9 th International Symposium on High Mountain Remote Sensing Cartography September 2006 Graz, Austria

Transcription

1 An Automated Method to Delineate the Ice Extension of the Debris-Covered Glaciers at Mt. Everest Based on ASTER Imagery Manfred F. Buchroithner and Tobias Bolch Summary Institute for Cartography, Technische Universität Dresden, D Dresden, Germany An automated morphometry-based approach to outline the debris-covered glaciers of the Mt. Everest Area (Nepalese Himalaya) based on ASTER data is presented. In order to obtain a satisfying ASTER DEM of this extreme topography unnaturally peaks were automatically identified and eliminated, and holes were filled with elevation data from topographic maps. The developed morphopetric glacier-mapping method includes altitude, slope, plane- and profile curvature as well as thermal information. The result clearly indicates the area of the debris-covered glaciers. Problems occur mainly in the distal parts of the glaciers tongues where the glacier ice is inactive and thick debriscovers show no signs of movement. Here, the resolution of the ASTER DEM is not sufficient. However, the results show the proof of this concept, and available multitemporal DEMs of higher resolutions in near future will still improve them. Keywords: Debris-covered glaciers, glacier mapping, morphometric analyses, ASTER DEM, Khumbu Himal, Mt. Everest 1. Introduction Glaciers are climate indicators because they interact sensitively with the climate. Since the end of the Little Ice Age, and especially since the end of the last century, a nearly global glacier recession is noticeable (Barry 2006). Recent glacier shrinkage results in an increasing coverage of the ice with debris (Bolch & Kamp 2006; Stokes et al. 2007). This, again, hampers the mapping of the actual ice snout by means of spaceborne imagery due to the spectral similarity to the surrounding bedrock (Hempel 2005). This is especially true for many glaciers in the Himalaya. On the other hand, satellite imagery basically represent an ideal tool to develop an automated way of outlining the glacier-ice extension in remote mountain areas. Consequently, recent approaches focus on a more accurate automated mapping of debris-covered glaciers. One promising method to distinguish between a debris-covered glacier and surrounding bedrock could be by analysing surface temperatures [1], because the underlying ice cools the supraglacial debris. This is measurable on the surface with the thermal signature from ASTER and Landsat imagery if the debris-cover does not exceed a certain density and/or thickness. Ranzi et al. (2004) showed at the Miage Glacier (Alps) the capability of the thermal band to identify the glacier if the debris cover does not exceed 0.5 m. Another approach for debris-covered glacier mapping is to include morphometric parameters. Bishop et al. (2001) presented a two-fold hierarchical model including elevation, slope, aspect and curvature derived from a SPOT DEM for debris-covered glaciers at Nanga Parbat (Pakistan) [2]. Paul et al. (2004) followed a semiautomated approach to map a debris-covered glacier in the Swiss Alps from a TM4/TM5-ratio image and using the slope gradient [3]. Further improvements of the glacier map were then reached by integrating a vegetation classification using multispectral data, neighborhood analyses and change detection. Zollinger (2003) used a multidimensional approach for debris-covered glaciers in Khumbu Himal [4]. It includes multispectral classification, slope (threshold 25 ) and filters. Bolch & Kamp 2006 presented a simple but robust method for alpine valley glaciers using clustering of curvature features [5]. All these methods show a potential for an automated mapping but are mostly developed for a special glacier type and still include several problems and inaccuracies. This is especially true if applied for the debris-covered glaciers at the study area at the Nepalese side of Mt. Everest (Chomolungma). The presented work is a contribution to the international program GLIMS (Global Land Ice Measurements from Space, Kargel et al. 2005). Page 1 of 8

