Glacier mapping from multi-temporal optical remote sensing data within the Brahmaputra river basin R. Frauenfelder a, b, A. Kääb b * b a Presently at: Norwegian Geotechnical Institute, Pb. 3930 Ullevaal Stadion, 0806 Oslo, Norway – rf@ngi.no Department of Geosciences, University of Oslo, PO Box 1047 Blindern, 0316 Oslo, Norway – andreas.kaab@geo.uio.no Abstract – Glacier distribution and glacier changes were investigated in the Upper Brahmaputra River Basin using remote sensing and GIS in order to determine the influence of melting glaciers on the long-term runoff of the Brahmaputra River (EU-project BRAHMATINN). Current glacier distribution and glacier changes since the 1970/80s were mapped using multi-temporal optical remote sensing data from the Landsat series and the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER). Repeat glacier outlines were then combined with the digital elevation model from the Shuttle Radar Topography Mission (SRTM) and analyzed within a GIS in order to assess spatio-temporal gradients in change in glacier area. Our findings yield total glacier area losses between -7 to -13% per decade in the Upper Brahmaputra River basin (UBRB) for the period from 1970/80 to 2000; c. 175 km3 w.e. glacier mass was lost, which equals c. 7 km3 w.e. glacier mass loss per year. This corresponds to a yearly average mass balance of more than -0.3 m w.e. 1.2 Upper Brahmaputra River Basin (UBRB) The Brahmaputra River Basin (Figure 1) is the biggest transHimalayan river basin (A = 651’334 km2; L = 2’896 km), encompassing parts of the territory, ecosystems, people, economies and politics of China, Bhutan, Nepal, India and Bangladesh. The Upper Brahmaputra River Basin (UBRB) in this study is defined upstream of the town Guwahati in NE-India and drains about 500’000 km2. The geography of the UBRB is characterized by the alpine mountain system of the Himalaya and the tropical Monsoon climate. The hydrology of the UBRB is driven by the Monsoon cycle and has a positive annual water balance with runoff regimes ranging from glacial-nival to pluvial. http://www.brahmatwinn. uni-jena.de/ Keywords: glacier distribution, glacier change, Landsat, SRTM, GIS, Brahmaputra 1. INTRODUCTION AND PRESENTATION OF STUDY REGION 1.1 Motivation and aims Alpine mountains are a key factor for the overall hydrological regime, and major rivers in the world, such as the Brahmaputra in Asia have their headwaters in alpine mountains. In humid parts of the world, mountains provide 30% to 60% of downstream freshwater, and in semi-arid and arid environments, this figure adds up to 70% to 95%, in particular in the dry season when glaciers provide large volumes of melt water. Glacier retreat and permafrost thaw in high mountains have presently reached an extent and speed that are without historical precedence. Glaciers in the southern and central parts of the Himalayas are expected to be especially sensitive to present atmospheric warming due to their summer-accumulation type. This is likely to have substantial impacts on the hydrological dynamics, resulting in a greater variability in precipitation and stream flows, increasing intensity of extreme events comprising water quantity and water quality. The overall objective of the BRAHMATWINN** project is, therefore, to enhance and improve capacity to carry out a harmonized integrated water resources management (IWRM) for the Brahmaputra River Basin. Figure 1. Catchment area of the Brahmaputra river. Source: www.earthtrends.wri.org 2. METHODS In our study, current glacier distribution and glacier changes since the 1960s are mapped using multi-temporal optical remote sensing data from the Landsat series, the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) and CORONA. Repeat glacier outlines are then combined with the Digital Elevation Model (DEM) from the Shuttle Radar Topography Mission (SRTM) and analyzed within a GIS in order to assess spatio-temporal gradients in glacier tongue retreat and change in glacier area. In this contribution we mainly focus on our findings based on Landsat ETM+ data from around the year 2000 and SRTM data. As a reference for the derived 2000 glacier data serves the Chinese Glacier Inventory (CGI), which is based on airphotos from the 1970s to 1980s in our study areas. In addition, the analyses presented here focus on two detail areas. One was of special interest for the overarching