Final Report for GIS19: Ecological Analysis of Bear River Range.

Final Project Report: SPATIAL ANALYSIS OF SNOW COVER AND TOPOGRAPHY FOR PREDICTING VEGETATION TYPES IN UTAH'S NORTHERN BEAR MOUNTAIN RANGE.

 PREPARED BY: Edward Bork, Eric Duffin, and Ghassan Mikati:
	Geography 673: Geographic Information Analysis.
	Spring Project, 1995.
	Prof: Dr. Doug Ramsey

Introduction and General Goal of Project:

What follows is a review of the specific objectives, methods, and analysis procedures that our group followed, as well as the results and conclusions that we discovered.

Background Information:

One such situation where the use of a surrogate variable may be useful, is for the inventory and mapping of rangeland plant communities. While inventory is usually done using aerial photography and its subsequent interpretation, large areas of rangelands may be more efficiently inventoried using synoptic, remotely sensed data (Lillesand and Kiefer 1987). Evans et al. (1989) used this technique to investigate the spatial relationships of plant communities, snow cover, and topography. In particular, they used the relative location and depth of snow to account for differences between the plant communities that they observed. That analysis demonstrated the analytical power and advantages of a Geographic Information System for analyzing terrain-geobotanical interrelationships.

A significant proportion of the western United States consists of vast foothills, mountains, and plains. In these areas, ground reconaissance and mapping is time consuming, expensive, and generally inefficient. Using remote sensing directly, or indirectly with the aid of surrogate variables, may prove very advantageous in the management of this resource. This project explores the applicability of using surrogate variables to predict natural resource variables of interest. In particular, it evaluates the use of mapped snow cover for mapping general vegetation types, and explores the interrelationships between topography and the previous two variables.

The area that was choosen for this project was the Bear River Range northeast of Logan in northern Utah. This area is highly variable, with complex interactions between slope, elevation, aspect, climate, geomorphology, and disturbance producing a highly diverse and interwoven patchwork of vegetation across the landscape. Managers require an understanding of how these factors interact to produce this vegetation mosaic if they are to manage the resources for particular uses such as recreation, wildlife production, forestry, and water management.

Specific Objectives:

To determine the spatial relationship(s) between topographical digital elevation model (DEM) data, including aspect and elevation, and vegetation types.
To determine the spatial relationship between DEM data and residual spring snow cover.
To evaluate the correlative, and therefore, predictive, relationship between snow cover and vegetation types.

Data Requirements, Sources, and Preparation:

Introduction:

A. Preparation of Vegetation Coverages.

The source for the vegetation data was the Utah GAP inventory at 30 meter resolution. This data was already rectified and georeferenced. To prepare this data, first the area of interest was clipped out using the Subset command in ERDAS. Next, because the original GAP file contained 38 specific vegetation classes, these were combined into broader classes. Our group decided to combine all the vegetation classes into general vegetation growth form classes including tall and short coniferous trees, tall and short deciduous trees, and a shrubland/grassland mixture. We felt these categories were particularly important to recognize and practical because of the different ecological requirements of these growth forms. For example, coniferous forest, aspen forest, and shrublands are usually each associated with very different types of ecological sites. Classes from the GAP data that did not fit into any category were masked and by coding them to 0. These classes included urban areas, agricultural land, open water, and bare rock.

Once the vegetation classes were identified on paper, two image files were created using the Recode tool in Imagine. One final adjustment was made which included masking out all elevations below 1500 meters. The Mask option in Imagine was used with the contour elevation map acting as the masking file. This removed most of the areas on the lower mountain slopes and valley benches that were not georectified very well on the snow cover image (discussed later).

The first vegetation image produced for analysis with other data coverages contained 5 vegetation cover classes including the following (Percentage is the proportion of the study area covered by that class (color in parenthesis is for the image that follows):

Tall Coniferous Forest (13.9%) - includes stands with species such as Engelmann spruce, subalpine fir, Douglas fir, or pondersoa pine as dominants (Dark Green).
Tall Deciduous Forest (18.2%) - includes stands dominanted by trembling aspen (Yellow).
Short Coniferous Forest (7.9%) - includes stands dominated by shorter evergreen trees such as Rocky Mtn. juniper (Light Green).
Tall Shrub & Short Deciduous Forest (32.9%) - includes stands with species such as mtn. mahogany, mtn. oak, or mtn. maple (Red).
Sagebrush & Low Shrub & Grassland (27.1%) - consists of open or semi-open stands of big sagebrush, perennial grasses and forbs, alpine species, or sedges (Purple).

