 5x5 filter.

Unsupervised classification of the 5x5 smoothed TM-SPOT merged image.

If the differences were due to a simple multiplication error, the results of the 5x5 merged image classification should be much closer in agreement to the original TM image than to the TM-SPOT merged image. They were clearly not what we had hoped for.

Summary table comparing area and percentage of potential waterfowl habitat for each method.

Accuracy assessments were then done on the images.

Accuracy assessment of the unsupervised classification for the TM-SPOT merged image (columns) in relation to the TM unsupervised classification (rows). Comparisons are based on area (hectares) and percentages.

Accuracy assessment of the unsupervised classification for the smoothed 5x5 TM-SPOT merged image (columns) in relation to the TM unsupervised classification (rows). Comparisons are based on area (hectares) and percentages.

Conclusions

The buffer method overestimated waterfowl habitat by almost 50% compared to the unsupervised TM classification.

While the TM-SPOT merged image classified permanent features (water) basically the same as the original TM image.

However, variable features such as vegetation were not classified similarly. This was probably due to the fact that the images were taken on different dates and years.

The merging technique might provide useful high resolution multispectral data if images are taken on the same date.

Bibliography


Baldassarre,G.A., and E.G.Bolen. 1994. Waterfowl Ecology and Management. John Wiley & Son, Inc., New York. 609 Pp.

Johnsgard,P.A. 1975. Waterfowl of North America. Indiana University Press, Indiana.575 Pp.

Kadlec, J.A., and L.M. Smith. 1989. The Great Basin Marshes. Pp. 451-474 in L.M.Smith, R.L.Pederson, and R.M.Kaminski, eds. Habitat Management for Migrating and Wintering Waterfowl in North America. Texas Tech. University Press, Lubbock, Texas.

We would like to thank a the people without whose guidance and patience this report would have not been possible. Number one on that list is Dr. R. Douglas Ramsey.