An Application of Remotely Sensed Imagery to the Study of Avalanche Phenomenon

RS21 Project Presentation, June 1st

Project Goals

The purpose of this project is to utilize remotely sensed imagery as a tool to aid in identifying avalanche terrain, specifically those characteristics of vegetation and topography that contribute to the formation of avalanches. In no way is it an attempt to supersede the historical collection of data and direct observation methods that predominantly occur in the field. Nor does this project attempt to predict stability or the level of safety for a particular area, which should in the authors opinion be based on good routefinding skills, and rigorous field evaluation of hazard. Rather, the design of this project roughly follows a format for cataloging and mapping terrain features proposed by McClung et. al. , specifically focussing on those elements that can be identified in remotely sensed imagery that are related to avalanche events, similar to an avalanche atlas. According to McClung et. al. methods such as contour maps, aerial photos and ground observations aid in determining if slopes meet the criterion for avalanche formation, and an investigation of clues such as younger, different species of trees, vegetation composition such as the composition of species, and damage clues such broken limbs, scars on foliage, leaning or fallen trees, and piles of debris can indicate the paths that avalanches follow or create in the landscape (1993). The authors strongly agree that an important supplement to this analysis of remotely sensed imagery would include field collection and observation methods that could be integrated into a GIS data base and graphic model, thus allowing for the historical cataloging of direct observations, avalanche frequency based on vegetation analysis, extent, and size to be integrated into an expanded project.

A number of definitions are applicable when studying so called avalanche terrain that give a common terminology for the parts of an avalanche. An avalanche path is the area in which the factors that contribute to avalanche formation can occur, and in which the avalanche is propagated and can move. An avalanche path has a starting zone, in which it is propagated or originates (this is where the snowpack fails and begins to move), a track which is the area below the starting zone in which the avalanche flows, connecting the starting zone with the runnout zone or zone of deposition in which debris are deposited (McClung et. al., 1993). These areas can vary from season to season and from avalanche event to event, as well as in frequency, with a smaller avalanche occurring in a larger path or track.

Knowing these terms, one then can think about the terrain factors (particularly looking for those we can identify in remotely sensed imagery) necessary in order for an avalanche to occur or be propagated, especially those that effect the parts of an avalanche such as the starting zone, which must be on a slope angle that will allow the slide to start and accelerate. A general rule of thumb for identifying hazardous slope angles are those slopes between twenty five degrees (an intermediate ski run) and fifty degrees (an extreme decent), with the highest frequency of avalanches occurring at thirty eight degrees (an expert ski run), though this does not exclude steeper or gentler slopes from sliding as well (McClung et. al., 1993). Other important factors include the aspect of the slope, which limits the amount of solar radiation that can affect the pack (generally sun shaded aspects will have a higher frequency of avalanche activity), orientation to the wind, with those terrain features that contribute to the deposition and drifting of snow into starting zones (lee side of ridges) being of interest, and finally vegetation, which can serve an anchor in the starting zone to hold the snowpack in place, also serving to give clues as to the course that previous avalanches have taken in that location. Generally if vegetation is to serve as an anchor it must be dense enough that one could not ski through it, about 500 conifers per hectare on gentle slopes, while slopes with less dense vegetation, or low under-story vegetation such as young growth can be prone to avalanches (McClung et. al., 1993). Avalanche tracks can be "open", that is without lateral boundaries, and "channel" in shape such as a gully, or other depression in the terrain, with any combination of the two being possible as well. Most tracks, and therefore avalanche debris will follow the fall line of the slope, with the track angles (not starting zone angle) being between fifteen and thirty degrees (McClung et. al., 1993). Runnout or deposition zones occur where the topography and slope angle equal the static friction angle for the avalanche, typically around fifteen degrees (McClung et. al., 1993)

The affect of avalanches on vegetation includingTree-rings, and Avalanche Trim Lines

An interesting affect of avalanche paths on vegetation is called trimline in which the avalanche debris literally cuts a swath through young and old vegetation. Additionally, avalanche frequency can be determined by studying the composition of trimlines, specifically looking at the age and species composition of vegetation, noting the growth of "pioneer species" bordering and within the path (McClung et. al., 1993). Field analysis of tree rings, tree scars, and broken vegetation are the best tools for "dating" avalanche frequency for a path and area. Specifically through the study of tree rings in live vegetation, one can see when a tree was bent down the fall line by avalanche debris, causing the formation of "reaction wood" in the following season to right the growth of the tree. Yet, for the mapping of avalanche terrain, the classification of vegetation and analysis of trimline, paths through vegetation and slope can provide adequate information. The fundamental goal of this project was to determine if 20 meter SPOT imagery could be used to identify avalanche trimlines and paths, particularly focussing on a classification that would separate bare ground, young vegetation, and sparse vegetation from older growth. This information, in combination with DEM or digital elevation data such as contour and slope would hopefully give an indication of starting zones and avalanche paths for the logan peak area.