Fig. #1: GAP Vegetation Map with 5 Classes.

The original vegetation class map of the same area with all 38 classes is included below for comparison.

Fig. #2: The Original GAP Vegetation Map with 38 Classes.

The second vegetation map we produced was similar to the first except that the 5 vegetation classes were combined into only 3. This file was created to simplify the analysis with aspect and elevation combined, because of the already large number of classes in the latter (ie. 20 classes). Below is a listing of these three classes used, their areas and image colors, and an example of the image itself.

Coniferous Forest (21.8%) - includes stands with species such as Engelman spruce, subalpine fir, Douglas fir, pondersoa pine, or juniper as dominants (Green).
Deciduous Forest (51.1%) - includes stands dominanted by trembling aspen, mtn. mahogany, mtn. oak, or mtn. maple (Yellow).
Sagebrush & Low Shrub & Grassland (27.1%) - consists of open or semi-open stands of big sagebrush, perennial grasses and forbs, alpine species, and sedges (Red).

Fig. #3: GAP Vegetation Map with Only 3 Classes.

B. Preparation of Snow Coverage.

The map of snow cover proved to be the most difficult and challenging component of the data preparation process. Since our objective was to predict vegetation types using residual snow cover, a panchromatic SPOT image with 10 meter resolution taken on April 14th, 1994, was used to produce a snow cover map for our study area consisting of the Bear River Range north of Logan Canyon.

The initial step in this procedure was to georectify the image. Our first attempt used a georeferenced TM image of Cache County. We used two viewers in Imagine, with the SPOT image as the source and the TM image as the destination. After entering 35 ground control points (GCPs), we conducted a second order transformation to rectify the image. Although our RMS error appeared to be quite low (a total of 2.39), when we overlayed and compared the SPOT image to the existing road coverage for Cache Valley, the SPOT image appeared to be poorly registered. This continued, despite utilizing all possible transformations. There were probably two principles reasons why this failed. First, the SPOT image was apparently taken at an oblique angle, making higher elevation areas, opposite side slopes (from the sensor), and locations further from nadir, more distorted. Second, no GCPs could be located in the mountainous areas on the TM image, which exacerbated the poor georectification within our study area.

As a result of this failure, our group decided to subset the SPOT image for only our study area and attempt to rectify this area from hardcopy orthophoto and quadrangle maps. Subsetting was done using ERDAS. Once this was done, we obtained 5 orthophotos and 4 quad maps and began entering GCPs in Imagine, with the SPOT image as the source image and the tablet derived coordinates as the destination. Unfortunately, there were some transformation problems with the digitizing pad that really slowed us down, including the loss of some 100 destination GCPs. We finally decided to manually record the UTM location of each source GCP from the hardcopy maps. With the assistance of the Instructional Technology department and the USU computer graphics lab, we were able to print a 0.5 kilometer grid on a sheet of mylar. This was going to be laid over orthophoto quadrangles and used as a reference to measure GCP's and manually enter them into ERDAS IMAGINE. Before this could be done, however, a solution to the digitizer problem was found.

We then went back and used the digitizer and tablet to input 147 GCPs, primarily in mountainous areas. We used roads and intersections mostly, but also used readily distinguishable conifer stands, mountain peaks, and rock slides where they were available. This was done in order to help ensure that the image was actually rectified at higher elevations; the area of primary interest to begin with. Several points were observed to have residual errors that were very high, over 40 pixels. These were deleted, bringing the total number of GCPs used to 143.

At this point a second order transformation was done on the image. The resulting SPOT image was then examined for error by overlaying the Cache Valley road map on top of it and looking for differences between the two images. It quickly became evident that the road coverage was consistently to the right and above roads found on the SPOT image. Differences within the SPOT image also appeared to be greater on east aspects, and increase at higher elevations, probably as a result of the SPOT image being taken at an oblique angle above the earth's surface. Since the area of interest for this project was that area that had snow cover and this was generally at higher elevations (ie. above 6500 feet or so), we used the inquire cursor at mid slope elevations to get an approximation of what type of adjustment would be the most appropriate.

To properly compensate and complete the georeferencing, we performed a linear adjustment on the image by adding 250 meters and 125 meters, respectively, to the x and y coordinates of the upper left corner of the image. This resulted in a relatively good fit for the north end of the image but not the south. Since the majority of snow cover in our study area was located in the north end of the SPOT image, we eliminated everything south of Logan Canyon using Subset in Imagine, deciding to focus our efforts on the northern part of the original SPOT image. A sample of the georectified SPOT image is provided below.

