ABSTRACT
Glacier inventory is important both to provide estimates of freshwater storage and as an indicator of climatic variability. Glacier records in Canada are not as extensive as in some countries owing to the large landmass and relatively recent development of glacierized areas, although there is a significant history of glacier inventory efforts using conventional analogue methods.  Landsat Thematic Mapper (TM) data were combined with regional digital elevation data to classify and map the Illecillewaet Icefield area in Glacier National Park, B.C. The best results were obtained by utilizing the second, third and fourth principal components of analysis of the glacierized area, isolated under a mask created from a PCA of the entire scene, combined with TM4/TM5 ratio and NDSI (normalised difference snow index) images as input to a maximum likelihood classification. Qualitative assessment of this method suggests that it can successfully avoid problems associated with sensor saturation and shadowed areas and can discriminate debris mantled ice and ice-marginal water bodies.

     The glaciers of the Columbia Mountains in British Columbia represent a significant reservoir of water in the Columbia River Basin. Various water supply interests depend to a significant degree on runoff from these glaciers, especially during dry periods, while long-term glacier change is considered an effective index of overall climate change. Periodic inventory of glacier attributes is an important component of glacier volume assessment. Attributes such as equilibrium line altitude and accumulation area ratio of these glaciers can indicate mass balance trends. These attributes can be extracted from the classification of LANDSAT data

A single Landsat TM scene captures an area covered by hundreds of airphotos, minimizes relief displacement, and can afford sufficient resolution to discriminate the features of interest. Landsat scenes have been available for almost three decades (since 1972 for MSS, 1982 for TM) and can be geocoded, orthorectified and combined with digital elevation data to provide an accurate integrated database for deriving important glacier inventory attributes using image processing techniques.

      Landsat Thematic Mapper (TM) data were combined with regional digital elevation data to classify and map the Illecillewaet Icefield area in Glacier National Park, B.C. The best results were obtained by utilizing the second, third and fourth principal components of analysis of the glacierized area, isolated under a mask created from a principal components analysis ( PCA ) of the entire scene, combined with TM4/TM5 ratio and NDSI ( normalised difference snow index ) images as input to a maximum likelihood classification. Qualitative assessment of this method suggests that it can successfully avoid problems associated with sensor saturation and shadowed areas and can discriminate debris mantled ice and ice-marginal water bodies

Pixel saturation is recognized as a typical problem over glaciated and snow covered scene areas, particularly in the visible bands: TM1, 2 and 3 (Hall et al 1988 ). PCA reduced this by identifying most of the scene brightness variance and thus the saturation, with the first principal component.            Subsequent principal components, especially the second, third and fourth, were found to depict strong, unsaturated contrast over the glaciated areas, enhancing surface features and facies ( larger image ).

            Further image processing involved ratioing and a normalized difference snow index. The TM4/TM5 ratio is cited by Hall et al (1987 ) as effective for discriminating the ice and snow facies in glaciological studies, particularly through areas of shadow. The NDSI helps distinguish snow from similarly bright soil, rock and cloud (Dozier, 1989 ). It is calculated through image arithmetic using the following relationship:

This has been shown to be an effective index for mapping snow cover in rugged terrain (Hall et al, 1995 ).

      Challenges facing automated mapping of glacier areas include the discrimination of the ice and snow facies of the glacier, identification of debris covered ice, topographically and cloud shadowed areas and water bodies marginal to the glaciers. Glacier facies are fundamentally divided into ice and snow facies, with the border between the two describing the transient snowline. Late in the mass balance year, the transient snowline can be regarded as approximating the glacier equilibrium line.

       The difference between water saturated snow and wet firn or ice at the transient snowline can be difficult to discriminate, particularly when physical and radiometric conditions vary through a scene. Debris covered ice includes supraglacial moraine , ice-cored marginal moraine and buried ice. A thin supraglacial debris cover significantly alters the spectral signature of ice, while thickly covered ice cannot be discriminated spectrally from surrounding moraine.