BRAHMATWINN project * Corresponding author: Andreas Kääb, andreas.kaab@geo.uio.no ** The BrahmaTWinn project is funded through the Sixth Framework Programme of the European Commission (FP6), SUSTDEV-20053.II.3.6, contract no. 036592 (GOCE). (Lhasa river basin), the other was chosen to test the representativeness of the findings in the latter area (NW Himalaya). The results are compared to a third study area (Northern Buthan) which is one of the BRAHMATWINN sites, but on which we do not report here in detail. Glaciers are automatically mapped using ratios between Landsat Thematic Mapper (TM and ETM+) near and short-wave infrared channels, following an approach presented in detail by Paul et al. (2002) and Kääb et al. (2002). Hereby, the different spectral properties of glacier ice in the channels TM4 (near infrared) and TM5 (short-wave infrared) is exploited (Figure 2, a–d). First, ratio images from a channel division TM4/TM5 are computed (a). From these ratio images, binary glacier masks (black = “glacier’’, white = “other’’) are created by interactive thresholding (b). In the dry northern slopes of the Himalayas, the threshold value for the ratio image turns out to be c. 2.5. The glacier maps are then filtered with a 3x3 median filter in order to reduce noise (c) and the resulting raster areas are finally converted into vector data (d, e). This vector data consists basically of large contiguous ice masses which have to be divided into individual glaciers prior to data retrieval. This is done by intersecting this data with a vector layer of glacier basins. The glacier basin layer is digitized manually (on-screen digitizing) using the Landsat ETM+ as base data to discern the glacier basins and the ice divides. Wherever possible, the CGI data was also used as reference. Visual inspection of the glaciers and associated glacier basins in GoogleEarth was used as an additional information source. The intersection of the ice mass vector data with the glacier basin layer yields individual glacier outlines, again in vector format (f). The final outlines of individual glaciers were compared to the glacier outlines from the CGI and any suspect disagreements were eliminated from the sub-set used for change analysis. Reasons for such disagreements were, for instance, obvious geo-referencing or digitizing errors in the CGI. The accuracy of the method is high for snow covered and clean glacier ice outside shadow areas. Misclassification typically occurs for wetlands, lakes and rivers erroneously classified as ice, and missing classification on middle- and side-moraines. These errors can easily be eliminated by post-processing (i.e. manual exclusion/inclusion of wrongly classified/missing areas). More problematic are wrong classifications in cast shadow (induced by the steep relief) and on debris-covered glaciers. Also these errors have to be corrected manually during post-processing, but need a high degree of glaciological expert knowledge. 3. RESULTS AND DISCUSSION Figures 3 and 4 show the location of the two study areas, example subsets of produced glacier outlines and the size distribution of glaciers found in these areas. Tables A–D show the glacier changes (in area and volume) found for both study areas. In the NW-Himalayas the airphotos that form the base of the Chinese Glacier Inventory (CGI) stem from 1980, the airphotos for the Lhasa area are from 1970. Figure 2. Glacier segmentation with Landsat Thematic Mapper (TM and ETM+) imagery in the western Himalayas: (a) ratio image from a division of channel TM4 and TM5, (b) glacier mask (black = “glacier’’, white = “other’’) created from the ratio image by interactive thresholding, (c) filtered (3x3 median filter) glacier map, (d) resulting glacier areas, resp. ice mass areas; (e & f) zoom-in of (d): (e) glacier areas (hatched) before intersection with glacier basin layer and post-processing, (f) glacier areas (hatched) after intersection and post-processing (see text for further explanations). We found significant differences in area change between the two regions studied (cf. Figure 3 & 4; Tables A & B). These differences can partly be attributed to differences in glacier size distribution in the different study areas, different climatic conditions (from the north slope of the Himalaya that is influenced by Monsoon to the dry-continental plateau areas of Tibet), and different glacier hypsography