2 2. General Characteristics of the Debris-Covered Glaciers at Mt. Everest Figure 1: Pseudo 3D-view (ASTER an ASTER DEM) of the Nepalese side of the Chomolungma Massiv with its debris-covered glaciers. The debris-covered glaciers at Mt. Everest ( Himalaya Type ) have special characteristics. In contrast to the Alps, where debris-cover occurs mainly at the distal part of the glacier tongue and sometimes in the form of middle moraines, the tongues of many large glaciers in the Himalaya are often several kilometres long, covered by a thick layer of debris (Moribyashi & Higuchi 1977, Fig. 1 & 2). The thickness of the debris-cover reaches up to several meters (Watanabe et al. 1986). These tongues are characterised by a rough surface, sharp declines of sometimes more than 20 meters and the occurrence of cryokarst formations and supraglacial lakes. The average slope of the whole debris-covered tongue is comparatively low. During field surveys the existence of small secondary moraines could be observed (Fig. 2). In addition, all investigated tongues contain parts of stagnant ice at the front, which are probably connected with the active glacier ice, but differ in their surface characteristics. These parts are smoother, and partly even open grass vegetation is able to grow. Richter et al. (2004) described the occurrence of grass vegetation and even trees on several debris-covered glaciers in mountain ranges around the world. Some terminal moraines show even the structures of a rockglacier. The characteristics are well described for the Khumbu Glacier (Iwata et al. 1980, Watanabe et al. 1986, Gades et al. 2000). The parts of stagnant ice are commonly included in the glaciers, e.g. also for the Nepalese Glacier Inventory (Mool et al. 2001). Here, it has to be mentioned that a clear delineation of the glacier termini is not possible or only with high uncertainties, even with ultra-high resolution images (e,g, IKONOS, QuickBird) or in the field. Page 2 of 8

3 Figure 2: Left: Debris-covered tongue of Lohtse Nup Glacier in the Imja Valley (Photo: T. Bolch 2006); Right: The same glacier in the ASTER Image (3-3-1). The arrow indicates the recent marginal moraine. 3. ASTER DEM ASTER data allow the generation of DEMs with a horizontal resolution of m, and in addition multitemporal data sets allow comparisons of changing DEMs of the same location. Although in high mountains the extreme topography makes DEM generation in general more difficult, ASTER DEMs can still be of good accuracy and useful for geomorphologicel and glacier mapping (Bolch & Kamp 2003, Bolch et al. 2005, Eckert et al. 2005, Gspurning & Sulzer 2004, Kääb et al. 2003, Kamp et al. 2005, Paul et al. 2004). The base for the morphometic analysis presented in this study is a self-generated ASTER DEM. Three ASTER scenes of the years 2001, 2002, and 2003 have been chosen. Using 18 ground-control points based on the map Mount Everest of the National Geographic Society (National Geographic) and approx. 100 tie points for each scene a DEM was generated from each stereo-pair using the software Geomatica 9.1. Mainly due to the extreme relief conditions these preliminary DEMs contained several data gaps. These occurred around the peaks and steep slopes. In addition, several abnormal spikes occur. The debris-covered glacier tongues, however, are well represented (Fig. 3 left). In order to minimize these errors postprocessing was performed for orthophoto generation and visualisation purposes. Most of the spikes could be treated by a subtraction of the DEMs. These areas were then cut out with a small buffer. In the next step several gaps could be removed through combination and averaging of all three generated DEMs. In the last step the remaining gaps were filled with relief data based on the contour-lines of the National Geographic Map. SRTM data could not be used, because they have also large gaps around the highest peak of the world. Subsequently lakes were automatically identified and integrated in the DEM, each with a constant height, based on the approach of Hänel (2002). The overall quality of the final DEM is promising: no gaps, and the relevant surface characteristics are well represented (Fig. 3 right). Page 3 of 8