        Shadowed areas are less spectrally varied than illuminated areas, resulting in greater classification difficulty. Additionally, it is noted that cloud shadow on snow in the study scene shows a signature very similar to illuminated ice, leading to their confusion in many classification attempts. Water bodies also have a signature very similar to that of glacier ice, leading to potential misclassification of marginal lakes as ice.

       The supervised classification was trained on eleven classes representing three glacier facies, snow, firn and ice; bedrock and moraine forefield facies and water each under both illuminated and shadowed conditions. Training classes were not established for vegetation and clouds. Supervised classification for glacier mapping of unprocessed TM scenes were found to be badly hindered by cloud and topographic shadows.

       Classification using the second, third and fourth principal components yielded results which avoided significant misclassification of shadowed areas and water bodies. However, the overall glacier area was slightly under-represented, recognized by overlaying the classification result on a TM5-4-3 composite. The addition of the NDSI and ratio TM4/TM5 promoted the inclusion of virtually all glacier area. Nunataks are correctly identified under all illumination conditions, as are medial and dispersed supraglacial moraines. Ice marginal water bodies are correctly discriminated.

      Areas of cloud shadow on snow are committed erroneously to the firn class, as are some very steep topographically shadowed parts of the snowfield (see "1" on Figure 2 ). Ice is felt to be accurately represented but challenges were encountered in discriminating ice from heavily shadowed bedrock areas and water bodies. Areas of highly fractured ice, such as crevasse fields and the base of icefalls are easily mistaken for firn. An example of what is regarded as a successfully classified shadow area is marked " S" while an erroneously committed area of topographic shadow without glacier is marked "X" .

      Debris covered ice is recognized as a significant challenge in glacier inventory mapping ( Whalley and Martin, 1986 ). Automatic classification can only discriminate areas whose spectral character is influenced by the underlying ice. In this study area, there is a dispersed cover of debris through which ice spectral characteristics can be seen, and where mixed pixels of debris/moraine/bedrock and pure ice are found. The latter case occurs around the glacier terminus, explaining the margin of red pixels seen on the Illecillewaet Glacier

DIRECTIONS FOR FUTURE WORK

      Refinement of the classification method and rigorous accuracy assessment promise to facilitate the production of accurate maps of glacier extent and facies. Integration of the digital elevation model with the dataset will facilitate the derivation of important glacier inventory attributes. The location and extent of each glacier are implicit to the dataset, as are accumulation and ablation areas, from which the accumulation area ratio (AAR) is derived.

     Maximum, minimum and median elevations, hypsography, orientation of accumulation and ablation areas, and elevation of the transient snowline are all important factors of mass balance which can be easily extracted from the dataset (Ommanney, 1980 ); (Ostrem and Haakensen, 1980 ). Maintenance of the inventory within a GIS environment will allow queries to be made and reports generated for any scale of enquiry, from individual glaciers, to icefield and glacier regions. This reduces the need to compile the extensive tabular reports which characterize past glacier inventories.  

CONCLUSIONS

      Supervised classification of Landsat TM scenes in the mapping of glacier extent for glacier inventory purposes appears to be a reasonable expectation, and through time for change detection studies and the projected influence of climate change and global warming. Principal components analysis, image ratioing and image differencing produce superior classification input channels compared to unprocessed TM bands.
       A secondary set of components with loadings and generated images based on glacier surfaces alone provide the most useful image channels for classification, highlighting local variations and reducing the influence of the surrounding terrain. Lower components provide image data that may yield further information related to glacier surface topography and is not currently available from digital elevation files. Identification of the transient snowline under varying radiometric conditions is difficult, but may be improved by refinement of the discrimination between  and wet snow classes.

ACKNOWLEDGEMENTS

The study was in collaboration with Drs. Brugman and Pietroniro of the Columbia Mountain Institute of Applied Ecology and the cryospheric systems research initiative ( CRYSYS ), a Canadian Department of Environment and University contribution to the NASA Earth Observing System (EOS ) program. The authors wish to acknowledge CRYSYS for providing the operating funds for this study.