Yet another example of trimline, this time through dense conifers.

Methodology

Among the available imagery, we chose to use SPOT multi-spectral imagery of Cache Valley and Wasatch Cache National Forest over the LANDSAT TM. As a group we decided that this platform would provide us the best spectral resolution, which is about 20x20 meters. The LANDSAT TM on the other only had a spectral resolution of 30 meters making it to course for the level of analysis needed to differentiate between vegetation types, trimline, and bare soil. In the multi-spectral configuration, SPOT senses 3 spectral regions: Band 1 0.51-0.59 micrometers (green) Band 2 0.61-0.68 micrometers (red; chlorophyll absorption) Band 3 0.79-0.89 micrometers (Near IR) and takes a 60km swath but uses 3,000 samples for each line (Campbell, 1987).

For our study site we chose Logan Peak based on the coverage of imagery and that the site had characteristics such as a history of avalanches and readily observable trimlines that the group was interested in.

The Logan Peak study area is located in beautiful (and somewhat green according to our image) Cache Valley...and yet another view

The previous image of Logan Peak was created using the perspective view utility in ERDAS Imagine using our classified SPOT 20 meter image and a 30 meter DEM file clipped to the study area as layers for generating the perspective view. This is the target area we used to create the Perspective View and here is the Perspective View tool.

Rectification Process

In order to compensate for distortion caused by the altitude, attitude, and velocity of SPOT, and the 20 meter images lack of a projection, we had to first rectify the 20 meter image of Cache Valley and Cache National Forest prior to classification and analysis. To do this, one must first acquire a 1)rectified image of the area or 2)acquire orthophotos, topographic maps, and/or up-to-date aerial photos. After the necessary ancillary data has been gathered, ground control points (GCP) must be located on both the unrectified image and other image source. For our discussion, we used a rectified 10 meter SPOT image of Cache Valley. After locating GCP's common to both images, over 120 points were entered on both images using the GCP editor in ERDAS Imagine. The transformation editor was used in conjunction with the GCP editor to predict points onto the rectified SPOT 10 meter image. The difference between the predicted points and the actual ground locations is called RMS error, which increases for each GCP point as they are moved to their appropriate locations, also increasing the over-all RMS error. Our total RMS error was less than 19 meters, after which we resampled the SPOT 20 meter data to the new transformation matrix.

Here it is, a rectified SPOT 20-Meter Multi-Spectral Image!

Image Classification

Digital image classification is the process of assigning pixels to classes. By comparing these pixels to each other and pixels of known identity, it is possible to assemble groups of similar pixels into classes the informational categories of interest to users of remotely sensed data (Campbell, 1987).

Unsupervised Classification was used in our project to determine vegetation types along avalanche trimlines in our study area. This is the identification of natural groups or structures within multi-spectral data, as well as definition, identification, labeling, and mapping of natural classes (Campbell, 1987). This method was used because of our ability to interpret the resulting classes after classification, using airphotos. The benefits to using unsupervised classification are minimizing the possibility of human error, the fact that we had no extensive prior knowledge of the region, and this method allows small unique classes to be identified (Campbell, 1987). Due to the particularly bad weather and snow pack that we have encountered this year, we were unable to ground-truth the Logan Peak area for a supervised classification.

We began this process by using the Unsupervised Classifier tool in Imagine. Our rectified Logan Peak image was used as the input raster file and we specified our output cluster file and signature set. The number of classes was set at 20 with 6 iterations, while convergence threshold and zeros were left as is (Ramsey, 1995).

We then relabeled the new classes, which can be displayed and labeled using the Raster Attribute Editor, and then selected eighteen different shades of green to specify our classes in the editor. This can be done by highlighting the number of the desired Row, then clicking on the corresponding color Row with the third mouse button and using the color wheel to design your own color (Ramsey, 1995).

We completed our classification by creating a Feature Space Plot, using the Signature Editor, to help determine the spectral reflectance of surface features in our image with greater accuracy. This can be done by clicking on the Feature/Create/Feature Space Layers menu and using our rectified Spot image as the input raster layer file. Then highlight the Red and the NIR bands in the Create Feature Space Maps dialog, and click on the output to viewer button to display the feature space plot. To link these two images use the Feature/View/Select Viewer and Cursor, in that order, select the desired viewers to link and click on the Link button. The crosshairs generated allow you to determine vegetation types by using previously known vegetation types as sites (Ramsey, 1995).

Classification tools: info tool, signature editor, and unsupervised classification.

Classification of the image using the attribute editor and, color selector.

Our classified Image

The group decided to re-classify our image from 30 classes to 20 classes for a number of reasons, primarily our image was "pixilated" when we had to many classes. Our initial choice of colors for each class made the image difficult to read.