4 Figure 3: Left: ASTER DEM with data gaps; Right: Final ASTER DEM. 3. Assessment of Existing Delineation Methods The exclusive use of thermal information [1] fails for glaciers with massive debris-cover. This is true for the glaciers in the Khumbu region. In addition, shaded non-glacier areas have surface temperatures similar to the clean ice areas, but debris-covered ice with direct radiation is somewhat warmer. This hampers especially in shaded moraine complexes a clear differentiation of glacier and non-glacier area. Another problem is the relatively coarse resolution of the thermal bands (Landsat 60 m, ASTER 90 m). Following the approach of Bishop et al. (2001) [2] for the Everest Area similar results as the ones described by the authors for the Nanga Parbat Area were obtained: The hierarchical model using slope, aspect, plane- and profile curvature proves its capability for glacier delineation, but still contains several inaccuracies. The main problems occur at the glacier termini and the glacier sides, if the transition to non-glacier zones is smooth, e.g. if the lateral moraine is missing or not represented in the DEM. The approach presented by Paul et al. (2004) [3] fails if transferred to the Himalayan debriscovered glaciers in its original form. However, if the gradient threshold is shifted to around 12 when using an ASTER DEM many parts of the debris-covered tongues can be included. Multitemporal data to delineate the termini are of no use due to the stable position of the glacier termini. Another problem occurs because of vegetation cover on some debris-covered spots. Therefore the NDVI as a threshold has to be used with caution. The multidimensional approach by Zollinger (2003) [4] shows promising results. However, several missclassifications occur, where bare rock-surfaces appear due the melting of the smaller debris-free glaciers since the end of the Little Ice Age. The morphometric mapping method based on curvature characteristics as presented by Bolch & Kamp (2006) [5] is very useful to describe the surface characteristics. However, due to the different characteristics of the Himalayan Type debris-covered glacier a distinct delineation is also not possible. Hence, to be successful a more complex method has to be applied. 4. Developed Mapping Method Visual interpretation of the ASTER thermal bands confirms that, despite the above mentioned problems, large areas of the debris-covered glaciers are identifiable. This means that thermal information can be used as additional information. Analysing the morphometric parameters in more detail one finds that the slope gradient (Fig. 4 left) and the curvature features (Fig. 4 right) are optimum to describe the surface characteristics and the transition to the surrounding moraines, mountain slopes and glacier forefields. The average slope gradient of the areas of interest is < 6, and except for very small areas with higher rates which reflect the typical rough surface, it does not exceed 10 in the ASTER DEM. The profile curvature, which represents the curvature in direction of the steepest slope, well indicates the mountain ridges and the lateral moraines. The plane curvature, which Page 4 of 8

5 represents the curvature in aspect direction aspect, is especially useful for hydrology features, e.g. to obtain areas with convergent water flow. Figure 4: Left: Slope gradients; Right: Profile curvature distrubution. Thus, a combination of the described features for the delineation of the debris-covered glaciers seemed to be promising. Therefore we defined a number of training areas distributed over several glaciers, which represent the typical surface characteristics. Based on these polygons statistical parameters like mean, minimum, maximum and standard deviation were computed, and then the characteristic interval was calculated for each parameter separately. Finally the determined areas were intersected (Fig. 5). In postprocessing steps isolated areas smaller then 0.1 km² and areas which are not connected with clean ice detected by means of ratio images were eliminated. Small gaps within the final areas were filled. The method was realised using ErdasImagine 9.0 and AML. NO ASTER-TIR Reference-Debris Plancurvature Glacierpolygon Reference-Debris Original-DEM Slope Glacierpolygon Reference-Debris Original-DEM Curvature Glacierpolygon Reference-Debris Determination of statistical Original-DEM values Glacierpolygon Reference-Debris Determination of (MIN, Original-DEM MAX, Sigma) Glacierpolygon statistical Determination values of (MIN, statistical MAX, Sigma) Determination values value of (variable) (MIN, statistical MAX, Sigma) values value (variable) Determination of (MIN, statistical MAX, Sigma) [MIN+value*sigma; NO non-potentialvaluesvalue (variable) (MIN, MAX, Sigma) MAX-value*sigma] [MIN+value*sigma; NO Debris-Glacier non-potential- value (variable) MAX-value*sigma] [MIN+value*sigma; NO Debris-Glacier non-potential- value (variable) MAX-value*sigma] YES [MIN+value*sigma; NO Debris-Glacier non-potential- MAX-value*sigma] YES [MIN+value*sigma; NO Debris-Glacier non-potentialpotential-debris-max-value*sigma] Debris-Glacier YES Glacier potential-debris- YES Glacier potential-debris- YES Glacier potential-debris- Glacier potential-debris- Glacier non-debris- Glacier YES Debris-Glacier Figure 5: Scheme for the morphometric glacier mapping (MGM) in Khumbu Himal. Page 5 of 8

