The study presented here was part of an interdisciplinary project to assess the vulnerability of a hydropower plant in Peru in relation to natural hazards. It was concerned with potential hazards and risks related to glaciers. The catchment of the hydropower plant, located at the eastern (Amazonian) slope of the Altiplano, includes two important glacierized areas: the Cordillera Carabaya in the South and the Cordillera Vilcanota in the North.
Due to the remoteness of the area, available data was scarce in general. With respect to glacier data, the only information available consisted in the Glacier Inventory of Peru from 1962. Therefore, it was necessary to acquire new, present-day data in order to assess the current hazard situation. Satellite data was found to match best the study requirements due to its area-wide, cost-effective and most current information characteristics. Relevant hazards to the plant were outbursts from glacier lakes, possibly in combination with ice avalanches, and, in more economic terms, decreasing runoff for hydropower production because of glacier shrinkage.

Landsat-7 ETM+ image (09-08-1999) showing the Quelccaya Ice Cap using the spectral bands TM5, TM4 and TM3 (RGB). Glaciers appear in light blue, lakes in dark blue and vegetation in green.
 
 

In a first step, the spectral information of Landsat-7 ETM+ could be used to extract lake surfaces automatically. The Normalized Difference Water Index (NDWI) uses two spectral bands:
(TM4 -TM1)/(TM4 + TM1)
Lakes could thus clearly be extracted as black areas.
 

Once lakes in a glacierized area had been detected, a search had to be made for potentially hazardous lakes. Lakes in direct contact with glaciers should be considered as potentially critical. Fusion of Landsat-7 multispectral (30 m ground resolution) and panchromatic data (15 m) allowed for a closer examination of the detected area (image below). Comparison with aerial photography from 1962 facilitated the estimation of the development of one potentially dangerous site (photograph below).
 
 

Estimation of the lake volume from satellite data and examination of the flow path (slope, debris reservoirs, further lakes) in satellite data and topographic maps enabled the estimation of the potential reach of a lake outburst event. Based on the above hazard assessment, a field campaign to study the lake (above) and the whole flow path down to the inlet into the main river was carried out, and mitigation measures could be proposed.

Outburst flood modeling

The case of an outburst of the lake was investigated in more detail on the basis of the July 2001 ASTER scene. As for the ETM+ scene, collection of ground control points had to made on the basis of 1:100'000 topographic maps since no better information was available. With roughly 20 ground control points distributed evenly over the whole scene a root mean square error of about 3 pixels was achieved. In view of the scarcity of data in such a remote area, this was considered to be sufficient to compute a digital elevation model (DEM) for outburst flood modeling. The potential of ASTER data for derivation of DEMs is due to stereo viewing of a separate sensor. For DEM calculation, an epipolar image pair with the co-registered nadir-looking (3N) and back-looking channel (3B) is computed. A DEM of the whole scene with coarser ground resolution (60 m) for orthorectification of the multispectral channels 3N,2,1 and a DEM of a subset with higher resolution (30 m) for outburst flood modeling was then calculated (15), (16).

Input data to the outburst flood model is basically the DEM derived from ASTER data and the area of the lake extracted from the NDWI image. The pixels of the lake represent the starting point of the outburst model. The modeling is performed within a GIS-environment and is based on hydrological flow modeling. The direction of the downhill flow is according to the next steepest neighbour in a 3x3 window (D8 method), (17). In addition to the flow along the steepest path, a factor is introduced which allows the flow to divert horizontally up to 45° on both sides. A linear function defines that the more the flow diverts from the steepest downslope direction the greater is the resistance. Following the specific terrain downvalley from the outburst source, a certain area is covered by the outburst flood. The probability that a pixel at the very side of this area be affected is lower than for a corresponding pixel along the steepest downslope direction. Fig. 4 shows the outburst flood model starting from the lake. The colors relate to the probability that a certain pixel be covered by the outburst flood. The outburst flood can be simulated with satisfactory level of detail. Downvalley from laguna Huañunacocha, the model fails due to errors in the DEM caused by cloud cover. The main part of the distance is marked by a medium-probability hazard.