Fig. #4: Georectified Mid-April 1994 SPOT Image of the Bear River Range.

The next step was to examine the pixel values in the SPOT image using the inquire cursor to determine the range of values covered by areas of snow and no snow. Examination of the histogram for the SPOT image's pixel values did not provide much clear direction (below).

Fig. #5: Histogram of Pixel Values from the SPOT Image.

We found that some areas of the SPOT image with obvious snow cover had pixel values as low as 110 (ie. probably mixed with conifers) while other areas free of snow cover (assumed to be sparse grass) had values as high as 135. Further examination indicated these latter areas were found predominantly on south-facing slopes, specifically in areas near ridgelines. We originally had decided to classify the SPOT image into areas of no snow, marginal snow, and heavy snow (pixel values < 110, between 110 & 135, and > 135, respectively). In order to compensate for the influence of south-facing slopes, we attempted to reclassify the image by constructing a model (using Imagine Modeler) which would classify pixels with values between 110 and 135 on south-facing aspects into the no-snow category. Our efforts with the Imagine modeller proved unsuccessfull, forcing us to go with the three classes of no snow, positively snow, and probably snow (ie. between 110 and 135). An example of the classified image using the 110 and 135 cutoffs is provided below. The snow class is white while the probable snow class is blue. Because the area of snow coverage is above 1500 meters anyways, masking of this data layer was not necessary.

Fig. #6: Map of Mid April, 1994 Snow Cover for the Bear River Range with 2 Classes.

After the initial analysis was done with the snow coverage using 2 classes, our subsequent analysis indicated that only a moderate proportion of the data was in the questionable snow class. In particular, the area of snow (ie. the class over 135) was 11,698.6 Ha., while the questionable class (ie. from 110 to 135) was 2,648.2 Ha. Furthermore, a significant proportion of the questionable snow class was located at upper elevations. Therefore, to avoid wasting the pixels that fell into the second class and for ease of data processing (ie. to reduce the number of classes), we decided that the amount of error introduced by the brighter crests and south facing slopes was acceptable and pooled the second and third classes into one. The resulting map produced a coverage of snow with only 1 type as follows.

Fig. #7: Revised Map of Mid April, 1994 Snow Cover for the Bear River Range.

C. Preparation of Topographic Coverages.

The topographic coverages for the study area were obtained from the Digital Elevation Model (DEM) 7.5 minute quadrangles. First we went to the library and looked up the names and lower right UTM coordinates of the 9 1:24,000 quadrangles we required to cover the northern portion of the Bear River Range. Next, we imported the appropriate files into our workspace and uncompressed them. Following that, we used Latticemerge in ARC to create our GRID file, 9 quadrangles in size. This file we were able to view in both ARCVIEW and Imagine. All the other topographic coverages were created from this Grid file.

An initial elevation map or coverage was produced with 32 classes. To acheive this, the DEM grid file of 9 quads was used. First the zeros in the elevation data had to be removed from the file. This was done using the GRID subprogram of ARC. A table was created using the word editor for setting the elevation themes (50 meters intervals between 1500 meters and 3000 meters). Two additional classes were added, one below 1500 meters and one above 3000 meters elevation. After this, the DEMTOPO command in the ARC GRID subprogram was used to perform the reclassification, with the text file providing the necessary criteria. Finally, the new image was imported into IMAGINE, Recoded with the Image Interpreter into 300 meter classes beginning at 1500 meters, and recolored for visual effect. This file is shown below. Concentric rings inwards represent progressive increments of 300 meters elevation as follows (Relative size of each interval within the study area is in parenthesis):

1500 meters to 1799 meters (12.7%) - Red.
1800 meters to 2099 meters (29.3%) - Green.
2100 meters to 2399 meters (33.4%) - Purple.
2400 meters to 2699 meters (20.3%) - Light Blue.
over 2700 meters (4.3%) - White.

Fig. #8: Elevation Map of Bear Mountain Range with 300 Meter Contours.

After importing the Grid file into Imagine as .IMG files, we created an Aspect coverage using the Aspect function in the Topographic Analysis option of the Interpreter. This initial file was in continuous aspect format (ie. degrees) and was reclassified using Recode into 4 primary aspect classes (eg. North, South, East, and West) according to the following criteria (Relative size of each aspect within the study area is in parenthesis):

North Aspects (21.4%) - Slopes facing 316 to 45 degrees (Dark Blue).
East Aspects (23.1%) - Slopes facing 46 to 135 degrees (Green).
South Aspects (27.0%) - Slopes facing 136 to 225 degrees (Light Blue).
West Aspects (28.6%) - Slopes facing 226 to 315 degrees (Red).