6 5. Result and Discussion The final result of the morphometry-based glacier mapping well represents the outline of the debriscovered glaciers (Fig. 6). The deviation between the calculated and the manually digitized polygons is in the range of 2 to 5%. One spot, where there is actually no glacier anymore, was misinterpreted. The main problems occur at the distal and on some lateral parts of the debris-covered glaciers, where no clear indicators exist. Field survey showed that even on-site a definite borderline is not identifiable. This is especially true for areas where the debris-cover is several meters thick and stagnant ice exists. Here the transition to the forefield or to the adjacent non-glacier area is smooth. The small secondary moraines, which border the active glacier, are too small to be represented in the ASTER-derived DEMs. However, where distinct lateral moraines exist, the developed method could clearly delineate the glacier. Hence, the developed method shows very well the capability of the morphometry- and temperature-based glacier delineation, and the generated ASTER DEM turned out of reasonable use. However, due to the medium resolution of the DEM the typical glacier surface characteristics are only little reflected, and this limits the accuracy of the mapping results. In result one has to state that additional information has to be included to fulfil the requirements to describe the very complex topography of the glaciers and their surroundings. Therefore, further work will concentrate on the improvement of the knowledge about the surface characteristics if the debriscovered glaciers and their adjacent moraines. For this, ultra-high resolution satellite images (IKONOS and QuickBird) will be used in addition. Moreover, in the near future high-resolution multitemporal DEMs will be available, e.g. from the TanDEM-X SAR mission. Figure 6: Automatically delineated areas of debris-covered glaciers (yellow polygons). Page 6 of 8

7 Conclusion A successful approach to map the debris-covered glaciers at the Nepalese side of Mt. Everest based on ASTER imagery was presented. The generation of ASTER DEMs in areas with pronounced relief is with difficulties, e.g. data gaps and abnormal spikes occur. These faults, however, could be corrected, and the improved DEM reflects reality very well. The developed method based on morphometric parameters derived from the ASTER DEM and ASTER thermal information could delineate the debris-covered glaciers with satisfying accuracy. Problems occur mainly at the distal parts of the glaciers where stagnant ice exists and where it is even in the field difficult to distinguish between glacier and non-glacier. Further improvements are expected when using higher-resolution DEMs. Aknowledgements The authors like to thank Ulrich Kamp (University of Montana, USA) for the initiation of this project, André Kunert (Inst. f. Cartography, TU Dresden) for ASTER-DEM generation and algorithm realising and Basanta Shrestha (ICOMOD, Nepal) for the cooperation. ASTER data were provided at no cost through NASA/USGS under the umbrella of the GLIMS project. The project is funded by the German Research Foundation (DFG) under the code BU 949/15-1. Literature Barry, R. G., The status of research on glaciers and global glacier recession: a review. Progress in Physical Geography 30(3): 285. Bayr, K. J., D. K. Hall & W. M. Kovalick, 1994: Observations on glaciers in the eastern Austrian Alps using satellite data. International Journal of Remote Sensing, 15: Bishop, M. P., R. Bonk, U. Kamp & J. F. Shroder, Terrain analysis and data modeling for alpine glacier mapping. Polar Geography 25: Bolch, T. & U. Kamp, Qualitätsanalyse digitaler ASTER-Geländemodelle von Hochgebirgsregionen (Cerro Sillajhuay, Chile/Bolivien). Kartographische Nachrichten 5: Bolch, T. & U. Kamp, Glacier mapping in high mountains using DEMs, Landsat and ASTER data. Proceedings of the 8th international symposium on High Mountain Remote Sensing Cartography, March 2005, La Paz, Bolivia: im Druck. Bolch, T., U. Kamp & J. Olsenholler, Using ASTER and SRTM DEMs for studying geomorphology and glaciation in high mountain areas. M. Oluic (Hg.): New Strategies for European Remote Sensing, Millpress, Rotterdam: Eckert, S., T. Kellenberger & K. Itten, Accuracy assessment of automatically derived digital elevation models from aster data in mountainous terrain. International Journal of Remote Sensing 26(9): Gades, A., H. Conway, N. Nereson, N. Naito & T. Kadota, 2000, Radio echo-sounding through supraglacial debris on Lirung and Khumbu Glaciers, Nepal Himalayas. IAHS Publication 264: Gspurning, J. & W. Sulzer, DEM-Generierung aus ASTER-Daten und Evaluierung. J. Strobl, T. Blaschke & G. Griesebner (Eds.) Angewandte Geoinformatik Beiträge zum 16. AGIT- Symposium Salzburg.. Heidelberg, Wichmann Verlag: Hänel, C., Integration von Topographiedaten in ein Hochgebirgs-GIS des Zentral-Altai. Unpubl. semester thesis, Institute for Cartography, TU Dresden. Hempel, A., Vergleich von Methoden zur Abgrenzung schuttbedeckter Gletscher im Khumbu Himal mit ASTER- und Landsatdaten. Unpubl. semester thesis, Institute for Cartography, TU Dresden. Kääb, A., C. Huggel, F. Paul, R. Wessels, B. Raup, H. Kieffer & J. S. Kargel, Glacier monitoring from ASTER imagery. Accuracy and applications. EARSeL eproceedings 2 - Observing our cryosphere from space Page 7 of 8

