At the beginning of glacier studies with Landsat, (1) false colour composites (FCCs) from MSS digital data (later also TM) were used to delineate the accumulation and ablation area of a glacier or to classify different ice and snow facies. Afterwards and also recently glacier mapping with TM was carried out by (2) manual delineation of the glacier outline with a cursor on a FCC from MSS data and combination with a TM image. Further studies used (3) segmentation of ratio images with reflectance thresholds and (4) different supervised classification techniques. Another aspect (5) is the calculation of glacier reflection with TM and the comparison with ground measurements. Furthermore some studies about snow mapping with TM exists (e.g. DOZIER 1989). They can be used for calculation of the accumulation area ratio (AAR) of a glacier to estimate its mass balance. All methods and corresponding authors are summarized in the table below.

(1) FALSE COLOUR COMPOSITES

ØSTREM (1975) was one of the first userof Landsat images (MSS data) for glaciological studies. He generated prints of FCCs to distinguish between accumulation and ablation area of numerous outlet glaciers from the Jostedalsbreen ice cap in Norway. He already pointed out the large capabilities of Landsat data for mass balance estimations of a large number of glaciers at the same time. ROTT (1976) used contrast enhanced prints from MSS data to estimate the height of the snow line of about 50 glaciers in the Tyrolean Alps and calculated their AAR. DELLA VENTURA et al. (1983) developed a technique (MSS 5 * MSS 6) for automatic glacier classification (in ice, snow, other) for a small region in Tyrol and revealed quite good results.WILLIAMS (1987) used a MSS winter image with low solar elevation for morphological mapping of the Vatnajøkull ice cap and a MSS late summer image with a special contrast enhancement for delineation of different ice and snow facies in a FCC. A comparison of these facies was given in WILLIAMS et al. (1991), together with an analysis of the distinguashable facies with Landsat TM (ice, slush, snow).

(2) MANUAL DELINEATION

The second method was applied especialy to Landsat MSS data and the derived glacier outlines were afterwards combined with a false colour composite (FCC). After coregistration of both images changes in glacier length were derived, and showed good agreement in comparison with in-situ measurements (HALL et al., 1992). Some more studies situated in the Mt. Wrangell area and in the College Fjord (Alaska) with MSS derived glacier boundaries combined with a TM FCC are discussed in WILLIAMS and HALL (1993). Also WILLIAMS et al. (1997) used cursor tracking of the glacier margin of Vatnajökul to calculate changes in glacier length. ROTT and MARKL (1989) had shown that interactive tracking of the glacier boundary with a cursor by an expert reveal more accurate results than an automatic algorithm. Nevertheless, for a larger number of glaciers this method is not very practical, allthough some problems (debris cover on the glacier tongue) were removed by this method.

(3) SEGMENTATION OF RATIO IMAGES

Furthermore both authors concluded that an algorithm of type (3) using TM 3 / TM 5 ratio images revealed better results in shadow zones than TM 4 / TM 5, whereas the latter showed better performance in mapping glacier areas facing towards the sun. Also ROTT (1994) applied method (2) and calculated spectral reflactances (R) from TM channel 3 and 5 before segmentation of the image into different classes. He assigned values of [R(TM 3) / R(TM 5)] > 1.3 to the snow and ice class and separated this class into snow for R(TM 3) > 0.48, firn (0.32 < R(TM 3) < 0.48) and ice (R(TM 3) < 0.32). Furthermore, JACOBS et al. (1997) used a segmentation threshold of R(TM 4) / R(TM 5) > 1.0 to delineate the perimeter of the Barnes Ice-cap. HALL et al. (1988) calculated the planetary reflectance at the satellite sensor to delineate three different zones (bare ice, wet snow and fresh snow) for glaciers in Austria and Alaska. BAYR et al. (1994) used ratios of the digital numbers from TM 4 and 5 to delineate the glacier boundary in the ablation zone for two different glaiers in Austria. Furthermore they calculated the average reflectance of the glacier basin and the area of each glacier in three different years.