This aspect map was then recolored using the Raster Attribute Editor (Colors in parenthesis above are for the image that follows later) and Masked using the Elevation Map to remove all elevations below 1500 meters. The Mask option in Imagine was used with the contour elevation map as the masking file. This was done to simplify the subsequent analysis because the area below 1500 meters is primarily in the valley and is of little interest in the analysis as it contains no snow cover. The final aspect coverage is depicted below.

Fig. #9: Aspect Map of the Bear Mountain Range.

These recoded images were then used to produce a combined coverage of Aspect and Elevation for all areas above 1500 meters. To achieve this, the Matrix function in the Imagine Interpreter was used. The output image combined the 4 aspect and 5 elevation classes to produce an image with 20 new classes, which were then recolored for convenience. Although it is difficult to interpret visually, this image is depicted below. Various shades of red are west facing slopes, green are east facing, yellow-brown are south facing, and blue are north facing. For all aspects, the darkest shades represent the lowest elevation classes with progressively lighter shades representing increasing contour intervals.

Fig. #10: Map of Combined Aspect and Elevation Coverages for the Bear Mountain Range.

Results, Interpretation, and Discussion:

Part 1: Vegetation and Topopgraphy:

Before dealing with snow cover and how it relates to topography and vegetation, our group felt it would be interesting to do some analysis of the topography and vegetation, and see how they relate together. This analysis was divided into three parts:

Aspect with vegetation.
Elevation with vegetation.
A combination of aspect, elevation, and vegetation classes.

#1. Aspect and the Five Vegetation Classes:

The combination of the aspect data map and the vegetation map with five classes provided a meaningful way to correlate physical temperature and moisture factors with vegetation growth and distribution. The table shows how the vegetation is distributed among the four aspect classes. (see also the graphical depiction below):

The above data clearly shows that the North and East slopes had more tall coniferous forests with 27.5 and 16 % respectively, whereas the warmer and less moist South and West slopes had more of the short coniferous 15 and 25.5 % respectively, shrub types 36 and 37 % respectively, and grasslands 34 and 27 %respectively. This is basically due to the fact that the latter species don't need as much water as the tall coniferous trees, and thus grow better on the drier slopes. Furthermore, it can be noticed from the graph above that the tall deciduous forests are located on the East slopes.

#2. Elevation and the Five Vegetation Classes:

This part of the analysis deals with the same five vegetation classes, but this time compared to the elevation data map where contour lines were drawn at 300 meters intervals starting at the elevation of 1500 meters. The table below shows the distribution of the different vegetation classes with each successive elevation interval (see also the graphical depiction below):

Fig. #12: Graph of the Vegetation Classes (% Cover) Relative to Elevation.

The tables indicate that the Tall Conifer type was more abundant (a maximal presence at 2400 to 2700 meters) at higher elevations whereas the Sagebrush & Grass were present at all elevations, but predominantly above 2700 m. In addition, it is important to note that the Tall Deciduous trees were concentrated at lower elevations (especially below 1800 meters) because this vegetation type consists of species adapted to warmer and drier conditions.

#3. A Combination of Aspect, Elevation and Three Vegetation Classes:

The last part of this comparison involves all three of the types of data available, although only the vegetation data with three classes was used. The data shows a similar pattern to the previous analysis, with much the same conclusions reached concerning the distribution of vegetation. For more information, check the referenced tables below for the elevation of your choice.

Aspect and Vegetation at 1500 to 1800 meters.

Aspect and Vegetation at 1800 to 2100 meters.

Aspect and Vegetation at 2100 to 2400 meters.

Aspect and Vegetation at 2400 to 2700 meters.

Aspect and Vegetation at Elevations > 2700 meters.

Conclusions for the Topography versus Vegetation Analysis:

Tall coniferous forest are present on the cooler and moister North and East slopes at the higher elevations.
Shrubs, Sagebrush and grasses are concentrated in the lower elevations of the warmer and dryer South and West slopes.
Tall decideous are mainly present at lower elevations as those species are usually less cold resistant.