8 Kamp, U., T. Bolch & J. Olsenholler, Geomorphometry of Cerro Sillajhuay, Chile/Bolivia: comparison of DEMs derived from ASTER remote sensing data and contour maps. Geocarto International 20(1): Kargel, J. S., M. J. Abrams, M. P. Bishop, A. Bush, G. Hamilton, H. Jiskoot, A. Kaab, H. H. Kieffer, E. M. Lee, F. Paul, F. Rau, B. Raup, J. Shroder, D. Soltesz, D. Stainforth, L. Stearns & R. Wessels, Multispectral imaging contributions to global land ice measurements from space. Remote Sensing of Environment 99(1-2): Iwata, S., O. Watanabe & H. Fushimi, Surface morphology in the ablation area of the Khumbu glacier. Journal of the Japanese Society of Snow and Ice 41 (Special Issue): Mool, P. K., S. R. Bajracharrya & S. P. Joshi, Inventory of glaciers, glacial lakes, glacial lake outburst floods: Monitoring and early warning systems in the Hindu Kush-Himalayan Region, Nepal. ICIMOD, Kathmandu, Nepal, pp. 386, Moribyashi, S. & H. Higuchi, Characteristics of glaciers in the Khumbu region and their recent variations. Journal of the Japanese Society of Snow and Ice 39 (Special Issue): 3-6. National Geographic, Map Mount Everest Sargarmatha (Nepali) Qomolangma (Chinese), scale 1: Paul, F., C. Huggel & A. Kääb, Combining satellite multispectral image data and a digital elevation model for mapping of debris-covered glaciers. Remote Sensing of Environment 89(4): Paul, F., A. Kääb, M. Maisch, T. Kellenberger & W. Haeberli, The new remote sensing derived Swiss Glacier Inventory: I. Methods. Annals of Glaciology 34: Ranzi, R., G. Grossi, L. Iacovelli & T. Taschner, Use of multispectral ASTER images for mapping debris-covered glaciers within the GLIMS Project. IEEE International Geoscience and Remote Sensing Symposium Vol. II: Richter, M., T. Fickert & F. Grüninger, Pflanzen auf schuttbedeckten Gletschern - wandernde Kuriosiäten. Geo-Öko XXV: Stokes, C. R., V. Popovnin, A. Aleynikov, S. D. Gurney & M. Shahgedanova, Recent glacier retreat in the Caucasus Mts, Russia, and associated increase in supraglacial debris cover and supra/proglacial lake development. Annals of Glaciology 46. in press. Watanabe, O., S. Iwata & H. Fushimi, Topographic characteristics in the ablation area of the Khumbu Glacier, Nepal Himalaya. Annals of Glaciology 8: Zollinger, S., Ableitung von Parametern für die Identifikation und Beobachtung gefährlicher Gletscherseen in Nepal aus ASTER Satellitendaten. Diploma thesis, Geographisches Institut, Universität Zürich, pp Page 8 of 8