distributions. Especially the significantly different glacier size distribution characteristics are important for the up-scaling of our results. Figure 5 shows percentual decadal glacier area changes in three study areas investigated within the BRAHMATWINN project. In the NW Himalayas (area 1) percentual decadal glacier area change amounted to c. -8%/10 yr, in the SW Nyainqentanglha range near Lhasa (area 2) the value is c. -7%/10 yr. In the area of Northern Buthan (area 3, not presented here in detail) percentual decadal glacier area change was -9 to -13%/10 yr for the period 1976– 2000. Table B. %-Glacier change 1970–2000 per glacier size class in the SW Nyainqentanglha mountain range (see also Figure 4). *) Note that in this study area, glacier size class no. 7 contains only 1 glacier. Glacier size class 1 (< 0.1 km2) 2 (> 0.1 km2 to < 0.5 km2) 3 (> 0.5 km2 to < 1.0 km2) 4 (> 1.0 km2 to < 5.0 km2) 5 (> 5.0 km2 to < 10.0 km2) 6 (> 10.0 km2 to < 20.0 km2) 7 (> 20.0 km2) *) Area change 1970–2000 (%) -63.1 -41.1 -30.5 -18.5 -11.2 -5.8 +1.9 Figure 3. Satellite-derived glacier area delineation, exemplified for a site within the study area in the NW Himalayas (area 1). Symbology: yellow lines = glacier outlines derived by using ratios between TM4 (near infrared) and TM5 (short-wave infrared) channels; green lines = Chinese Glacier Inventory (CGI) from c. 1980. The pie chart visualizes the number of glaciers in this study area per size class. Table A. Glacier change in % over 1980–2000 per glacier size class in the NW Himalayas (see also Figure 3). Glacier size class 1 (< 0.1 km2) 2 (> 0.1 km2 to < 0.5 km2) 3 (> 0.5 km2 to < 1.0 km2) 4 (> 1.0 km2 to < 5.0 km2) 5 (> 5.0 km2 to < 10.0 km2) 6 (> 10.0 km2 to < 20.0 km2) 7 (> 20.0 km2) Area change 1980–2000 (%) -68.5 -45.1 -30.9 -22.7 -14.8 -16.7 -10.0 Figure 5. Mean glacier area changes in the two study areas reported here (1 = NW Himalayas, 2 = SW Nyainqentanglha) and in Northern Buthan (= 3; part of the BRAHMATWINN project but not reported in detail here) shown as decadal percentual changes. Our results are in good agreement with numbers reported from other studies: In Pumqu (Tibet), mean glacier area changes amounted to -8%/10yr for the period 1970-2001 (Jin et al. 2005). S. Donghui reports glacier area changes of up to -3.5 (-2.5 to 4.5)%/10yr for the Tibetean plateau (personal communication, 2006). Karma et al. (2003) studied glacier changes in Buthan and found mean glacier area changes in the order of -9%/10yr for the period 1963-1993. Glacier volume cannot be measured directly from space. We use two different empirical relations between glacier area and mean glacier thickness to estimate glacier volumes. The large discrepancy between both methods underlines that these approaches for glacier volume estimation can only yield rough first-order estimates. We estimate the according ice volume loss to be on the order of 20%, or roughly -0.3 m water equivalent per year as average over both study areas and both volume estimate methods (Table D). Figure 4. Satellite-derived glacier area delineation, exemplified for a site within the study area in the SW Nyainqentanglha mountain range, northwest of Lhasa (area 2). Symbology: yellow lines = glacier outlines derived by using ratios between TM4 (near infrared) and TM5 (short-wave infrared) channels; green lines = Chinese Glacier Inventory (CGI) from c. 1970. The pie chart visualizes the number of glaciers in this study area per size class. Using the CGI and our inventory results we upscale the above glacier change to the entire UBRB and estimate it to be on the order of 15% area loss per decade since the 1970s, or, in other words, 7 km3 w.e. per year (0.015 mm a-1 sea level equivalent). Problems encountered during our analyses are largely connected to uncertainties of the Chinese Glacier Inventory (CGI). The CGI dataset is an invaluable source for comparison, yet it has some deficiencies which complicate change detection analyses. The CGI data were produced from published maps, dating from 1981 through 2002 (personal communication B. Raup, GLIMS). The airphotos used for producing these maps are from 1970 and 1980 for our two study areas. Little is known, however, about the used methodology to digitize the original map data, about the date when the glacier outlines contained in the maps were mapped, and about the datasets’ general accuracy. Errors found during