Detection of glacier lakes

According to the structure of the project, the first level covers large areas and performs the analysis to allow the second level to focus on the selected hazard sites. In the third level, detailed small-scale studies are applied.

Level 1)
The coarse level of mere detection of glacial lakes over large areas. This is motivated by the fact that the location of many lakes, especially in remote high mountain areas, are not known. This level aims at a complete area-wide application.
In a first phase, the detection of glacial lakes itself by multispectral imagery is basically a matter of water / non-water discrimination.
Applying basic techniques of multispectral classification similarly to the widely used Normalized Difference Vegetation Index (NDVI) a Normalized Difference Water Index (NDWI) is introduced. According to the idea of two spectral channels with maximum reflectance difference for an object (i.e. water) a blue channel (maximum reflectance of water) and a near-infrared channel (minimum reflectance of water) were chosen, i.e. for Landsat-TM:

NDWI = (TM4 - TM1) / (TM4 + TM1)
 

The above images show the NDWI image from Fletscherhorn area (Wallis, Switzerland) such as calculated with the TM data. The disturbing shadow areas originally present on the image (left) were filtered out by the integration of digital elevation data (right). With the reconstruction of sun elevation angle and azi-muth at the exact time of the satellite data acquisition, a cast shadow mask was computed using a digital elevation model (DEM) of 25 x 25 m ground resolution. The mask was then overlaid on the NDWI image. It could thus be assured that only lakes appeared as black spots. Similar tests with a DEM of coarser resolution and minor quality indicated that resolution ad quality of the DEM are crucial to obtain satisfactory results

 Level 2)
Once potential hazard sources (i.e. periglacial and glacial lakes) have been detected in downscaling step 1) a more detailed analysis of the extracted lakes is required in step 2) in order to evaluate the hazard potential.
In a first approach, the spatial resolution of the remote sensing data was enhanced by integration of a panchromatic SPOT image of 10 x 10 m ground resolution (SPOT-2 scene recorded on 27-8-1994). However, the improvement of the spatial resolution came thereby along with a loss in spectral information. Aiming at the full exploitation from the potential of both remote sensing systems (i.e. SPOT-Pan and Landsat-TM), fusion techniques were considered to be a promising method.
The algorithm chosen in the present study is based on a method proposed by Munechika et al. (1993) and takes the spectral sensitivity of the input channels into account. The spectral characteristics of the remote sensing data can thus be better preserved. The enhanced spatial resolution together with the spectral information of the image after the fusion allows for a more detailed analysis of the potential hazard sites. Within the primary factors in determining the hazard potential of a lake is lake size (area, volume) since it determines the amount of water available in case of an outburst. Vegetation, which can give important indications on the stability of a moraine, can be distinguished from non-vegetated terrain due to its high reflectivity in near-infrared ranges. Large debris reservoirs, in most cases visually detectable, can be mobilized during a lake outburst. Their assessment can thus contribute to estimate the potential debris flow volume. The clear detection of moraine dams on a visual basis can be a tough challenge, even in high resolution images, as long as stereo view is not available. Sucessful detection also depends on the physical size (height, width) of the dam. The integration of high resolution DEMís - if available - can substantially facilitate the problem because of their abil-ity to extract geomorphometric object characteristics of moraine dams.
 
 
 

A fusion image of Landsat-TM (4,3,2) and SPOT-Pan of Belvedere glacier and Lago delle Locce. Vegetation is in red, the moraine breach of the lake outburst and the corresponding outburst flood path is clearly recognizable.
 
 

Level 3)
The third downscaling level includes the most detailed and specific hazard evaluation. It is based on the analysis of the previous levels and becomes relevant in areas where an important hazard potential has been found in downscaling steps 1) and 2).
Technically, photogrammetric methods are usually applied at this levels. In the Gruben area extensive photogrammetric studies on glacier lakes have been carried out. High-precision photogrammetric techniques allow for monitoring lake levels and changes in ice thickness (important for ice-dammed lakes). In the case of moraine dams, the lake level in relation to the dam height determines the freeboard and thus relates to the risk of an upcoming outburst. Dam width and height are deducible from standard photogrammetric procedures and have both implications for the stability of the dam and its vulnerability to overtopping and erosion by displacement waves from ice or rock falls. Additionally, debris and rock slope instabilities and ice movements in hanging glaciers can be continously observed. The objective is thereby to foresee potential trigger events of lake out-bursts.
Most recently, digital photogrammetry has enabled the development of special techniques for the measurement of surface deformation (Kaeaeb and Vollmer, in press).
It is important to mention that in any situation recognized as critical through level 1) to 3), field studies (e.g with geophysical techniques) are inevitable for an ultimate assessment of the hazard situation.