LINKS

A guide to day hikes in Glacier National Park  Columbia Mountain Institute of Applied Ecology CRYSYS Glossary of glacial terminology  GLIMS -Global Land Ice Measurement from Space  LANDSAT NASA Parks Canada
	UNBC Geography

REFERENCES

Adam, S. 1996. Snow line mapping using radar imagery, Place Glacier, B.C.. Unpublished M.Sc.       Thesis, University of Saskatchewan, Canada. 132 pp.

Boresjö-Bronge, L. and Bronge C., 1996. Landsat TM data and ground radiometer measurements for snow and ice type classification in the Vestfold Hills, East Antarctica. Proceedings of the Fourth Circumpolar Symposium on Remote Sensing of the Polar Environments, Lyngby, Denmark. ESA SP-391. pp. 71-80.

Brugman, M.M., Pietroniro, A. and Shi, J., 1996. Mapping alpine snow and ice using Landsat TM and SAR imagery at Wapta Icefield. Canadian Journal of Remote Sensing, .22, 1, 127-136.

Champoux, A. and Ommanney, C.S.L., 1986. Evolution of the Illecillewaet Glacier, Glacier National Park, B.C., using historical data, aerial photography and satellite image analysis. Annals of Glaciology, 8, 31-34.

Civco, D.L., 1989. Topographic normalization of Landsat Thematic Mapper digital imagery. Photogrammetric Engineering and Remote Sensing, 55, 9, 1303-1309.

Dozier, J., 1989. Spectral signature of Alpine snow cover from the Landsat Thematic Mapper. Remote Sensing of Environment, 28, 9-22.

Dozier, J. and Marks, D., 1987. Snow mapping and classification from Landsat Thematic Mapper data. Annals of Glaciology, 9, 97-103.

Hall, D.K., Ormsby, J.P., Bindschadler, R.A., and Siddalingaiah, H., 1987. Characterization of snow and ice reflectance zones on glaciers using Landsat TM data. Annals of Glaciology, 9, 104-108.

Hall, D.K., Chang, A.T.C., and Siddalingaiah, H., 1988. Reflectances of glaciers as calculated using Landsat-5 TM Data. Remote Sensing of Environment, 25, 311-321.

Hall, D.K., J.L. Foster, J.Y.L. Chien and G.A. Riggs, 1995. Determination of actual snow covered area using Landsat TM and digital elevation model data in Glacier National Park, Montana. Polar Record, 31, 177, 191-198.

Howarth, P.J. and Ommanney, C.S.L., 1986. The use of Landsat digital data for glacier inventories. Annals of Glaciology, 8, 90-92.

Lodwick, G.D. and Paine, S.H. (1985) A digital elevation model of the Barnes ice-cap
derived from Landsat MSS data. Photogrammetric Engineering and Remote Sensing, 51, 12, 1937-1944.

Ommanney, C.S.L., 1980. The inventory of Canadian glaciers: procedures, techniques, progress and applications. World Glacier Inventory (Proceedings of the Riederalp Workshop, September 1978): IAHS-AISH Publ. no. 126. pp. 35-44.

Ommanney, C.S.L., 1986. Mapping Canada?s glaciers since 1965. Annals of Glaciology, 8, 132-134

Orheim, O. and Luccitta, B.K., 1987. Numerical analysis of Landsat Thematic Mapper images of Antarctica: surface temperatures and physical properties. Annals of Glaciology, 9, 109-120.

Østrem, G., 1975. ERTS data in glaciology - An effort to monitor glacier mass balance from satellite imagery. Journal of Glaciology, V15, 73, 403-415.

Østrem, G., and Haakensen, N., 1980. The Scandinavian glacier inventory - two glacier atlases. World Glacier Inventory (Proceedings of the Riederalp Workshop, September 1978): IAHS-AISH Publ. no. 126. pp. 161-171.

Whalley, W.B. and Martin, H.E., 1986. The problem of hidden ice in glacier mapping. Annals of Glaciology, 8, 181-183.

BACK to UNBC GIS Lab Project Home Page