(4) CLASSIFICATION

GRATTON et al. (1990) used a GIS to improve the classification accuracy of Landsat MSS and TM data for mountain glacier mapping. Within the GIS environment they performed automated selection of training areas for a Maximum-Likelihood classification. Although high accuracy is achieved for most classes (snow, clean ice, vegetation, rock, water), regions with debris had to be classified by visual inspection to improve the correlation coefficient of the classification results with a digitized map from 0.74 to 0.87 (with TM) and from 0.61 to 0.85 (with MSS). Also BINAGHI et al. (1993) used a GIS to improve the quality of glacier mapping with TM data and to update glacier inventory parameters more efficiently. For classification they used Fuzzy-Set Theory and the Dempster-Shafer Theory to attribute probability distributions to sets of classes (Basic Probability Assignments). They combined them with ancillary data to calculate the final classification (snow, ice, other). An inventory of the Southern Patagonian Icefield (SPI) based on TM data from 14. January 1986 was done by ANIYA et al. (1996). They carried out a cluster analysis (ISODATA) with TM bands 1, 4 and 5 on Moreno Glacier and grouped the 20 clusters into three classes (snow, ice and rock). For the whole SPI they performed a parallelepiped or maximum-likelihood (in cases of ambiguity) classification. If misclassification was apparent (shade, moraine and supraglacial debris) manual correction was applied. They also calculated an average AAR of 0.75 for the entire SPI. An evaluation of different glacier mapping algorithms was carried out by SIDJAK and WHEATE (1999). They applied supervised maximum-likelihood classification to different combinations of input bands. These bands were made of principle components (1-4 or 2-4), partly under a glacier mask derived from PC2 by thresholding, a NDSI (TM2-TM5/TM2+TM5), a TM4/TM5 ratio image or a TM 5,4 and 3 composite image. The best classification was revealed with a combination of PC 2-4, TM4/TM5 and the NDSI as the input bands. Even nunataks, medial and dispersed supraglacial moraine was correctly mapped. Another evaluation of snow and ice mapping methods is carried out by BRONGE and BRONGE (1999) for Antarctica. They created a mask of snow and ice covered regions by grey level slicing of a TM 3 image and performed a PCA with all TM reflective bands under these mask. A Maximum-Likelihood classification was performed with PC 1-4 as input and 55 training areas with different illumination conditions for ice and snow. Analysis of the results showed good coincidenceagreement with simultaneous field measurements. The TM 3 / TM 4 ratio proved to be a good tool for blue ice/snow separation, because it is independent of thin clouds and cloud shadows.

(5) REFLECTION

A comparison of Landsat TM derived snow and ice reflectance with field measurements at the same time, is carried out by HALL et al. (1990). By applying an atmospheric correction for scattering and absorption, reflectance increased by 5-17% and correspond with field measurements within 6%. During snow metamorphosis and with decreasing viewing angle the anisotropy of snow reflectance increased, so that TM-derived reflectances are only good estimates for nadir viewing conditions. Also KOELEMEIJER et al. (1993) compared Landsat TM derived reflectance from the tongue of the Hintereisferner (Austria) with in-situ measurements, but from a different year. They also corrected for atmospheric scattering and absorption and obtained a good coincidence between the measurements. They pointed out, that the decrease in reflection towards the glacier front depends on crevasses, dust, debris and concentration of air bubbles in the ice. WINTHER (1993) compared TM derived relectance from August 1987 and 1988 with field measurements from August 1992 and 1993 over two glaciers in Svalbard. He discovered large variations in albedo between 5 days, the two investigated years and spatialy on both glaciers. The derived albedo was influenced by atmospheric scattering and absorption, the topography (slope & aspect), the pixel itself and adjacent pixels, and the anisotropy of the snow reflectance. A different approach was followed by GRATTON et al. (1993). They determined the net radiation balance of the Athabasca Glacier basin from TM data, using a DEM and the radiative transfer code LOWTRAN 6. They produced a surface-cover map from a supervised classification procedure and calculated albedo from TM band reflectance with different weighting schemes for different cover types. Furthermore they computed TM 6 brightness temperatures to derive the terrain emitted radiation. As a supplement also the work carried out by JACOBSEN et al. (1993) should be mentioned: They derived surface albedo of the Mitdluagkat Glacier in Greenland from SPOT HRV data and a DEM (resampled to a spatial resolution of 60m). Their values corresponded very good withto field measurements from other authors at other sites. They also map the spatial distribution of accumulation and ablation zones on the glacier.