Part 2: Topography and Snow Cover:

In order to more clearly define the relationships between topography and residual snow cover, the original DEM data was reclassified into meaningful aspect and elevation classes as decided upon by our group. The individual classes were then analyzed for snow cover. The general pattern of snow cover in our study area is illustrated below. Snow cover is represented by white, all other shades represent elevation data in 300 meter contour intervals. Visual analysis of the original Spot image indicated no significant snow coverage below 1500 m. As a result, all analysis of DEM data with respect to snow cover included only those areas with elevation > 1500 m.

Fig. #13: Map of Snow Cover and Elevation Combined for the Bear River Range.

The percentage of residual snow cover was classified by four aspects. The results showed patterns that are typically found throughout the Northern hemisphere where East and North-facing slopes retain higher levels of snow for longer periods than West and South-facing slopes. This is believed to be the result of higher levels of radiant exposure on West and South Aspects (Gray, et al, 1981).

Unfortunately, the images of snow coverage used in this project did not contain information about snow depth. It is believed that snow depths on North and East facing slopes were greater than depths found on the other two aspects. The prevailing wind direction along the Bear River Mountain Range is from the west. Moist air being driven by these winds will orographically deposit precipitation in greater amounts along west facing slopes. Throughout the year, snow can be scoured from these slopes and redeposited on leeward, or east facing slopes (McClung et al, 1993). This process is also thought to be responsible for some of the differences in snow coverage. Redeposition of snow, particularly along sharp breaks in slope can be responsible for creating dense snow formations that will be more resistant to the melting process.

Fig. #14: Graph of Percentage Snow Cover vs. Aspect.

Analysis of percentage snow cover at 300 m elevation intervals showed significant increases above 2100 m. Snow covered areas of less than 1% are believed to represent areas of bare soil or light colored grass.

Fig. #15: Graph of Percentage Snow Cover vs. Elevation (300 m increments).

Elevation and Aspect data were then combined into a matrix of 20 classes and analyzed for snow cover. Increases in snow cover above 2100 m were evident for the East and North facing aspects where a minimum of 10% greater coverage was exhibted over West and South slopes. This trend continued to 2700 m where snow cover was essentially 100% on all aspects.

Fig. #16: Graph of percentage Snow Cover vs. Aspect and Elevation (300 m increments).

Elevation data was further classified into 50 m increments to more clearly identify relationships between elevation, aspect, and snow cover. Significant increases in snow cover were not apparent until 2150 - 2000 m on North facing slopes only. All aspects began to exhibit increased snow cover after an additional 100 m in elevation (2250 - 2300 m). The general trends exhibited in the aspect vs snow classification were apparent to an elevation of 2850 m where snow cover appeared to be relatively homogeneous between aspects.

Fig. #17: Graph of percentage Snow Cover vs. Aspect and Elevation (50 m increments)

Conclusions for the Snow Versus Topography Analysis:

Overall trends indicated that percent snow cover on East and North facing aspects was higher than that occuring on West and South facing slopes. Further analysis of elevation data indicated that this trend was followed for elevations up to approximately 2850 m. Areas above this elevation appeared to have approximately equal levels of snow coverage (approaching 100%) on all aspects. Further analysis of snow coverage would be enhanced if information regarding snow depth and/or water equivalence were available.

Part 3: Vegetation and Snow Cover:

This portion of the analysis outlines the results of our group`s attempt to address the third question: can snow cover be used as a predictor for vegetation type? A summary of our vegetation coverage with 5 principle growth form classes and our snow coverage indicated the following correlations (see also the graphical depiction below):

Fig. #18: Graphical View of the Relationship Between the 5 Vegetation Types and Snow Cover.

A map of the resulting area is depicted below. Colors for the five vegetation types are the same as in Figure #1, except that snow covered areas are depicted by a lighter shade of the same color. This color scheme is as follows:

Fig. #19: Map of Five Vegetation Types Under Snow and Snow-Free Conditions.

Each percentage is the proportion of that vegetation type that was classified as snow. Thus, the higher this proportion, the greater the potential for using snow cover as a surrogate variable to classify vegetation. The data indicates that none of the correlations for any of the vegetation types were very high. This may have been because of, (1) the difficulty we had rectifying the image (ie. because of the oblique angle of incidence for the SPOT image), (2) an inability to classify areas with very little snow as snow, or (3) the image we used being taken at a time of the year when the correlations we were looking for were inherently poorer (ie. too little or too much snow to find a solution).

The numbers did suggest, however, that the Short Conifer type is strongly negatively correlated with snow, probably because the tree species in this type (ie. Pinyon and Juniper) are generally found at lower elevations in this part of the state.