Here's an older classified image!

Use of Aerial Photography In Identifying Avalanche Terrain

The use of aerial photographs was pertinent in helping to determine cover types, and avalanche paths around the Logan Peak area in conjunction with the SPOT 20 meter imagery since spectral classification alone was not "fine" enough in terms of resolution to discern the small details needed in our classification. The photos we used were 1:12000 black and whites taken in August of 1981. Although these were not the best photos to use, they were the only ones available to us. Also, the use of orthophotos would have been a better choice but the Logan Peak quadrangle was again not available.

The first process in identifying vegetation cover types on aerial photos is to obtain some form of a stereoscope so that the viewer may view the photos pseudoscopically. With the level of accuracy we needed, our group opted in using a reflecting stereoscope. Once set up we began by locating Logan Peak, from there we began by identifying the vegetation around Logan Peak. This was somewhat a very difficult task because we were using 14 year old photos and cover types do tend to change in that amount of time. Nevertheless, with the SPOT imagery and the use of recognition elements we were able to identify avalanche paths and the vegetation surrounding them.

The methodology used to identify vegetation cover types was adopted from Avery and Berlin pages 52-56. Most people have some level of difficulty identifying vegetation from vertical photographs, but every aerial photo has characteristics known as recognition elements. These elements include shape, size, pattern, shadow, tone or color, texture association, and site (Avery & Berlin, 1992). Shape refers to the external form or configuration of an object. Size is merely the surface dimensions of an object. Pattern refers to the overall spatial form of related features. Shadows cast by objects are important because their shapes provide profile views of certain features that can facilitate their identification. Tone or color relates to the reflective characteristics of objects within the photographic spectrum. Texture is the visual impression of coarseness or smoothness caused by the variability or uniformity of an image. Association is one of the most useful tools in identifying objects on the ground. Site is the location of an object relative to its environment. With the use of these recognition elements and past photo interpretation experience we were able to identify the aspen, conifers, and avalanche paths on or near the Logan Peak area.

On the air photos, areas with high reflectancy (bare ground), sparse vegetation, and clear distinctions between vegetation, trimline and chutes characterized potential avalanche paths. Of particular interest were those areas that had a definitive line between conifer and small less dense vegetation in the steep chutes.

Our use of air photos also involved the classification of vegetation and other ground cover types near the city of Logan.

Topographic Analysis

An analysis of the study site using slope, elevation and aspect was conducted for the logan peak study area by generating .img files from the 30 meter DEM file.

Contour lines for the study area and contour interval information all came from 30 meter DEM files created using Imagines' image interpreter topographic analysis tools

One problem that was experienced with the contour lines was that they did not match the classified .img files coordinates. To correct this problem it was necessary to change the map information for the contour layer, specifically changing the X and Y coordinates to move the contours so they fit the layer below it.

In order to generate the slope file, particularly a color gradient by steepness, the attribute editor is used on the pseudo color .img file of the slope file created above. You can add a color gradient to the image showing differences in slope angle, with higher values being denoted by hot colors and low slope angles denoted by cool colors.

Here is a layer composition of the contour lines overlaid on the classified logan peak image and a larger version of the contour lines overlaid on the classified image

Conclusions

The following avalanche paths were identified using a combination of spectral image classification and aerial photography. It becomes evident that the level of detail needed to discern an edge affect within a classified image such as the trimline and bare ground interfaces will require imagery more detailed than the 20 meter data used, unfortuneately alternatives are currently unavailable . Possibly 10 meter panchromatic imagery in conjunction with the multispecral 20 meter data and orthophotos would allow more accurate identification despite pixilizing of the imagery, additionally making possible more of a GIS layer by layer analysis and overlay of these themes. A more detailed analysis that incorporates a GIS analysis using ARC INFO, particularaly the INFO database capabilities to record historical information on avalanche frequency and extent for a particular path, as well as to record field observations would be useful in creating a more detailed "avalanche atlas". Other useful analysis could include calculating the proportion of vegetation types within each avalache path AOI. With proper funding incentives we would be more than happy to tackle these tasks.

Avery, Thomas E., Berlin, Graydon L. 1992. Fundamentals of Remote Sensing and Airphoto Interpretation. Macmillan Publishing Company. New York, NY.

Campbell, James B. 1987. Introduction to Remote Sensing. The Guilford Press. New York.

ERDAS Field Guide. Third Edition Revised and Expanded. ERDAS Inc. Atlanta Ga. 1994

Lellesand, Thomas M., Kiefer, Ralph W. 1987. Remote Sensing And Image Interpretation. John Wiley & Sons, New York.

McClung, David And Schaerer, Peter 1993. The Avalanche Handbook. The Mountaineers. Seattle, Washington.

Ramsey, Doug.: Homework Assignments Winter Qtr. 1995.