our analyses were: (a) local distortions in geo-referencing, and (b) missing glaciers. The distortions manifested themselves mainly in two ways: either as differences in geospatial accuracy on different sides of mountain ranges, or as “shifting” geospatial accuracies within a given region. In addition, the watershed delineation in the CGI is quite coarse, which caused problems for the delineation of the individual glacier basins. Our findings show that the Upper Brahmaputra river basin has lost roughly 20% of its water reserves bound in glaciers during the last 20–30 years. While such glacier melt leads to increased dryseason runoff on the short term, the long-term consequences of decreasing ice resources could be a decline of river runoff contribution from glaciers in the dry season. Modelling of and adaptation strategies to these changes in the physical environment and their socio-economic consequences is the purpose of the BRAHMATWINN project. Table C. Absolute change in glacier area and glacier volume 1970/80–2000 in the NW Himalayas (area 1) and the SW Nyainqentanglha mountain range (“Lhasa”; area 2). Methods for volume estimation: 1) Maisch 1992, 2) Driedger & Kennard 1986. Jin, R., X. Li, T. Che, L. Wu, and P. Mool, “Glacier area changes in the Pumqu river basin, Tibetan Plateau, between the 1970s and 2001”, Journal of Glaciology, vol. 51, no. 175: p.p. 607–610, 2005. Area 1 Area 2 Area 1 Area 2 Glacier area 1970/80 Glacier area 2000 (km2) 449.4 535.3 (km2) 372.6 429.5 Glacier volume 1970/80 Glacier volume 2000 (km3) 12.91)/27.72) 13.0/19.6 (km3) 10.6/22.4 10.4/15.6 Area change 1970/80– 2000 (km2) -76.7 -105.8 Volume change 1970/80– 2000 (km3) -2.3/-5.3 -2.6/-4.0 Area change 1970/80– 2000 (%) -17.1 -19.8 Volume change 1970/80– 2000 (%) -17.6/-19.2 -20.0/-20.3 Table D. Total and yearly mean net glacier mass balance 1970/80–2000 in the NW Himalayas (area 1) and the SW Nyainqentanglha mountain range (“Lhasa”; area 2) as derived from the volume change estimates shown in Table C. Methods for volume estimation: 1) Maisch 1992, 2) Driedger & Kennard 1986. Area 1 Area 2 Net mass balance1) mm w.e. 1970/80– per year 2000 -5.49 -0.27 -5.40 -0.18 Net mass balance2) mm w.e. 1970/80– per year 2000 -12.97 -0.64 -8.25 -0.27 REFERENCES Driedger, C.L. and P.M. Kennard, “Glacier volume estimation on Cascade volcanoes: an analysis and comparison with other methods”, Annals of Glaciology, vol. 8: p.p. 59–64, 1986. Karma, Y. Ageta, N. Naito, S. Iwata, and H. Yabuki, “Glacier distribution in the Himalayas and glacier shrinkage from 1963 to 1993 in the Bhutan Himalayas”, Bulletin of Glaciological Research, vol. 20: p.p. 29–40, 2003. Paul, F., A. Kääb, M. Maisch, T. Kellenberger, and W. Haeberli. “The new remote-sensing-derived Swiss glacier inventory: I. Methods”, Annals of Glaciology, vol. 34: p.p. 355–361, 2002. Kääb, A., F. Paul, M. Maisch, M. Hoelzle, and W. Haeberli, “The new remote-sensing-derived Swiss glacier inventory: II. First results”, Annals of Glaciology, vol. 34: p.p. 362–366, 2002. Maisch, M., “Die Gletscher Graubündens: Rekonstruktionen und Auswertung der Gletscher und deren Veränderungen seit dem Hochstand von 1850 im Gebiet der östlichen Schweizer Alpen”, Physische Geographie, vol. 33A/B: 324 p./128 p., 1992. Brahmatwinn project – Twinning European and South Asian River Basins to enhance capacity and implement adaptive management approaches. http://www.brahmatwinn.uni-jena.de/ Chinese Glacier Inventory. http://wdcdgg.westgis.ac.cn/DATA BASE/Glacier/glacier_inventory.asp World Glacier Inventory. http://nsidc.org/data/docs/noaa/ g01130_glacier_inventory/ Watersheds of the world: Asia and Oceania – Brahmaputra watershed. www.earthtrends.wri.org 4. CONCLUSIONS Extrapolation of our findings to the entire UBRB yield total glacier area losses from 1970/80 to 2000 between -7 to -13% per decade. In the period 1970/80–2000, c. 175 km3 w.e. glacier mass was lost in the UBRB, which equals c. 7 km3 w.e. glacier mass loss per year. This corresponds to a yearly average mass balance of more than -0.3 m w.e. which would have contributed c. 0.4 mm to sea level rise since 1970/80 (0.015 mm a-1). ACKNOWLEDGEMENTS Landsat data from United States Geological Survey (USGS), orthorectification: Geodata Solutions, Jena. Chinese Glacier Inventory (CGI) from Institute of Tibetan Plateau Research through Global Land Ice Measurements from Space database (GLIMS).