Of the remaining types, some useful information could still be gathered. For example, both the Tall Conifer and Aspen types had the highest correlations with snow. This was expected since these species have higher moisture requirements for growth. The later greenup and increased water availability later into the spring and summer would allow these species to dominate the landscape. This indicates that the potential to use snow cover as a surrogate variable to map these vegetation types probably does exist, although the conditions under which this would be most effective remains unknown. A potential danger to using snow as a surrogate variable, is that the correlations we are seeing here are partly an artifact of many types being ecologically limited to upper elevations where snow cover is bound to be much higher. Regardless, mapping for general vegetation types does appear to be feasible with the aid of snow cover. Although the open Shrublands & Grasslands type had a relatively high correlation, this type included a large portion of alpine and subalpine meadows, most of which were probably positively correlated with snow cover. A similar situation was likely with the Deciduous Tree & Shrub type. A diverse mixture of types spanning many different ecological conditions was probably pooled into one category here. The result was a moderate correlation.

To determine whether any of the original vegetation types (ie. from Figure #3) were more distinctly correlated with snow cover, this coverage was compared to the snow map. The result indicated the following correlations (ie. percent of the vegetation class classified as snow cover). Only the types over 1 percent are listed. This information is also provided graphically below.

Fig. #20: Graphical View of the Relationship Between Local Vegetation Types and Snow Cover.

These results further tend to support the conclusion that Tall Conifer (particularly those with Spruce and Fir) and Aspen Types are more highly correlated with snow cover. In particular, the Spruce-Fir, Ponderosa Pine, Spruce-Fir & Mtn. Shrub, and Aspen & Conifer Mix were correlated the highest. Other notables included the diverse and varied mixture of shrub types such as Mountain Mahoghany, Oak, and the generic Shrub type. As concluded earlier, many of the shrub and grassland types show positive responses as well, including Sagebrush, Grassland, and the Wet and Dry Meadow types, which are generally found across many elevations.

Conclusions for the Snow by Vegetation Analysis:

None of the general vegetation types were highly correlated with snow cover, although the pattern of the correlations did make ecological sense, thereby indicating snow cover could possibly be used to predict and map vegetation types in the future
Of the general vegetation types, Short Conifer had the lowest correlation with snow cover while Tall Vegetation types (ie. Aspen & Conifer) had the highest.
There likely was significant information lost by combining vegetation types into general classes (eg. the Grassland and Shrubland type).
Of the specific vegetation types, correlations with snow cover were highly variable.
Again, certain Tall Conifer types (eg Spruce-Fir & Ponderosa Pine) and Tall Deciduous types had good correlations. However, many shrubland, grassland, and meadow types were inexplainably also quite high.
While the potential exists for using snow cover to help map vegetation types, considerable work is needed on how to improve the accuracy, reliability, and consistency of the results, particularly in ecologically complex areas.
Note that higher correlations for upper elevational areas may simply be autocorrelation because of the overwelming coverage of snow.
There were many other possible reasons for the low correlations including the following explanations pertaining to the data sources:
- Poor rectification of the SPOT image (ie. the Oblique angle of incidence).
- An inability to classify correctly, either the areas with very little snow as snow (ie. under Conifers), OR areas with no snow as snow-free (ie. Crests and Open Grasslands).
- Selection of a poor year to conduct the study, OR poor time of year to inventory snow.
- Excessive ecological complexity within the study area.
- Inadequate Resolution or Accuracy of the GAP Vegetation Coverage.
- Incorrect Georectification of the GAP Vegetation Coverage.

Literature Review:

Evans, B.M., D.A. Walker, C.S. Benson, E.A. Nordstrand, and G.W. Petersen. Spatial relationships between terrain, snow distribution, and vegetation patterns at an arctic foothills site in Alaska, Holartic Ecology, 12: 270-278.

Friedel, M.H. 1994. How Spatial and Temporal Scale Affect the Perception of Change in Rangelands, Aust. Rangeland Journal, 16: 16-25.

Gray, D.M., D.H. Male. 1981. Handbook of Snow: Principles, Processes, Management and Use. Pergamon Press Canada Ltd., 776 pp.

Lillesand, T.M. and R.W. Kiefer. 1987. Remote Sensing and Image Interpretation, 2nd ed., John Wiley and Sons, 721 pp.

McClung, D., S. Schaerer. 1993. The Avalanche Handbook. The Mountaineers, 1011 SW Klickitat Way, Seattle, Washington, 98134. 271 pp.